WinNonlin User's Guide 5.2.1 - Free Download PDF Ebook

WinNonlinUser’s Guide ® Version 5.2.1 WinNonlin Version 5.2.1 WinNonlin® copyright ©1998-2008 Pharsight Corporation. All rights reserved. This software and documentation are owned by Pharsight Corporation, and may be used only as authorized in the license agreement controlling such use. No part of the software nor the accompanying documentation may be reproduced, transmitted, or translated, in any form or by any means, electronic, mechanical, manual, optical, or otherwise, except as expressly provided by the license agreement or with the prior written permission of Pharsight Corporation. This product may contain the following software that is provided to Pharsight Corporation under license: Actuate™ Formula One® copyright 1993-03 Actuate, Inc. All rights reserved. SentinelLM™ 6.0 copyright 1998 Rainbow Technologies, Inc. All rights reserved. Microsoft® XML Parser version 3.0 copyright 19982000 Microsoft Corporation. All rights reserved. IMSL® copyright 1993 Visual Numerics, Inc. All rights reserved. Information in the documentation is subject to change without notice and does not represent a commitment on the part of Pharsight Corporation. The documentation contains information proprietary to Pharsight Corporation and is for use by Pharsight Corporation customers only. Use of the information contained in the documentation for any purpose other than that for which it is intended is not authorized. NEITHER PHARSIGHT CORPORATION NOR ANY OF THE CONTRIBUTORS TO THIS DOCUMENT MAKES ANY REPRESENTATION OR WARRANTY, NOR SHALL ANY WARRANTY BE IMPLIED, AS TO THE COMPLETENESS, ACCURACY, OR USEFULNESS OF THE INFORMATION CONTAINED IN THIS DOCUMENT, NOR DO THEY ASSUME ANY RESPONSIBILITY FOR LIABILITY OR DAMAGE OF ANY KIND WHICH MAY RESULT FROM THE USE OF SUCH INFORMATION. Destination Control Statement All technical data contained in the software and documentation are subject to the export control laws of the United States of America. Disclosure to nationals of other countries may violate such laws. It is the reader's responsibility to determine the applicable regulations and to comply with them. United States Government Rights The software and documentation constitute “commercial computer software” and “commercial computer software documentation” as such terms are used in 48 CFR 12.212 (Sept. 1995). United States Government end users acquire the software under the following terms: (i) for acquisition by or on behalf of civilian agencies, consistent with the policy set forth in 48 CFR 12.212 (Sept. 1995); or (ii) for acquisition by or on behalf of units of the Department of Defense, consistent with the policies set forth in 48 CFR 227.7202-1 (June 1995) and 227.7202-3 (June 1995). The manufacturer is Pharsight Corporation, 321 E. Evelyn Ave., 3rd Floor, Mountain View, CA 94041. Trademarks PK Automation Application, PKS, Pharsight Knowledgebase Server and Trial Simulator are trademarks of Pharsight Corporation. Pharsight, WinNonlin, IVIVC Toolkit for WinNonlin, WinNonMix, Drug Model Explorer and DMX are registered trademarks of Pharsight Corporation. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. SigmaPlot is a registered trademark of Systat Software, Inc. S-PLUS is a registered trademark of Insightful Corporation. NONMEM is a registered trademark of The Regents of the University of California. Actuate is a trademark and Formula One is a registered trademark of Actuate Corporation. SentinelLM is a trademark of Rainbow Technologies, Inc. Microsoft, the Internet Explorer logo, MS-DOS, the Office logo, PowerPoint, Visual C++, Visual Studio, Windows, the Windows logo, Windows NT, the Windows Start logo and XL design (the Microsoft Excel logo) are registered trademarks of Microsoft Corporation. Pentium and Pentium III are trademarks or registered trademarks of Intel Corporation. Adobe, Acrobat, Acrobat Reader and the Adobe PDF logo are registered trademarks of Adobe Systems Incorporated. IMSL is a registered trademark of Visual Numerics, Inc. Compaq Visual Fortran is trademark of Compaq Computer Corporation. All other brand or product names mentioned in this documentation are trademarks or registered trademarks of their respective companies or organizations. Pharsight Corporation 321 E. Evelyn Ave., 3rd Floor, Mountain View, California 94041 Voice +1-650-314-3800 • Fax +1-650-314-3810 http://www.pharsight.com • [email protected] Contents Chapter 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 WinNonlin. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 WinNonlin editions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 In-vitro in-vivo correlation toolkit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 WinNonlin user documentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Modeling and nonlinear regression. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Model fitting algorithms and features of WinNonlin . . . . . . . . . . . . . . . . 5 Summary of regression methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Pharsight Corporation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Scientific and consulting services . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Pharsight contact information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Technical support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Licensing and product upgrades. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Customer feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Data Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .13 WinNonlin windows and files. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 File types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 WinNonlin and WinNonMix file compatibility . . . . . . . . . . . . . . . . . . . 17 Hidden windows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Workspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Printing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Page setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Page setup for workbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Page setup for text windows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Page setup for charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 Print preview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 iii Chapter 2 Pharsight WinNonlin User’s Guide Print . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 Printing workbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 Printing text objects and models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 Print all . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 Print all options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 WinNonlin options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Directories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Workbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 Licensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 Data dependencies and origins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 Saving dependencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 Refreshing results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 Unlocking derived data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Object origins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Data history worksheets and log files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 History worksheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 WinNonlin Log Viewer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Chapter 3 Workbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .41 Editing data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 Inserting or deleting worksheets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Inserting and deleting rows, columns or cells . . . . . . . . . . . . . . . . . . . . 43 Using formulas and functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 Filling adjacent cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 Formatting cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 Clearing cell values and formats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 Undo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 Find, find next and replace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 Go to . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 Column names and units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Units builder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 Data manipulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 Freezing rows and columns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 Sorting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 Merge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 Data exclusion and inclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 Re-including data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 Transformations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 iv Contents Multiple transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Transform action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Multi-transformation filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Multi-transformation equation manager . . . . . . . . . . . . . . . . . . . . . . . . Rank transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Status code transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rule sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a rule set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 4 64 66 66 67 68 69 71 73 Data import and export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .75 Microsoft Word export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 Word export document options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 Word export chart options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 Word export workbook options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 SAS Transport file export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 NONMEM export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 Data file structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 Exporting to NONMEM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 Dependent variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 Other data items . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 Dose information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 Occasions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 SigmaPlot export. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 S-PLUS export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 ODBC import . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 Creating an import . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 Final settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 Connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 Creating a new connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 Loading a saved connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 Queries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 Building a query . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 Schema filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 Query builder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 Loading a saved query . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 Loading an existing import . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 Refreshing an imported data set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 The custom query builder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 Select a builder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 Study selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 v Pharsight WinNonlin User’s Guide ODBC export. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 Creating an export. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 Export definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 Building an export definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 Field selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 Loading a saved export definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 Loading a saved export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 The Enterprise Software Development Kit . . . . . . . . . . . . . . . . . . . . . . . . . 107 Import file format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 Usage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 Export file format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 Custom query builder interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 VB example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 Chapter 5 Charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .115 Chart Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 X-Y variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 Scatter plots and bar charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 Histograms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 Sort and group variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 Error bars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 Chart titles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 Axis options. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 Chart Designer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 Frequently used chart formatting options . . . . . . . . . . . . . . . . . . . . . . . . . . 124 Changing the plot type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 Changing line style and markers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 Changing axis scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 Creating a semi-log plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 Setting grid lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 Removing or adding walls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 Chart templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 Copy and paste. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .131 Table Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 Missing and excluded data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 WinNonlin table templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 Joining data sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 Summary statistics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144 Chapter 6 vi Contents Significant digits and column alignments. . . . . . . . . . . . . . . . . . . . . . . . . . Table formatting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Table definition files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a table definition file . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Loading a table definition file . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 7 145 146 148 148 149 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .151 Computing summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 Output and computational formulae. . . . . . . . . . . . . . . . . . . . . . . . . . . 153 Weighted summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 Output and computational formulae for weighted calculations . . . . . . 157 Noncompartmental Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . .159 NCA model selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 Model variables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 Urine models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 Lambda Z or slope estimation settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 Lambda Z or slope range selection. . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 Calculation of Lambda Z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 Calculation of slopes for effect data. . . . . . . . . . . . . . . . . . . . . . . . . . . 163 Setting Lambda Z or PD slope ranges . . . . . . . . . . . . . . . . . . . . . . . . . 164 Partial areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 Therapeutic response. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168 Dosing regimen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 Time deviations in steady-state data. . . . . . . . . . . . . . . . . . . . . . . . . . . 172 Model options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 Output options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 Weight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 Title . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176 NCA settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178 Parameter names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 Computational rules for WinNonlin’s noncompartmental analysis engine 181 Data checking and pre-treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181 AUC calculation and interpolation formulas . . . . . . . . . . . . . . . . . . . . 183 Partial areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184 Therapeutic response windows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186 Sparse sampling computations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188 Chapter 8 vii Pharsight WinNonlin User’s Guide Chapter 9 Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .193 Loading the model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193 Model properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196 Data variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196 Model parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 Dosing regimen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200 Dosing data variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202 Model options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203 Output options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204 Weight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 Title . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206 PK settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206 Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208 Running the model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 Stop modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 Troubleshooting modeling problems . . . . . . . . . . . . . . . . . . . . . . . . . . 209 Errors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210 Text (ASCII) output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 Workbook output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214 Chart output. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215 User Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .217 ASCII user models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217 Creating a new ASCII model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218 User model parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219 Model structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220 ASCII model examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226 WinNonlin variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231 Temporary variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232 The WinNonlin programming language . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 Comparison operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 Functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 Bioassay functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236 Compiled user models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236 Using FORTRAN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237 Example using Digital (now Compaq) Visual FORTRAN 6.0 . . . . . . 237 Using C++ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238 Example using Microsoft Visual C++ 6.0 . . . . . . . . . . . . . . . . . . . . . . 239 Chapter 10 viii Contents Chapter 11 Weighting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .241 Weighted least squares . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241 Iterative reweighting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242 Reading weights from the data file . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242 Scaling of weights. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242 Weights in ASCII models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243 The WinNonlin Toolbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .245 Nonparametric superposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245 Data and assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246 Computation method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246 Performing nonparametric superposition . . . . . . . . . . . . . . . . . . . . . . . 248 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249 Semicompartmental modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250 Data and assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250 Computation method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250 Performing semicompartmental modeling . . . . . . . . . . . . . . . . . . . . . . 252 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253 Crossover design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254 Data and assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255 Descriptive analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256 Hypothesis testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257 Using the Crossover Design tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260 Numeric deconvolution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260 Deconvolution methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261 Using deconvolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269 Deconvolution variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 Deconvolution dosing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272 Deconvolution settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273 Deconvolution output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275 The IVIVC Toolkit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .277 Wagner-Nelson and Loo-Riegelman methods . . . . . . . . . . . . . . . . . . . . . . 278 Wagner-Nelson inputs and calculations . . . . . . . . . . . . . . . . . . . . . . . . 278 Loo-Riegelman inputs and calculations . . . . . . . . . . . . . . . . . . . . . . . . 279 Using the Wagner-Nelson or Loo-Riegelman method. . . . . . . . . . . . . 280 Data variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281 Dosing regimen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282 Loo-Riegelman model parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 283 ix Chapter 12 Chapter 13 Pharsight WinNonlin User’s Guide Lambda Z ranges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284 Output options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287 Weighting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288 Title . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290 Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291 Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293 Data variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295 Convolution settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297 Convolution output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298 Convolution units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299 Levy plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302 The IVIVC Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302 Minimizing the IVIVC Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304 Project Status. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304 Saving and loading IVIVC projects . . . . . . . . . . . . . . . . . . . . . . . . . . . 305 In-vitro tab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306 In-vivo tab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307 Correlation tab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310 ASCII user model for correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312 Prediction tab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313 Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315 Using the IVIVC Wizard with PKS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319 Chapter 14 Linear Mixed Effects Modeling . . . . . . . . . . . . . . . . . . . . . . . . . .323 An example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323 The linear mixed effects model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326 Construction of the X matrix. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328 Linear combinations of elements of β. . . . . . . . . . . . . . . . . . . . . . . . . . 332 Determining estimability numerically . . . . . . . . . . . . . . . . . . . . . . . . . 333 Substitution of estimable functions for non-estimable functions . . . . . 334 Fixed effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336 Output parameterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336 Variance structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338 Repeated effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339 Random effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341 Least squares means . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344 x Contents Contrasts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Joint versus single contrasts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Nonestimability of contrasts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Degrees of freedom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Other options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LinMix Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LinMix limits and constraints. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fixed effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Variance structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Random coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Repeated measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Contrasts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Least squares means . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LinMix general options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LinMix output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Linear mixed effects text . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Linear mixed effects workbook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Tests of hypotheses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Akaike’s Information Criterion (AIC) . . . . . . . . . . . . . . . . . . . . . . . . . Schwarz’s Bayesian Criterion (SBC) . . . . . . . . . . . . . . . . . . . . . . . . . . Hessian eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Predicted values and residuals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fixed effects parameter estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . Coefficients. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Iteration history . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Parameter key . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Computational details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Restricted maximum likelihood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Newton’s Algorithm for maximization of restricted likelihood. . . . . . Starting values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Convergence criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Satterthwaite’s approximation for degrees of freedom . . . . . . . . . . . . Comparing LinMix to ANOVA and SAS’s PROC MIXED . . . . . . . . . . . LinMix and ANOVA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LinMix and PROC MIXED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 15 346 347 348 349 349 349 350 350 352 353 355 356 356 358 359 360 361 361 361 362 363 363 364 364 364 365 365 365 366 366 368 369 369 370 372 372 373 Bioequivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .375 xi Pharsight WinNonlin User’s Guide Introduction to bioequivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375 Basic terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375 Bioequivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376 The model and data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378 Linear model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378 Variance structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378 Missing data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378 Variable name limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378 Average bioequivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379 Study designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379 Recommended models for average bioequivalence . . . . . . . . . . . . . . . 379 Least squares means . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381 Classical and Westlake confidence intervals . . . . . . . . . . . . . . . . . . . . 383 Two one-sided T-tests. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384 Anderson-Hauck test. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386 Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386 Individual and population bioequivalence. . . . . . . . . . . . . . . . . . . . . . . . . . 387 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387 Bioequivalence criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388 Details of computations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390 Bioequivalence Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395 Using average bioequivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396 Fixed effects for average bioequivalence . . . . . . . . . . . . . . . . . . . . . . . 396 Random effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398 Repeated variance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399 Bioequivalence options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400 Bioequivalence general options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401 Using population/individual bioequivalence . . . . . . . . . . . . . . . . . . . . 401 Fixed effects for population/individual bioequivalence . . . . . . . . . . . . 402 Bioequivalence options for population/individual bioequivalence . . . 402 Bioequivalence output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403 Average bioequivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403 Population/Individual bioequivalence. . . . . . . . . . . . . . . . . . . . . . . . . . 404 Ratio tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406 Chapter 16 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .409 Simulating responses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409 xii Contents Simulation of a compiled library model . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating the input data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Setting up the model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Data variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Model parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dosing regimen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Model options. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Running the simulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparing dosing schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dosing regimen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 17 409 410 410 411 412 413 414 414 415 416 Scripting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .417 Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417 Writing and editing a script. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418 Executing a script . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420 Within WinNonlin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420 From the command line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420 From file manager or explorer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421 Scripting object properties and methods. . . . . . . . . . . . . . . . . . . . . . . . . . . 421 Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421 Engines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423 Script file components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426 Note on saving files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 428 Processing multiple profiles in script files . . . . . . . . . . . . . . . . . . . . . . . . . 428 Special file formats. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430 Scripting the Pharsight Knowledgebase Server . . . . . . . . . . . . . . . . . . . . . 440 Using the Pharsight Knowledgebase Server . . . . . . . . . . . . . . .445 The PKS information structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446 Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446 Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 447 Other documents. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448 Dosing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448 Dosing worksheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448 Append dosing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449 Dosing dialog . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449 Accessing the PKS from WinNonlin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449 PKS sessions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450 Connecting to the PKS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451 Logging in . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452 xiii Chapter 18 Pharsight WinNonlin User’s Guide Creating a study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452 Study definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453 Study Creation Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 454 Subject identifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455 Subject ID attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455 Subject data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456 Subject data attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457 Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457 Sample data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 458 Sample attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 458 Dosing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459 Dosing attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 460 Data Collection Point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461 Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461 Study creation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462 Loading a study or scenario. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463 Select observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464 Select dose data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464 Scenario search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464 Creating scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465 Adding data columns to a scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . 466 Adding non-WinNonlin documents to a scenario . . . . . . . . . . . . . . . . 466 Optimizing PKS/WNL performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467 Change tracking and audit information . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470 Audit information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470 Update reason . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471 Name objects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472 Dosing information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472 Changing the dose regimen. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473 Appending or updating a study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474 PKS libraries and object shares . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475 PKS share . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475 Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476 Loading a library from the PKS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477 Adding or updating a library in the PKS . . . . . . . . . . . . . . . . . . . . . . . 477 Library and share status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 478 PKS Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479 WinNonlin and PKS differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480 Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480 Studies and scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481 xiv . . . . . . . . . . . 533 Model 101 . . . . . . . . . . . . . . . . . . . . . . . 520 Model 3 . . . . 530 Model 18 . . . . . . . . . . . . . 505 NCA output parameters and their computation formulas. . . 519 Model 1 . . . . . . . . . . . . . . . . . . . . 522 Model 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522 Model 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 527 Model 13 . . . . . . . . . . . . . . . . . . . . . . . . . . . 524 Model 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .483 Appendix B WinNonlin Model Libraries . . . . . . . . . . . . . . . . . 520 Model 2 . . 528 Model 15 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Contents Appendix A Glossary . . . . . . . . . . . . . . . 522 Model 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 506 Plasma or serum data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529 Model 17 . . . . 532 Pharmacodynamic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 536 xv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 526 Model 12 . . . . . . . . . . . . . . . . . . . . . . 528 Model 16 . . . . 536 Model 107 . . . . . 514 Drug effect data (model 220) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534 Model 102 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523 Model 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535 Model 105 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519 ASCII models and files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525 Model 11 . . . . . . 533 ASCII library . . . . . . . . . . . . . . . . . . . 521 Model 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534 Model 103 . . 531 Model 19 . . . . . . . . . . . . . 504 Noncompartmental analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .503 Notation and conventions . . . . . . . . . . 507 Urine data . 535 Model 106 . . . . . . . . . . . . . 528 Model 14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 524 Model 9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512 Sparse sampling (pre-clinical) data . . . 515 Pharmacokinetic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 536 Model 108 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535 Model 104 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541 Michaelis-Menten models . . . . . . . . . . . . . . . . . . . . . . . . . . 548 Model 602 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544 Model 304 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 545 ASCII library. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549 Model 603 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .579 Warnings and Error Messages . . . . . . . . . . . . . . . . .637 Compartmental modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 649 Table Wizard . . . . . . . 637 LinMix and Bioequivalence wizards. . . . . . . . . . . . . . . . . . . . . . 541 Model 54 . . . . . . . . . . .Pharsight WinNonlin User’s Guide PK/PD link models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 545 Simultaneous PK/PD link models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 546 Sigmoidal/dissolution models . . . . . . . . . . . 541 Running an indirect response model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 542 Model 301 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 540 Model 53 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Arrays and Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539 Models . . . . . . . . . . . . . . . . . . . . . . . . . . 540 Model 52 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537 Linking PK and PD models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549 Model 604 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .551 Appendix D Commands. . . . . . . . . . . . . . 652 xvi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 550 Appendix C Worksheet Functions . . . . . . . . . . . . . . . . . . . . . 537 Output parameters . . . . 544 Model 303 . . . . . . . . . . . . . . . . . . 547 Model 601 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543 Model 302 . . . . . 545 Data . . . . . . . . . . . . . . . . . 538 Indirect response models . . . . . . . . . 545 Linear models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 652 Descriptive statistics . . . . . . . . . . . . . . . . . . .557 Appendix E Appendix F Scripting Commands . . . . . 536 Notation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 540 Model 51 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Warnings. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IVIVC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .657 Bioequivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Contents Noncompartmental analysis . . . . . . . . . . . . . 663 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Scripting language. . . . . . . . . . . . . . . . . . . . Error messages . . . . . . . . . . . . . . . . . . . . . . . . 653 653 653 654 655 Appendix G References . . . . . . . . . . . . . 658 Pharmacokinetics . . . . . 657 In vitro-in vivo correlation and formulation science . . . . . . . . . . . . . . . . . . . . . . . . . . . . 660 Sparse data analysis methods . . . . . . 665 xvii . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 659 Regression and modeling . . . . . . . . . . . . . . . . . . . . . 662 Statistics . . . . . . Pharsight WinNonlin User’s Guide xviii . The 1 . WinNonlin editions WinNonlin is distributed in two editions. tables. shares data with ODBC-compliant databases. PK/PD link. and in addition. nonlinear modeling and Pharsight contact information Welcome. management and tracking of clinical and pre-clinical data. noncompartmental. • • • “WinNonlin” on page 1 “Modeling and nonlinear regression” on page 3 “Pharsight Corporation” on page 9 WinNonlin WinNonlin is a sophisticated tool for nonlinear modeling. This chapter introduces WinNonlin. user documentation. WinNonlin Professional and WinNonlin Enterprise. pharmacodynamic. Pharsight's solution for secure storage.Chapter 1 Introduction WinNonlin editions. It also supports creation of custom models to enable fitting and analysis of virtually any kind of data. It is particularly suited to pharmacokinetic and pharmacodynamic (PK/PD) modeling and noncompartmental analysis to evaluate data from bioavailability and clinical pharmacology studies. and indirect response models. WinNonlin includes a large library of pharmacokinetic. and listings required for regulatory submission. and serves as a portal to the Pharsight® Knowledgebase Server™ (PKS). It supports the analyses and generates the figures. WinNonlin Enterprise provides all the features of WinNonlin Professional. WinNonlin Professional is made for scientists involved with nonlinear modeling. nonlinear modeling and the Pharsight suite of products and services under the following headings. pharsight. noncompartmental analysis. and a step-by-step hardcopy with CD shipment test to confirm that installation is complete and functional. enhanced tools for convolution...pharsight.\WINNONLIN\USER DOCS Operating procedures for each software feature. See “References” on page 657 for a bibliography of references on the underlying theory. an addendum to the release notes will be posted to support_sflogin.. Step-by-step examples illustrating use of WinNonlin features in realistic scenarios. information on the calculations performed by WinNonlin. Known issues and work-arounds in the current version.. nor statistical analysis. Examples Guide Online help . .com/support/ If issues and work-arounds are discovered after product release. WinNonlin user documentation WinNonlin includes the following user documentation. deconvolution and plotting as well as a wizard to streamline setup of common IVIVC work flows.. and appendices detailing WinNonlin functions.php the customer support web site with any software patches. dissolution models. F1 key..1 WinNonlin User’s Guide combination of WinNonlin Enterprise and PKS supports compliance with the Food and Drug Administration’s (FDA) 21 CFR part 11 regulations. support_sflogin.\WINNONLIN\USER DOCS and Installation and licensing instructions.php Release notes . The IVIVC suite includes a variety of correlation models..com/support/ Frequently asked questions and solutions for WinNonlin. WinNonlin user documentation Document Getting Started Guide User’s Guide Default location Description .\WINNONLIN\USER DOCS Addendum to release notes FAQ The manuals are not intended as a comprehensive text on nonlinear estimation.. In-vitro in-vivo correlation toolkit With purchase of a special license. Table 1-1. or Help button ence material on WinNonlin’s computational engines.\WINNONLIN\USER DOCS Accessed from WinNonlin via the Operating procedures for the software features and referHelp menu. http://www. http://www. 2 . WinNonlin can be upgraded to include a suite of tools for analysis of in-vitro in-vivo correlations (IVIVC). at least one of the parameters appears as other than a coefficient. Y = A1 + A2X1 + A3X2 + A4X3 + ε These examples are all “linear models” since the parameters appear only as coefficients of the independent variables. X. A simple example is the decay curve: Y = Y0 exp(-BX) This model can be linearized by taking the logarithm of both sides. behavioral. the intercept and B. 3 . Models can be used to help interpret data. Most scientists have encountered linear models. biological. the slope. finding a mathematical equation and a set of parameter values such that values predicted by the model are in some sense “close” to the observed values. Two other common examples are polynomials such as: Y = A1 + A2X + A3X2 + A4X3 + ε and multiple linear regression. The simplest example is a line (linear regression): Y = A + BX + ε where Y is the dependent variable.Introduction Modeling and nonlinear regression 1 Modeling and nonlinear regression The value of mathematical models is well recognized in all sciences— physical. and with the power and availability of computers. but as written it is nonlinear. Models are used in the quantitative analysis of all types of data. the independent variable. Note: WinNonlin’s LinMix module handles linear regression. and others. models in which the dependent variable can be expressed as the sum of products of the independent variables and parameters. In nonlinear models. and to predict future results. to test hypotheses. as well as Analysis of Variance and Analysis of Covariance. A. mathematical models provide convenient and powerful ways of looking at data. The concern here is “fitting” models to data—that is. When the basic differential equations can be integrated to algebraic equations. This manual is not intended to be a comprehensive text on nonlinear estimation theory.1 WinNonlin User’s Guide Another example is the Michaelis-Menten equation: V max C V = ---------------Km + C This example can also be linearized by writing it in terms of the inverses of V and C. the two are equivalent. A discussion of using nonlinear least squares to obtain maximum likelihood estimates can be found in Jennrich and Moore (1975). pages 304-305). These are maximum likelihood and least squares. and many have been suggested. two good references for the topic of general nonlinear modeling are Draper and Smith (1981) and Beck and Arnold (1977). (1981) and Gabrielsson and Weiner (2001). or nonlinear regression models. All pharmacokinetic models derive from a set of basic differential equations. The books by Bard (1974). differential equations or a combination of the two types of equations. However. but better estimates are obtained if the nonlinear form is used to model the observations (Endrenyi. There are many models that cannot be made linear by transformation. (1981). and a set of data. In least squares fitting the “best” estimates are those that minimize the sum of the squared deviations between the observed values and the values predicted by the model. Modeling in the context of kinetic analysis and pharmacokinetics is discussed in Endrenyi. 4 . Ratkowsky (1983) and Bates and Watts (1988) give a more detailed discussion of nonlinear regression theory. but only two are much used. ed. ed. One such model is the sum of two or more exponentials. Given a model of the data. how does one “fit” the model to the data? Some criterion of “best fit” is needed. it is most efficient to fit the data to the integrated equations. such as Y = A1 exp(-B1X) + A2 exp(-B2X) All such models are called nonlinear models. With certain assumptions about the error structure of the data (no random effects). But it is also possible to fit data to the differential equations by numerical integration. WinNonlin can fit models defined in terms of algebraic equations. well-known techniques can be used to solve these linear equations for the parameter estimates. Since it is not possible to know exactly what the minimum is. In ML. WinNonlin will compute values of the dependent variable and also estimate the variance-inflation factors for the estimated parameters. That is. however. provide some information about the “goodness of fit” and about how well the parameters are estimated. such as population PK/PD models. any method for computing the least squares estimates in a nonlinear model must be an iterative procedure.e. because the system of equations that results by setting the partial derivatives to zero is a system of nonlinear equations without general solution methods. In the case of models with random effects. estimates that result in a smaller sum of squared deviations. WinNonlin can also be used to simulate data from a model. WinNonMix fits models of this variety. The iteration continues until hopefully the minimum (or least) sum of squares is reached. A more complete discussion of the theory underlying nonlinear regression can be found in the books cited previously. the likelihood of the data is maximized with respect to the model parameters. This gives a set of linear equations with the estimates as unknowns.Introduction Modeling and nonlinear regression 1 The sum of squared deviations can be written in terms of the observations and the model. Consequently. The program does. some stopping rule must be given at which point it is assumed that the method has converged to the minimum sum of squares. In the case of linear models it is easy to compute the least squares estimates. initial estimates of the parameters are made and then in some way modified to give better estimates. the more general method of maximum likelihood (ML) must be employed. using the information contained in a set of observations. That is. i. This method will not work for nonlinear models. As such it can only take the data and model supplied by the researcher and find a “best fit” of the model to the data. Model fitting algorithms and features of WinNonlin WinNonlin is a computer program for estimating the parameters in a nonlinear model.. given a set of parameter values and a set of values of the independent variables in the model. A computer program such as WinNonlin is only a tool for the modeling process. The program cannot determine the correctness of the model nor the value of any decisions or interpretations based on the model. One equates to zero the partial derivatives of the sum of squares with respect to the parameters. 5 . WinNonlin avoids this difficulty by using singular value decomposition rather than matrix inversion to solve the system of linear equations. if the data contain very little information about one or more of the parameters. WinNonlin provides the “least squares” estimates of the model parameters. Many nonlinear estimation programs. with the additional requirement that at every iteration the sum of squares must decrease. especially for those problems where some of the parameters may not be welldefined by the data and the sum of squares surface is complicated in the region of the minimum. but rather move on to points where the problem may be better defined. To speed convergence. the singular value decomposition algorithm will always find a solution to the system of linear equations. Much of the difficulty with this algorithm arises from singular or near-singular equations at some points in the parameter space. The Gauss-Newton algorithm uses a linear approximation to the model. As such it must solve a set of linear equations at each iteration. the Gauss-Newton algorithm with the modification proposed in Hartley (1961). the adjusted parameter 6 . its usefulness has formerly been limited by the extensive amount of computation it requires. Beck and Arnold (1977) compare various least squares algorithms and conclude that the Box-Kanemasu method (almost identical to Hartley's) is the best in many circumstances. The simplex method does not require the solution of a set of equations in its search and does not use any knowledge of the curvature of the sum of squares surface. WinNonlin uses the modification to the Gauss-Newton method proposed by Hartley (1961) and others. and a Levenberg-type modification of the Gauss-Newton algorithm (Davies and Whitting (1972)). as discussed above. have found Hartley's modification to be very useful. However. When the Nelder-Mead algorithm converges. with a choice of three algorithms for minimizing the sum of squared residuals: the simplex algorithm of Nelder and Mead (1965). The parameter estimates that are obtained after the algorithm converges a second time are treated as the “final” estimates. The power and speed of current computers make it a more attractive choice. (See Kennedy and Gentle (1980). WinNonlin restarts the algorithm using the current estimates of the parameters as a “new” set of initial estimates and resetting the step sizes to their original values. The simplex algorithm is a very powerful minimization routine. including the original NONLIN. As indicated.1 WinNonlin User’s Guide It is assumed that WinNonlin users have some knowledge of nonlinear regression.) One result is that the iterations do not get “hung up” at a singular point in the parameter space. This modification helps the algorithm locate the global minimum (as opposed to a local minimum) of the residual sum of squares for certain difficult estimation problems. One way to avoid this is by using a 'trust region' solution. points outside the bounds are assigned a very large value for the residual sum of squares. Note: The Gauss-Newton method with the Levenberg modification is WinNonlin’s default estimation method. (2) More importantly. To fit those models that are defined by systems of differential equations. if any. In such a case it may be preferable to give up some properties of the unconstrained estimation in order to obtain parameter estimates that are physically realistic.. For more information on trust region methods. For this reason. The Levenberg and Marquardt algorithms are examples. that is. This algorithm is a 5th order Runge-Kutta method with variable step sizes. 1976). When the simplex method is used. (1) They require more computation. however. a solution to the system of linear equations is not accepted unless it is sufficiently close to the parameter values at the current iteration. Trust region methods tend to be very robust against ill-conditioned data sets. and thus are not efficient with data sets and models that readily permit precise estimation of the parameters. Then the minimization algorithm may fail due to any number of numerical problems. but the data set may contain so much error that the actual least squares estimate is negative. setting reasonable limits on the parameter may prevent the algorithm from wandering off and getting lost. resulting in “constrained optimization. evidence that any nonlinear estimation problem is better or more easily solved by the use of the exact partial derivatives. it is possible for users to ignore this information and use the estimates as though they were really valid. In WinNonlin the partial derivatives required by the Gauss-Newton algorithm are approximated by difference equations. Murray and Wright (1981) or Davies and Whitting (1972). it is recommended that users always set limits on the parameters (the means of doing this is discussed in the modeling chapters of this document). the trust region methods obtain parameter estimates that are often meaningless because of their large variances. In WinNonlin two different methods are used for bounding the parameter space.” For example.Introduction Modeling and nonlinear regression 1 vector may be so far from the least squares solution that the linearization of the model is no longer valid. the model may contain a parameter that must be non-negative. Although WinNonlin gives indications of this. 7 . see Gill. This sends the algorithm back into the admissible parameter space. the RKF45 numerical integration algorithm is used (Shampine et. There are two reasons. why one may not want to use them. At other times. It is often desirable to set limits on the admissible parameter space. al. There is little. it often performs very well on ill-conditioned data sets. This method of bounding the parameter space has worked extremely well with a large variety of models and data sets. but it is important enough that WinNonlin is supplied with a library of the most commonly used pharmacokinetic and pharmacodynamic models. the main WinNonlin library has been built in. Users can also create custom libraries. Because of WinNonlin’s flexibility and generality. plus an additional utility library of models. The second transformation uses the inverse of the normal probability function to go to an infinite parameter space. Since it doesn't require the estimated variance-covariance matrix as part of its algorithm. (See Draper and Smith (1981) or Ratkowsky (1983) for discussions of reparameterization. many options and specifications may be supplied to the program.1 WinNonlin User’s Guide With the Gauss-Newton algorithms. It is well known that a reparameterization of the parameters will often make a problem more tractable. However.) When encountering a difficult estimation problem. as explained in “User Models” on page 217. or compiled. To make WinNonlin easy to use. the user may want to try different bounds to see if the estimation is improved. It must be pointed out that it may be possible to make the problem worse with this kind of transformation of the parameter space. To speed execution time. The first transformation is from the bounded space as defined by the input limits to the unit hypercube. Fitting pharmacokinetic and pharmacodynamic models is a special case of nonlinear estimation. ASCII library files corresponding to the same models as the built-in models. experience suggests that this will rarely occur. two successive transformations are used to affect the bounding of the parameter space. there isn't enough information contained in the data to precisely estimate all of the parameters in the model or that the sum of squares 8 . Summary of regression methods Method 1: Nelder-Mead simplex Method 1 is a very powerful algorithm. Bounding the parameter space in this way is in effect a transformation of the parameter space. and the complexity of nonlinear estimation. are also provided. Ill-conditioned indicates that for the given data set and model. Reparameterization to make the models more linear also may help. The model libraries and their uses are described in “WinNonlin Model Libraries” on page 503. WinNonlin is capable of estimating the parameters in a very large class of nonlinear models. many of the most commonly used options are set as WinNonlin defaults. However. 9 . including ill-conditioned data sets. More information is available on the web at www. For further information. Pharsight Corporation is a public company headquartered in Mountain View. Pharsight Corporation helps companies make more informed and objective decisions that lead to more successful and predictable trial outcomes. it can be extremely slow.Introduction Pharsight Corporation 1 surface is highly curved. Method 3: Gauss-Newton with Hartley modification Method 3 is another Gauss-Newton method.com. California. it can be slow. Method 3 is recommended when speed is a consideration or when maximum likelihood estimation or linear regression analysis is to be performed. Method 2: Gauss-Newton with Levenberg and Hartley modification Method 2 is the default. The company's proprietary technology and world-class consulting expertise enable pharmaceutical and biotechnology companies to make more confident decisions in dealing with complex issues of drug development and product portfolio management. It is not recommended for fitting complex models. It is not as robust as Methods 1 and 2 but is extremely fast. the Levenberg modification suppresses the magnitude of the change in parameter values from iteration to iteration to a reasonable amount. including WinNonlin®. WinNonMix®. When used to fit a system of differential equations. Pharsight Corporation Pharsight Corporation provides software and services designed to improve the efficiency of drug development.pharsight. please visit Pharsight on the web at www.pharsight.com. Although the procedure works well. Products Pharsight offers a comprehensive suite of software products. Drug Model Explorer™ and the Pharsight Knowledgebase Server™ suite of products. Although the Gauss-Newton method does require the estimated variance-covariance matrix of the parameters. Trial Simulator™. and also performs well on a wide class of problems. This enables the method to perform well on ill-conditioned data sets. planned and delivered in accordance with specific needs and agreed-upon objectives. Contact Pharsight for schedules. Consultants can facilitate: • • • • • Gathering and interpreting data Building drug and population models Guiding variability testing and trial optimization Addressing specific development program objectives Process automation Training Pharsight Corporation offers workshops and training courses at regular intervals.com/support/support_sflogin.com (fastest response time) Web: http://www. please e-mail a complete description of the problem. If further assistance is needed. including copies of input data.pharsight. pricing and availability. 10 . fax or post.com/training/. phone. Suite 205 Cary. E-mail: support@pharsight. detailing the error. please contact Pharsight technical support by e-mail or web (preferred). Pharsight Scientific Services are project-centered.php Phone/voice mail: 01-919-852-4620 Fax: +1-919-859-6871 Post: Pharsight Corporation 5625 Dillard Drive. On-site training is available by request.pharsight. North Carolina 27518 For the most efficient service.1 WinNonlin User’s Guide Scientific and consulting services Our scientific consultants can design and optimize clinical trials and clinical development programs to meet an individual company's objectives. The training schedule is also available on Pharsight’s corporate web site at www. Pharsight contact information Technical support Please consult the documentation (see “WinNonlin user documentation” on page 2) to address questions. as well as the ASCII output. model file and instructions to WinNonlin. if applicable. php +1-919-859-6871 licensing@pharsight. as follows.com http://www. fax or through Pharsight’s customer support web site.pharsight.com +1-919-852-4620 11 .com Telephone: +1-888-708-7444 (from US only) Fax: +1-650-314-3811 Request assistance with Pharsight software licensing by e-mail or telephone: E-mail: Telephone: Customer feedback Please submit requests for product enhancements and defect corrections by email. please contact Pharsight sales: E-mail: [email protected] Pharsight contact information 1 Licensing and product upgrades For license renewals and product upgrades.com/support/support_sflogin. E-mail: Web: Fax: support@pharsight. 1 WinNonlin User’s Guide 12 . script. Text windows contain ASCII text representing a model. See “Charts” on page 115 for instructions on creating and formatting plots. “Workspaces” on page 19: saving and reloading WinNonlin analyses. and modeling and analysis default settings. “Printing” on page 21 “WinNonlin options” on page 28: default file locations. text and chart windows and their supported file types. “Data dependencies and origins” on page 33: refreshing analysis output when source data change. “Data history worksheets and log files” on page 37: tracking changes to data and operations in WinNonlin. WinNonlin windows and files WinNonlin provides three window types for managing data objects: • • • Workbooks hold tabular data in one or more worksheets that support spreadsheet operations and formats. See “Workbooks” on page 41.Chapter 2 Data Management Working with WinNonlin data windows and files This chapter covers data handling in WinNonlin. and unlocking output (removing dependency on source data). • • • • • “WinNonlin windows and files” on page 13: WinNonlin workbook. worksheet behaviors. or modeling output. Chart windows contain plots. • See also: “Data import and export” on page 75. under the following topics. 13 . Click OK. (Before using New Chart. *.JPG) Windows Metafile (*.DTA) Microsoft Excel 5. as noted in the last row of Table 2-1. 5.CSV) Microsoft Excel 4. The New dialog appears. 97 (*.0/95.DTA. *. New Chart will open the Chart Wizard to create the plot(s).WMF) Chart windows 14 . *. New Text or New Chart from the drop-down menu that appears. Choose Tools>Copy Workbook from the WinNonlin menus. N). Select the appropriate radio button.PRN. Open the workbook in WinNonlin. Choose File>New from the menus (Alt+F.0.PWO) SAS® Transport (*.PCO) WinNonlin Graph Object (*. Table 2-1.CSV) ASCII model parameters (*.XLS) Pharsight Workbook Object (*. *. 97. analysis settings entered in the WinNonlin user interface can be saved to file. *.WDO) Pharsight Chart Object (*. 2000 (*. Click the New (left-most) button on the WinNonlin tool bar and select New Workbook. *.XPT.STX) WinNonlin Data Object (*.XPT.PCO) Bitmap image (*.WGO) File formats: save/export ASCII data (*. In addition.BMP) Vector graphic (*.0/95. The new window opens. Any window can be hidden to reduce clutter in the WinNonlin parent window (see “Hidden windows” on page 18). 2.PWO) SAS® Transport (*. 2. Supported file formats Window Workbooks Example Subject data Modeling output Dosing data Model parameters File formats: load/import ASCII (*.PRN.DAT.DAT. To create a copy of an existing workbook: 1. 3. *. File types Each window type can load and save the file types listed in Table 2-1. open the workbook window containing data to be plotted.XLS) Pharsight Workbook Object (*.2 WinNonlin User’s Guide To create a new window: 1.STX) Pharsight Chart Object (*.) or 1. PTO) 2 Text windows Modeling output ASCII models Scripts Notes ANOVA settings Model settings (Load/ Bioequivalence Save in mod. *. Pharsight chart and workbook objects can save multiple worksheets in a workbook.PWO. Changes must be saved to a more recent Excel file type or Pharsight file (*.WTO. A Pharsight Chart Object (*. *.PSC) Pharsight Text Object (*. or multi-chart windows.TXT) Library model(s) (*.RTF) Pharsight Script (*.WGO.WMO) WinNonlin Command File (*. *.CMD) and script files (*.RTF) Pharsight Script (*. *.WDO. A Pharsight Model Object (*. Pharsight objects (*.PCO. *.PSC) Pharsight Text Object (*.LML) Pharsight Model Object (*.PWO.PCO) preserve all formatting.LIB) Rich text (*. *.WSC) WinNonlin Text Object (*.0 files and WinNonlin object files (*.PWO) saves all information in a workbook. constants (dosing). *. A Pharsight Workbook Object (*. *.PSC). the source ASCII model is updated to include any changes in WinNonlin. A Pharsight Text Object (*. inclusions/exclusions and formatting.TXT) Library model(s) (*. Supported file formats (continued) Window Example File formats: load/import ASCII text (*. Non-Pharsight file formats may limit the information saved. data sort order. command files (*.WTO) ANOVA Command File (*. including formatting and links to the graph data. including all worksheets. units. sorting and units applied in WinNonlin.ANV) LinMix or Bioequivalence Command File (*.PTO. When an ASCII model is used with a Pharsight Model Object. data variables.WMO). as well as the graph attributes. as follows. 15 .Model settings eling dialogs) LinMix or Bioequivalence Command File (*.PMO.PTO) WinNonlin Script (*.PTO.WSC) are import-only. including values for the model parameters.PTO) saves all text and formats in a text window. data settings. and modeling options.PMO) contains the current model (the ASCII code or the number of the compiled model) and any associated data.LML) Pharsight Model Object (*.PMO) Note: Excel 4.Data Management WinNonlin windows and files Table 2-1.PMO) WinNonlin Model Object (*.CMD) File formats: save/export ASCII text (*.PCO) saves all charts in a chart window. *.LIB) Rich text (*. data exclusions. Copy the new column. delete the other two columns. Non-numeric data may cause problems in WinNonlin calculations.RTF) preserve text and formatting from text windows. Note: When a chart is saved as a Windows Metafile (*. However. click Yes. non-numeric data may appear in workbook cells that should contain numeric data. units and column headers are not saved. others will not. . formatting. Saving a chart window as a Pharsight Chart Object saves all plots in the window and their attributes (formatting. etc. when transferring data from other packages. a snapshot of the plot that appears on-screen is saved. owing in part to the different sort order inherent in nonnumeric data. Missing data If a data set includes missing values.JPEG or . If you will open the chart in a Microsoft application. One indicator that the data are non-numeric is that non-numeric data will be left justified rather than centered in the cells. This may be caused by leading or trailing blank spaces in some fields.” Some applications that read Metafiles will expect this information. or units. Last.2 WinNonlin User’s Guide ASCII data files save the text of the active worksheet or text window. Multiple worksheets cannot be saved to a single ASCII file.WMF. Excel files preserve data. without any formatting. Rich text files (*. such as New = Old+0. then check that the character denoting missing values was specified when the file was opened (as set in the Options dialog accessed from the Open dialog). To convert non-numeric data to numeric data. and the cells with the missing values do not have a gray background. Troubleshooting: loading tabular data Non-numeric data Occasionally.WMF) WinNonlin prompts whether to “Include placeable header information. 16 .). When a chart is saved to a . then paste only the values (not the formula) to another column. such as Word or Excel.BMP file. and multiple worksheets in a workbook. create a new variable using a formula or function. data inclusions/exclusions. pmo Not available *.prn.pto *.xpt.anv *.Data Management WinNonlin windows and files Column problems 2 If all of the data appear in the first column.dta. *. WinNonlin and WinNonMix file compatibility File Type ASCII Data Excel Data files WinNonlin *.pwo *.0 (*.stx *. WinNonlin and WinNonMix file compatibility The following table outlines file types that can and cannot be shared between WinNonMix and WinNonlin.xls) *.wto *.dat.wdo *.lml Compatibility Identical Identical WinNonMix *.rtf *.pco *.wto *.xls) Excel 5. *. *.0/95 (*.rtf *.xls) Excel 97 (*.csv Excel 4. If a number of blank columns or cells appear where they do not belong.xls) *.0/95 (*.txt.pwo *.xls) Excel 5.csv Excel 4. check the Options (accessed from the Open dialog) to ensure that the correct delimiter is specified when loading the data.pco *.wgo Not available *.pto *. check the Treat Consecutive Delimiters as One option on the Options dialog.prn.wmo *.xls) Excel 97 (*. *.wdo *. *. *. Table 2-2.anv Not available Text files Rich text files WinNonlin files: Data Text Graph Model Pharsight files: Data Text Chart Model SAS Transport files ANOVA command filesa LinMix command files Identical Identical Identical Identical Identical NA Identical Identical Identical Not compatible NA Identical NA 17 . *.lib *.0 (*. *.pmo *. *.dat.wgo *.lib *.txt.dta. imp. Hidden windows cannot be saved to files or workspaces. The Hide Window command leaves the active window open. *. Hidden windows do not appear in list boxes (for modeling. New *.dat.psc.dat. *. for example).exp *. WinNonlin and WinNonMix file compatibility (continued) File Type Bioequivalence command files Model object filesb Dosing files Model parameters Lambda Z range Partial areas Graph template Table definition files Command files Script files ODBC import and export files PKS files WinNonlin *.ANV files cannot be created.imp.dat.wsc *.dta ASCII data: *.dat.dta ASCII data: *.dat. but hidden from the WinNonlin workspace and commands. WinNonMix model object files (*. *.pks Compatibility NA Not compatible Not compatible Not compatible NA NA Identical Identical NA NA Identical NA WinNonMix Not available *. and Chart Wizard.gtp *. b. *. *. *.dta *.dat. *.pmo ASCII data: *.tdf Not available Not available *. Table Wizard.cmd (import only) *. *.dta ASCII data: *. ANOVA command files can be imported into the Linear Mixed Effects Wizard or the Bioequivalence Wizard in WinNonlin.tdf *. Descriptive Statistics.PMO) include more information than WinNonlin model object files.exp Not available a. *. 18 .dta Not available Not available *.gtp *.dta ASCII data: *.lml *.pmo ASCII data: *. Hidden windows Windows of all types can be hidden so as not to clutter the parent window.2 WinNonlin User’s Guide Table 2-2. WinNonlin may prompt the user for the location of some source data files. 2. To un-hide a window: 1. and all data object windows that are open in WinNonlin. All workbooks. 19 . This information can be saved to a workspace file (*.Data Management Workspaces To hide a window: 2 1. Choose File>Save Workspace (Alt+F. To save current work to a workspace: 1. Workspaces A workspace is a snapshot of all current dose and model settings. such as modeling output. if not previously-saved. Choose Window>Hide Window from the WinNonlin menus.PMO) files may use absolute paths to data files.. then loaded. which are named after the workspace. the Save Workspace dialog will appear. If the workspace is new. The workspace file contains the relative paths to these objects. Model settings. 2.PMO). If a workspace containing model information is moved. and model settings that have not yet been saved to disk are implicitly saved to separate files. are saved to a Pharsight Workbook Object (*. to be reloaded in future WinNonlin sessions. are saved to a Pharsight Model Object (*. from the WinNonlin menus. Open or select the window to be hidden. Unsaved workbook data.PWO). V) from the WinNonlin menus. text windows. Check the window(s) to unhide and click OK. Choose Window>Unhide Windows. Note: Pharsight Model Object (*. charts..WSP). To create a new. K) from the WinNonlin menus. 3. 2. and start a new.2 WinNonlin User’s Guide 2. In the Load Workspace dialog. Click Save. empty workspace: 1. 20 . dash and space characters. 2. Select File>Save Workspace As (Alt+F. Navigate to the appropriate directory and enter a name for the workspace. L) from the WinNonlin menus. navigate to the appropriate directory and double-click the desired workspace. empty. Click Save. To save an existing workspace to a new workspace name or location: 1. Choose File>New Workspace (Alt+F. Choose File>Load Workspace (Alt+F. Navigate to the appropriate directory and enter a name for the workspace. WinNonlin will prompt the user to save any unsaved work. Limit the name to alphanumeric. dash and space characters. The Save Workspace dialog appears. 3. close all windows. workspace. To load a saved workspace: 1. W) from the WinNonlin menus. Limit the name to alphanumeric. Select Page Setup from the File menu (Alt+F. Page setup To set page settings for print output: 1.Data Management Printing 2 Printing Customize print output using the following features: • • • • “Page setup” on page 21 to set formatting and layout options. 21 . “Print” on page 25 to print the current window. The page setup dialog appropriate to the active object appears. “Print preview” on page 25 to check print output before printing. “Print all” on page 27 to print selected items from any open windows. G). See: • • • “Page setup for workbooks” “Page setup for text windows” on page 24 (including model windows) “Page setup for charts” on page 25 Page setup for workbooks Page setup options for workbook windows are specified on four tabs: Page Tab Orientation: Select Portrait (vertical) or Landscape (horizontal) page layout. 2 WinNonlin User’s Guide Scaling: The workbook can be scaled as a percentage of the normal size or select Fit to in order to make it fit on a certain number of pages. Any entry not preceded with the ampersand “&” will be reproduced as 22 . in inches or centimeters. Page Numbering: Select Automatic to start numbering pages at 1. Paper Size: Select the size paper to use. Headers and Footers Tab Enter the alignment. Check Horizontally and/or Vertically to center the output on the page horizontally (across). or check the Start with page number check box and enter the first page number. formatting and content for page headers and footers using the codes listed in the tables below and/or text to be printed verbatim. Margins Tab Enter margins for the worksheet. vertically (up and down) or both. including the header (top) and footer (bottom) margins. as selected under Units. Uses the specified font size . Underlines the header. Prints the Current Time. Special character codes for printing worksheet-specific information such as worksheet titles must appear last. If the codes are entered in the wrong order WinNonlin may ignore some of them. Prints the Page Number.must be a two digit number. Table 2-3. Left-aligns the characters that follow.number && Prints the Workbook Name. Prints the page number plus number. Alignment codes &L &C &R Left-aligns the characters that follow. Strikeout font. Uses an italic font. Codes and text are centered unless &L or &R is specified. Table 2-4. Font codes &B &I &U &S &”fontname” &nn Uses a bold font. Codes must be space-separated. Centers the characters that follow.Data Management Printing 2 entered. Prints the page number minus number Prints an ampersand 23 . Table 2-5. Uses the specified font. Alignment codes must precede fontrelated codes. Right-aligns the characters that follow. Centers the characters that follow. Special codes for workbook information &F &A &P &D &T &L &C &R &P + number &P . Right-aligns the characters that follow. Prints the Current Date. Prints the Worksheet Name. Page Order: Specify the print order: Top to Bottom (left columns print before those farther right) or Left to Right (top rows print before the lower rows). Special codes for workbook information &N Prints the total number of pages in the document Sheet Tab Print Area: Enter the range of columns and rows to print using absolute or relative cell references. Margins: enter the page margins in inches.2 WinNonlin User’s Guide Table 2-5. so that items with shading or color do not use grayscale (gray “missing” cells will appear white). as absolute or relative cell references. the title is printed at the left edge of each page. but they must be adjacent. Multiple rows or columns may be selected. Page setup for text windows Page setup options for text windows sets the page orientation and margins. Output can be specified to print in black and white. If a row is entered. Separate non-contiguous ranges with commas. Print Options: Set whether or not grid lines and row and column headers will print. 24 . Orientation: Select Portrait (vertical) or Landscape (horizontal) page layout. If a column is selected. the title is printed at the top of each page. Print Titles: Specify rows or columns for which titles should be printed on each page. Separate non-contiguous ranges with commas. Multiple charts per page: If multiple charts are to be printed per page. 4. Once this option is selected. Print To print the active window using the default print settings: 1.Data Management Printing 2 Page setup for charts Page setup for charts sets chart size and the number of charts per page. 5. 2. A Print Preview window appears with the layout from the Page Setup dialog. left and right margins of the page. Scaling: Select Actual Size to print the chart at its original size. Click Next to scroll through multiple pages. Select Layout for Printer to expand the scale of the graph to the size and layout of the selected paper orientation (similar to the Scaling function). Layout: Select Screen Layout to print the graph at the size it displays on the screen. as charts will be scaled to fit the selected number across and down the page. 25 . Make sure that the object to print is the active window in WinNonlin. bottom. 3. Select File>Print Preview from the WinNonlin menus. depending on the Windows settings. Click Close to close the window and return to WinNonlin. Margins: Enter a value in inches or centimeters. select this check box and select the format from the list box. the controls for scaling and layout are disabled. Select Best Fit to scale the graph proportionally to fit the page (recommended). to specify the top. Print preview To preview print output: 1. Orientation: Select whether the graph is to be printed on the page vertically (Portrait) or horizontally (Landscape). Click the Print button in the WinNonlin tool bar. Use a hyphen to indicate ranges. 4. Click Preview to view the expected print output or OK or Print. Note: To select multiple sheets in a workbook. text). hold down the Ctrl key and click on each worksheet tab. Set options as appropriate for the object type (workbook. Selected Pages: Select Pages in the Print Range box and enter the page number(s) as a comma-separated list. chart. Note: Page numbers are printed for text objects only when the printing is initiated via the print button on the tool bar. change line colors to black. P) from the WinNonlin menus. The Print dialog appears. Select the Print Range: • • The entire worksheet: Select All in the Print Range box. Under Print What. 2. Printing workbooks 1. Printing text objects and models 1. Choose File>Print (Alt+F. See also: “Printing workbooks” on page 26 and “Printing text objects and models” on page 26. 26 . select whether to print all worksheets (Entire Workbook) or only the currently selected worksheet(s) (Selected Sheet(s)). 3. Note: When printing charts. use the Page setup dialog first in order to specify multiple charts per page and chart size. 2. Enter the number of copies to print.2 WinNonlin User’s Guide To print the active window: 1. 2. Select the appropriate printer and paper size. Select the window to be printed. If lines appear fuzzy in printed plots. 2. c.) 6. (See “Print all options” below. To select specific items to print: a. 5. Print all options Multiple charts per page: Check this box and select the layout below to put more than one chart on a page. Choose File>Print All from the WinNonlin menus. click the Print button in the Print All dialog. To print all open windows. Open the window(s) to print. 4.Data Management Printing 2 Print all To print a selection from the open windows: 1. Uncheck Select all objects. Click Print. Click the “+” symbols to view individual windows. including all worksheets and charts. worksheets or charts. Use the Options button to set up the page layout for charts. with page breaks between sets of charts. 3. b. 27 . Check the box next to the items to print. charts. Click Apply to set these options and keep the dialog open or OK to keep these settings and close the WinNonlin options dialog. 2. Choose Tools>Options from the WinNonlin menus and open the Workbooks tab of the WinNonlin options dialog. Choose Tools>Options from the WinNonlin menus and click on the Directories tab of the WinNonlin options dialog. Workbooks To set or change default workbook options: 1. 4. and/or text with or without page breaks • default orientation of the Final Parameters output table • default method for the calculation of AUC in noncompartmental analyses • treatment of group variables in the table generator • default use of page breaks in tables • display of units with column headers in worksheets. click the Browse button for that directory. if it is not already open.2 WinNonlin User’s Guide WinNonlin options WinNonlin provides options to set default file locations. To set the default file location for WinNonlin model library or user model library files. worksheet behaviors. 3. WinNonlin options Category Directories Options • default directory for WinNonlin Library and User Library files Workbooks • whether to display formulas in worksheets • direction in which the Enter key moves the cursor in a worksheet • value used to denote missing data in worksheets Models • default output: workbooks. and modeling and analysis default settings as described in Table 2-6. Select the desired drive and directory. 28 . Table 2-6. chart labels and tables • disable/enable warning message for impending license expiration Tables Units Licensing Directories To set or change the default model library directories: 1. The picture of the folder opens. Models To set default model options: 1. 3. Choose the direction of cursor movement between worksheet cells when the Enter key is pressed: check Enter Moves Down to move the cursor down the column. in individual worksheet cells. Text: check to include text output listing model settings. Note that the formula bar at the top of each worksheet displays any formula in the current cell.Data Management WinNonlin options 2 2. A second worksheet called Non-transposed Final Parameters will present parameters in separate rows. Set the layout for the Final Parameter output worksheet: Checking Transpose Final Parameter Table will present each parameter in a separate column. Any cell containing this string will be grayed and treated as missing in (excluded from) modeling and analyses. enter it in the box labeled Missing Value. This layout is a useful for performing additional analyses. Figure 2-1 and Figure 2-2 present an example of each layout. 2. rather than the values computed by formulas. 29 . 5. fitting progress and final results. Click Apply to set these options and keep the dialog open or OK to keep these settings and close the WinNonlin options dialog. 4. Charts: check to include charts of model data and results. Use the check boxes to select the default form(s) of modeling output: • • • Workbook: check to include tabular results in a workbook. 3. Check the Show Formulas check box to display formulas. such as descriptive statistics. To specify a character or string to represent missing data. the cursor will move to the right when the Enter key is pressed. else. Check Page breaks for paginated output. Choose Tools>Options from the WinNonlin menus and open the Models tab in the WinNonlin options dialog. on modeling output. 7 of Gabrielsson and Weiner (2001). entitled Transposed Final Parameters. the Final Parameters worksheet will use the untransposed format. Figure 2-2. will contain the transposed version. See Section 3. 30 . Example modeling output: un-transposed Final Parameters. using linear interpolation between untransformed data through Clast. to be used in computing area under the curve (AUC) in noncompartmental analyses. Linear trapezoidal (linear interpolation): linear trapezoidal rule. from among the methods below. • • • • Linear/log trapezoidal: log-trapezoidal rule. Select the default NCA calculation method. using linear interpolation between log-transformed data points through Clast. log-linear interpolation between data points from Cmax. and between log-transformed data from Cmax through Clast. This option is provided primarily for WinNonlin scripts that depend on the transposed format the Final Parameters worksheet. for detailed discussion of AUC computation methods. and the second worksheet. 4. Example modeling output: transposed Final Parameters. extrapolated to Tinf. If the Transpose Final Parameters option is not checked.2 WinNonlin User’s Guide Figure 2-1. Linear up/log down: linear interpolation between untransformed data up to Cmax. Linear trapezoidal (linear/log interpolation): trapezoidal rule with linear interpolation up to Cmax. 2. Choose Tools>Options from the WinNonlin menus and open the Tables tab. Note that these settings can be overridden in individual tables and charts. Units for specific output parameters can be changed for a modeling session (see “Model options” on page 203).Data Management WinNonlin options 2 5. Choose Tools>Options from the WinNonlin menus and select the Units tab of the Options dialog. 2. Choose whether to have units displayed on chart labels and/or in tables. 3. Use the check box to set the default for creating page breaks when the value of the group variable(s) changes. To specify default units display: 1. Tables To specify default table options: 1. check Place Page Break Data Outside Table Body. Select the Show Units with Column Headers check box to include units in column headings of output workbooks. or choose File>Options from the Table Wizard menus. Units The display of units in workbooks. rather than as a column in the table body. 3. WinNonlin will generate a warning message upon launch if your license to use the software is within 30 days of expiration. 4. 4. Click Apply to set these options and keep the dialog open or OK to keep these settings and close the WinNonlin options dialog. The Licensing tab 31 . charts and tables is set on this tab of the WinNonlin options dialog. Licensing Be default. Click Apply to set these options and keep the dialog open or OK to keep these settings and close the WinNonlin options dialog. To print group variables outside and above the table. Click Apply to set these options and keep the dialog open or OK to keep these settings and close the WinNonlin options dialog. Best Available: WinNonlin will attempt to secure an Enterprise license. 2. 4. b. To prevent WinNonlin from ever producing the warning. (See “WinNonlin editions” on page 1 for information about WinNonlin editions. Choose Tools>Options from the WinNonlin menus and open the License tab of the Options dialog. then. To allow WinNonlin to produce a warning message on launch within 30 days or license expiration. select Enable under Warnings.2 WinNonlin User’s Guide of the WinNonlin options dialog provides a control to disable. Professional. To prevent WinNonlin from producing the license warning for a set number of days (beginning on the day this selection is made). Enterprise: WinNonlin will only accept Enterprise edition licenses. Click OK to apply your selection. 5. 32 . 2. Click OK to apply your selection. Professional: WinNonlin will accept only Professional edition licenses. c. 3. The new setting will apply the next time you launch WinNonlin. select Postpone for under Warnings and select the appropriate number of days in the drop-down list. or the best available from the network license server. The new setting will apply the next time you launch WinNonlin. Choose Tools>Options from the WinNonlin menus and open the License tab of the Options dialog. Under Features. The Licensing tab also provides a means for floating license users to specify a preference to obtain a license for WinNonlin Enterprise edition. 3.) To disable or enable license expiration warnings: 1. select Disable. if no Enterprise license is available. a Professional license. choose the license type that WinNonlin should request from your license server (see “WinNonlin editions” on page 1): a. delay or enable this warning message. To set a preference for type of floating license: 1. 2. To install a node license key code.16. You can refresh 33 . pages 14 . Click Tools > Options > Licensing tab. text windows. 4. pages 12 and 13. pages 11-16. 2. Data dependencies and origins In order to protect the integrity of analysis results. 1. To install a floating license key code. WinNonlin marks each derived product as out of date. follow the instructions in the WinNonlin Getting Started Guide. WinNonlin locks derived data products—including modeling or analysis output in workbooks. Go to the main menu. Check the IVIVC Checkbox. If the source data changes. 3. 5. After the license is installed. 1. Click OK. follow the instructions in the WinNonlin Getting Started Guide.Data Management Data dependencies and origins Installing an IVIVC license 2 Note: Installing an IVIVC License requires the same steps and listed descriptions as those for WinNonlin in the Getting Started Guide. and charts—against changes. Open WinNonlin. by coloring it pink. WinNonlin must be configured to recognize IVIVC. with derived objects as branches.2 WinNonlin User’s Guide the derived objects based on updated source data. 34 .PWO (top item) provided the source data for the descriptive statistics results in the workbook THEOPH24_NCA_DSTATS.PWO. Its tree shows dependencies in reverse of the Parent View. 2. the modeling output workbook THEOPH24_NCA_OUTPUT. make output independent of the source so that it is unlocked and editable. The Parent View (below) shows the items’ source objects.PWO is dependent on its source workbook. The Child View shows objects that are derived from a selected item. if desired. In the example below. Select an object from the drop-list at the top of the Dependency View dialog. choose File>Dependencies from the WinNonlin menus. THEOPH_12VS24. and the model THEOPH_NCA. To view dependencies: 1. the noncompartmental analysis output THEOPH24_NCA_OUTPUT.PWO. or. in a tree with source objects as branches. In this example. After an analysis.PMO. if any. create the desired output. 2. Make the necessary edits to the source data and/or analysis settings. Click the Origins button to view pedigree information for derived objects. Refreshing results If source data or analysis settings are modified while their derived objects (output) remain open in WinNonlin. Saving dependencies Dependencies persist only while related objects remain open in WinNonlin. with dependencies intact (see “Data dependencies and origins” and “Saving dependencies” above). 35 .” Note: Opening and using the OK button to close any modeling or analysis dialog marks derived objects as out of date regardless of changes to analysis settings. To save objects with dependency information. or when they are saved together in a workspace or PKS scenario. 4. Click the Refresh button to view pedigree information for derived objects. each output window is shaded and its title bar includes: “Source Changed. Select an output item from the analysis to be refreshed at the top of the Dependency View dialog. Choose File>Dependencies from the WinNonlin menus. 3. and save the set to disk as a workspace (see “Workspaces” on page 19) or to the Pharsight Knowledgebase Server as a scenario (see “Using the Pharsight Knowledgebase Server” on page 445). the derived objects become out of date. Open all input and output in WinNonlin.Data Management Data dependencies and origins 2 3. keep all objects open. To update output to reflect changes to source data or analysis settings: 1. See “Refreshing results” on page 35. See “Object origins” on page 36. 4. 2 WinNonlin User’s Guide 5. making its data editable and removing object dependencies. Selecting either NCA output object for refresh would update both NCA objects but not the descriptive statistics. as above. will update that output and the NCA chart and workbook on which it depends. a change to data exclusions in THEOPH_12VS24. and statistics on the NCA workbook (THEOPH24_NCA_DSTATS. 2.PWO). Changes to the source data will no longer mark this derived work as out of date.PWO and THEOPH24_NCA_OUTPUT. Click this button to view a pedigree for each child object open in WinNonlin. Unlocking derived data Analysis and modeling output can be unlocked. After running an analysis. In the example above. including: 36 .PWO deprecated the NCA chart and workbook output (THEOPH24_NCA_OUTPUT. and any derived objects that lie between the source data and the selected item. Click the Detach button. 3. Object origins The Origins button appears when an object that is dependent on any other object(s) is selected in the Dependencies View dialog. and Refresh will no longer be available. Click the Refresh button to re-generate all output from that analysis. Selecting the descriptive statistics for refresh.PCO). To unlock a derived product: 1. Select the item to unlock at the top of the Dependency View dialog. choose File>Dependencies from the menus. The sheet then logs edits to the workbook. and indicates the number of worksheets. e. and only the Annotations column can be edited (enter notes here). modeling or LinMix Source ID: identifier for any source data from which the object derives. a report of analysis settings and variables. This pedigree information is saved as an XML dependencies file (*. PKS operations. Two known cases that are not captured in the History page are filling a range of cells and moving a range of cells. It cannot be deleted. if applicable. The first line always notes creation of the workbook. WinNonlin also includes a separate application. Data history worksheets and log files WinNonlin provides two means to track data operations. with the worksheet and. the WinNonlin Log Viewer.g. all data manipulations.DEP) when the source and derived objects are saved in a workspace. workbook name for source data For certain analyses. This file is always saved as part of a PKS scenario. including application of units.Data Management Data history worksheets and log files 2 • • • • Object type and file or window name WinNonlin engine used to create the object. See “WinNonlin Log Viewer” on page 39.. History worksheet The History worksheet logs operations made on the workbook data. and the analysis operations listed in Table 2-7. including sorts and transformations. Each WinNonlin workbook includes a History worksheet that logs most operationsA on data from that workbook. that provides access to a log file that tracks all data operations performed in WinNonlin. A. 37 . notification that all parameters set to Missing • For NCA.• Source data workbook and worksheet tistics • Missing/excluded values • For weighted analyses.2 WinNonlin User’s Guide Table 2-7. and the profile number at which each occurred • If warnings generated. notification if any value(s) of the dependent variable <= 0 • Notification of any regressor/covariate found to be alphanumeric • Warnings or errors from the analysis. if a least square mean is <0. and note that all values for that profile set to missing • Success or failure of the LinMix/BioEq operation LinMix and • Source data workbook and worksheet Bioequivalence • Missing/excluded values (BE) • In BE. Analysis operations logged in the History worksheet Operation Deconvolution Items logged • • • • Source data workbook and worksheet Missing/excluded values Notification of any failed or invalid profiles Success or failure of the deconvolution operation Descriptive Sta. notification of weights < 0 (all values for that profile will be missing) • Notification of any floating point error. lambda_z exclusions (or the lack of any) for each profile • Modeling halted using Stop button • Success or failure of the modeling process Toolbox tools • Source data workbook and worksheet • Success or failure of the toolbox engine operation 38 . notification of any profile with all zero values for volume or concentration • For NCA. and it occurs at time=0. notification that some calculation(s) couldn't be made • If Ln transformation included. but no errors: “warnings generated on one or more profiles of data” • Success or failure of the LinMix/BioEq operation Merge • • • • • • • Source data workbook and worksheet #1 Missing/excluded values from source worksheet #1 Source data workbook and worksheet #2 Missing/excluded values from source worksheet #2 Sort variable used from each worksheet Include variables from each worksheet Success or failure of the Merge operation PK and Non• Source data workbook and worksheet compartmental • Missing/excluded values analyses (NCA) • Profiles with less than 2 usable records noted: “cannot be processed” • For urine NCA. if only 1 non-zero concentration is found for any profile. not cell-by-cell values) • Changing WinNonlin registry settings • Transform operations • Specifying column units • Clearing units • PK/PD modeling errors • Cut/Copy/Paste operations • Include/Exclude data operations (include/ exclude parameters. • WinNonlin startup and finish • Open/save workbook • Replace/Remove data (replace/remove parameters.2>Log Viewer. for reviewing a record of operations performed in the application. the Log Viewer. When the log file reaches its size limit. From the Windows Start menu. the file is purged to a backup file. including messages of success/failure 39 . choose Programs>Pharsight>WinNonlin 5. This log file size can be specified or changed using View>Options in the Log Viewer application menus.Data Management Data history worksheets and log files 2 WinNonlin Log Viewer WinNonlin has a separate application. WinNonlin writes this information to a separate log file. To run the WinNonlin Log Viewer: 1. The application log contains entries for the following. not cell values) • Insert column/row • Delete column/row • Undo operations • Units conversions. 2 WinNonlin User’s Guide 40 . shown as individual tabs within a workbook window. Every workbook contains at minimum one data worksheet and a History worksheet. • “Inserting or deleting worksheets” on page 42 41 . merging workbooks.Chapter 3 Workbooks Data editing. which details many of the changes and operations made on the data (see “Data history worksheets and log files” on page 37). Editing data WinNonlin worksheets support spreadsheet-like data entry and formatting. or imported (see “Data import and export” on page 75). Data can be typed or pasted into a new worksheet. and manipulation in WinNonlin worksheets Input and output data for WinNonlin analyses are stored and manipulated in Excel-compatible worksheets. excluding and including data. formatting. copied. • • • • “Editing data” on page 41: editing and formatting worksheet data using formulas. “Data manipulation” on page 56: sorting. Worksheet operations are covered in the following sections. “Transformations” on page 62: creating new workbook columns as functions of other columns. Numerical and text values may be entered directly into worksheet cells. copy and paste and fill-down. and/or pasted via the Windows Clipboard. and transforming non-numeric values. and cut. “Status code transformations” on page 69: replacing non-numeric codes with specified values. Worksheet editing operations are detailed in the following sections. functions. Choose Edit>Insert Sheet from the WinNonlin menus.3 WinNonlin User’s Guide • • • • • • • • “Inserting and deleting rows. 3. The Sheet Name dialog appears. Other worksheets can be added and deleted. Select the worksheet to be named. Click Yes to confirm deletion of the current worksheet. To name a worksheet: 1. Choose Edit>Delete Sheet from the WinNonlin menus. 2. 42 . find next and replace” on page 50 “Go to” on page 52 Inserting or deleting worksheets Every workbook must contain at least two sheets: a data worksheet and a worksheet labeled Data History (see “Data history worksheets and log files” on page 37). 3. To delete a worksheet: 1. Open the workbook to the insertion point for the new worksheet. The new worksheet will be inserted in front of the currently-selected worksheet. Open the worksheet to be deleted. Double click on the tab for the sheet. These worksheets cannot be deleted. A new blank worksheet is inserted on top of the current worksheet. Enter the name to be displayed on the tab and click OK. columns or cells” on page 43 “Using formulas and functions” on page 44 “Filling adjacent cells” on page 46 “Formatting cells” on page 46 “Clearing cell values and formats” on page 49 “Undo” on page 50 “Find. 2. To add a new worksheet: 1. 2. In the Insert dialog that appears. select how to move the surrounding cells. To delete cells: 1. 3. I) or right-click the selection and choose Insert from the shortcut menu. columns or cells To insert a row or column: 1. select the same number of rows or columns as the number to be inserted. Select Insert from the Edit menu (Alt+E. To insert a cell: 1. 2. A row or column will be inserted above or left of the active row or column. 2. D) or right-click the selection and choose Delete from the shortcut menu. A single cell is inserted. 2. 43 . I). select Entire Row or Entire Column. 3. Highlight a cell within the row. then choose Edit>Insert. Select Insert from the Edit menu (Alt+E. 3. Select Delete from the Edit menu (Alt+E. Highlight a row or column by clicking on the row or column label. I) or right-click and select Insert from the shortcut menu. Select Insert from the Edit menu (Alt+E. 2.Workbooks Editing data 3 Inserting and deleting rows. select either Shift Cells Right or Shift Cells Down. Select a cell with the mouse. From the Delete dialog. Select the cell(s) to be deleted. Note: To insert multiple row or columns quickly. or 1. In the dialog that appears. to create a variable equal to the value in column B minus the value of cell A1. A3 denotes the third cell in the first column. 2. C3 = B3A3. D) or right-click the selection and choose Delete from the shortcut menu. 2. “=. rather than relative to the currently-selected cell. From the Delete dialog. See Table 3-1 for a list of valid operators. D). For example. select Entire Row or Entire Column. 44 . you could set cell C1 = B1-A1. Using formulas and functions Worksheet values may be entered as mathematical formulas. A1:B4 calls the first four cells in the first two columns (starting from the top left cell in the worksheet). etc. Cell C2 would equal B2-A2. Cell references Formulas and functions refer to worksheet cells by location in the worksheet grid. or 1. Pasting this formula down column C would cause C1 = B1-$A1. Select the row(s) or column(s) to be deleted by highlighting the row or column label(s). Select Delete from the Edit menu (Alt+E. A complete listing and syntax are provided in “Worksheet Functions” on page 551. 3. for copy and paste operations. separated by a colon. C2 = B2-$A1.3 WinNonlin User’s Guide To delete rows or columns: 1. Select a cell in each row or column to be deleted. enter the top-left and bottom-right cells. In contrast.” This is useful for creating new columns as a function of other columns. For example. To identify a range of cells. to create a new column of values equal to the difference between values in columns B and A. etc. then use the fill-down feature to paste this formula to all cells in column C. For example. by beginning the cell’s value with the equals sign. WinNonlin worksheets also support an extensive list of functions. C3 = B3-$A1. Select Delete from the Edit menu (Alt+E. A “$” makes a reference position absolute. you would cell C1 = B1-$A1. using row number and column letter. . =SQRT(A1) or =A1+5. type the formula. Press the Enter key to enter the formula into the cell. Select the cell and position the cursor over the small black square at the lower right hand corner of this cell until the cursor changes to a small black cross. In the edit bar at the top of the worksheet. select the first cell in the column of the new variable. 2. Using the mouse. 4.g. 3. To enter cell references. 45 . e. either type them or click on the desired cell (drag to select multiple cells) to insert the reference into the formula. including the operator and an expression representing the value.Workbooks Editing data 3 Table 3-1. Workbook operators Operator + / * ^ % = > < >= <= <> Description Addition Subtraction Division Multiplication Exponentiation Percentage Equal to Greater than Less than Greater than or equal to Less than or equal to Not equal to To create a new variable using a formula: 1. To copy to adjacent cells in a selected block of cells: 1. 4. 46 . or to the left or right. Edit>Fill>Right or Edit>Fill>Left from the menus. Edit>Fill>Up. Press and hold the mouse button. Select the cell to copy. Select the cells to be filled.3 WinNonlin User’s Guide 5. Filling adjacent cells A value or formula from one cell can be filled to adjacent cells up. down. Position the cursor over the small black square at the lower right hand corner of this cell until the cursor changes to a small black cross. The Format Cells dialog has five tabs for specifying the following. enter the value or formula to copy. To copy to adjacent cells using the mouse: 1. 2. and press the Enter key. and drag the cursor down the column to copy the formula. If necessary. and drag the cursor down and/or to the right over the target cells to copy into them. Formatting cells Choose Data>Format (or click the Format tool bar button with a workbook in the active window) to apply formatting to the currently-selected worksheet cell(s). Select Edit>Fill>Down. 2. including the cells to be copied as the left-most cells in the selection. Press and hold the mouse button. 3. 3. but not non-numeric entries. to align positive values with negative values enclosed in parentheses. Choose format settings from the tabs detailed below. 2. An underscore “_” at the beginning of symbols indicates left-alignment of left parentheses. 4. Select Data>Format from the WinNonlin menus or click the Format tool bar button. Category: Select a number category or All to show all types. Type: Select a format here. and indicates whether thousands are separated by commas. To set the format for a cell or selection of cells: 1. The Sample field will display an example. An underscore “_” at the end inserts a space. # represents digits to the thousands. • • • • 0 (zero) represents the a digit or exponent and decimal places. Select the cells to format. Number format The Number tab of the Format Cells dialog sets the handling of numbers. The Format Cells dialog appears. if any.Workbooks Editing data 3 • • • • • Note: Number format Alignment Font face and format Borders Fill color and pattern Data formatting is not saved in ASCII (text) files. Syntax in the Type box is as follows. To select whole rows or columns. in the selected cells. Symbols appearing after a semicolon indicate handling of negative numbers. Click OK to apply the changes or Cancel to close the dialog without changes. click on the row or column header(s). 47 . y represents years. 48 . Left: left-aligns cell content. Horizontal: alignment from left to right can take the following values: • • • • • • General: left-aligns text and right-aligns numbers. wraps text onto multiple lines to fit the cell width. Vertical: aligns content at the top. Choose Category = All and Type = General to display entries as they are typed in. E indicates scientific notation. or bottom of each cell. s represents seconds. d represents days. Alignment The Alignment tab of the Format Cells dialog sets alignment and line wrapping to the selected cells. Wrap Text: when text content is longer than the cell width allows. h represents hours on a 24-hour clock. but using scientific notation for numbers >= ten billion. Center: centers cell contents. Center across cells: centers content in the left-most cell across a multicolumn selection. center. assuming adjacent cells to the right are empty. ?/? indicates display as fractions rather than decimals. Merge Cells: joins selected cells into a single cell that spans multiple columns and/or rows. Fill: repeats cell contents enough time to fill the width of the cell. Right: right-aligns cell contents.3 WinNonlin User’s Guide • • • • • • • • Note: * indicates that zero values will be replaced with a dash. m represents minutes for times or months for dates. and merges cells. see “Inserting and deleting rows. but not formatting. red line around the outermost edges of the block of selected cells. color and format to the selected cell or cells. blue lines between cells within the selection. select a line style and color. To delete a cell from the grid. Clearing cell values and formats When cells are cleared. and thin. use the buttons or click between the marks in the diagram below to indicate border locations. Fill color and pattern Use this tab to set background color and pattern for the selected cell or cells. Next. 49 . from cells: 1. First. To clear data values. The example below will put a thick. Borders Use this tab to create borders around the selected cell(s). the values and/or formats are deleted. Select the cells to be cleared. columns or cells” on page 43. but the cells remain in the grid.Workbooks Editing data 3 Font face and format Use this tab to assign a font face. 1. Undo The Undo capability allows the single. Choose Edit>Clear>Values from the menus. Select the cells to be cleared. Select Edit>Clear>All from the menus. when editing workbooks. To clear data values and formats from cells: 1. To clear formatting from cells: 1. but will not remove the cells. f or Ctrl+f). Undo will appear as a normal selection on the Edit and shortcut menus. below. U). Choose Edit>Find from the WinNonlin menus (Alt+e. Select the cells to be cleared. Select Undo from the Edit menu (Alt+E. or press the Delete key.1 to 0. find next and replace To locate specific data values within a worksheet: 1. 2. right-click in the worksheet and select Clear Values from the shortcut menu. This will delete the information from the cells. most-recent action to be reversed. 3. 2. To undo an operation: 1. Select Edit>Clear>Formats from the menus. use the keyboard shortcut Ctrl+Z or right click in the window and select Undo from the shortcut menu. Alternately. The following example will find concentration values from -0. Enter a numeric or text search string in the box labeled Find. Select the appropriate options. Alternately. When an operation has been performed that can be undone. 2. 50 . charts or text windows.3 WinNonlin User’s Guide 2. Find. The Undo command is disabled if the most-recent action cannot be undone. – Less than the search string (<). 51 .Workbooks Editing data 3 Case Sensitive Checking this box causes only occurrences with the exact combination of uppercase and lowercase letters to be found. or – Not equal to the search string (<>). – Greater than the search string (>). Operation This option searches for a value: – Equal to the search string (=). Select Replace from the Edit menu (Alt+E. the currently-selected column or a selection of cells. E). the column Conc is one potential search area because the a cell in that column was selected before opening this dialog. N). Click Find. In this example. Search Area Allows searching over the entire data set. The Replace dialog appears. Select Find Next from the Edit menu (Alt+E. – Greater than or equal to the search string (>=). To repeat the designated search: 1. 4. Tolerance Finds values not different from the (numeric) search string by more than the tolerance (optional). To replace text or data in a grid: 1. – Less than or equal to the search string (<=). Enter the grid location of the cell. 52 . G). Options are the same as for Find. for the third cell (3) in the first column (A). Click Goto. 3.3 WinNonlin User’s Guide 2. enter: A3. Click Replace to review each occurrence before replacement or click Replace All to replace all occurrences without reviewing each. Select the appropriate options. Go to The Go to function jumps to a specific cell in any worksheet within the current workbook. Choose Data>Goto from the menus (Alt+D. 3. For example. 4. 4. with one addition: Remove Row Matching Criteria Check this box to remove any row that contains the search string within the search area. The Goto dialog appears. 5. since the Case Sensitive option is checked. above. values of “bql” will not be replaced. The example above will replace all concentrations with a value of “BQL” with zero. Enter a numeric or text replacement string in the box labeled With. 2. Note that. To move to a given cell using a cell reference: 1. Enter a numeric or text search string in the box labeled Replace. Select any worksheet in the active workbook under Sheet. Type the name. 4. WinNonlin will display them in braces: {furlongs}. Note that long column names can affect performance. 53 . for the new worksheet that will contain the transformed data (see “Multiple transformation” on page 64). Press Enter or click Close. such as descriptive statistics. Each column in a worksheet can use different units. Note: WinNonlin does not allow spaces in column headers. Nonstandard units—units enclosed in braces—will be carried throughout any operations. but not used in any calculations. the data cannot be used in analyses and models. Highlight the column to be named.Workbooks Column names and units 3 Column names and units The Column Properties dialog sets column headers and units for the current worksheet or. otherwise. Click the Edit Header button. The new column header appears in the worksheet. See “WinNonlin options” on page 28 for units display and default options. 3. Permissible units include the following. If the units are not standard or recognizable. 6. To specify column headers: 1. for multiple transformations. Click the Column Properties tool bar button or double click a column header to open the Column Properties dialog. The units appear in the column header. 2. in parentheses. If a data set is loaded that has column names with spaces WinNonlin will ask to substitute an underscore (_) for the spaces (and offer the opportunity to save the data set to a different name). Highlight a cell in the column to be named. and will be used in data calculations. 5. To enter units in the Column Properties dialog: 1. Select the column name in the column header field. Table 3-3. Enter the units for that column in the New Units field.3 WinNonlin User’s Guide Table 3-2. 2. Unit types Unit type Time day hour minute second millisecond microsecond nanosecond Mass gram mole pound IU Volume Liter Unit hr min s ms us ns g mol lb IU L Abbreviation day Prefixes can be used for units of mass and volume. 54 . Prefixes Prefix femto pico nano micro Abbreviation f p n u Prefix milli centi deci deca kilo Abbreviation m c d dk k Note: Units are not case-sensitive. The Units Builder assists with units selection. Click the Specify Units button. Note: Worksheet units can be saved only in Pharsight Workbook Objects (*. 4. Click the Convert Units button. Select a cell in the column to operate on. Units builder Units are edited in either the Column Properties dialog or the Units Builder. 3. Click Close to close the dialog.PWO). Converting units Units can be converted automatically using the Column Properties dialog. See “Units builder” on page 55. Click the Column Properties tool bar button. 3. The current units appear in the list box.Workbooks Column names and units 3 3. Enter the new units in the Specify Units field. 4. The Column Properties dialog appears with the column selected. Select a cell in the column to be converted. The Column Properties dialog appears with the column selected. 4. Other file types lose unit assignments when the file is closed. To convert units: 1. 2. The current units appears in the list box. 2. Click Close to close the dialog. Click the Column Properties tool bar button. 55 . Click the Clear Units button. The data in the column are converted to use the new units. To clear the units associated with a column: 1. In the Column Properties dialog click the Units Builder button. To build units with operators (grams/ml. Mass. 3. Once the unit (with appropriate prefix) displays correctly click the Add button to the right of that kind of unit. Click Close to close the Column Properties dialog. or Volume) and the appropriate unit. Select the kind of unit (Time. To freeze a row or column: 1. 56 . With the data open in WinNonlin. enter the units and operators in order. 4. milli. etc. 6. 8. or one cell at the outer-most row/column to freeze. The Units Builder dialog appears. Click OK to close the Units Builder. Data manipulation WinNonlin worksheet operations include the following: • • • • “Freezing rows and columns” on page 56 “Sorting” on page 57 “Merge” on page 57 “Data exclusion and inclusion” on page 60 Freezing rows and columns Freezing one or more rows and/or columns creates a non-scrolling region that remains on the screen while you scroll through the remaining data. click the Column Properties tool bar button or select Data>Column Properties from the WinNonlin menus. 2. 7. Prefixes (kilo. 9. Select the column or row. Click the Specify Units button to associate the units with the data column.3 WinNonlin User’s Guide To use the Units Builder: 1. for example). The appropriate abbreviation displays in the New Units field.) can be selected from the Prefix field. 5. Sorting Sorting provides a quick means of ordering worksheet data. The Sort dialog appears. Choose Window>Freeze Panes from the WinNonlin menus.Workbooks Data manipulation 3 2. Merge The Merge command joins selected data from any two worksheets into a new workbook. 2. If the data have been sorted previously. Select the sort order (Ascending or Descending) and click OK. Repeat for any other sort variables. Open the worksheet and choose Tools>Sort from the WinNonlin menus or click the Sort button on the tool bar. 57 . To sort the data in a worksheet: 1. To unfreeze rows and/or columns: 1. Choose Window>Unfreeze Panes from the WinNonlin menus. Note: Exclusions are ignored when merging data sets. the sort variables are listed here. 3. To merge data sets: 1. Select a variable name and drag it to the Sort box. Open the two data sets to merge in one or two workbook windows. 4. or highlight the variable then type the letter S while holding down the SHIFT key. 7. These variables will appear in individual columns of the merged data set. Select Tools>Merge from the WinNonlin menus. 6. Be sure that the corresponding variables appear in the same order for each data set. An example is provided below. selected below. Drag the corresponding variable for data set 2 from its variables list to its sort variables list. The merged data set appears in a new workbook. 4. 58 . This applies in the case that one data set has more observations than the other for a given sort level. Select the appropriate worksheet for each data set under Sheet1 and Sheet2. one per data set. 3. on the right side of the dialog. 5. 8. Check Carry along data for similar sort keys if you wish to fill-in empty cells for given sort level with the last-available value. drag a variable on which to join records from the data set 1 variable collection to its sort variables list. For each data set.3 WinNonlin User’s Guide 2. the name from data set 1 will be used as the column header in the merged data. If a secondary sort variable is needed. It is divided into two areas. Select the workbook containing each data set from the pull-down menus for Dataset 1 and Dataset 2. Click Merge. repeat. If the names differ. The Merge dialog appears. For data set 1. They need not match in data sets 1 and 2. The merged data records will be matched on values of the sort variables. drag variables to be copied to the output workbook from the Variable Collection box to the Include Variables box. 35 0. the other two are Include Variables.76 Effect 0.1 0.1 0.98 1. Data sets to merge Data set 1 Subject A A A Time 0 5 10 Conc 2. Note that the two datasets must be selected before loading settings. Subject and Time are sort variables.35 0. Table 3-4. Carry along data for like sort levels The following example shows the effect of selecting to carry along data for like sort levels when merging data.95 Table 3-6. Merged data set without carry along for like sort levels Subject A A A A Time 0 5 5 10 0.27 1.98 1.Workbooks Data manipulation 3 The settings in this dialog can be saved and re-loaded for later use by way of the Load and Save buttons.95 59 .27 Effect 0.27 0. and data set 1 or 2 must contain the variables used in the file.35 0. Merged data set with carry along for like sort levels Subject A A A A Time 0 5 5 10 Conc 2.1 0.95 Table 3-5.98 1.76 Data set 2 Subj A A A Time 0 5 5 Effect 0.27 0.76 Conc 2. S). Select any desired options from those detailed below. 3. E. I. For example. Note: Excluded cells are treated as missing data in analyses. Select Data>Include>Selection from the WinNonlin menus Selection (Alt+D. from the WinNonlin menus (Alt+D. The data selections are made using the Exclude and Include commands on the Data menu.3 WinNonlin User’s Guide Data exclusion and inclusion Specific cells. all data are merged or transformed. Select Data>Exclude>Selection from the WinNonlin menus (Alt+D. E. Excluding and including a selected range of cells To exclude a selected range of cells: 1. Enter the value(s) to search for in the Find field. S). Select entire row(s) or column(s) by highlighting the row or column label(s). Select entire row(s) or column(s) by highlighting the row or column label(s). Select the cells using the mouse. Select excluded cells to be re-included using the mouse. rows and/or columns in a worksheet can be excluded from analyses. The selected region now is no longer marked in red. Excluding and including data based on criteria To exclude data: 1. 2. C). To re-include a selected range of cells: 1. if only a single subject must be refit from a dataset holding many subjects. 2. Select Data>Exclude>Criteria... 2. Exclusions are ignored in merges and transformations. 60 . you can exclude all but the desired subject and refit the data. The selected region is highlighted in red to indicate that it has been excluded. The Exclude dialog appears. check this option. – greater than or equal to the search string (>=). a single column. – less than or equal to the search string (<=).Workbooks Data manipulation 3 Search Area Allows searching over the entire data set. 4. – greater than the search string (>). To exclude each row containing the search value in the search area. – less than the search string (<). Tolerance A value not different from the search string by more than the tolerance will be treated as a match (applicable only when searching for numeric data). In the example below. Criteria Using this option search for a value: – equal to the search string (=). Case Sensitive Checking this box causes only occurrences with the exact combination of uppercase and lowercase letters to be found. or – not equal to the search string (<>). or the selected cell(s). 61 . Exclude Row Matching Criteria To exclude cells containing the search value. Click OK. the column Time is shown as a potential search area because Time was the active column when the dialog was opened. un-check Exclude Entire Row Matching Criteria. N).3 WinNonlin User’s Guide Re-including data To re-include specific data: 1. WinNonlin will use the first as baseline. then click OK. WinNonlin worksheets support the following transformation types: Natural log Log base 10 Square root Absolute value Change from baseline Percent change from baseline Ratio versus baseline X·Y X/Y X+Y X-Y X*Y ex 1/X Xn X*n X/n X+n X-n where X and Y are columns in the current worksheet and n is a number. which are carried into the new column. C). To remove exclusions from the current worksheet: 1. Select Data>Include>Criteria from the WinNonlin menus (Alt+D. for multiple transformations (see “Multiple transformation” on page 64). If a profile has more than one value at time 0. Other worksheets in the same workbook will not be affected. I. Select Data>Reset Selections (Alt+D. See the examples below. WinNonlin will use the value corresponding to Time = 0 as baseline for each profile. 2. Transformations A transformation creates a new column in the current worksheet. Resetting selections Resetting the data selections clears all exclusions in the active worksheet. In baseline transformations. Individual profiles are identified using one or more sort variables. All data in the source column are transformed regardless of exclusions. Select the inclusion criteria from those described for exclusions under “Data exclusion and inclusion” on page 60. 62 . or in the destination worksheet. WinNonlin will seek a baseline value (at time zero) for each unique combination of sort variable values. 3. or No to calculate change from a single baseline value for the data set. Select the desired transformation from the pull-down menu for Transform. Click Yes to choose sort variables. enter a value for n in the box labeled n. It will also sort the worksheet on the sort variables. 5. 4. Select the variables to be used from the pull-down menus for Column x. to compute change from baseline separately for each profile. as needed for the transformation type. 7. WinNonlin queries whether to specify sort variables. drag variables that identify individual profiles to the sort variables list. Other data may require more than one variable. Sort variables define unique profiles. Select the position for the new variable in the worksheet: – Adjacent places the new variable to the right of the 'Column x' variable. 2. 6.Workbooks Transformations To transform a column. 63 . If sort variables are included: In the Sort dialog. The example below creates a new column to the right of the BP column. The example below creates a new variable called Ln_Conc. If required. For baseline transformations. – Append places the new variable after the last variable in the data set. a single variable representing subject IDs suffices. Open the data and choose Tools>Transform from the WinNonlin menus or click the Transform tool bar button to open the Transform Data dialog. Time and Column y. such as Subject and Period. In most cases. creating a new column in the current worksheet: 3 1. 8. in the order presented here. Replace [New] with a column name for the new variable. and places it in a new column at the right of the existing worksheet columns. Supported operations include transformation of multiple columns. You may specify column headers and units for the new worksheet as part of the multi-transform settings.3 WinNonlin User’s Guide This example is from a crossover study. custom equations and rank transformations. The transform will seek a baseline value at time zero for each unique combination of values for Subject ID and Formulation. data inclusion/exclusion. 9. as well as search/replace/delete. 64 . The results appear in a new workbook. Click OK to complete the transformation and create the new variable. Multiple transformation The multiple transformation feature creates a copy of a source worksheet with specific data operations applied. check the box next to the transform(s). under Destination Workbook Name and Destination Sheet Name. Enter a name for the new workbook and worksheet. 5. 2. Open the data to transform in WinNonlin and choose Tools>Multi-Transform.Workbooks Transformations 3 All operations listed under Transforms in the Multi-Transform dialog will be applied to a copy of the source worksheet data. 6. Note: Use the Save button to capture all settings in XML format for later re-use. To set up transformations. 3. After adding transformations. Select the workbook and worksheet containing the data to transform under Source Workbook and Source Sheet in the Multi-Transform dialog.. 65 . To copy and transform data in an existing worksheet: 1. from the WinNonlin menus. To delete one or more transformation operations.. which will contain the results. use Add>Column Properties to set the column names and/or units for new columns. then click the Delete button. click the Add button and follow the instructions for the Transform action dialog. 4. Transform actions Action Description See: Add Filter Searches the source data for a specified value and “Multi-transformation filter” includes.“Column names and units” Properties tion worksheet. equation manager” on page 67 Column Sets the column headings and units for the destina. Table 3-7. include/exclude or search/remove operation to the output of a Multiple transformation: 66 .3 WinNonlin User’s Guide 7. Click OK to create the new workbook. “Multi-transformation using a formula or function. 8. excludes. removes or replaces it in the on page 66 destination sheet. page 62 Multi-transformation filter This dialog applies a search/replace. select one of the following actions in the Transform Action dialog and click OK to proceed. To change an operation. page 68 Transform Creates a new column in the destination sheet as a “Transformations” on transformation of a source worksheet column. on page 53 Rank Creates a new column in the destination sheet as a “Rank transformations” on rank transformation of a source worksheet column. Add Equation Creates a new column in the destination worksheet. Transform action To add an operation to a Multiple transformation. check the box next to that transformation (and no others) and click the Edit button. Select the appropriate operation(s) to find values equal to and/or less or greater than the search value. Select the desired action under Operation: – Replace: to find a value (or range) and replace it with another. See also “Data exclusion and inclusion” on page 60. For Replace operations. Select the Search Area: the whole source worksheet or a specific column. Multi-transformation equation manager This dialog creates a new column in the destination worksheet of a Multiple transformation using a formula or function of columns in the source work- 67 . 5. enter a value under Tolerance. Check Case Sensitive to search for strings that match not only letters but also capitalization. Click OK to add the operation to the Multiple transformation. – Remove: to delete rows containing specified values. enter the value with which to replace the search string under With.Workbooks Transformations 3 To filter data values to appear in the destination worksheet: 1. 3. 7. 4. Numeric values that are +/. – Include/Exclude: to (re-)include or exclude specified values (or rows containing them) in the destination worksheet.tolerance of the search value will be treated as matching the search value. Enter the value or string to search for in the topmost field. 6. To search for values within a range of a numeric search value. 2. using Add>Column Properties from the Multi-Transform dialog. Descending order ranks the highest value as 1 and increments by one for each smaller value. Order: Ascending order ranks the lowest value as 1.3 WinNonlin User’s Guide sheet. Enter the formula or function in the field at the top of the dialog. See: • • “Using formulas and functions” on page 44 for instructions on entering formulas and functions. Rank transformations This dialog creates a new column in the destination worksheet of a Multiple transformation as a rank listing of a column from the source worksheet. Group By: WinNonlin will generate a separate set of ranking values for each unique value of the Group By variable. Set the column name and units separately. and increments by one for each larger value. and “Worksheet Functions” on page 551 for functions and syntax. 68 . a substitution rule is defined by the list of possible nonnumeric representations for concentration and the values to be substituted for each. Table 3-8. plotting. Standard laboratory procedures may require that such data be reported as a character string rather than a numerical value. Common status code substitution rules Conditional substitution Use data for: Nonnumeric code BQL BQL BQL 0 "BQL LLOQ/2 0 " Unconditional substitution Before Tmax 0 After Tmax Missing First of consecutive after Tmax LLOQ/2 After first of consecutive after Tmax Missing PK analysis Summary statistics Listing of individual data Plotting of individual BQL data on log-scale Plotting of individual BQL data on linear scale The substitution rules are typically defined in standard operating procedures or method sheets for a given company or department. Some common substitution rules for BQL samples are presented in the table below. In general. The rules may become more complex when one wishes to make a distinction between concentrations that are below the quantification limit (BQL) and below the detection limit (BDL). Representing such a concentration as 0 (zero) is not always appropriate. and substitution rules vary between companies and departments. Possible substitutions for PK analysis Non-numeric code BQL Unconditional substitution Before Tmax After Tmax 0 Missing First of consecutive After first of conseafter Tmax cutive after Tmax LLOQ/2 Missing 69 . as doing so can introduce a statistical bias or misrepresentation in some analyses. tabulation.Workbooks Status code transformations 3 Status code transformations Data analysis. Different substitution rules are required depending on the intended use of the data. and summarization are problematic when the data contain concentrations that fall below the lower limit of quantification (LLOQ) of the assay. One possible substitution rule for PK analyses is presented below: Table 3-9. by copying over selected columns.g.3 WinNonlin User’s Guide Table 3-9. Possible substitutions for PK analysis (continued) Non-numeric code NS ISV NA … Unconditional substitution Missing Missing Missing … Before Tmax After Tmax First of consecutive After first of conseafter Tmax cutive after Tmax WinNonlin provides a tool for transforming observation values from nonnumeric status codes to number values for use in analyses and plots. Open the data set in WinNonlin.. in the WinNonlin menus to launch the Status Code Transformation Wizard. The Status Code Transformation Wizard creates a new workbook based on an existing one. BQL). if any. or identify an existing column containing the LLOQ for each observed concentration. Choose Tools>Status Codes (e. The wizard also stores information about the lower limit of quantification (LLOQ) for one or more sampling assays. 70 .. To select data to be transformed: 1. can be used in defining how status codes are transformed. The LLOQ. 2. It transforms values as specified by user-defined rules. The user may specify an LLOQ and the new column header into which it will be placed.. Numeric concentration data and the non-numeric concentration status codes may be the same or different columns of the data set. The new columns will use the units (if any) of the original concentration data. a fixed LLOQ value. or conditionally depending when the data point occurs relative to Tmax. 71 . Each rule set will generate a new column of transformed observation data. Concentration Variable: column containing the data to be transformed. 4. Drag and drop columns from the Variables list to assign the variables to be included and that to be transformed. Rule sets Up to three rule sets may be included in the Status code transformations. LLOQ Variable: (optional) column containing the lower limit of quantification (LLOQ) associated with each measurement in the Concentration Variable.Workbooks Status code transformations 3 At any point in this wizard. alternately. 3. Status Code Variable: (optional) column containing the non-numeric status codes associated with each datum in the Concentration Variable (above). • 5. A rule set includes a list of non-numeric codes and the value substitutions to be made when each code is encountered in the data. will determine transformation of those concentration values. as follows: • • • • • Sort Keys: each unique combination of sort key values defines one profile from which T-max will be calculated. Time Variable: column containing times for the data to be transformed. Make sure the appropriate data set is selected under Data Set. The substitutions may apply unconditionally across the data. and whether two or more consecutive observations of the same code occur. the settings and transformation rules may be loaded or saved from/to a file using the Load and Save buttons. Click Next to proceed to the Rule sets. set by checking Static Value and entering the value under LLOQ Value. Carry Alongs: variables to be copied into the output data set. stored locally. and click the Finish button to create the new workbook. Review the summary of the transformation settings.3 WinNonlin User’s Guide To set up the status code transformation rules: 1. To save a rule set to the PKS as a system object. and follow the instructions under “Creating a rule set” on page 73 to create a new rule set. 2. Use the Back button to make changes. 4. Click Next to proceed to the final dialog of the wizard. check PKS Load and click Save. Repeat to create up to three rule sets. Each rule set will generate a new data column in the output. The unidentified codes will be listed in an alert dialog. click Save. The wizard will detect the presence of unidentified non-numeric codes in the specified column. 72 . 5. 3. To save a rule set to an external file. Click the Add button under Rule Sets. For each code.) It is possible to load rule sets from a local file or the PKS. the applied rule(s) will be stored with the scenario Creating a rule set Status code transformations can include as many as three rule sets. Conditional Substitutions: if the transformation depends on the status code’s position relative to Tmax and/or whether the code is repeated in consecutive observations. click Load. Unconditional Substitution: if the transformation is the same regardless of whether the status code occurs before or after Tmax. or is repeated. stored locally. then enter the value to which the status code will be transformed here. (See “Status code transformations” on page 69. Enter each status code to be transformed in a separate row. check PKS Load and click Load. To load a rule set from the PKS. To define a rule set: 1. 73 . enter the following: a. 3. To import a rule set from an external file. This name will be the default name for the column generated by this rule. b.Workbooks Status code transformations 3 Note: If code-transformed data are saved to the PKS as a scenario. under NonNumeric Code. Enter a name under Rule Set Name. Enter any descriptive text under Rule Set Description (optional). then leave Unconditional Substitution blank and enter all four of the following: – Before Tmax: value to assign if the status code occurs before Tmax. 2. One row or zero rows may be set to Yes. Click OK when the rule set is complete to return to the Status Code Transformation wizard. The LLOQ value is set in the first dialog of the Status Code Transformation Wizard. this value will be assigned to the first of those contiguous observations. c. 4. The values entered may include numbers. operators. Syntax for the rules: The status rules are not case-sensitive. Yes indicates that when numeric values less than the LLOQ are encountered. above. – After First of Consecutive after Tmax: if the status code appears in two or more contiguous observations. and the variable LLOQ. – First of consecutive after Tmax: if the status code appears in two or more contiguous observations. 74 . if a LLOQ value or data column was assigned (see “Status code transformations” on page 69). after Tmax. after Tmax.3 WinNonlin User’s Guide – After Tmax: value to assign if an isolated (not more than one consecutive) status code occurs after Tmax. this value will be assigned to all but the first of those contiguous observations. Use when < LLOQ: click to toggle between Yes and No. those values will be transformed as though they contained the status code in that row. See “Rule sets” on page 71. etc.. (optional) run a script and (optional) return script output to WinNonlin. “S-PLUS export” on page 86: export data. The Enterprise edition of WinNonlin can. in addition. Oracle. and save analysis output back to a table in that database. as follows. and can re-import S-PLUS and Sigma Plot script output. (optional) run a script and (optional) return script output to WinNonlin. • • • “ODBC import” on page 87 “ODBC export” on page 102 “The Enterprise Software Development Kit” on page 107 75 . graphics and tables. “SigmaPlot export” on page 85: export data. “SAS Transport file export” on page 78: save data in SAS-ready format. “Supported file formats. (See Table 2-1.g. WinNonlin Enterprise can read rows of PK data from an ODBC-compliant database. See the following for details. SAS.Chapter 4 Data import and export Sharing data with other applications WinNonlin opens and saves a variety of data. “NONMEM export” on page 80: export data and dosing to NONMEMcompatible files. exchange data directly with Open Database Connectivity (ODBC) compliant databases (e. • • • • • “Microsoft Word export” on page 76: automatically transfer selected text.) WinNonlin also exports data to several software packages. text and graphics file types for convenient interchange with other applications. charts and worksheets into Word text.) For example.” on page 14. WinNonlin Enterprise includes a Software Development Kit (SDK) for building custom ODBC query tools. if any. they cannot be edited in Word. After completing model setup. uncheck Select all objects in the Word Export dialog as shown below. To export modeling output to Word automatically: 1. 2. If Word is not open. management and security. Microsoft Word export WinNonlin can export any or all open windows to a Microsoft Word document. Open the objects and choose File>Word Export from the WinNonlin menus.4 WinNonlin User’s Guide WinNonlin’s Enterprise edition integrates with the Pharsight Knowledgebase Server to provide secure data storage. or append the data to the end of the open document. chart or text window to export. To export selected objects to a new Word document: 1. charts cannot be edited after export. 76 . Worksheets are exported as Word tables. Text windows become text in Word’s “normal” paragraph style. 3. Apply any chart formatting needed. a useful feature for automated analyses. click the Start Modeling and Export to Word tool bar button. WinNonlin will launch the analysis. In WinNonlin. WinNonlin will launch it and export the active window to a new document. To select specific objects for export. To export the active object to a new or existing Word document: 1. Charts and model objects export as Windows bitmaps. thus. Click the Export to Word tool bar button. open the worksheet. open that document in Word. 2. It can also automatically export modeling or NCA output. To append data to an existing document. create the selected WinNonlin output and export all output to Word. See “Using the Pharsight Knowledgebase Server” on page 445. below the data table. 77 . Word export document options The following options apply to Microsoft Word export using File>Word Export in the WNL menus. Add source line to objects: check this option to include text describing the source file and object name to each object exported as follows. Workbooks: adds the worksheet name and absolute path. • • • Charts: adds the chart name and absolute path. Text: adds a heading above the text that reads. 5.Data import and export Microsoft Word export 4 3. Orientation: Portrait (long side of the page is vertical) or Landscape (long side of the page is horizontal). if unsaved) in square brackets. “Text -” followed by the file path and name (or window name. Use the Options button to set options for page layout. and export the selected objects to a new . Checking a window or window type will export all objects within that window or window type. Click the Export button. WinNonlin will launch Word. format.DOC file. 6. below the chart picture. if needed. 4. Check the objects to export. as detailed under “Word export document options”. file name and most-recent save date for the chart. Click the + markers to expand the lists of windows and objects as needed. file name and save date for the data. headings and object numbering. 2. Word export workbook options Preserve formatting for tables: includes formatting for workbooks created in the WinNonlin Table Wizard. Add page break: forces a page break before and after each chart. The Save File dialog appears. 3.xpt. Each log chart appears below the original linear chart. To export data to SAS (via SAS Transport files): 1. Name the file and click Save. Start figure numbers at N: adds figure numbers to charts. 78 . With the desired worksheet active in WinNonlin.stx) in the Save as Type list. Start table numbers at N: adds table numbers to exported worksheets. Select SAS Transport (*. with page breaks between sets of charts. Add log charts: creates and exports a chart with a logarithmic Y-axis for each linear scatter chart. Each worksheet is one table in the Word export. SAS Transport file export Any worksheet can be saved to a SAS Transport file. Check this option and enter a number for the first chart to include “Figure N” above each chart. Because SAS field names are more restrictive than WinNonlin column names. This option can slow the export significantly. select Save As from the File menu. *. Check this option and enter a number for the first table to include the text “Table N” above each table. WinNonlin will request new names for any that would be invalid in SAS.4 WinNonlin User’s Guide Word export chart options Multiple charts per page: Check this box and select the layout below to put more than one chart on a page. Chart format: export charts as Windows metafiles or bitmaps. and are limited to alphanumeric characters and the underscore. or reload previously-saved settings. Scroll right and click on a cell under Precision Type to toggle between significant figures (Sig. 7. If the Rename Columns dialog appears. 79 .MAP). Enter the desired number of significant figures or decimal places under Precision. Optionally. You may also edit the label names and precision for exported values. 5. edit data set name. enter new names under Variable Name for any names that appear in red text. 6. Digits) and decimal places. Use the Load and Save buttons to save these settings to an ASCII text mappings file (*. Click OK to create the SAS Transport file. The new names must be no longer than 8 characters.Data import and export SAS Transport file export 4 4. Data file structure The NONMEM data file contains two sections: the headers and data. NONMEM places a four-character restriction on names. e. the value in DV represents DEPQ.” etc. Thus. the header might contain: # Variable GEND # male = 1 # female = 2 3. The DVID column disambiguates between dependent variables..” “B. This will appear as a comment like that below: # ID TIME WT GEND AGE BP DV MDV DVID AMT RATE EVID 2. whenever DVID is 2. Comments in the header can indicate how WinNonlin variable names map to those used in NONMEM. Abbreviations.DAT). All NONMEM data are numeric. 4. discrete values stored as strings. The following four items can appear in the header of a NONMEM data file exported from WinNonlin: 1. or “male” and “female” must be encoded as numeric values for use in NONMEM. Typically.g. Names of columns.4 WinNonlin User’s Guide NONMEM export WinNonlin provides a NONMEM Export function to save observation and (optional) dose data in a NONMEM-compatible data file (*. The header can include the DVID column mappings. For example. NONMEM export can also generate a NONMEM control stream containing an outline of required commands or commands copied from a user-specified template. DVID. The exported DV column can hold one or more dependent variable(s). Encoding. this encoding is explained in the header. The first line in the header contains the names of the columns in the data file. 80 . For example: # Variables in DV column # DEPC = 1 # DEPQ = 2 Thus. “A. -2 means that a zero order bolus for the rate is calculated. For infusions. for simplicity. Description Subject identifier. If more than one dependent variable (DV item) is exported. then MDV can be skipped. this column will equal 1 for the first.) Provided only when more than one DV item is present. User may include as many as are available. 2 for the second. -1 indicates a zero-order bolus for which NONMEM calculates the duration. and 3 for the third.Data import and export NONMEM export 4 NONMEM data may include the following columns. RATE = 0 for bolus dosing. A positive value indicates an infusion. Frequently provided. but. The amount value of the dosing variable. Nominal or actual time. There will be one column for each independent variable. (If DV is never missing. the dose rate. Usually required. 81 . Table 4-1. Thus. Always provided. Indicates the compartment into which a dose was administered or from which DV was observed. 9 10 CMT EVID Usually provided. The status of the dependent variable: 0: The row includes a valid value of DV 1: The row does not include an observation of DV or the value is missing. The dependent variable’s values. Required. this column disambiguates. The event type: 0: An observation event (for DV) 1: A dosing event 2: An other-type event 3: A reset event 4: A reset-and-dose event 6 DVID 7 8 AMT RATE Required for PK. one should always provide MDV. The name DV is a fixed name for NONMEM. NONMEM columns Position Name 1 2 3 4 5 ID TIME independent variables DV MDV Presence Always provided. if three DV items are chosen. Optional. The RATE column shows a dot when no value is administered. The value of the independent variables. This form export should always include it. Choose File>NONMEM Export. including the full (absolute) path. header information. Tab name Files Dependent variables Other data items Dose information Occasions Description Export file path and name. To be available for export. 5. 3. Dose data for NONMEM export is drawn from the WinNonlin model. as an additional means to explain variability 6. Specify the dosing (if any). Set up the model variables as detailed under “Modeling” on page 193. If a control stream is output with the data. PK/PD link or indirect response model.4 WinNonlin User’s Guide Exporting to NONMEM To export data to NONMEM: 1. Comment character: Enter a character to precede and identify comments in the output file. The Export to NONMEM dialog appears. 2. comment identifier Dependent variables and dependent variable ID column Independent variables Dosing variables and compartment to dose into (optional) Identifiers for within-subject occasions to be considered separately. 82 .DAT) to be created. 4. dose data must be either entered the Model Properties dialog (Dosing Regimen tab) or available on the Dosing worksheet of PKS study data (WinNonlin Enterprise only). then this string will appear in the $PROB statement. this string will appear in the $DATA statement. Load a data set and a compiled compartmental PK. Set the export data and settings using the following tabs. This character will precede the header rows. When all settings are complete. If a control stream is output with the data. click OK to create the NONMEM file. It is useful to enter descriptive comments here to make it clear what is in the data file. Files Problem (optional): This string becomes the heading for the NONMEM export file. The default is: #. and appears in the $INPUT COMMENT= command in the control stream. or use the Browse button to set it. Data File: Enter the name of the output data file (*. Generate skeleton control stream generates a skeletal control stream. If a single dependent variable is specified. consisting of the $PROB. All dependent variables must use numeric values. Inhibits the creation of a new control stream. and $DATA commands only.Data import and export NONMEM export 4 Control stream: Select the appropriate option: • • Do not generate control stream: available for users who wish to write their own control stream. 83 . If the dependent variable is associated with a particular compartment of a PK model. usually less than 10. and $DATA commands at the beginning. for the file to create (above) and the template file (below) or use the adjacent Browse button to locate each. CMT must be a non-negative integer. Additional header comments will indicate which column (NONMEM abbreviation) is associated with each DVID value. DVID: If multiple dependent variables are included. or use the Browse button to set it. all will appear stacked in a single column identified in the headers as DV. enter the compartment number here. • Dependent variables Specify one or more dependent variable(s) for the NONMEM export by checking the box under Select? for the appropriate columns. $INPUT. Enter the names. including full paths. the header comments in the NONMEM data file will indicate “DV=COL” where “COL” is the NONMEM abbreviation entered in this dialog box. time and other independent variables. Specify the file name. including the full path. $INPUT. WinNonlin will create a dependent variable ID (DVID) column to disambiguate them. and to set their names in the exported data. CMT stands for compartment. Other data items Use this tab in the NONMEM export dialog to identify the subject ID. Enter a unique positive integer value for DVID for each dependent variable. Generate control stream from template copies a user-supplied control file template and inserts the appropriate $PROB. If more than one dependent variable is included. Dose information Dose data for NONMEM export is drawn from the WinNonlin model. TIME for the time variable. These options include: • • • • • • Derive RATE from the data. b. In this case. Treat as bolus. 2. Edit the values under NONMEM Abbreviation as necessary to read: a. Associate dose with a compartment: To dose to a specific PK compartment. Have NONMEM calculate duration (RATE=–1). Each four-character abbreviation must be unique and cannot be a NONMEM reserved name. 84 . check this box then enter the compartment number under CMT. Case will be preserved as entered. one variable representing time (required) c. and c. b. Have NONMEM calculate rate (RATE=–2). this allows estimation of the rate parameter. one variable representing subject identifiers (required). For infusions.4 WinNonlin User’s Guide To identify independent variables for export: 1. this allows estimation of the duration parameter. numerals and/or underscore characters for each other independent variable. a unique name up to four letters. Check the box under Select? for: a. no dose data are exported. any number of independent variables (optional). For infusions. The Dose Information tab of the NONMEM export sets options for exporting this dose information. Include no dose information. All doses will be given a RATE of 0. and must be specified in either the Model Properties dialog (Dosing Regimen tab) or via the Dosing worksheet included with PKS study data (WinNonlin Enterprise only). ID for the subject identifier. CMT must be a non-negative integer less than 10. See the NONMEM NM–TRAN documentation to determine which numbers to enter. choose its text window (or file path) from the list under Optional Script.Data import and export SigmaPlot export 4 Occasions Including occasional variation in a model can sometimes explain additional variability in the data. SigmaPlot export WinNonlin provides options to export worksheet data to SigmaPlot. from the WinNonlin menus. Open the data to export in a WinNonlin workbook. The Occasions tab in the NONMEM export dialog exports a separate column to differentiate occasions (e. separate profiles within subjects). To include and run a script. from earliest to latest (do not include time 0). At the beginning of an occasion. NONMEM will reset all compartment concentrations to 0. if any. into WinNonlin: 85 . and select a method for creating occasions: • • list the time values at which new occasions start. All columns in the worksheet will be exported to a SigmaPlot notebook.g. Choose File>Sigma Plot Export. or indicate how long each occasion lasts. 7. 3. Enter a name for the new notebook under Note Book Name. 4. WinNonlin can automatically import script output. Note: SigmaPlot export requires installation of SigmaPlot 8 on the same machine as WinNonlin. To export data to SigmaPlot and (optional) run a script: 1. open or create it in a WinNonlin text window. Use occasions: check this box to generate an occasion identifier. if any.. 5. 6. To include and run a script. and (optionally) when the first occasion begins. To automatically load script output.. 2. with or without a script to run. The script must be open in a WinNonlin text window. Select the data to export under Workbook and Sheet.. 4 WinNonlin User’s Guide a. if any. Choose File>S-PLUS Export.. do not perform any actions in WinNonlin until the script run completes and its output appears in WinNonlin. Note: S-PLUS export requires that S-PLUS version 6.” button to select the folder. All columns in the worksheet will be exported to an S-PLUS data frame. 4. Enter the full path to the SigmaPlot output folder under Output or click the “. To include and run a script. 86 .. 8. with or without a script to run. WinNonlin can automatically import script output. If you chose to return script output to WinNonlin.. 3.. Check Wait for completion. b. S-PLUS export WinNonlin provides options to export worksheet data to S-PLUS. open or create it in a WinNonlin text window. from the WinNonlin menus.1 is installed on the same machine as WinNonlin. Select the data to export under Workbook and Sheet. Open the data to export in a WinNonlin workbook. 2. Click OK to export the data and execute the script. To export data to S-PLUS and (optional) run a script: 1. IMP). including Watson version 6. 7. Check Wait for completion. for later re-use. into WinNonlin: a. Click OK to export the data and execute the script.. The connection and query settings can be saved to an import file (*.0 DMLIMS. then defining a query to draw specific fields and records from it. To automatically load script output.Data import and export ODBC import 4 5. Enter the full path to the output folder under Output or click the “. An import involves establishing a connection to the database. the ODBC import command provides the option to update the data to reflect the current database content. or import. This string can be cre87 . Enter a name for the new data frame under Data Frame Name. Instructions are provided under: • • “Creating an import” below “Loading an existing import” on page 96 WinNonlin also provides a Custom Query Builder for creation of dynamic link library (DLL) files that can access additional data source types. See also “ODBC export” on page 102.. choose its text window (or file path) from the list under Optional Script. See “The custom query builder” on page 97. b. ODBC import Use ODBC import to load data from an ODBC-compliant database directly to a WinNonlin worksheet.IMP) contains all information required to draw data from a remote data source. To include and run a script. Database connection: Both ODBC import and export use a connection string that defines the database type and location. if any. See “Refreshing an imported data set” on page 97 for instructions. 6. do not perform any actions in WinNonlin until the script run completes and its output appears in WinNonlin. The script must be open in a WinNonlin text window. It includes two parts: 1. Once ODBC-imported data are loaded in WinNonlin. If you chose to return script output to WinNonlin. Creating an import An ODBC import specification (optionally saved as *.” button to select the folder. if any. 8. “Creating a new connection” on page 90 b. Choose File>ODBC import from the WinNonlin menus. Proceed to the Final settings dialog. Create or load a connection string as detailed under: a. 88 . Create or load a query as detailed under: a. Final settings The Final Settings dialog summarizes the ODBC import. each can be loaded separately. Select Create a new import in the ODBC Import wizard and click Next. To create and save (optional) a new import: 1. field(s) and records to import. Its top section displays the connection string. an import requires a query to set which data table. 2. WinNonlin translates the query into Structured Query Language (SQL). The lower section determines data formatting in WinNonlin. 2. One connection string can be re-used with any number of different queries. Query: Once the database connection is established. “Loading a saved query” on page 96 5. which can be edited before executing the import. The middle section contains the SQL string for the query. or loaded from a previously-saved import or export as detailed under “Connection” on page 90. “Loading a saved connection” on page 92 4. “Building a query” on page 93 b. 3. The query may be created in the ODBC Import wizard or loaded from previouslysaved import as described under “Queries” on page 92.4 WinNonlin User’s Guide ated by pointing to the database type and file. 3. WinNonlin will then execute the import. so a password saved within is not secure. • • • • Set Column Widths adjusts the destination worksheet's columns to match the widths of the fields in the database. Save Password: This option includes the database user name and password in the import file with the connection string. If Check Number of Records? was checked. CAUTION: The import file is not encrypted. Choose settings in this dialog (below). This allows scripting to use ODBC connectivity without end user intervention (see “FileType” on page 593). 89 . and dates. Click Finish to import the selected data to a WinNonlin workbook. Else. currency. and edit the SQL if desired.Data import and export ODBC import 4 1. click No. 4. Check Number of Records? provides a count of filtered records prior to the import. Set Column Names adjusts the destination worksheet to have the same column names as the source fields.IMP). and the option to return and edit the query if needed. 2. Set Column Formats applies the appropriate number formats to columns containing times. a dialog appears with the number of records returned by the filter. To save the import (*. click Yes and save to the appropriate directory. Click OK to proceed. In either case. The File Data Source tab (below) lists all file Data Source Names (DSN’s) and subdirectories on the system. Each cell can hold up to 16KB of text. Select Create a new connection and click Next.4 WinNonlin User’s Guide WinNonlin can import up to 65. “Loading a saved connection” on page 92 extracts the connection string from a previously-saved import or export file. defining the database type and location. The ODBC drivers might impose additional limits. to be used with a new or saved query or export. 3. 90 . 4. Choose File>ODBC Import (or Export) from the WinNonlin menus. Connection The first step in an ODBC import or ODBC export is to establish a connection to a specific database. These are file-based data sources that can be shared among all users who have the same ODBC drivers installed. This connection string may be saved as part of an import or export settings file for re-use. Select a data source from one of the two tabs in the Select Data Source dialog. This establishes a link to the database. The import wizard is shown below.000 rows into one worksheet. Creating a new connection To create a new connection string: 1. The data sources need not be dedicated to a single user or local to a specific computer. Select Create a new Import (or Export) in the ODBC Import (or Export) wizard. 2. and click Next. “Creating a new connection” on page 90 creates a new connection string. Click OK to proceed with the import or export. 6.Data import and export ODBC import 4 New data sources can be set up using the New button. enter the appropriate log in ID and password for the data source. See the Microsoft Windows documentation for details. Select a data source and click OK. 5. 91 . which provides the majority of the connection string information. The Machine Data Source tab (below) uses information stored in the machine registry. If a log in dialog appears. These are set up by the Database Administrator or Information Technology personnel. Select Load an existing connection and click Next. it can be loaded for use with different import queries or export definitions. Select the import file (*.IMP) or export definition file (*.EXP) containing the connection string for the desired database and click Open. See “Queries” on page 92 or “Export definition” on page 103. Proceed with the import or export. Choose File>ODBC Import or File>ODBC Export from the menus. Queries Once a database connection string is established (see “Connection” on page 90). An Open dialog appears. • • “Building a query” on page 93 “Loading a saved query” on page 96 92 . field(s) and records to be imported. Select Create a new Import (or Export) in the first dialog and click Next. Loading a saved connection Once a connection string is saved in an import (*. ODBC import requires a query specifying the data table. 2. 5.4 WinNonlin User’s Guide The next step is to set up a query to pull the desired data from the database. as detailed under “Queries” on page 92 for imports or “Export definition” on page 103 for export. To load an existing database connection: 1. 4.IMP) or export (*.EXP) file. 3. as detailed in the following sections. or an export definition to map WinNonlin data columns to database fields. Select Build a Query and click Next to proceed to “Schema filter” below. Choose File>ODBC Import from the WinNonlin menus. If the Schema Filter dialog appears while building a query (“Building a query” on page 93). select the schema to use in filtering the list of data tables then click OK to proceed to “Query builder” below. Schema filter The WinNonlin Query Builder will access the database. To create a new query: 1. It provides a means to filter the available database tables by schema. It can also include filters to extract specific records from the table. as detailed under “Creating a new connection” on page 90 and “Loading a saved connection” on page 92. a metadata organizational tool. If the appropriate ODBC drivers are available. These schema were created by your database administrators. 93 . the Schema Filter dialog appears. Load a new or saved connection string. 3. and load a list of the available tables.Data import and export ODBC import 4 Building a query A query notes the database table and fields (optional) to import to WinNonlin from an ODBC-compliant database (see “ODBC import” on page 87). 2. To filter the records. select a table from which to import data. To import all the records from those fields. 4. 3. Select one or more fields to import. click the Done button and proceed to step 7 below. Click Next to select specific fields to be imported. 94 . click Next. When the Source Tables list appears. 2.4 WinNonlin User’s Guide Query builder To select source tables and fields while building a query: 1. or click Done to load all fields in the table and proceed to step 7 below. Begin the query setup as described under “Building a query” on page 93. d. select the And or Or radio button and return to step a above. 95 . b. Select a Field Name to filter on from that list box. click the List Distinct Values button to select from among all existing values for that field. The Import: Final Settings dialog appears.Data import and export ODBC import 4 5. e. Click Done when the filters are complete. To create the filter: a. If more than one criterion will be applied. 7. Filters are applied in the order created. Follow the instructions under “Final settings” on page 88 to complete the import. Choose an operator from that field's list box. c. Click the Add button to set the filter. Enter a value in the Field Value field. 6. The ODBC Import: Final Settings dialog appears with the connection string and query loaded. 3. respectively.IMP) file and click Open. as detailed under “Creating a new connection” on page 90 and “Loading a saved connection” on page 92. An Open dialog appears. The ODBC Import: Final Settings dialog appears with all settings for that import loaded. It also provides a convenient template for a new ODBC import. To load an existing import: 1. 2. In the ODBC Import dialog select Load an existing Import and click Next. Loading an existing import An import setting file (*. any saved query can be loaded so long as it uses tables. See “Final settings” on page 88 to complete the import and load the data into a WinNonlin workbook. An Open dialog appears. 3.4 WinNonlin User’s Guide Loading a saved query Once a connection string (new or existing) has been established as part of Creating an import. 2. The default file type is *.IMP. 96 . Choose File>ODBC Import from the WinNonlin menus. fields and record types in the database being connected to. In the ODBC Import: Query dialog select Load a query and click Next. Use File>ODBC Import from the WinNonlin menus to connect to a new or previously-saved database. Select the appropriate import (*.IMP) contains the connection string and query to quickly re-load data from an ODBC database into WinNonlin. Select the file with the desired connection string and click Open. To load an existing query: 1. Even if the workbook has been saved locally. click the Finish button. To use the custom query builder with Watson v. Select File>ODBC Import from the menus. 97 . those data can be refreshed to reflect the current the database source by choosing the Import ODBC command from the File menu.0 DMLIMS.Data import and export ODBC import 4 4. including Watson version 6. The custom query builder The Custom Query Builder provides a means to build custom dynamic link library (DLL) files that access additional data source types. Further. it can be re-opened and refreshed following this same procedure.0: 1. and using the Back button. Refreshing an imported data set When data from an ODBC import are open in WinNonlin. 5. 6. When the settings are correct. WinNonlin Enterprise provides templates for creating the DLL files to access custom data sources. Makes changes as needed in the Import: Final settings dialog. Click Yes to refresh the data in the worksheet from the database. The data from the database are then loaded into a WinNonlin worksheet. e. To add resources to the Query Builder. 2. In the Configuration dialog. click Add.4 WinNonlin User’s Guide 2. Select a builder The Select a Builder dialog appears as part of the Custom Query Builder (see “The custom query builder” on page 97) to set the data resources to access. If the Custom Query Builder has not been used before. select that item and click Edit. this dialog may be empty. 98 . enter the name of the Watson LIMS tables owner. To configure the Watson DMLIMS custom resource to ensure that data are imported correctly. then double-click the appropriate DLL file.g..DLL. WATSON_DMLIMS. 1. 3. Enter the full directory path for the Watson templates or use the Browse button to locate the directory on your local machine or on a network server. Select Use Custom Query Builder and click Next to proceed to “Select a builder” below. 4. A log on dialog appears. password. Note: The user name. The import is complete. 99 .Data import and export ODBC import 4 5.) in order to find a particular listing more quickly. Study selection The Study Selection dialog lists the resources available in the data source for a custom query (see “The custom query builder” on page 97). password. and click OK to proceed to the Study selection dialog. Enter user name. Select a file (file type *. and alias for any custom database system (such as the Watson DMLINS) are set by the database administrator or IT personnel. click Yes. Select a study by clicking on the row.. Use the Order Study By field to sort any category (e. a. Request this information from that person or department. Click OK to close the Configuration dialog. 7. etc. study number/ID.WTS for Watson LIMS) and click Open. click No and proceed with the following steps.g. study director. Highlight Watson DMLIMS (or another custom query builder) in the Select a Builder dialog and click the OK button. To load a saved query. 6. 8. and database alias. If no query has been saved for this builder. 1. b. The Open File dialog appears and provides the opportunity to select an existing query. A dialog asks whether to load a query. The query is read and the file data is loaded into a worksheet. drag each variable to the Selected Variables field. contact your IT department. containing the list of variables available in the selected database. For a subset. Click Next. The Import Data Variables dialog appears. 3. If all available variables are desired. click the Select All button. 2. 100 .4 WinNonlin User’s Guide Note: If no projects are listed in this dialog. A particular combination of variables can be saved as a template. operator. A filter is defined by specifying one field name. Then enter a name for the template. To create a new template drag the variables for the template to the Selected Variables field and click the New button. 101 . Click Done if all the data in the selected fields are to be imported. Note: The fields are imported in the order shown in the Selected Variables list. and field value.Data import and export ODBC import 4 4. The Filter dialog appears. The data can also be filtered further by clicking Next. 5. Alternately. To set a filter: Select the field to filter on from the Field box. Age > 12 and Gender = Male). ODBC export WinNonlin Enterprise edition can export selected worksheet columns to specific fields within a table in any ODBC-compliant database. Export definition: Once the database connection is established. See: • • “Creating an export” below “Loading a saved export” on page 107 Creating an export An ODBC export specification (optionally saved as *. click the List Distinct Values button and select a value. an export requires a mapping of columns from the WinNonlin worksheet to fields in the target database table. Both the connection string and export definition can be saved to an export file (*.g. Like ODBC import.EXP) for later re-use. It includes: 1. If Check # of rows returned was checked. WinNonlin will report a count of records returned. Database connection: Both ODBC import and export can use the same connection string to define the database type and location. This string is created in the ODBC Export wizard or loaded from a previously-saved import or export as detailed under “Connection” on page 90.EXP) contains all information required to save data directly to an external database. Click Done to load the data into a WinNonlin worksheet. 2. One connection string can be re-used with any number of different export definitions. Click the Add button to set the filter.4 WinNonlin User’s Guide 6. the export requires a connection string defining the database type and location. 102 . 7.. each can be loaded separately. Use the And/Or operators to create compound filters (e. It also requires an export definition to map columns in the source worksheet to fields in the target database table. Either enter a value in the Value field or select from all specified values of that field. This export definition may be created in the ODBC Export wizard or loaded from a previously-saved export as shown under “Export definition” on page 103. Choose the operator below. EXP). 2. Create or load an export definition as detailed under: a. The settings can be saved to an export definition file (*. Select Create a new export in the ODBC Export wizard and click Next. to what tables and fields in the selected database. Choose File>ODBC export from the WinNonlin menus. 6. After the export completes.Data import and export ODBC export To create and save (optional) a new export: 4 1. “Loading a saved export definition” on page 106 5. “Creating a new connection” on page 90 b. “Building an export definition” on page 104 b. • “Building an export definition” on page 104 specifies which data are to be exported. • 103 . Make any adjustments needed in the Field selection dialog and click OK to complete the export. Create or load a connection string as detailed under: a. “Loading a saved connection” on page 92 4. “Loading a saved export definition” on page 106: One connection string can use any number of different export definitions. Click Yes to create a new export (*. Export definition An export definition maps WinNonlin worksheet columns to fields in one table of the target database for an ODBC export. WinNonlin will ask whether to save the export settings. Else. 3.EXP) file. click No. As with the ODBC import. 4. The ODBC Export dialog appears. 2. To create a new ODBC export definition: 1. Select Build a Definition and click Next. if any have been set up by your IT personnel. “Creating a new connection” on page 90 b. 104 . The Schema Filter dialog displays available schemas. the schemas to use in selecting the data variables can be set. to connect to the target database. Select Create a new Export and click Next.4 WinNonlin User’s Guide Building an export definition Export definitions require that a connection string be loaded first. “Loading a saved connection” on page 92 After a connection string is set. 3. the ODBC Export: Definition dialog appears. Open the set to export in WinNonlin and choose File>ODBC Export from the menus. Create or load a connection string as detailed under: a. check the Fixed check box for that variable and enter the value under the field name. Drag the variables from the Variables list to the appropriate database fields under Field Definitions. 3. 2. Field selection This dialog determines the core of an Export definition. 4. If this is a new export. Using schema filters can speed up export operations using saved export files. To export one constant value for all records in a given field. This file contains all information needed to reproduce the same export. This method can be used to differentiate one export from another. Select the database table to receive the exported data under Table Name. WinNonlin will ask whether to save the connection string and export definition to a new export file (*. The data type and field size are set by your Database Administrator. If the Schema Filter dialog appears. it sets which data columns are exported to which database tables and fields. select one item and click OK. excluding any constants in fixed fields. Proceed to the Field selection dialog to complete the export. Click OK to export the data to the database. 1. 105 .EXP). 6.Data import and export ODBC export 4 5. use the Schema Filter to narrow the set of available data tables (optional) and click OK. Select Load a Definition and click Next. 5.EXP) can be re-used with any database connection. 2. 3. To load an export definition: 1. if necessary. 8. If this is a new export. Click Yes to create a new export (*. “Loading a saved connection” on page 92 Once the connection string has been loaded. 6. Create or load a connection string as detailed under: a. WinNonlin will ask whether to save the export settings. In the Open dialog. 9.EXP) file. click No. Else. Make any adjustments needed in the Field selection dialog and click OK to complete the export. 106 . The ODBC Export dialog appears. Open the set to export in WinNonlin and choose File>ODBC Export from the menus. If the Schema Filter dialog appears. assuming the field and table names in the export file exactly match a table in the selected database. Log on to the data source.EXP) and click the Open button. “Creating a new connection” on page 90 b.4 WinNonlin User’s Guide Loading a saved export definition The field mappings in any ODBC export definition file (*. select the export file (*. the ODBC Export: Definition dialog appears. 4. Select Create a new Export and click Next. 7. Data import and export The Enterprise Software Development Kit 4 Loading a saved export Once an ODBC export has been created and saved. To load both a connection string (pointer to a specific database) and export definition (mappings). Choose File>ODBC Export from the WinNonlin menus. The Enterprise Software Development Kit The Enterprise edition of WinNonlin includes a Software Development Kit (SDK) for Developers.EXP). the export settings. 4. including the connection string and/or the export definition. Make any needed edits to the export settings. select Load an Export and click Next. 107 . and click OK to export the data. To load an existing export: 1. Log on to the data source if needed. An Open dialog appears. 5. System Administrators and Database Administrators who wish to create custom query builders for ODBC connectivity. 3. This feature extracts both the connection string and the export definition from a saved export file (*. The ODBC Export: Export dialog appears.EXP) and click Open. can be reloaded for reuse with any data set containing the column names used in the export file. Select the desired export definition file (*. 2. The Field Definitions are loaded automatically in the Field Selection dialog. Network=DBMSSOCN." "Select Subject.Conc from bguide1a where (Subject = '2 ')" "0" "0" "1" "1" END FILE This would allow WinNonlin Enterprise to define the connect string. 108 . the creation of an import file for loading should leave the connect string blank. 6 lines contained in quotes. At query completion.Conc from bguide1a where (Subject = '2 ')" "0" "0" "1" "1" END FILE Usage Generally. potentially. since the connect string is dependent on the drivers installed on the machine. So.SERVER=thunder. this file is loaded. contain the following: BEGIN FILE "" "Select Subject. an import file distributed by an IT department would. 1 2 3 4 5 6 BEGIN FILE (Note 6 lines) Connect string Query Set Column Formats (0 or 1:True or False) Set Column Names (0 or 1:True or False) Set Column Widths (0 or 1:True or False) Set Max RC (Always set to 1) END FILE Example: BEGIN FILE "DRIVER=SQL Server.Time1.4 WinNonlin User’s Guide Import file format The Import file is an ASCII file. and let the application define the connect string. When the end-user creates or loads a query.UID=dan.Time1.WSID=THUNDER.APP=Visual Basic. Each column-field definition is represented by three (3) consecutive lines. Export file format An Export file is an ASCII file that contains 2+3*n lines. the table field that the column is exported to.Data import and export The Enterprise Software Development Kit 4 the import could be saved to store the connect string generated by WinNonlin Enterprise. The new code will not compile unless all interfaces in the DLL are implemented. In some cases. the developer can reference it and implement its interface. it may be necessary to code to a specific database API set to optimize performance when determining access rights and database structure.. When this DLL is registered on a machine. 109 .DLL] is essentially a stubbed out API.DLL The customized query building feature allows site-specific and database-specific customization of query building. Current State: This ActiveX in-process server [QUERYBUILDERINTERFACES. the exported column in the worksheet is left blank. where n is the number of column-field definitions. BEGIN FILE 1 Connect String 2 Table Name (Table that data is exported to) 3 Worksheet Column for definition 1 4 Table Field for definition 1 5 Fixed Field Flag for definition 1 6 Worksheet Column for definition 2 7 Table Field for definition 2 8 Fixed Field Flag for definition 2 … 2+3*n-2) Worksheet Column for definition 1 2+3*n-1) Table Field for definition 1 2+3*n) Fixed Field Flag for definition 1 END FILE Custom query builder interface File: QUERYBUILDERINTERFACES. The definition consists of the column in the worksheet that is exported. If a field is fixed. “True” or “False”). The interfaces: There are two interfaces: one accessed by WinNonlin (IQUERYBUILDER) and one accessed by the Custom Query Builder (ICONSUMER). and another line to determine if this field is fixed or not (i. a Flag. At this point a complete import file has been created.e. It is not required to. Methods Method Name GetQuery GetError 110 Type Long String Function This tells the custom query builder to get the query (i. This string is then passed to WinNonlin for audit purposes.4 WinNonlin User’s Guide IConsumer is the consumer of the imported data. build it). This interface is implemented by the Custom Query Builder. =0 (No error). String This is used to set or get the help file name used by the Custom Query Builder. It is stored in the registry when the Custom Query Builder is added to WinNonlin. Properties Property Name Type ConnectString QueryString QueryBuilderDescription QueryBuilderVersion Function String Connect string used by the generic ODBC query builder. Currently. This should point to the exact location of the help file. This enables refreshing of the data source. String A description of the query builder. it takes the path of the DLL. The Custom Query Builder could build its own string. QueryBuilderListName AppHelpFile String Name to be used in the list of available Custom Query Builders. String The version of the custom query builder. the help file is located in the same directory as the DLL. . It is also stored in the registry when the Custom Query Builder is added to WinNonlin. QueryBuilderExt String The extension used by the query builder for saving import files. It is up to the developer of the Custom Query Builder whether to use this file is used as the help. Table 4-2.HLP on to the end of the name to create the help file name to be used by the Custom Query Builder. The return values are: <0 (Cancel). When the resource manager registers the DLL.e. >0 (Error) Returns the error message if one exists. String Final query string to be accessed once the custom query builder is done. the name of the DLL and appends . IQueryBuilder interface The interface comprises the properties and methods in Table 4-2 and Table 43. The Custom Query Builder is responsible for importing data into the WinNonlin object by implementing the IConsumer interface. This object is set by WinNonlin via IQueryBuilder. Table 4-3. VB example 1. The extension of the file should be the same as returned in QueryBuilderExt. =0 (No error).IQueryBuilder. This tells the Custom Query Builder to populate the worksheet object handed to it in the WorkSheet subrountine. The structure of the import file is up to the developer. As Icon.DLL. The return values are: <0 (Cancel). the base table must be defined.DLL. and attempts to Import data based on the contents of the Import File. Table 4-4. it is required that the class implement111 .ImportViaImportFile File as string ConfigureString CustomQueryBuilder This function takes an import file name (usually built and saved by the Custom Query Builder).DLL 3. Make it public. Create a class named CQueryBuilder. 2. >0 (Error) PopulateWorkSheet IConsumer interface The interface is accessed by the Custom Query Builder. col As Long) Variant Sets/Gets the value for the given row and column. Add a reference to QueryBuilderInterfaces. Held for future use. the owner of the Watson tables must be defined if the log on ID is not the owner. For instance. for the WATSONLIMS. Not implemented yet.Data import and export The Enterprise Software Development Kit Table 4-3. if necessary. IConsumer properties Property ColumnHeader(idx As Long) Units(idx As Long) Type String String Function This sets/gets the column header for the column. This is also stored in the registry. Methods (continued) Method Name Type Function 4 ImportData. For the Integraware LIMS. Create a new project of type ActiveX DLL. This information is stored in the registry by the WATSONLIMS. Cell (row As Long. highlight the Custom Query Builder on the Custom Query Builder list and click Edit. Implement QueryBuilderInterfaces.The consumer object is an interface to the worksheet within WinNonlin. This is a hook to allow the Custom Query Builder Manager to configure the DLL. Currently. This is used for scripting purposes. To access this method. sumer Long This function is called after the GetQuery function and the WorkSheet subroutine. WorkSheet wrhSht WinNonlin calls this subroutine to set the consumer object in the Custom Query Builder. The calling application needs to know the name of the class to create that implements the interface. the object list has IQueryBuilder. Figure 4-1.4 WinNonlin User’s Guide ing the interface is named CQueryBuilder. Option Explicit Implements QueryBuilderInterfaces. stubbed code is generated.IQueryBuilder 4. If the selected property/method is not defined. 112 . Since IQueryBuilder is implemented. and the other list has its properties and methods. Now each required method and property can be defined as illustrated below. Next. A Simple Object Diagram is shown to the right. finish the code required to generate a valid SQL Query. IQueryBuilder Properties: ConnectString QueryString QueryBuilderDescription QueryBuilderVersion QueryBuilerExt QueryBuilderListName AppHelpFile Methods: GetQuery GetError ConfigCutsomQueryBuilder ImportDataViaImportFile WorkSheet PopulateWorkSheet IConsumer Properties: Cell ColumnHeader Units 113 .Data import and export The Enterprise Software Development Kit 4 5. no data loaded rc< 0 C all P opulateW orksheet m ethod of the C Q B rc> 0 S how error M essage. no data loaded rc< 0 C all G etQ uery on the C Q B rc> 0 S how error M essage. no data loaded rc= 0 C all W orkS heet m ethod of C Q B w ith the IW krS ht as param eter C anceled. no data loaded rc= 0 R eturn to W inN onlin m ain processing 114 . C onnectS tring properties of the CQB C anceled. A ppH elpfile.4 WinNonlin User’s Guide T yp ica l u sa ge of th e C u stom Q ue ry B u ild e r (C Q B ) C reate object im plem enting IC onsum er: IW rkS ht C reate an Instance of the C ustom Q uery B uilder (C Q B ) w ith interface IQ ueryB uilder S et Q ueryS tring. Modeling output: Use the Model Options to include chart output from compartmental modeling or noncompartmental analysis. See the following for additional chart options. See “File types” on page 14 for details. and other chart options WinNonlin provides high quality plots of input data and modeling or NCA output. See “Chart Designer” on page 123. Modify type and formatting of existing charts in the Chart Designer.Chapter 5 Charts Using the Chart Wizard and Chart Designer to create and format plots. • • • “Frequently used chart formatting options” on page 124 “Chart templates” on page 129 “Copy and paste” on page 130 WinNonlin can save charts to many different file formats. See “Output options” on page 204. There are three means of creating and editing plots. For additional plot types. Chart Designer. Chart Wizard: Use this wizard to plot data in any worksheet as a XY plot. The Chart Designer supports additional plot types not available in the Chart Wizard. bar chart. 2. create a plot and edit it in the Chart Designer (below). Chart Wizard Use the Chart Wizard to plot worksheet data. You can then change the plot type and formatting using the Chart Designer as detailed under “Chart Designer” on page 123. Information on printing charts appears under “Printing” on page 21. See “Chart Wizard” on page 115. 3. 1. or histogram. 115 . Open the worksheet(s) containing data to plot. Multiple pairs of XY variables can be included. then use the Chart Designer to change the plot type. 3.5 WinNonlin User’s Guide To create a plot: 1. Make a scatter plot. 2. The worksheet must be the active worksheet in the workbook window. or histogram with the desired variables. Note: Additional chart types are available after a plot is created. 4. The Chart Wizard appears. Click the Next button to procede to the X-Y variables. 116 . The workbook must not be hidden. bar chart. for overlay plots. Click the Chart Wizard tool bar button or choose Tools>Chart Wizard from the WinNonlin menus. These settings differ for Scatter plots and bar charts versus Histograms. See “Chart Designer” on page 123. Choose a chart type. X-Y variables The X-Y variables tab of the Chart Wizard sets the independent and dependent variables to be plotted. Charts Chart Wizard 5 Scatter plots and bar charts To select variables for a scatter plot or bar chart: 1. Make sure that the workbook to be plotted is displayed in the Data Set field (shown below). The columns in the active worksheet appear under Variable Collection. 2. To specify each X or Y variable, highlight a variable under Variable Collection and drag it to the X or Y Variable box. 3. To create an overlay chart, showing more than one pair of X and Y variables: a. If all variables are in one workbook, drag paired sets of X and Y variables to the appropriate variable boxes, so that pairs are listed side-by-side. b. If the variables come from more than one workbook, select a second workbook under Data Set to select X and Y variables from that data set. 4. (Scatter plots only) check Automatic Sorting to sort the data in ascending Xvalue order before drawing plot lines. This option would be inappropriate for a hysteresis curve. Figure 5-1 below provides an illustration of a simple data set plotted with and without automatic sorting. 117 5 WinNonlin User’s Guide Figure 5-1. Plotting data with (middle) and without (right) automatic sorting 5. To create multiple plots or multiple series within a plot, click the Sorting tab and follow the directions under “Sort and group variables” on page 119. 6. (Scatter plots only) To set up error bars, open that tab of the dialog and see “Error bars” on page 120. 7. Click Next to proceed to the Chart titles. Histograms To select variables for a histogram: 1. Make sure that the workbook to be plotted is displayed in the Data Set field. The columns in the active worksheet are listed in the Variable Collection list. 2. To specify the variable(s) for which frequencies will be displayed, highlight a variable under Variable Collection and drag it to the Y Variable box. Repeat to add variables on the same set of axes. 3. Enter the number of bars for the X axis (optional) under Number of Intervals. 118 Charts Chart Wizard 5 WinNonlin will determine the range of the Y variable, divide this range into the number of intervals, then count the number of observations that fall into each interval, i.e., the frequency for each interval. 4. To create multiple plots, click the Sorting tab and follow the directions under “Sort and group variables” on page 119. 5. Click Next to proceed to the Chart titles. Sort and group variables To sort data into multiple plots, or group data series within a plot: 1. Open the Sorting tab of the Chart Wizard. 119 5 WinNonlin User’s Guide 2. Drag and drop variables from the Variable Collection list to the Sort or Group variables box: • • Sort variable(s): a separate plot (set of axes) will be created for each unique combination of sort variable values. Group variable: each plot will contain a separate line (scatter plot) or set of bars (bar charts) for each unique combination of group variable values. The example illustrated above will generate one plot for each drug formulation, with an individual line (scatter) or set of bars (bar chart) for each subject. 3. (Scatter plots only) To set up error bars, open that tab of the dialog and see “Error bars” on page 120. 4. Click Next to proceed to the Chart titles. Error bars To add error bars to a scatter plot: 1. On the Error Bars tab of the Chart Wizard and set the following: a. Up variable: error values in the upward direction (Y-axis increase). b. Down variable: error values in the downward direction (Y-axis decrease). c. Relative error bars: error values are expressed as absolute difference from the mean; that is, each error value is added to (up variable) or subtracted from (down variable) the corresponding datum. d. Absolute error bars: Error data are expressed as Y-axis co-ordinates for the top (up variable) or bottom (down variable) of each error bar. 120 Charts Chart Wizard 5 CAUTION: In overlay charts (more than one set of X-Y variables per chart), the selection of absolute versus relative error bars applies to all data series in the chart. The example above uses relative error bars for a plot of average concentrations plus or minus the standard error at each time point. In this case, the standard error supplies both the Up and Down variables. 6 5 4 3 2 1 0 0 1 2 3 4 5 6 Mean = 3 Standard Error = 1 UP variable = Standard Error Down variable = Standard Error Relative Error Bars Figure 5-2. Plotting mean and standard error using relative error bars Absolute error bars are useful for plotting more than one value at each X-axis point, e.g. minimum, mean and maximum, as illustrated below. 6 5 4 3 2 1 0 0 1 2 3 4 5 6 Mean = 3 Minimum = 1 Up variable = blank Down variable = Minimum Absolute Error Bars Figure 5-3. PLotting mean and minimum values using absolute error bars 2. Click Next to proceed to the Chart titles. 121 5 WinNonlin User’s Guide Chart titles 1. To include a title at the top of each chart, enter it in the Title field in the final dialog of the Chart Wizard. 2. Select or enter the other options: • • • Axis titles by default use the variable names from the data set. To change an axis title, select and replace the text in the appropriate axis title field. Units: Enter or edit the units for the axis labels. See “Units” on page 31 to select whether column units are displayed as axis units by default. Sort key title: If sort variables are used the sort key title appears below the plot title to identify which value of the sort variable(s) appears in the plot, e.g., “Formulation = Tablet.” Edit the display name of each sort variable (in this case, Formulation) by selecting its number in the sort key field. Use group headers: Check this option to include the group variable name(s) in the chart legend, e.g., “Subject = DW.” Legend: When a chart includes more than one pair of X and Y variables, the Legend field provides a means to edit the text that will identify data from each data set in the chart legend. • • 3. Click Finish to create the chart. Axis options Once a plot is created using the Chart Wizard or as part of modeling output, adjust the X and Y axes as follows. See also “Chart Designer” on page 123. 122 Charts Chart Designer To set axis options 5 1. Choose Chart>Axis Options from the WinNonlin menus to open the Axis Options dialog. 2. Select the appropriate options from the following. • • • Uniform Axis: If a sort key was used to identify multiple profiles, this optional applies the same axis range and divisions to all profiles. Logarithmic: Check this box to plot on a natural-log scale. Intervals: For a histogram, enter the number of intervals under Intervals. 3. Click OK. Chart Designer Use the Chart Designer to format and enhance charts created using the Chart Wizard (see “Chart Wizard” on page 115) or as part of modeling or NCA output. The Chart Designer provides extensive plot formatting options and a selection of additional chart types, including three-dimensional plots. See the Chart Designer online help system for detailed usage instructions. To use the Chart Designer: 1. With a chart in the active window, choose Chart>Chart Designer from the WinNonlin menus, or double click on the chart or a specific plot element. Alternately, right-click on the chart and choose Chart Designer from the shortcut menu. 123 5 WinNonlin User’s Guide The Chart Designer provides a tree view in the left pane. Click an + symbol to expand an element to view sub-elements. The tabs to the right change as different elements are selected in the tree. When an entire group of elements is selected from the tree view—all chart series, for example—the controls to the right apply to all of the selected items. 2. Select an item or item(s) to format from the tree view, and choose settings to the right. See the Chart Designer Help or “Frequently used chart formatting options” on page 124 for additional information. 3. Click Apply to execute the changes or OK to execute the changes and close the Chart Designer. Frequently used chart formatting options This section highlights some commonly-used chart formatting options: • • • • • • “Changing the plot type” on page 124 “Changing line style and markers” on page 125 “Changing axis scaling” on page 125 “Creating a semi-log plot” on page 126 “Setting grid lines” on page 127 “Removing or adding walls” on page 128 See also the Chart examples in the Examples Guide. Changing the plot type To select the type of plot: 1. In the Chart Designer, click Chart at the top of the tree view. 2. Select the Type tab on the right side of the Chart Designer. 3. Select the 2D radio button for two-dimensional plots; 3D for three-dimensional. 4. Select the chart type from those available under Chart. See the Chart Designer Help for details and data requirements for each chart type. 124 Charts Frequently used chart formatting options 5 Changing line style and markers Chart colors, line styles, and markers (i.e., symbols) can be changed in the Chart Designer. The formatting may be applied to all data series or an individual series in the chart. To format series lines: 1. To select a particular line for formatting, do one of the following. a. Double-click on that line in the chart. The Chart Designer opens to the appropriate tab, or b. With the chart in the active window choose Chart>Chart Designer from the WinNonlin menus. The Chart Designer opens. Then click Series in the tree view for all series, or use the + to open the list of series and select one. 2. Click the Lines tab. 3. To hide the line, uncheck the Show Series Line check box. 4. Otherwise, select the desired line style, width or color. 5. To set how line segments join at data points use the Join menu. 6. To specify how the ends of lines are displayed select from the menu for Caps. 7. Click Apply or OK. Changing axis scaling Charts use automatic scaling by default. As a result, axis scales may change when the chart window is resized. The axis scaling can be set to a fixed scale. 125 5 WinNonlin User’s Guide To change the scale of an axis: 1. Double-click on the axis or choose Chart>Chart Designer from the WinNonlin menus. The Chart Designer opens. 2. Select the appropriate axis in the tree view pane. 3. To change the scale sizes use the Value Scale tab. 4. Make sure that the Automatic check box is unchecked to set a fixed scale. 5. Enter the minimum and maximum value to appear on the axis. 6. Enter the total number of major divisions. 7. Enter the number of minor divisions per major division. This example creates an X axis that ranges from 0 to 4, with a major division at every multiple of 0.5, and a minor division at every quarter. 8. Click Apply or OK. Creating a semi-log plot A semi-log plot can be created quickly by using the Axis Options dialog or Chart Designer. Use the Chart Designer to set the log base. To create a semi-log plot setting the Axis Options: 1. Select Axis Options from the Chart menu. The Axis Options dialog appears. 2. Select the Logarithmic check box for the Y-axis. 126 Charts Frequently used chart formatting options 5 3. Click Apply or OK. To set the log base using the Chart Designer: 1. Open the chart and double-click on the Y axis or right-click on the chart and choose Chart Designer from the pop-up menu. 2. Select the Y axis in the tree view as shown below. 3. Open the Scale Type tab to the right. 4. Select the Logarithmic check box and enter the desired log base. 5. Click Apply or OK. Setting grid lines To remove the grid lines from a chart: 1. Open the chart and double-click on the Y axis or right-click on the chart and choose Chart Designer from the pop-up menu. 2. Select the X-axis in the tree view. 127 5 WinNonlin User’s Guide 3. On the Pens tab change the Major Grid Pen and Minor Grid Pen Style to the desired line style. Use NULL to show no grid lines. 4. Click Apply. 5. Repeat for the Y-axis. 6. Click Apply or OK. Removing or adding walls Chart walls form a border at the maximum X and Y values, enclosing the chart in a rectangle. To set chart walls: 1. Double-click on the chart or right-click on the chart and choose Chart Designer from the pop-up menu. 2. Select Plot from the tree view menu. 3. Select the Walls tab. 128 Charts Chart templates 5 4. In the Pen group box, select the line style or change it to NULL for none. 5. Click Apply or OK. Chart templates WinNonlin provides the option to create chart templates containing customized settings, including chart type and formatting set in the Chart Designer. To save a chart template: 1. Use the Chart Wizard or Chart Designer to create and format a plot with all the options to be saved in the template. 2. With the chart window selected, choose Chart>Save Template from the WinNonlin menus. 3. Set the file location and name. Save the file with the extension .GTP. 4. Click Save. To load a graph template: 1. Create a chart with the desired content. The chart must contain the same number of X, Y, sort, group and error variables as the template. 2. With the chart selected, choose Chart>Load Template from the WinNonlin menus. 3. The Load Template dialog appears. Select the location and file that contains the graph template (*.GTP). 129 5 WinNonlin User’s Guide 4. Click Open. The graph is now updated with the formatting established in the template. Note: If a template that contains formatting for one data series is applied to a chart that contains multiple series, the template is applied only to the first series. Copy and paste WinNonlin provides the capability to copy a graph paste it into another Windows application in Windows Metafile (*.WMF) format. Note: The Export to Word tool bar button exports the active object directly to Microsoft Word, and can be used to export an image of the current chart. To copy and paste a chart: 1. Select the chart window containing the chart to copy. 2. Select Copy from the Edit menu. A copy of the chart image is placed on the clipboard in Windows Metafile (*.WMF) format. 3. Switch to the other application and (depending upon which application is being used), use the Paste or Paste Special function to paste the image. Note: In order to support cut and paste, the Windows Character Map is limited to the 255 characters noted in the Windows 98 ANSI version. Windows 2000 has multiple Character Mapping options (use Programs>Accessories>System Tools>Character Map in the Windows Start menu to view them). Windows: Western must be selected as the default option in order to cut and paste the characters supported in WinNonlin. Look to the lower, right corner of the Character Map to see if “Keystroke: Alt+XXXX” is within the 255 range for any special characters (characters not included on the computer keyboard). 130 Click on the sample to expand it. with or without calculating summary statistics. with flexible formatting options. 1.and analysis-ready table of worksheet data and (optional) summary statistics in a new WinNonlin workbook. Select Tools>Table Wizard from the WinNonlin menus or click the Table Wizard tool bar button to open the Table Wizard. Select a table template from those listed in Table 6-1. Open one or two data sets to provide table data. click on it again return to the Table Wizard. 2. Nine table templates provide a choice of layouts and summary statistics. A sample of the selected template appears at the top of the Table Wizard. Make sure that the desired data are selected as the active worksheet in the workbook(s). See “WinNonlin table templates” on page 134 for details and examples of each template. • • • • • “Table Wizard” on page 131 “Summary statistics” on page 144 “Significant digits and column alignments” on page 145 “Table formatting” on page 146 “Table definition files” on page 148 Table Wizard Follow these steps to create a new workbook containing a formatted table.Chapter 6 Tables Creating formatted tables in WinNonlin The Table Wizard generates a report. 131 . 4. See the following for instructions and options. 3. Note: Template 9 is available only when at least two workbooks are open. Form Capsule Subject 1 1 1 1 1 Time 0. Choose File>Options from the Table Wizard menus to set options for page breaks between group variable values (optional). As template 6. joining data from two workbooks. As template 6. See also “Tables” on page 31 for details on each table option. By default. set the join variables as described under “Joining data sets” on page 143. As template 6.6 WinNonlin User’s Guide Table 6-1.00 340. demarcating the separate data groupings. 6.92 Alternately. Summary statistics for one data set by cross variable values. with the group variable within the table or above it as shown below. with summary statistics in columns instead of rows. For template 9. 5. Assign variables: Drag the variables to be included from the list at the bottom left to the appropriate boxes to the right. Raw data from one data set by cross variable values. select the workbook containing raw data in the left list.17 1470.00 2069.00 5.32 1914. and the workbook containing computed values in the right list. Raw data from one data set.00 10.00 15. each value of the group variable(s) can generate a page break. 7. with columns ordered by cross variable within variable. group variables appear within the table.00 20. Raw data and summary statistics from one data set by cross variable values. Select the dataset to be used from the drop-down list near the lower left corner of the Wizard. 132 . Raw data and summary statistics for one data set. For template 9. Table templates Template Template 1 Template 2 Template 3 Template 4 Template 5 Template 6 Template 7 Template 8 Template 9 Description Summary statistics for variables in one data set.00 Conc 0. as shown for the variable Form below. 17 1470. 13. indicating the page break settings. 11.00 10. the table cell designated for the result will be left blank. that row will not appear in the table. 133 . the cell that would contain SD is blank in the resulting table. Click Create to create the table in a new WinNonlin workbook.00 5. 9. If an entire row is excluded.92 A button will appear to the left of each group variable in the Table Wizard. To toggle the setting for a specific variable.00 20. go to “Summary statistics” on page 144.32 1914. Missing and excluded data If a value needed for any statistical calculation is missing. Click Finish to create a preview of the table. click the button next to that variable. 10.) Excluded values in the data set are treated as missing in the calculation of summary statistics.Tables Table Wizard Form= Capsule 6 Subject 1 1 1 1 1 Time 0. (The representation of missing values is set in the Tools Options dialog. Click the Prev button to modify table settings if needed.00 Conc 0. Set formatting as detailed under “Table formatting” on page 146. As a result. Click Next. 8. 12. If the template includes statistics. Set significant digits as described under “Significant digits and column alignments” on page 145.00 340. See “Workbooks” on page 28. For example. a page break will be inserted when the value of that group variable changes. when N = 1 the SD cannot be calculated.00 15. If the button contains an X. on the Workbook tab.00 2069. Template 1 This template computes and displays summary statistics for each column included under Variables. for each drug formulation. if any. median and maximum concentration and blood pressure for each subject.6 WinNonlin User’s Guide WinNonlin table templates The Table Wizard provides the following nine templates with different layouts and summary statistics usage. The statistics are computed separately for each unique combination of values for the group variable(s). 134 . This example computes the minimum. Template 9 joins data from two workbooks. then by ID variable values. It does not generate summary statistics. Data are sorted alphanumerically by group variable values. for each drug formulation. This example lists concentration and blood pressure measurements for each subject. at each time in a crossover study. in the order listed.Tables Table Wizard 6 Template 2 This template lists raw data for each column selected under Variables. 135 . 6 WinNonlin User’s Guide Template 3 Template 3 adds summary statistics to template 2. if any. The example below lists concentrations and blood pressure measurements at each time. 136 . and summarizes it for each unique combination of group variable values. It lists the raw data for each variable at each value of the ID variable(s). and summarizes those values for each subject and formulation. 137 . this template generates summary statistics and displays them without the raw data. It computes separate statistics for each unique combination of group and cross variable values.Tables Table Wizard 6 Template 4 Like template 1. 6 WinNonlin User’s Guide This example reports minimum. median and maximum concentration and blood pressure for each subject and each drug formulation. 138 . at each ID variable value. This example reports raw concentration and blood pressure data for each subject at each time. with cross variable values in adjacent columns. with the two formulations in adjacent sets of columns. Subjects appear in separate table rows with each formulation in a separate set of columns. A separate listing is included for each unique combination of values for the group and cross variables. Template 5 This template reports raw data for each column in the Variables list. Tables Table Wizard 6 Template 6 This template adds computation of summary statistics to the listing of raw data as presented in template 5. Statistics are computed separately for each unique combination of group and cross variable values. 139 . the regular variable columns are grouped together within values of the cross variable. 140 . all cross variable values are grouped together. In template 6. with one set for each regular variable. In template 7.6 WinNonlin User’s Guide Template 7 This template presents the same information as template 6 with the columns re-ordered. Tables Table Wizard 6 Template 8 Most templates present raw data in columns. Statistics are calculated for each ID variable value. 141 . Template 8 presents raw data in rows with summary statistics in the right-most columns. within group and cross variable values. 6 WinNonlin User’s Guide Template 9 Template 9 joins data from the active worksheet in each of two workbooks. it can combine raw PK data from an input workbook with PK parameter estimates from a modeling output workbook. The two data sets are “joined” on these group and ID variable values. with a row for each subject (ID variable). The template is designed to present raw data (aggregate variable) at different sample times (DS1 cross variable) and PK parameters (DS2 variables) in columns. For example. It generates summary statistics for each unique combination of group and ID variable values. 142 . Joining data sets Template 9 merges two workbooks into one table..Tables Table Wizard 6 The workbook containing raw data must be selected from the left drop-down list under “Please select datasets.” in the wizard The workbook holding PK parameters or other statistics must be selected from the right drop-down list. 143 . The group and ID variables are selected in matched pairs.. The Table Wizard joins the datasets by matching values for group and ID variables. one from each workbook. 4. In the Table Wizard. The values of these variables will be used to merge records from the two data sets. They need not have identical names. ID variables: Click Define Join for ID variables at the lower right and repeat the steps above to match variables that uniquely identify individual data profiles. One set of statistics will be computed for each unique combination of group variable values. but must represent the same entity with an overlapping set of values. (See “Template 9” on page 142.) 144 . which are a part of most table templates. 6. Click Add Join. (See “WinNonlin table templates” on page 134. select template 9 and select the two data sets that will provide table data. then click Done to return to the Table Wizard. Repeat to set all needed pairs. The Select Join Variables dialog appears. 5. Summary statistics The Table Wizard can compute summary statistics. Group variables: Click Define Join for group variables at the lower right. 3.6 WinNonlin User’s Guide To define a join: 1. Select one variable from each data set to make a matched pair. to match group variables from the two data sets.) 2. This selection is made separately for each column and summary statistic in the table. enter the desired percentage in the Confidence Interval box. To select all of the available summary statistics. 3.Tables Significant digits and column alignments To select the summary statistics to be included in a table: 6 1. When finished. See “Output and computational formulae” on page 153 for details on each statistic. click All. Check the check box preceding each summary statistic to be included in the table. Significant digits and column alignments The Table Wizard allows the user to select the number of decimal places or the number of significant digits to be used. click Next to set the Significant digits and column alignments. 145 . Note: The statistics in the table will appear in the order listed here. 2. To include a confidence interval at a confidence level other than 95%. After the settings are complete. borders or formatting. Note: For variables with integer values. including page numbers and footnotes.g.6 WinNonlin User’s Guide Significant digits: For each table column and summary statistic listed. Check Data Only at the top left corner of this dialog to create a table (worksheet) without page breaks. 146 . Column alignments: For each column and summary statistic. Table formatting The final dialog of the Table Wizard provides the means to format table cells. headings and page layout. click the button under Significant digits/Decimal places to select which you will enter. Subject or Period. click Next to proceed with Table formatting. click the button under Alignment to select the alignment within table cells. choose Decimal places and enter 0 as the number of places. then enter the number of significant digits or decimal places to the right. column headers. e. 9. 3. 2. Make adjustments and click OK. Click the Set Font button to change the font. header or footer relative to the page. Variable (Unit) to present units in parentheses after the variable name or Variable. Insert line breaks by pressing the ENTER key. or font style for that item. Unit to present units after a comma. Select the appropriate radio button in the Page Layout group box: Portrait to orient the long side of the page vertically. 4. To include page numbers at the bottom of each page. options for units display appear at the bottom right.. To include a table title on each page. 147 . following the variable name. 6. Heading). check the Underline box. 8. To include text at the bottom of each page. Landscape to orient it horizontally. font size. Check None to exclude units from the table.g. 5. If the input workbook included units. select Page Headers under Item and type a title under Header Text. select Page Footers under Item and check Page numbering below the footer text and enter the first page number under Start at. To underline text for an item (e. Select an item to format from the pull-down list called Item. Check the appropriate box under Alignment to align the table title. The Font dialog appears.) 7. select Page Footers under Item and enter the text under Footer Text. (Click the Prev button to set table cell alignments.Tables Table formatting 6 1. TDF. variables. Choose File>Save Definition As (Alt+F. Click OK. 4. Table definition files are also useful for generating formatted tables via scripting. 2. 148 . Select a drive and directory and enter a descriptive file name. then saving the settings from within the Table Wizard. The table definition file includes: • • • • • • Table template number Names and mappings of table variables (data columns) Summary statistics to be included (if applicable) Titles (up to five lines) Footers (up to five lines) Formatting: – Alignment and decimal places or significant figures for the data – Table and page formats from the final dialog of the Table Wizard Creating a table definition file Table definition files must be created in WinNonlin. a data set must include the exact column names used in creating the definition file. To create and save a table definition file (*. by creating a table with the desired formatting.6 WinNonlin User’s Guide Table definition files A table definition file (*. The Save Current Settings dialog appears.TDF) stores a user-defined table format. Follow the instructions under “Table Wizard” on page 131 up to the table preview (just before clicking the Create button). To use a table definition file. using the file extension . statistics (optional) and formatting. S) from the Table Wizard menus. 3. The table definition file can then be reused to quickly create tables with that format. to set up a table with the desired template.tdf): 1. tdf) in the Table Wizard: 1. as detailed under “Table Wizard” on page 131.Tables Table definition files 6 Loading a table definition file To load a table definition file (*. 5. Select the appropriate workbook. Choose File>Open Definition from the Table Wizard menus. Open or create the worksheet containing data for the table. 3. 149 . The table can be edited if needed. 2. Specifications made via the table definition file will be imported into the Table Wizard. Make sure it uses the exact column names used in the Table Definition File. The Table Wizard appears. Choose Tools>Table Wizard from the menus or click the Table Wizard button on the tool bar. 4. Select the appropriate table definition file and click Open. 6 WinNonlin User’s Guide 150 . to summarize modeling results.Chapter 7 Descriptive Statistics Summary statistics and weighted summary statistics WinNonlin can compute summary statistics for variables in any workbook. or to test for normal distribution of data. 151 . for preclinical summaries. or click the Descriptive Statistics button on the tool bar. Open the input data worksheet. The data set must be open and not hidden. Computing summary statistics To set up descriptive statistics: 1. Separate statistics for subgroups are obtained through the use of sort variable(s). This feature is frequently used to create data to plot means and standard errors. Choose Tools>Descriptive Statistics from the WinNonlin menus. 2. Create Sheets for Summary Variables: When there are multiple summary variables the output can be sent to one or multiple worksheets in the output workbook. If there are fewer than five data points. 5. The median is plotted as an asterisk (*). Select the options for the statistics: • • Confidence intervals: Enter the desired confidence interval in the % field. The box hinges are the 25th and 75th percentiles. typing the underlined letter in the name of the destination box. tenth. fifth. ninety-fifth. Select sort variable(s) if desired.7 WinNonlin User’s Guide The Descriptive Statistics dialog allows settings to be saved to an ASCII file for later re-use. If output is combined onto one worksheet. while pressing the Shift key. If there are no outliers beyond this range. Note: Select variables either by dragging the variable name to the appropriate box. no units will appear with the output. Click Calculate. 4. 3.5 times the H-spread (the distance between the hinges) of the hinges. Drag the variable(s) to be summarized to the Summary Variables box. 152 . fiftieth. or by highlighting the variable name then. the whiskers end at the maximum and minimum points. seventy-fifth. • • • 6. through the Save Setup and Load Setup buttons.5 and 3 H-spreads from the hinges are marked by an o. Percentiles: Selecting the check box means that the first. The whiskers end at the data points farthest from the median that are still within 1. and the those outside 3 H-spreads are marked by x. Box and Whiskers plot: The Box and Whiskers plot appears in a separate text window. ninetieth. and the ninetyninth percentiles will be included in the output. twenty-fifth. Data points between 1. Weight Variable: (Optional) See “Weighted summary statistics” on page 157. A separate analysis is done for each unique combination of sort variable values. See “Units” on page 153. the data are graphed individually rather than as a box. Nmiss. certain of the calculations will have units. CI Upper Var Everything else Units No units No units No units No units No units x-unit2 x-unit2 x-unit Output and computational formulae Summary statistics include the following computations.Descriptive Statistics Computing summary statistics 7 Units When summary statistics are calculated on a variable with units. Table 7-1. Descriptive statistics output Statistic N Nmiss Nobs Mean SD Description Number of observations with non-missing data Number of observations with missing data Number of observations Arithmetic Average Standard Deviation: Variance SE Standard Error: SD ------N 153 . Assuming that the variable summarized is x and has x-units specified. SD log Variance CI Lower Var. CV% Geometric Mean Skewness Kurtosis Mean log. the units for the summary statistics are: Statistic N. Nobs CV%. 7 WinNonlin User’s Guide Table 7-1. Descriptive statistics output (continued) Statistic Variance Unbiased sample variance: N Description ∑ ( Xi – X ) N–1 Min Median Max Range CV% Geometric Mean Harmonic Mean Minimum value Median value Maximum value 2 i=1 ------------------------------- Range of values (maximum value minus minimum value) Coefficient of variation: (SD/Mean)*100 Nth root of the product of the N observations Reciprocal of the arithmetic mean of the reciprocals of the observations 154 . + ---. x1.∑ H i n i=1 n and (n – 1) H i = ---------------------------------------------------------------------------------------------1... Descriptive statistics output (continued) Statistic Pseudo SD Description 7 Jackknife estimate of the standard deviation of the harmonic mean. xn the pseudo standard deviation is: n pseudo SD = where ( n – 1 ) ∑ ( Hi – H ) i=1 2 1 H = -. .+ … + ---------.+ … + ----⎞ ⎝x x x x x ⎠ 1 2 i–1 x+1 n (n – 1) = ----------------------------⎛ n ⎞ 1 1 ⎜ ---⎟ – --⎜∑ x⎟ x j⎠ i ⎝j = 1 Mean log SD Log CV% Geometric Mean Arithmetic average of the natural logs of the observations Standard deviation of the natural logs of the observations Coefficient of variation of the geometric mean: 2 exp ( ( SD_log ) ) – 1 × 100 SD Mean – t α ⁄ 2 ------N CI X% Lower Mean Lower limit of an X% confidence interval for the mean: 155 ..11 1 1⎛ ---.+ ----------.Descriptive Statistics Computing summary statistics Table 7-1. For n points. and t α ⁄ 2 is from the t-distribution with N-1 degrees of freedom. χL2 cuts off a lower tail. P(ercentiles) The Pth percentile divides the distribution at a point such that P percent of the distribution are below this point. the t-distribution is close to the normal distribution.05. Coefficient of skewness: Skewness ∑ wi ( xi – xw ) ⁄ N -------------------------------------------------------3⁄2 2 [ ∑ wi ( xi – xw ) ⁄ N ] 3 156 . (integer part) f = p*(n+1) . Thus. (fractional part) x(j) is the j-th order statistic. Descriptive statistics output (continued) Statistic Description CI X% Upper Mean Upper limit of an X% confidence interval for the mean: SD Mean + t α ⁄ 2 ------N where ( 1 – α ) × 100 is the percentage given for the confidence interval.j. a 95% confidence level indicates that alpha = 0. of area α/2 where (1 . CI X% Lower Var Lower limit of an X% confidence interval for the variance: ( N – 1 ) × Variance ----------------------------------------------2 χU CI X% Upper Var Upper limit of an X% confidence interval for the variance: ( N – 1 ) × Variance ----------------------------------------------2 χL where χL2 and χU2 are from the χ2-distribution with N-1 degrees of freedom. Otherwise. Note that for N>30. For a sample size of n. and χU2 cuts off an upper tail.α)*100 is the percentage for the confidence interval.7 WinNonlin User’s Guide Table 7-1. The above is used if 1 <= j < n.f) * x(j) + f * x(j+1) where j = int(p*(n+1)). the empirical quantile is the smallest order statistic for j=0 or the largest order statistic for j=n. the quantile corresponding to the proportion p (0<p<1) is defined as: Q(p) = (1 . missing. it is set to zero.Descriptive Statistics Weighted summary statistics Table 7-1. and the statistics below. 2. See “Computing summary statistics” on page 151.– 3 2 2 [ ∑ wi ( xi – xw ) ⁄ N ] KS_pval Kolmogorov-Smirnov normality test p value. Output and computational formulae for weighted calculations The output for weighted summary statistics contains a column indicating the summary variable(s). using a column in the data set to provide weights. Make all specifications as for un-weighted summary statistics. Table 7-2. Weighted summary statistics output Statistic N Nmiss Nobs Description Number of non-missing observations (including those with weights = 0) Number of observations with missing data Number of observations (including observations with weights of 0) 157 . Drag the variable containing weights to the Weight Variable box (or highlight the variable to be used as the weight and press Shift+W). If a weight is non-numeric. one for each sort variable. 4 Weighted summary statistics Summary statistics can be weighted. To create weighted summary statistics: 1. or excluded. Descriptive statistics output (continued) Statistic Kurtosis Description Coefficient of excess (kurtosis): 7 ∑ wi ( xi – xw ) ⁄ N -------------------------------------------------. – 3 2 2 [ ∑ wi ( xi – xw ) ⁄ n ] 4 158 . using weighted mean and SD CI X% Lower Var CI X% Upper Var Skewness Lower limit of an X% confidence interval. Weighted summary statistics output (continued) Statistic Mean Description Weighted Arithmetic average: ∑ wi × xi ---------------------∑ wi SD Weighted Standard Deviation: Weighted Variance SE Weighted Standard Error: Weighted SD -----------------------------N Variance Weighted Variance: ∑ wi ( xi – xw ) ----------------------------------N–1 CV% 2 where x w is the weighted mean. using weighted mean and SD CI X% Upper Mean Upper limit of an X% confidence interval. using weighted variance Weighted coefficient of skewness: ∑ wi ( xi – xw ) ⁄ n ------------------------------------------------------3⁄2 2 [ ∑ wi ( xi – xw ) ⁄ n ] Kurtosis Weighted coefficient of excess (kurtosis): 3 ∑ wi ( xi – xw ) ⁄ n ------------------------------------------------. Weighted Coefficient of variation: (Weighted SD/Weighted Mean)*100 CI X% Lower Mean Lower limit of an X% confidence interval. using weighted variance Upper limit of an X% confidence interval.7 WinNonlin User’s Guide Table 7-2. “Model options” on page 173 NCA computations and output parameters are covered under “Computational rules for WinNonlin’s noncompartmental analysis engine” on page 181 159 . “Partial areas” on page 167 5. intravascular (IV) infusion or IV bolus dosing. using methods appropriate for either richly. and moments in richlysampled time-effect data. “NCA model selection” on page 160 2. WinNonlin provides the following noncompartmental analysis “models” for different types of input data. “Dosing regimen” on page 171 7. “Model variables” on page 161 3. areas. height. • The following sections detail the steps to set up a WinNonlin NCA. 1.Chapter 8 Noncompartmental Analysis Areas and slopes for plasma. “Lambda Z or slope estimation settings” on page 162 4. “Therapeutic response” on page 168 6. for concentration data from blood/plasma (single dose and steady state) or urine (single dose only). One PD model for analysis of slope.or sparsely-sampled data. • Six PK models: extravascular. These models support rich data sets as well as sparsely-sampled studies such as toxicokinetic studies. urine or PD data WinNonlin's noncompartmental analysis (NCA) engine computes derived measures from raw data. Sparse data sets can include up to 250 subjects. 2. Note that the Model 200 is highlighted and selected by default. 4. See “Sparse sampling computations” on page 188 for discussion. See “Noncompartmental analysis” on page 505 for details on the models and output parameters. Confirm that the correct model is selected. To analyze data with few observations per subject. 8. and click Finish to load the model. Click Next. 6. 5. check Use Sparse Sampling.8 WinNonlin User’s Guide NCA model selection To load a noncompartmental analysis model: 1. and select it as the active worksheet (click on that worksheet to bring it to the front). Click Next. 3. The model window will open with a sample graph. Open the worksheet containing data to be modeled. 160 . Select the appropriate NCA model. The next step in NCA is setting Model variables. Start the PK/PD/NCA Analysis Wizard by clicking on that tool bar button or selecting that command from the Tools menu. 7. Select the Noncompartmental radio button. and “Sparse sampling (pre-clinical) data” on page 514 for additional parameters computed. Drag and drop to specify the following (optional): • Sort Variable(s): categorical variable(s) identifying individual data profiles. subject ID and treatment in a crossover study. Volume: the volume of urine collected per time interval. e. A separate analysis is done for each unique combination of sort variable values. For sparse data sets. i. Urine models include the following variables. 2. drug concentration in urine. After NCA model selection. • • • Concentration: the dependent variable. choose Model>Data Variables from the WinNonlin menus to open the Data Variables dialog. drag and drop to assign the variable containing subject identifiers into the Subj_ID field. Lower times: the starting times for individual collection intervals. Y: the dependent variable. • Note: The output data set will automatically include the original values of the sort variable(s) and variables being modeled (X and Y for plasma models or midpoint and rate for urine models).Noncompartmental Analysis Model variables 8 Model variables To specify data columns for each variable in a noncompartmental analysis: 1. WinNonlin computes the midpoint and excretion rate (amount eliminated per unit of time) for each collection interval. in addition to the estimated parameters. drag and drop columns to assign variables to be modeled (see “Urine models” on page 161 for urine models): • • X: the independent variable. Carry Along Variables: data to be copied into the output workbook. 4.e. For plasma models. 5. observed values of interest. Urine models For urine models.. 161 .g.. Drag and drop columns to assign variables to be modeled as detailed under “Model variables” on page 161. 3. Click OK. Lambda Z or slope estimation settings are the next step in NCA setup. generally time. associated with the terminal elimination phase for concentration data. then the last four points. but is within 0. and calculated from at least three data points.0001 of the largest adjusted R2 value. 162 . then WinNonlin will determine the data points to include in λz or slope calculations as follows. Lambda Z or slope range selection If the user does not specify the data points to include in the calculation of λz or slopes. Points with a value of zero for the dependent variable are excluded. WinNonlin estimates λz using the regression with the largest adjusted R2 and: • • If the adjusted R2 does not improve. WinNonlin repeats regressions using the last three points with non-zero concentrations. If λz can be estimated. λz. For concentration data During the analysis. as selected in the Model Options (see “Model options” on page 173). λz must be positive. yet does not choose disable curve stripping as described under “Setting Lambda Z or PD slope ranges” on page 164. For each regression. etc. an adjusted R2 is computed: 2 ) Adjusted R = 1 – ( 1 – R ) × ( n – 1 -----------------------------------------(n – 2) 2 where n is the number of data points in the regression and R2 is the square of the correlation coefficient. or slopes of up to two selected data ranges for drug effect data.8 WinNonlin User’s Guide • Upper times: the ending times for individual collection intervals. NCA does not extrapolate beyond the observed data for drug effect. Points prior to Cmax or prior to the end of infusion are not used unless the user specifically requests that time range. parameters for concentration data will be extrapolated to infinity. last five. the regression with the larger number of points is used. These measures will be estimated using linear or linear/log trapezoidal rule. Lambda Z or slope estimation settings WinNonlin will attempt to estimate the rate constant. data points for Slope2 are labeled “2” and footnoted using an asterisk. WinNonlin will compute the best-fitting slope at the end of the data for Slope2. Calculation of Lambda Z Once the time points being used in the regression have been determined. For drug effect data WinNonlin will compute the best-fitting slopes at the beginning of the data and at the end of the data using the same rules that are used for lambdaZ— roughly. best adjusted R-square with at least three points. with linear or log regression. It is the user’s responsibility to evaluate the appropriateness of the estimated value. then WinNonlin will also compute the best-fitting slope at the beginning of the data for Slope1. Calculation of slopes for effect data WinNonlin estimates slopes by performing a regression of the response values or their natural logarithm. WinNonlin will almost always compute an estimate for λz. 163 . Data points for Slope1 are marked with “1” in workbook output and footnoted using “#” in text output. WinNonlin can estimate λz by performing a regression of the natural logarithm of the concentration values in this range on sampling time. The actual slopes are reported. “*”. If the user specifies the range only for Slope2. The data points included in each slope are indicated on the Summary table of the output workbook and text for model 220. λz is defined as: λz = -1 (estimated slope) Note: Using this methodology. then in addition to computing Slope1.Noncompartmental Analysis Lambda Z or slope estimation settings 8 For sparse sampling data The Lambda Z ranges for sparse data show the mean concentration at each time (plasma or serum data) or mean rate for each interval (urine data). according to the user's choice. If the user specifies the range only for Slope1. they are not negated as for λz. parameters that do not depend on λz (i.) will still be reported and the software will issue a warning in the text output. 164 . The user-specified range contains fewer than two non-missing observations. the Slopes tab of the Model Properties dialog (Model>Lambda Z Ranges or Model>Slopes). AUClast.8 WinNonlin User’s Guide Limitations of Lambda Z and slope estimation It is not possible for WinNonlin to estimate λz or slope in the following cases. Cmax. • • • • • There are only two non-missing observations and the user requested automatic range selection. For infusion data. for model 220. (Individual profiles are defined by the sort variable(s) in “Model variables” on page 161). indicating that λz was not estimable. In these instances. Setting Lambda Z or PD slope ranges To set the λz range for each PK profile. automatic range selection is activated. The Graph and Data buttons at the top right corner of the tab switch between a graphical or tabular interface to set time ranges for each data profile. or slopes to compute for each PD profile: 1. Automatic range selection is activated. The instructions below are for the graph view.e. The time difference between the first and last data points in the estimation range used is approximately < 1e-10. Tmax. and there are fewer than 3 points at or after the infusion stop time. but there are fewer than 3 positive concentration values at or after the Cmax of the profile. Open the Lambda Z Ranges tab or. etc. – Parameters that are extrapolated to infinity will not be calculated. select whether each slope should be computed using linear regression (Lin Reg) or regression on the log data (Log Reg). Select the axis scale for the graphical representation below the lower left corner of the plots: Linear or Log Linear. Note: To disable curve stripping for one or more individual profiles. For slopes. To turn off λz or slope estimation for all profiles. 4.Noncompartmental Analysis Lambda Z or slope estimation settings 8 2. enter a start time that is greater than the last time point in a given profile. 165 . and an end time greater than the start time. check Disable Curve Stripping at the bottom left and skip to step 8. With this option checked: – λz or slopes will not be estimated. 3. 8 WinNonlin User’s Guide 5. etc. Use the right/left arrows just below the plot view to move between profiles. These exclusions apply only to Lambda Z or slope calculations. Slope1 and Slope2. moments. User Time Range: Select the first data point to be included: left-click on a data point or enter a time value under Start time. Best Fit: Select this option to have WinNonlin determine the Lambda Z or Slope ranges as described under “Lambda Z or slope range selection” on page 162. For slopes. Data ranges (optional): For each profile. 166 . 6. Exclusions (optional): to exclude a data point from the calculations. use these steps to choose data ranges for up to two slopes. define the terminal elimination phase or slopes regions as follows. a. Repeat to reverse an exclusion. the excluded data points will still be included in computation of AUC’s. c. Select the last data point to be included: hold down the Shift key while left-clicking on a data point or enter a time value under End time. hold down the Ctrl key and left-click on that point. b. Note: The units for Start and End time are those of observation times in the data set. simply leave the start and end times on this tab empty. The start and end times need not appear in the data set. by setting start and end times for each.Noncompartmental Analysis Partial areas 8 7. To specify the partial areas under the curve to be calculated: 1. click OK.). 3. etc. 2. if λz is estimable. The time units are those of the input data set. and can be after the last observed time. 167 . Enter the start time in the cell labeled AUC1 Lower. and set values for each profile. repeat this step for the next area (AUC2. To save the settings and close the dialog. and reload them for re-use. 8. See “Partial areas” on page 184 for additional information on partial area computations. to exclude them. Use the arrow buttons directly below the plot to move between profiles. click Apply to save the settings and move to another tab of the Model Properties dialog. Partial areas The Partial Areas tab of the Model Properties dialog includes settings for computation of partial areas under the curve. For another partial area. Up to 126 partial areas can be computed per profile. Partial area computations are optional. Loading Lambda Z or slope ranges from an ASCII file The Load and Save buttons make it possible to store these time ranges to an external ASCII data file. Open the Partial Areas tab of the Model Properties dialog (Model>Partial Areas or click the Partial Areas tool bar button). When done. and the end time in the cell labeled AUC1 Upper. Parameters included depend on whether upper. Repeat steps 2 and 3 for other profiles or use the fill-down method (drag bottom right corner of selected cells) to copy from one profile to those below. lower. and re-load it for later use. Click OK when all settings are complete. Therapeutic response The Therapeutic Response tab of the Model Options dialog provides settings for computing times and areas during which observations fall above or below target values. WinNonlin will compute additional parameters noted in Table 8-1 below. See “Therapeutic response windows” on page 186 for additional computation details. 168 . 5. Use the Save and Load buttons to save all settings in the Partial Areas tab to an ASCII data file. or both limits are supplied.8 WinNonlin User’s Guide 4. For non-compartmental analysis of concentration data. Therapeutic window output for PK data Input values Parameters computed Lower and upper TimeLow TimeDuring TimeHigh AUCLow AUCDuring AUCHigh TimeInfDuring TimeInfHigh AUCInfLow AUCInfDuring AUCInfHigh TimeLow TimeHigh AUCLow AUCHigh TimeInfHigh AUCInfLow AUCInfHigh Time range for computation From dose time to last observation Description Total time below lower limit Total time between lower limit and upper limit Total time above upper limit AUC for time below lower limit AUC for time between lower limit and upper limit AUC for time above upper limit Total time between lower limit and upper limit Total time above upper limit AUC for time below lower limit AUC for time between lower limit and upper limit AUC for time above upper limit Total time below lower limit Total time above lower limit AUC for time below lower limit AUC for time above lower limit Total time (extrapolated to infinity) above lower limit AUC for time (extrapolated to infinity) below lower limit AUC for time (extrapolated to infinity) above lower limit Lower and upper From dose time to infinity using λz (if λz exists) Lower only From dose time to last observation Lower only From dose time to infinity using λz (if λz exists) For drug effect data (model 220).Noncompartmental Analysis Therapeutic response 8 Table 8-1. 169 . WinNonlin will make the computations detailed under “Drug effect data (model 220)” on page 515. To compute time and AUC spent within a therapeutic concentration range: 1. Open the Therapeutic Response tab of the Model Properties dialog (Model>Therapeutic Response or click the Therapeutic Response tool bar button). if no baseline value is entered. Threshold (optional): effect value for use in calculating additional times and areas. To set the baseline and (optional) threshold values for effect data (model 220): 1. For each profile. 2. if any. as baseline. b. Baseline: baseline effect value for each profile. (See “Additional parameters for model 220 with threshold” on page 517. a value must be entered here. If there is no data point at dose time. Enter the lower and upper (optional) concentrations delimiting the target range and click OK. Open the Therapeutic Response tab of the Model Properties dialog (Model>Therapeutic Response or click the Therapeutic Response tool bar button). WinNonlin will use the effect value at dose time. Required unless dosing data are provided in a Dosing worksheet (as when data are loaded from PKS).) 170 . In that case. enter the following: a.8 WinNonlin User’s Guide 2. WinNonlin will assume a time of zero. Current Units: Enter the units or select one item under Mass Prefix and Mass Unit then click the Add button. If dose time is not entered. Dose normalization affects units for all volume and 171 . For example. This will affect output parameter units. if doses are in milligrams per kilogram of bodyweight. select the appropriate factor in this drop-list. or after a final dose at steady state. For steady state. For example. Enter dose information in the Dosing Regimen dialog.Noncompartmental Analysis Dosing regimen 8 3. Click OK when finished. if Volume units were L. Choose Model>Dosing Regimen from menus or click the Dosing Regimen tool bar button to open the Dosing Regimen tab of the Model Properties. WinNonlin also assumes equal dosing intervals. 2. 3. Dose Normalization: If dose amount is normalized by subject bodyweight or body mass index. or read it from a worksheet as detailed in “Dosing data variables” on page 202. dose normalization of kg would change those units to L/kg. To enter dosing via the dosing regimen dialog: 1. Dosing regimen Models 200 to 202 (Plasma) assume that data were taken after a single dose. Models 210 to 212 assume single-dose urine data. as described below. enter Current Units of mg and Dose Normalization of kg. For drug effect data (model 220): Using the same time scale and units as the input data set. Click Apply to commit changes or OK to commit and close the dialog. and Percent_Recovered in NCA urine models. Time of last dose: If the dose was given at a time other than 0.8 WinNonlin User’s Guide clearance parameters in NCA and PK models. based on the tau value set above. in most studies. Excluded profiles appear in red (see “Data exclusion and inclusion” on page 60) 5. 4. 6. enter the dose amount under Value for each row in which Constant = Dose. check Steady State and enter the dose interval for each row in which Constant = Tau. 7. However. enter the time of the most recent dose for each profile under Value for each profile. 9. Dose: For concentration data (models 200-212). as well as AUC/D in NCA plasma models. Time deviations in steady-state data When using steady-state data. 8. 172 . Use the Save or Load button to save or re-load dosing information using an external ASCII file. Steady state: For steady state dosing (models 200-202 only). WinNonlin computes AUC from dose time to dose time + tau. See “Special file formats” on page 430 for file format. enter the dose time under Value for each row in which Constant = Time of Last Dose. one or more new windows is generated. and format of final parameters worksheet • Default units for output parameters Parameter names • Names. if dose time = 0 and tau = 24. Note: Set default output options via Tools>Options in the WinNonlin menus. Cmin and Tmin are found using observations taken at or after the dose time.Noncompartmental Analysis Model options 8 there will be sampling time deviations.083 hours. Table 8-2. ASCII and chart output • Method for computing area under the curve. 173 . The Model Options dialog appears. Click on a tab to select and adjust a group of options from those in Table 8-2. Click Apply to set options or OK to set options and close the dialog. NCA model options Category Output options Weight Title NCA settings Units Options • Workbook. 2. and/or ASCII text formats for modeling results • Uniform or weighted least squares • Text title for workbook. In this instance.975 or 24. chart. For steady state data. Cmax. with content as selected in the following NCA Model options. Select Model>Model Options from the WinNonlin menus or click the Model Options tool bar button. That is. 3. Output options After noncompartmental analysis is run. the program will estimate the AUC based on the estimated concentration at 24 hours. inclusion and precision for output parameters To set NCA model options: 1. and not the concentration at the actual observation time. Model options Options for NCA output are divided into the following sections. Tmax. but no later than dose time+tau. the last sample might be at 23. Summary of NCA worksheets Sheet name Final Parameters Dosing Summary Table Contents Estimates of the final parameters for each level of the sort variable The dosing regimen for the analysis The sort variables. points included in the regression forλz or slopes. Y. profiles with all or many missing parameter estimates are excluded from the results.8 WinNonlin User’s Guide Exclude profiles with insufficient data: If this option is checked. X. . User defined settings History of actions done in and to the workbook. and for each of the sub-areas from partial area computations. containing the same information as the workbook output. residual for the regression. 174 . The NCA workbook contains the worksheets summarized in Table 8-3. AUC. See “History worksheet” on page 37. Workbook: Check this option to generate an output workbook. Page Breaks: Check this box to include page breaks in the ASCII text output Output intermediate calculations: If this option is checked. profiles with insufficient data will output missing parameter values. Modeling errors are also reported in this file. These worksheets present the same information as the text output. See “Data deficiencies that will result in missing values for PK parameters” on page 182 for a listing of cases that produce missing estimates and are excluded using this option. If it is not checked. predicted Y for the regression. Text: This option provides an ASCII text version of the analysis results. Partial Areas User Settings History Chart: If workbook output is selected. the text output for the next analysis only will include values for iterations during estimation of λz or slopes. as shown below. Table 8-3. check this box to generate an NCA Chart window with a plot of observed versus predicted data for each profile. AUMC and weight used for the regression See “Partial areas” on page 167. but in tabular form. Noncompartmental Analysis Model options 8 Weight For NCA. selections on the Weight tab of the Model options dialog apply to the regression that estimates λz or slopes. Sparse sampling requires uniform weighting. 175 . Select Uniform Weighting from the pull-down list. the title is displayed at the top of each printed page in ASCII text output. Select Observed to the Power n button and enter the value of n. it already uses weighting implicitly. Enter the title text below. CAUTION: When a log-linear fit is done. It may include up to 5 lines of text. on the User Settings worksheet in the workbook output. See “Weighting” on page 241 for discussion of weighting schemes. However. and at the top of each chart in chart output. If it is not. The title will be centered on the page. To include a line break use Shift + Enter. one could argue that the weights should be 1/LogY. To use uniform weighting: 1. or use the Pre-selected scheme button to select the most common weighted least square options: 1/Y and 1/Y2. concentrations between 0 and 1 would have negative weights. To include a title: 1. the title will not display even if text is entered below. 176 . To use weighted least squares: 1. Note: When weighting by 1/Y and using the Linear/Log trapezoidal rule. 2. Title If included. 2. Select the Pre-selected Scheme radio button. rather than 1/Y. and therefore could not be included.8 WinNonlin User’s Guide Note that it is not the weights themselves that are important. rather it is the relative proportions of the weights. Make sure that Title Text is checked on the Title tab of the Model options dialog. if this were the case. approximately equal to 1/Yhat2. • Linear Trapezoidal rule (Linear Interpolation) [default and recommended for effect data (model 220)]. and sums up these areas. or after C0 for bolus models (if C0 > Cmax). is used to inject a concentration value for that endpoint. then the first maximum is used. WinNonlin inserts a final concentration value using: 1) the linear interpolation rule. As above except when a partial area is selected that has an endpoint that is not in the data set. All methods reduce to the log trapezoidal rule and/or linear trapezoidal rule. in the NCA Settings tab of the Model options dialog. This method (method 2) is recommended for Model 220 (PD data). applied to each pair of consecutive points in the data set that have non-missing values. but differ with regard to when these rules are applied. If Cmax is not unique. or 2) logarithmic interpolation if the endpoint is after Cmax. In that case. The default method is set on the Models tab of the WinNonlin options dialog (see “WinNonlin options” on page 28). • 177 .Noncompartmental Analysis Model options 8 NCA settings Calculation methods for AUC Four methods are available for the calculation of area under the curve. given under “AUC calculation and interpolation formulas” on page 183. It uses the linear trapezoidal rule. and partial area computations. The method chosen applies to all AUC. given below. If a partial area is selected that has an endpoint that is not in the data set. Linear Trapezoidal Method with Linear/Log Interpolation (method 4). AUMC. then the linear interpolation rule. and the Linear Trapezoidal (with Linear/Log Interpolation) methods all apply the same exceptions in area calculation and interpolation: If a Y value (concentration. If the Transpose Final Parameters option is not checked. entitled Transposed Final Parameters. The linear trapezoidal rule is used any time that the concentration data is increasing. Note: The Linear/Log Trapezoidal. 178 . rate or effect) is less than or equal to zero. the Linear Up/Log Down. If Cmax is not unique. if checked. if adjacent Y values are equal to each other. or C0 for bolus models (if C0 > Cmax). See “Transpose final parameter table” on page 206 for example output from compartmental modeling. the program defaults to the linear trapezoidal or interpolation rule. Log interpolation is used to create points for partial areas after Cmax. the Final Parameters worksheet will present the untransposed format. and linear interpolation is used otherwise. Units The Units tab of the Model options dialog (below) lists default units for output parameters and supports entry of preferred units. and the logarithmic interpolation rule if the concentration is decreasing. Points for partial areas are injected using the linear interpolation rule if the surrounding points show that concentration is increasing. Transpose Final Parameter Table This option. A second worksheet called Non-transposed Final Parameters will present one row for each parameter. and the logarithmic trapezoidal rule is used any time that the concentration data is decreasing. or after C0 for bolus (if C0 > Cmax). will create a Final Parameters worksheet with one column for each parameter. Linear Up/Log Down Method (method 3). Uses the log trapezoidal rule (given under “AUC calculation and interpolation formulas” on page 183) after Cmax. WinNonlin converts values as necessary to display the output in the preferred units. the program defaults to the linear trapezoidal or interpolation rule for that point. This option is provided primarily for WinNonlin scripts that depend on the transposed format in the Final Parameters worksheet. and linear trapezoidal otherwise. then the first maximum is used. and the second worksheet. • See “AUC calculation and interpolation formulas” on page 183 for equations. will present the transposed version.8 WinNonlin User’s Guide • Linear/Log Trapezoidal Method (method 1). Similarly. Y-variable. and.TXT or URINEUNITS. Use the Save and Load buttons to save preferred units as an ASCII text file. dosing. then load them back for later modeling. for plasma models. WinNonlin provides the option to specify the variable names and precision for the final model parameters. e..g. Note: Preferred units can be set only after units are set for the X-variable.TXT. Parameter names For noncompartmental analyses (NCA). Click Apply or OK. The units cannot contain space characters. PLASMAUNITS. 179 . Enter the unit under Preferred for the appropriate row.Noncompartmental Analysis Model options To set a preferred unit: 8 1. 2. Case is preserved. The file lists WinNonlin NCA parameter names paired with preferred names.8 WinNonlin User’s Guide Preferred name: Edit output parameter names as desired. The names cannot contain space characters. for example: MODEL202. Each line contains: 180 .NAM and MODELURINE. number of decimal places. if it exists. As the parameters vary by model and with steady state dosing. Default preferred parameter names Preferred parameter names can be set by a map file containing individual preferences or company standards. MODEL202SS.NAM. Precision: enter the number of significant digits or decimal places as (selected under Sig. Click on a cell under Include in Workbook to toggle between Yes (include that parameter) and No (exclude). save parameter information for specific models to different files. unless parameter names are set by company defaults (see “Default preferred parameter names” below). All settings on the Parameter Names tab can be saved to a text file for later reuse by clicking the Save button (and loaded using the Load button). Digits) for the parameter in that row. Case will be preserved.MAP and load preferred parameter names from that file. Include in workbook: WinNonlin provides the option to include only a subset of final parameters in NCA workbook output. it will search its installation directory for an ASCII text file called DEFAULTNCAPARAMETERNAMES.NAM. or nothing as the precision for that parameter. Sig digits: Click on a cell in this row to select whether to enter significant digits. When WinNonlin opens the Parameter Names tab for an NCA model. Noncompartmental Analysis Computational rules for WinNonlin’s noncompartmental analysis engine 8 ParameterName=PreferredName. 181 . Digits” or “Dec. Places” followed by a comma. Computational rules for WinNonlin’s noncompartmental analysis engine WinNonlin’s NCA computations are covered under the following headings. PreferredName is limited to alphanumeric characters and the underscore. Precision equals “Sig. then an integer indicating the number of significant digits or decimal places to use. • • • • • “Data checking and pre-treatment” on page 181 “AUC calculation and interpolation formulas” on page 183 “Partial areas” on page 184 and “Therapeutic response windows” on page 186 “Sparse sampling computations” on page 188 “NCA output parameters and their computation formulas” on page 506 Data checking and pre-treatment Prior to calculation of pharmacokinetic parameters. Precision where: ParameterName is WinNonlin’s default name for an NCA parameter. WinNonlin institutes a number of data-checking and pre-treatment procedures as follows. Include equals “Yes” or “No” and indicates whether to include that parameter wherever NCA parameters are output. Include. the lower time point is checked. IV bolus data: The program will perform a log-linear regression of first two data points to back-extrapolate C0. • Data exclusions NCA will automatically exclude unusable data points. Data points preceding the dose time: If an observation time is earlier than the dose time. Drug effect data: An effect value equal to the user-supplied baseline is inserted at dose time. if either the X. it is used in the regression. i.or Y-value is missing. Data points for urine models: In addition to the above rules.8 WinNonlin User’s Guide Sorting Prior to analysis. then the first observed positive y-value will be used as an estimate for C0.: • Missing values: For plasma models. for steady-state. For urine models. the associated record is excluded from the analysis. if the lower time is greater than or equal to the upper time. the minimum observed during the dose interval.e. data within each profile is sorted in ascending time order. that observation is excluded from the analysis. drug concentration. the data must be corrected before NCA can run. • • Data deficiencies that will result in missing values for PK parameters WinNonlin reports missing values for PK parameters in the following cases. or if the concentration or volume is negative. If a weighting option was selected. WinNonlin inserts a value using the following rules: • • Extravascular and infusion data: For single dose data. a concentration of zero. records with missing values for any of the following are excluded: collection interval start or stop time. 182 .) Insertion of initial time points If a PK profile does not contain an observation at dose time (C0). Dose time must be non-negative. For urine models. If the regression yields a slope >= 0 or at least one of the first two y-values is 0. (A profile is identified by a unique combination of sort variable values. or collection volume. it halts the NCA analysis and issues an error message. No output is generated. Percent_Recovered Bolus dosing and <=1 non-missing observation after dose time all final parameters Blood/plasma data with all zero values except one non-zero value at dosing time Urine data with all urine volumes equal to zero Urine data with all concentrations equal to zero Multiple observations at the same time point For NCA.Noncompartmental Analysis Computational rules for WinNonlin’s noncompartmental analysis engine 8 Profile issue(s) Profile with no non-missing observations after dose time Blood/plasma data with no non-zero observation values Parameters reported as missing all final parameters All final parameters except Cmax. Amount_Recovered. If it detects more than one. AUMClast all final parameters all final parameters All final parameters except Max_Rate. AUCall. AUC calculation and interpolation formulas Linear trapezoidal rule t2 t1 AUC t2 t1 C1 + C2 = δt × ------------------2 AUMC t1 × C1 + t2 × C2 = δt × --------------------------------------2 183 . WinNonlin permits only one observation per profile at each time point. AUClast. ⎠ C1 C1 where δt is (t2-t1).⎞ ⎝ C 1⎠ AUMC t2 t1 t2 × C2 – t1 × C1 C2 – C1 2 = δt × --------------------------------------.– δt × -----------------2 C2 ⎛ ----. then the linear trapezoidal rule will apply for that interval.8 WinNonlin User’s Guide Logarithmic trapezoidal rule AUC t2 t1 C2 – C1 = δt × -----------------C2 ln ⎛ ----. Linear interpolation rule (to find C* at time t*) t* – t 1 C* = C 1 + -------------. Partial areas Rules for interpolation and extrapolation Partial areas within observed data ranges are calculated as described under “AUC calculation and interpolation formulas” on page 183 and “Calculation 184 .⎞ ⎛ C 2⎞ ln ⎝ -⎠ ln ⎝ ----. or C1 = C2. C2 <= 0.× ( ln ( C 2 ) – ln ( C 1 ) )⎞ ⎝ ⎠ t2 – t1 Note: If the logarithmic interpolation rule fails in an interval because C1 <= 0. Note: If the logarithmic trapezoidal rule fails in an interval because C1 <= 0. or C1 = C2. then the linear interpolation rule will apply for that interval. C2 <= 0.( C 2 – C 1 ) t2 – t1 Logarithmic interpolation rule t* – t 1 C* = exp ⎛ ln ( C 1 ) + -------------. where C0 is the minimum value between dose time and tau. then a linear or logarithmic interpolation is done to estimate the corresponding Y. If both the start and end time for a partial area fall after the last numeric observation.Noncompartmental Analysis Computational rules for WinNonlin’s noncompartmental analysis engine 8 methods for AUC” on page 177. Both the start and end time for the partial area must be at or after the dosing time. the corresponding Y value is interpolated between the first data point and C0. according to the AUC Calculation method selected in the Model Options dialog. and except for models 200 and 202 at steady state. The following rules apply if some portion of the partial area falls after the last valid observation: • If a start or end time falls before the first observation. C0=0 except in IV bolus models (model 201). 185 . then the log trapezoidal rule will be used for the area from the last observation time to the end time of the partial area. the partial area will not be calculated. because a logarithmic decline must be assumed in order to obtain the predicted values.. • • • Partial area boundaries The end time for the partial area must be greater than the start time. where C0 is the dosing time intercept estimated by WinNonlin. then the log trapezoidal rule will be used. If a start or end time falls within the range of the data but does not coincide with an observed data point. (See “NCA settings” on page 177. not “missing” or “BQL”) and Lambda Z is estimable.e. Lambda Z is used to estimate the corresponding Y: Y = exp(alpha – Lambda-Z * t) = exp(alpha – Lambda-Z * tlast) * exp(–Lambda-Z * (t-tlast)) = (predicted concentration at tlast) * exp(–Lambda-Z * (t-tlast)) • • If a start or end time falls after the last numeric observation and Lambda Z is not estimable.) If a start or end time occurs after the last numeric observation (i. If the start time for a partial area is before the last numeric observation and the end time is after the last numeric observation. e. This is equivalent to these formulae: AUCLow = AUCLow + ylower * (ti+1 – ti) AUCDur = AUCDur + (yupper – ylower) * (ti+1 – ti) AUCHgh = AUCHgh + AUC* – yupper * (ti+1 – ti) If there was one boundary crossing. t*) is computed following the user’s specified AUC calculation method. If no boundary crossing occurred. Call this AUC*. the rectangle that is under the lower boundary is added to AUCLow.. To interpolate to get time t* in the interval (ti. or AUCHgh. i. the same rule is used as when inserting a point for partial areas. if the concentration values for this interval occur above the upper boundary. the difference ti+1– ti is added to the appropriate parameter TimeLow. The difference t* – ti is added to the appropriate time parameter 186 . depending on which window the concentration values are in.8 WinNonlin User’s Guide Therapeutic response windows For therapeutic windows. or TimeHgh. To compute these values. AUCDur. ti+1) and called the boundaries ylower and yupper. including the inserted point at dosing time if there is one. an interpolation is done to get the time value where the crossing occurred. for each pair of consecutive data points. The parts of this AUC* are added to the appropriate parameters AUCLow. ti+1) at which the concentration value is yw: Linear interpolation rule for time: t* = ti + [ (yw – yi) / (yi+1 – yi) ] * (ti+1 – ti) Log interpolation rule for time: t* = ti + [ (ln(yw) – ln(yi) ) / ( ln(yi+1) – ln(yi) ) ] * (ti+1 – ti) The AUC for (ti. TimeDur. The AUC for (ti. For example. The interpolation is done by either the linear rule or the log rule following the user’s AUC calculation method as described above in section 3. one or two boundaries for concentration values can be given. it is determined if a boundary crossing occurred in that interval. and the program computes the time spent in each window determined by the boundaries and computes the area under the curve contained in each window. the rectangle that is between the two boundaries is added to AUCDur and the piece that is left is added to AUCHgh. ti+1) is computed following the user’s specified AUC calculation method as described above in section 3. call this t*. Call the pair of time values from the data set (ti. If ylast is in the middle window for two boundaries: InfHgh values are the same as Hgh values. If the last concentration value in the data set is zero. the AUC rule after tlast is the log rule. ti+1). If lambdaZ is not estimable. Otherwise. AUCInfLow = AUCLow + ylast / lambdaZ. Then the process is repeated for the interval (t*. t**). Then the times and parts of AUC are added in for the interval (t**. and the time and parts of AUC are added in for the interval (t*. This is equivalent to: t* = (1/lambdaZ) * (alpha – ln(ylower)) TimeInfDur = TimeDur + (t* – tlast) AUCInfDur = AUCDur + AUC* – ylower * (t* – tlast) AUCInfLow = AUCLow + ylower * (t* – tlast) + ylower / lambdaZ 187 . For any of the AUC calculation methods. AUCInfLow is missing. (only if the last concentration was zero) If ylast is in the lower window: InfDur and InfHgh values are the same as Dur and Hgh values.Noncompartmental Analysis Computational rules for WinNonlin’s noncompartmental analysis engine 8 and the parts of the AUC are added to the appropriate AUC parameters. Add time from tlast to t* into TimeInfDur. add the rectangle that is under the lower boundary into AUCInfLow. If lambdaZ is not estimable. InfLow and InfDur parameters are missing. an interpolation is done to get t* as above for the first boundary crossing. the zero value was used in the above computation for the last interval. and add the piece that is left into AUCInfDur. ti+1). Add the extrapolated area in the lower window into AUCInfLow. If there were two boundary crossings. Compute AUC* from tlast to t*. t*) as above. Otherwise: Use the extrapolation rule to get t* at the lower boundary concentration. The extrapolated times and areas after the last data point are now considered for the therapeutic window parameters that are extrapolated to infinity. ti+1). The extrapolation rule for finding the time t* at which the extrapolated curve would cross the boundary concentration yw is: t* = (1/lambdaZ) * (alpha – ln(yw)). Then another interpolation is done from ti to ti+1 to get the time of the second crossing t** between (t*. and the time and parts of AUC are added in for the interval (ti. unless an endpoint for the interval is negative or zero in which case the linear rule must be used. but now ylast is replaced with the following if Lambda Z is estimable: ylast = exp(alpha – lambdaZ * tlast). add the rectangle that is in the middle window into AUCInfDur. the above procedure is followed and the InfDur parameters become the InfHgh parameters. The NCA treats this sparse data as a special case of plasma or urine concentration data.8 WinNonlin User’s Guide If ylast is in the top window when only one boundary is given. by taking the mean concentration value for each unique time value for plasma data. add the rectangle that is under the lower boundary into AUCInfLow. Use the extrapolation rule to get t** at the lower boundary concentration value. Add time from t* to t** into TimeInfDur. all extrapolated parameters are missing. rather than actual time. Add time from tlast to t* into TimeInfHgh. The standard error of the data is also calculated for each unique time or midpoint value. This is equivalent to: t* = (1/lambdaZ) * (alpha – ln(yupper)) t** = (1/lambdaZ) * (alpha – ln(ylower)) TimeInfHgh = TimeHgh + (t* – tlast) TimeInfDur = TimeDur + (t** – t*) AUCInfHgh = AUCHgh + AUC* – yupper * (t* – tlast) AUCInfDur = AUCDur + (yupper – ylower) * (t* – tlast) + AUC** – ylower * (t** – t*) AUCInfLow = AUCLow + ylower * (t* – tlast) + ylower * (t** – t*) + ylower / lambdaZ Sparse sampling computations The Noncompartmental Analysis (NCA) module provides special methods to analyze concentration data with few observations per subject. add the rectangle that is under the lower boundary into AUCInfLow. otherwise: Use the extrapolation rule to get t* at the upper boundary concentration value. Add the extrapolated area in the lower window into AUCInfLow. Compute AUC* for tlast to t*. Compute AUC** from t* to t**. for these analyses. it is recommended to use nominal time. It first calculates the mean concentration curve of the data. and add the piece that is left into AUCInfHgh. If there are two boundaries and ylast is in the top window: If lambdaZ is not estimable. and add the piece that is left into AUCInfDur. For this reason. 188 . or the mean rate value for each unique midpoint for urine data. i = m m = last observation time for AUCall. Specifically: ˆ SE ( AUC ) = ˆ Var ( AUC ) Since AUC is calculated by the linear trapezoidal rule as a linear combination of the mean concentration values. m ˆ AUC = where ∑ wi Ci i=0 C i = the sample mean at time i ⎧ ( t1 – t0 ) ⁄ 2 . and will account for any correlations in the data resulting from repeated sampling of individual animals. the sample standard error of the y-values at Tmax. For plasma data (models 200-202). i = 0 ⎪ wi = ⎨ ( ti + 1 – ti – 1 ) ⁄ 2 .Noncompartmental Analysis Computational rules for WinNonlin’s noncompartmental analysis engine 8 Using the mean concentration curve. and for the area under the mean concentration curve from dose time through the final observed time (AUCall). it will use the subject information to calculate standard errors that will account for any correlations in the data resulting from repeated sampling of individual animals. NCA will calculate all of the usual plasma or urine final parameters listed under “NCA output parameters and their computation formulas” on page 506. Standard error of the mean AUC will be calculated as described in Nedelman and Jia (1998). Standard error of the mean Cmax will be calculated as the sample standard deviation of the y-values at time Tmax divided by the square root of the number of observations at Tmax. … . using a modification in Holder (2001). or time of last measurable (positive) mean concentration for AUClast 189 . In addition. NCA will calculate the standard error for the mean concentration curve’s maximum value (Cmax). i = 1 . m – 1 ⎪ ⎩ ( tm – tm – 1 ) ⁄ 2 . or equivalently. ⎠ ⎝ 1 – ---. the area under the mean rate curve through the final observed rate.8 WinNonlin User’s Guide it follows that m ˆ Var ( AUC ) = where ∑ i=0 2 2 w i w j r ij s ij wi si ---------. An added constant in AÛC will not change Var(AÛC). NCA uses the unbiased sample covariance estimator. Note that the inserted C0 is treated as a constant even when it must be esti- 190 . w0 is multiplied by C0.⎠ ri rj ( C ik – C i ) ( C jk – C j ) For urine models (models 210-212). When computing the sample covariances in the above. In other words.0) is inserted. so SE_AUClast and SE_AUCall also will not change.vii) in Nedelman and Jia (1998). the AUC estimate contains an additional constant term: C* = C0(t1-t0)/2. the standard errors are computed for Max_Rate.+ 2 ∑ --------------------ri ri rj i<j r ij = number of animals sampled at both times i and j r i = number of animals sampled at time i s i = sample variance of concentrations at time i s ij = sample covariance between concentrations C ik and C jk for all animals k that are sampled at both times i and j. the maximum observed excretion rate. The above equations can be computed from basic statistics. and appear as equation (7. For cases where a non-zero value C0 must be inserted at dose time t0 to obtain better AUC estimates (see “Data checking and pre-treatment” on page 181). and for AURC_all. which can be found as equation (A3) in Holder (2001): r ij 2 s ij = k = 1 ( r ij – 1 ) ∑ --------------------------------------------------------------r ij⎞ ⎛ r ij⎞ ⎛ + ⎝ 1 – ---. instead of being multiplied by 0 as occurs when the point (0. then sij is set to 0 (zero). The AUC’s must be calculated using one of the linear trapezoidal rules. For the case in which rij = 1.) 191 . (Select as described under “NCA settings” on page 177. and ri = 1 or rj = 1. so that the variances and covariances of those other data points are not duplicated in the Var(AÛC) computation.Noncompartmental Analysis Computational rules for WinNonlin’s noncompartmental analysis engine 8 mated form other points in the data set. 8 WinNonlin User’s Guide 192 . Custom models are covered under “User Models” on page 217. Both model types can be used in scripts or loaded via the PK/PD/NCA Analysis Wizard. “Output” on page 210 WinNonlin provides an extensive library of models.LIB). ASCII models are human-readable and easily customized from within the WinNonlin interface. “Loading the model” 2. “Model properties” on page 196 3. To load a model: 1.DLL) files and ASCII text models (*. Custom compiled models can be created using FORTRAN or C++. Loading the model The WinNonlin uses two types of models: compiled (machine language) dynamic link library (*. Open the data to which the model will be fit in a WinNonlin worksheet. “Model options” on page 203 4. detailed under “WinNonlin Model Libraries” on page 503. “Running the model” on page 209 5. 193 . Compiled models provide faster performance.TXT or *. 1.Chapter 9 Modeling Model setup and output Setup and output for compartmental modeling in WinNonlin are covered under the following headings. Dissolutions models to assist in evaluation of in vivo-in vitro correlations for formulation science. Note: Hidden data sets are not available.) Note: The image above shows the wizard as it appears for a user who has an IVIVC licence installed. or first-order input. Models to perform linear regression. Geometric characterization of drug effect data or blood (plasma) or urine concentration data for IV infusion. Click the PK/PD/NCA Analysis Wizard button on the tool bar. (See “Hidden windows” on page 18. Select the type of model to load from those listed in Table 9-1. Table 9-1. WinNonlin model types Model type Sigmoidal/dissolution models Linear models Noncompartmental analysis 194 Description Available only to users who have purchased and installed a WinNonlin IVIVC license. 3. Click the links in the table for a listing of models and model details. Make sure that the correct workbook is selected at the top of the dialog. The model will be associated with the active worksheet in that workbook. Other users will not see the IVIVC models.9 WinNonlin User’s Guide 2. IV bolus. 4. . or choose Tools>PK/PD/NCA Analysis Wizard from the WinNonlin menus. Compiled models run faster. For pharmacokinetic or pharmacodynamic models. for ASCII models. 8. 6. 9. Click Next. stored as ASCII libraries in WINNONLN\LIB. 7. 5. Each includes a different PK model and PD model 105. stored as ASCII library files in WINNONLN\LIB\MM. specify whether to use the ASCII or a Compiled version. written and saved to an ASCII text file or compiled using C++ or FORTRAN. WinNonlin model types (continued) Model type Pharmacokinetic models Pharmacodynamic models PK/PD link models Description 1 to 3 compartment models with 0th or 1st order absorption. with or without a lag time to the start of absorption.LIB. See also “Creating a new ASCII model” on page 218. ASCII models appear in a WinNonlin text window and are easily customized. Indirect response models Michaelis-Menten models Simultaneous PK/ PD link models User Models Command file or Model Object Note: Only one model may be loaded in WinNonlin at a time. data and model settings. Select the model or. Four Michaelis-Menten (saturable elimination) models.Modeling Loading the model Table 9-1. Eight variations on Emax and Sigmoid Emax models.PMO) files can load ASCII or compiled user models. based on measured drug effect (inhibition or stimulation of response) and parameters defining either the build-up or dissipation of that effect. then loaded using this option. These models can be based on any WinNonlin PK model. Click Next. 195 . Simultaneously fit PK and PD data. Four indirect response models. 9 Any combination of WinNonlin’s compiled PK and PD models. enter the model information as described under “User model parameters” on page 219.WMO) files from prior WNL versions and Pharsight Model Object (*. and are available only in compiled form. Click Next. For compiled user models and new ASCII user models. Legacy command files and WinNonlin Model Object (*. treating PK parameters as fixed and generating concentrations at the effect sites for use by the PD models. load the model library file and enter the model number. Custom models. Study Number or Gender might be included as a carry-along variable to make it available for post-modeling analyses. 7. Drag and drop to select one or more sort variables (optional) to identify individual data profiles. Model parameters set parameter names and initial parameter estimates. 2. with a description of the model. To specify the roles of data columns in the model: 1. Proceed to set the Model properties. The Selection Summary dialog appears. Model options select output formats. Select the independent variable in the Variables list and drag it to the X Variable field. For example. Click Finish to load the selected model. as applicable to the model. the following can be set: • • • • Data variables tell the model which data columns to use. a crossover study might include Subject and Treatment as sort variables. For example. Dosing regimen or Dosing data variables set dose times and/or amounts. Choose Model>Data Variables from the WinNonlin menus or click the Data Variables button in the WinNonlin tool bar.9 WinNonlin User’s Guide 10. Simulations: No Y variable is needed for simulations. 5. WinNonlin performs a separate analysis for each unique combination of sort variable values. instead check the Simulate Data check box and enter units (optional) for the Y variable. 196 . Data variables Modeling requires that the model variables be identified before other settings. Load the model as described under “Loading the model” on page 193. and drag it to the Y Variable field. computational algorithms and units. 4. Select carry along variables (optional) to include in the analysis output. 6. Model properties Once a model has been loaded. Select the dependent variable. 3. and 2 for urine data. WinNonlin can compute initial estimates via curve stripping for single-dose compiled PK. 2. For example. every parameter requires both upper and lower bounds. WinNonlin Generated Initial Parameter Values with WinNonlin Bounds: WinNonlin will curve strip for initial estimates. Fixed parameters use a set value. the user must supply initial estimates or provide boundaries to be used by WinNonlin in creating initial estimates via grid search. Enter the initial estimate for each parameter in the column labeled Initial. The Model Parameters tab of the Model properties dialog appears. The FIXED command can be used for both modeling and simulating. To set up the initial parameter estimates: 1. This reduces the number of parameters in the model. and select WinNonlin or user-supplied bounds. • User Supplied Initial Parameter Values. The function variable might have the value 1 for plasma data. etc. See the note. Note: Data corresponding to the first (sorted) value of the function variable will be used with the first function. linear or IVIVC models. so are no longer considered parameters. the data set may include data on two compartments—plasma and urine. When user-supplied boundaries are used. Select Model>Model Parameters from the menus or click the Model Parameters tool bar button. Model parameters All iterative estimation procedures require initial estimates of the parameters. with the second function. PD. 9. If curve stripping 197 • . Select the function variable. For IVIVC models only (IVIVC license required). the second. below. If a model fits multiple functions simultaneously. some parameters may be given fixed values while others are estimated. if applicable for the model. Click OK and proceed to set the Model parameters. a function variable is required to tell WinNonlin which values of the Y variable apply to which function.Modeling Model properties 9 8. Statistics such as VIF are not computed for fixed parameters. PK/PD link. then produce boundaries on the model parameters for model fitting. For all other situations. Specify the type of Parameter Calculation and Boundaries. and so forth for each successive sort level. The Save and Load buttons can be used to store and re-use parameter settings in an external ASCII text file. To fix that parameter for all sort levels. When all values are set. This option is available when sort variables are included in the Data variables. For IVIVC models. To fix a parameter for a given sort level. When it is checked. 3. Propagate Final Estimates. some (but not all) parameters may be set to fixed values rather than estimated. The same type of parameter calculation and boundaries apply for all sort levels. but also limit the estimates during modeling. Multiple sort levels If the data has one or more sort variables. The final parameter estimates from the first sort level provide the initial estimates for the second. Enter the fixed value in the adjacent cell. Fixed parameters. • WinNonlin Generated Initial Parameter Values with User-Supplied Bounds: WinNonlin will do curve stripping for initial estimates. 6. This can be valuable if the values become unrealistic or the model won't converge. click in the cell in the Fixed or Estimated column and choose Fixed from the drop-down list. click the Fix all <Parameter Name> button at the lower right. 5. initial estimates and bounds are entered or calculated only for the first sort level. initial estimates must be provided for each level of the sort variable(s) unless Propagate Final Estimates is selected. See “Special file formats” on page 430. 198 .9 WinNonlin User’s Guide fails. click Apply (to leave the dialog open and Proceed to the Dosing regimen) or OK (to close the dialog). Note: Parameter boundaries not only provide a basis for grid searching initial parameter estimates. See also “Notes on initial parameter estimates and bounds” below. then grid search if that fails. 4. The structure of this file (PARMFILE) is described in the chapter on Scripting. the program will terminate. as WinNonlin cannot grid search for initial estimates without user-supplied boundaries. With the Nelder-Mead option (METHOD 1). boundaries are required. boundaries are required. When the Gauss-Newton method is used to estimate the parameters (METHODS 2 and 3). It is still recommended that the lower and upper boundaries be specified. even when WinNonlin is computing the initial estimates. Hold the cursor in the lower right-hand corner of this selection until the cursor turns into a small black + called a handle. the grid will comprise 34 = 81 possible points. Method 3 (Gauss-Newton (Hartley)) and Do not use bounds are recommended. For any model other than a compiled WinNonlin model. Enter values for the first profile. In this case. the residual sum of squares is set to a large number if any 199 • • • • • . Thus. WinNonlin will perform curve stripping for the corresponding macro constant model then use the macro constants to compute the micro constants. a grid search is performed to obtain initial estimates. The point on the grid associated with the smallest sum of squares is used as an initial estimate for the estimation algorithm. if WinNonlin is requested to perform initial parameter estimates (i. For linear regressions. The grid used in grid searching includes three values for each parameter. 3. but it may dramatically slow the parameter estimation process if the model is defined in terms of differential equations. Notes on initial parameter estimates and bounds • • The use of boundaries (user or WinNonlin supplied) is recommended. if four parameters are to be estimated. grid searching is requested).e. then select (highlight) all cells for the first profile with the mouse. the boundaries are incorporated by applying a normit transform to the parameter space. If multiple dose data is fit to a WinNonlin library model and WinNonlin makes initial parameter estimates. If the data are to be fit to a micro constant model. Release the mouse button..Modeling Model properties To copy model parameter information from the first profile to the remaining profiles: 9 1. 2. 4. WinNonlin uses curve stripping only for single-dose data. Note that this option has been added for convenience. Drag the handle over the target cells (to the bottom of the grid). WinNonlin dose information can be entered in the Regimen Dialog or read from worksheet columns using the Dosing Data Variables dialog (see “Dosing data variables” on page 202). The use of bounds with Nelder-Mead will. 11.) 2. 9.9 WinNonlin User’s Guide of the parameter estimates go outside the specified constraints at any iteration. which forces the parameters back within the specified bounds. Dosing regimen WinNonlin uses model constants to store dosing information. including the number of doses. 3-7. 12 # of doses dose 1 dose 2 (etc. 200 . 13. 19 # of doses dose 1 end time 1 start time 2 end time 2 (etc. 18 stripping dose # doses dose 1 time of dose 1 dose 2 time of dose 2 (etc. keep the parameters within the bounds. • Unlike the linearization methods. however. Note that PD and IVIVC models do not require dosing. 10. The use of bounds (either user-supplied or WinNonlin supplied) is always recommended with the Gauss-Newton methods. amount of the dose(s). and ND = number of administered doses. Specific models require the information described below. and dosing time(s). For some models the number of constants depends upon the number of doses administered. 14.) 15.) 8. Model Number Constant 1 2 3 4 5 6 7 8 9 NCON (2*ND)+1 (3*ND)+1 (2*ND)+2 3 4 1. NCON = the total number of constants to be specified by the user. the use of bounds does not affect numerical stability when using the Nelder-Mead simplex option. 16 bolus dose IV dose infusion length 17 stripping dose bolus dose IV dose infusion length time of dose 1 start time 1 time of dose 2 dose 2 For the table above. Note: If different profiles require different numbers of doses. dose normalization of kg would change those units to L/kg. enter the constants required for all levels of the sort variable(s). enter Current Units of mg and Dose Normalization of kg. Include time values that fall within the data but beyond the last real dose for these phantom doses. For example. Select Model>Dosing Regimen from the WinNonlin menus or click the Dosing Regimen button in the tool bar. if doses are in milligrams per kilogram of bodyweight. This will affect output parameter units. and enter doses of 0 for the extra doses. If a sort variable has been specified. Compiled models 2. The Dosing Regimen tab of the Model properties dialog appears.Modeling Model properties To enter dosing data in the Dosing Regimen dialog: 9 1. 4. Dose normalization affects units for all volume 201 . if Volume units were L. 3. select the appropriate factor in the Dose Normalization drop-list. If dose amount is normalized by subject bodyweight or body mass index. For example. enter the largest number. Enter units for the dose under Current Units. or select the unit prefix and unit below and click the Add button. Note: Use the Save and Load buttons to save the dosing values and units to an external ASCII file and reload settings for re-use. If the data set includes multiple profiles. instead of entering it manually in the dialog. 2. and Delete Row buttons. Un-check No Source. the Constant column of the Dosing Regimen dialog contains the names of elements of the CON array. Enter each constant value for each profile in the column labeled Value. to size the grid appropriately for the number of constants required by the model. 4. a row will be added for each profile.. 2.9 WinNonlin User’s Guide and clearance parameters in NCA and PK models. The structure of this file is described under “Special file formats” on page 430 4. Dosing data variables Use this dialog to extract dosing information from worksheet columns. Choose Model>Dosing Data Variables. Add Row. See also: • • “Dosing regimen” on page 200 for compartmental models “Dosing regimen” on page 171 for noncompartmental analysis To select columns containing dose information: 1. Click OK or Apply to enter the dosing information. Use the Insert Row. if needed. Open the worksheet with dose information in WinNonlin. ASCII models For ASCII models. and set the model variables as described under “Data variables” on page 196.. and Percent_Recovered in NCA urine models. from the WinNonlin menus or click the Dosing Data Variables tool bar button. Load the data and model in WinNonlin. 1. 3. as well as AUC/D in NCA plasma models. 3. 202 . Check the ASCII text in the Model window to see how many and which constants to enter. and column containing the dose-interval values (Tau). Select the column containing dose times under Time. Select the column containing dose amounts under Dose. respectively. 7. 203 . worksheet.Modeling Model options 9 5. Model options The options for compartmental modeling output and computational algorithm appear on five tabs in the Model Options dialog. Select the workbook and worksheet containing the dose information under Workbook and Worksheet. For noncompartmental analyses of steady-state data. 8. 9. 6. check Steady State and open the Tau Mappings tab. Click OK. Select the workbook. Output options The defaults output options are set in the WinNonlin Options dialog (see “WinNonlin options” on page 28).9 WinNonlin User’s Guide To set compartmental modeling options: 1. Select Model Options from the Model menu or click the Model Options tool bar button. The Output Options tab of the Model options dialog. Workbook: Tabular output in a workbook. Note: When attempting a difficult fit. 2. Output options Weight Title PK settings Units Select text. Text: This provides an ASCII (text) version of the output. Chart: This option generates plots showing predicted values versus observed values for each level of the ID variable. Page Breaks: Checking this starts a new page at each profile in text output. as well as any messages or errors that occur during modeling. see “Output” on page 210. below. containing all modeling settings and output in one file. Select the desired options using the following tabs. Text title to appear at the top of modeling output (text and charts) Model fitting algorithm Preferred output units for parameter estimates 3. Check or un-check the following options to include or exclude the output. For the contents of the workbook output. overrides the defaults for the current modeling session. The Model Options dialog appears. 204 . chart (plots) and/or workbook output forms Weighting of observation data. use the ASCII output to check for errors. Click OK or Apply. Select from the options discussed below. Programming statements can be used to define weights as part of a user defined model. Input the value of n in the box provided. scaling of the weights provides increased numerical stability. See “User Models” on page 217. Include Weight as a variable on the data file allows specific weight values to be applied to each observed value. Pre-Selected Scheme: to select from among the most common schemes: – weight by 1/observed Y – weight by 1/observed Y2 – weight by 1/predicted Y (iterative reweighting) – weight by 1/predicted Y2 (iterative reweighting) Observed to the Power n: to use weighted least squares.Modeling Model options 9 Weight For compartmental modeling. Iterative reweighting sets weights as the predicted value of Y to a specified power. See “Weighting” on page 241 for a detailed discussion of weighting schemes. Scaling of Weights: When weights are read from the data file or weighted least squares is used. other than uniform weighting: 1. 4. Weighted least squares weights by a power of the observed Y. However. 2. To set the weighting scheme for a modeling session: 1. Open the Weight tab of the Model options dialog (Model>Model Options). Input the value of n in the box provided. 2. there are four ways to assign weights. and click OK. 205 . 3. See “Scaling of weights” on page 242 for further discussion. Weights on File in Column: to read weights from a column on the data file. the weights for the individual data values are scaled so that the sum of weights for each function is equal to the number of data values (with non-zero weights). This scaling has no effect on the model fitting. as the weights of the observations are proportionally the same. Predicted to the Power n: to use iterative reweighting. with the Final Parameters worksheet displaying non-transposed data and a second worksheet. A second worksheet. the Final Parameters output worksheet will show each parameter in a separate column. use Method 3 without bounds. For linear regressions. only when it is printed. If this option is not checked. called Transposed Final Parameters containing the transposed version. check the No Scaling of Weights check box. The title also appears in workbook output.9 WinNonlin User’s Guide To to turn off scaling of weights. The use of bounds is especially recommended with Methods 2 and 3. the two worksheets will be reversed. Unchecking the Title check box suppresses the titles without deleting the title text from the Model options dialog. check Title and enter the title text below. Transpose final parameter table If this option is checked. will present each parameter in a separate row. Enter multiple lines using either CTRL + ENTER or SHIFT + ENTER to move to the next line. To include titles. PK settings The PK Settings tab of the Model options dialog (Model>Model Options) provides control over the model fitting algorithm and related settings. 206 . called Non-transposed Final Parameters. The transposed form is particularly useful when creating final report tables or when summarizing the final parameters using descriptive statistics. Minimization Select among the 3 minimization methods: – Method 1: Nelder-Mead simplex – Method 2: Gauss-Newton with Levenberg and Hartley modification – Method 3: Gauss-Newton with Hartley modification These methods are described in “Model fitting algorithms and features of WinNonlin” on page 5. Title Titles can be displayed at the top of each chart and each page of ASCII text output. See “Models” on page 29 for examples of transposed and non-transposed output. 0001. 207 . When fitting or simulating multiple dose data. For all three types. The program estimates these derivatives using a difference equation. the predicted data for the built-in PK models includes dosing times. The minimum allowable value is 10. Number of predicted values This setting determines the number of points in the predicted data. in addition to the concentrations generated.e. Note: To better reflect peaks (for IV dosing) and troughs (for extravascular. the predicted data plots may be much improved by increasing the number of predicted values. this command is ignored and the number of predicted points is equal to the number of points in the data set.Modeling Model options 9 Increment for partial derivatives Nonlinear algorithms require derivatives of the models with respect to the parameters.000. differential equations are used). This value may be increased only if the model has no more than one independent variable. Use this option to create a data set that will plot a smooth curve. Convergence is achieved when the relative change in the residual sum of squares is less than the convergence criterion. The default for user models is the number of data points in the original data set. Convergence criterion Use this option to set the convergence criterion. for infusion models.= -------------------------------------------------------∂P ( 1 ) Δ where Δ = the Increment multiplied by the parameter value. the maximum is 20. The default is 0. If the number of derivatives is greater than 0 (i. IV infusion and IV bolus dosing). For example: ∂F F(P(1) + Δ) – F(P(1)) -------------.. concentrations are generated at dosing times. data are generated for the end of infusion. For compiled models without differential equations the default value is 1000. in addition. dosing must contain units to use this option. where applicable. Enter the units in the Preferred column for that row. WinNonlin performs appropriate conversions as needed to report output in the preferred units. the mean square needs to be set to a fixed value. Alternately. The default units are displayed in the Default column. Select the row with the appropriate parameter name. For Method 1 (Nelder-Mead simplex). click the Units Builder button to construct and specify units in the Units Builder. the default is 500.9 WinNonlin User’s Guide Mean square The variance is normally a function of the weighted residual SS/df. each iteration is really a reflection of the simplex. 2. Y-variable and. For Methods 2 and 3 (Gauss-Newton). To specify preferred units: 1. Units for output parameters will be derived from these. If the number of iterations is set to zero. the default is 50. or the Mean Square. 3. Model size may range from 1 to 64. the default size is 4. See “Linear models” on page 546 or “Sigmoidal/dissolution models” on page 547 for model details. For Method 1. 208 . when computing the variance. Note: Linear or IVIVC models set preferred units for the independent (x) or dependent (y) variable. all output will be use the initial estimates as the final parameters. For certain types of problems. The input data worksheet columns for the X-variable. Model size Use this option to alter the default memory requirements when using ASCII models only. Iterations Use this option to change the maximum number of iterations. Select the Units tab of the Model options dialog. Units Specify the preferred units for output of modeling or simulation. Error messages may appear here as guidance in resolving the problem. Note that it is possible that a specified model and input values may be correct. apply the following steps in determining what may have gone wrong. In this case. and click OK. Check the history file associated with the PK Data Output window.Modeling Running the model 9 4. or terminate normally and converge to unreasonable values of the parameter estimates. The preferred units can be saved to a (*. Troubleshooting modeling problems Occasionally when modeling. 209 . Running the model To start modeling: 1. but there simply may not be enough information contained in the data to precisely estimate all of the parameters in the model. Note: The Select Data for Modeling dialog will appear if the original data set selected for modeling has become un-associated with the model. Stop modeling To stop modeling: 1. In case of an error the parameters can be reset to their original setting by clicking on the Reset to Defaults button. Click the Stop Modeling tool bar button. If that is the case.TXT) file for later re-use. If these types of problems occur. select the data set to be used for the modeling. using the Save and Load buttons. If no PK Data Output window appears. Select Model>Start Modeling from the menus or click the Start Modeling tool bar button. check the Text (ASCII) output for error messages. the only recourse is to obtain additional data. 5. WinNonlin will terminate abnormally. search for the word ERROR in the text output. One way to do this is to rerun the problem as a simulation. it will be written to the text output (if the user requested text output). error messages and warnings are written to the text output. try using METHOD 1 (Nelder . Carefully check the input data set and constants to make sure the values are correct. which may indicate that the model is ill-defined.. If an error occurs that does not prevent completion of modeling. 210 . Output One of the examples from the Examples Guide is used here to discuss the modeling output provided by WinNonlin. Also make sure that the model has not attempted any illegal arithmetic computations such as taking the logarithm of zero or the square root of a negative number. carefully check the model definition statements to make sure that there are no undefined or uninitialized variables. not one of the models in WinNonlin's built-in library). check the Simulate check box in the Y Variable group box in the Data Variables dialog box. Errors When modeling is run. warning messages may also be generated to alert the user that WinNonlin is having to work very hard to obtain the parameter estimates.9 WinNonlin User’s Guide If the model was defined by the user (i. errors and warnings will be written there. If the problem recurs.e. Note: To locate error messages. text output will be created. whether or not the user requests it. In addition.Mead . To do this.via the Settings tab in the Model Options dialog box). try rerunning the problem and add LOWER and UPPER constraints. A plot of the data can be very helpful. If an error occurs that prevents completion of modeling. If these messages occur. The data set may be ill-conditioned or the initial parameter estimates may be too far from the true values. This example uses a pharmacokinetic (PK) model. Check the initial values of the parameters to see if they are reasonable. Are the predicted values from the simulation reasonably close to the observed data? Often error messages occur because WinNonlin is having difficulty fitting the model to the data. as well as several measures of fit. Arrays and Functions” on page 557 for an alphabetic catalog of WinNonlin commands. this page shows the parameter values and the computed weighted sum of squared residuals at each iteration.Rank of the matrix of partial derivatives of the model parameters. and any errors that occurred during modeling. 211 . residuals. depending which method is used for minimization. If the matrix is of full rank. there is not enough information contained in the data to precisely estimate all of the parameters in the model. then the problem is ill-conditioned. “Workbook output” on page 214: worksheets listing input data.Modeling Output 9 Note: This section is meant to provide guidance and references to aid in the interpretation of modeling output.) If the model is defined in a user library. If the RANK is less than the number of parameters. • • • “Text (ASCII) output” on page 211: text version of all model settings and output. Minimization Process The second page varies. In addition. including any errors that occurred during modeling. Model Commands The first page is a copy of the input commands used to run the model. “Chart output” on page 215: plots of observed and predicted data. After Example 1 is run. That is. modeling iterations and output parameters. and other quantities. and is not a replacement for a PK or statistics textbook. If a Gauss-Newton method is used (Methods 2 or 3). the following two values are printed for each iteration: • RANK . then the statements that define the model are also printed. up to three new child windows appear in the WinNonlin window. depending on the model run. Text (ASCII) output The text output window contains a complete summary of the modeling for this PK example. (See “Commands. They are discussed in the following sections. RANK = # of parameters. sum of squares) point. then the estimation problem is very ill-conditioned. and the next-to-best point. The confidence interval labeled UNIVARIATE is the parameter estimate plus and minus the product of the estimated standard error and the appropriate value of the t-statistic. using bounds (LOWER and UPPER constraints) often helps reduce the condition number. Both are 95% confidence intervals. COND is the square root of the ratio of the largest to the smallest eigenvalue of the matrix of partial derivatives. say greater than 10e6. and two confidence intervals based on this estimated standard error. When using Methods 2 and 3. See Chapter 10 of Draper and Smith (1981) for a complete discussion of the true confidence intervals for parameter estimates in nonlinear models. If the condition number gets to be very large. but one or more parameter estimates will have standard errors that are large relative to the estimate. Many times a model will provide a good fit to the data in the sense that all of the deviations between observed and predicted values are small. while the PLANAR confidence intervals take into account the correlations among the parameter estimates. or smallest. final (best. The confidence interval labeled PLANAR is obtained from the tangent planes to the joint confidence ellipsoid of all the parameter estimates. as they are based on a linearization of the nonlinear model. the second page only shows the parameter values and the weighted sum of squares for the initial. 212 .9 WinNonlin User’s Guide • COND .Condition number of the matrix of partial derivatives. The UNIVARIATE confidence intervals ignore the other parameters being estimated. The estimated standard errors and confidence intervals are only approximate. The nearer the model is to being linear. The estimated standard errors are valuable for indicating how much information about the parameters is contained in the data. For a complete discussion of the UNIVARIATE and PLANAR confidence intervals see page 95 of Draper and Smith (1981). If the simplex algorithm is used. Only in the very rare event that the estimates are completely independent (uncorrelated) will the two confidence intervals be equal. the asymptotic estimated standard error of each estimate. the better is the approximation. usually the PLANAR intervals will be wider than the UNIVARIATE intervals. Final Parameters The third page of the output lists the parameter estimates. the sum of weighted squared residuals. The eigenvalues are another indication of how well the data define the parameters. Correlation Matrix. High correlations among one or more pairs of the estimates indicate that the UNIVARIATE confidence limits are underestimates of the uncertainty in the parameter estimates and. and anything less than 0.9. weights. the estimates may not be very reliable. although the data may be well fit by the model. the standard deviations (S) of the calculated function values and standardized residuals. then the eigenvalues will all be equal. calculated predicted values of the model function. Unlike the linear model case. Two measures which have been proposed for comparing competing models. Residuals Page five of the output lists the observed data.Modeling Output 9 Variance-Covariance Matrix. the correlation is not a particularly good measure of fit. it is usually not possible to drop one parameter from a nonlinear model. then WinNonlin generates a data point with time and concentration values of zero. area under the curve (AUC) is printed at the end of page five of the text output. A large ratio between the largest and smallest eigenvalue may indicate that there are too many parameters in the model. AIC (Akaike (1978)) and SBC (Schwarz (1978)). and the correlation between observed and calculated function values. residuals. If the parameter estimates are completely uncorrelated. If the data have been fit to a WinNonlin library model for extravascular or constant infusion input. Also. to compute AUC from zero to the last time. an estimate of the error standard deviation. If the first time value is not zero. Kuh and Welsch (1980). in the sense that the Gauss-Newton algorithm will have trouble finding the minimum residual sum of squares. Runs of positive or negative deviations indicate non-random deviations from the model and are indicators of an incorrect model and/or choice of weights. Also on page five are the sum of squared residuals. are also printed on page five. The AUC is computed by the linear trapezoidal rule. Note that this AUC in the text output differs from the AUC on the 213 . Unfortunately. For discussion of the use of eigenvalues in modeling see Belsley. data sets that result in highly correlated estimates are often difficult to fit. and Eigenvalues Page four of the output shows the correlation matrix of the estimates and the eigenvalues of the linearized form of the model. for most problems it will be greater than 0.8 probably indicates serious problems with the data and/or model. As such. . units. The secondary parameters are functions of the primary parameters whose estimates are given on page three of the output. Iteration number. univariate confidence intervals. Dosing Correlation Matrix Eigenvalues Condition Numbers 214 . along with their estimated asymptotic standard errors. they represent about the third level of approximation and must be regarded with caution. The standard errors of the secondary parameters are obtained by computing the linear term of a Taylor series expansion of the secondary parameters. In the case of multiple-dose data. Iteration number. Table 9-2. and planar confidence intervals for each level of the sort variables. Workbook output A PK Workbook window contains summary tables of the modeling data. for each level of the sort variables. The dosing regimen specified for the modeling. and value for each parameter. and lower and upper bounds for each level of the sort variables. rank and condition number for each sort level. Summary of PK worksheets Item Initial Parameters Minimization Process Final Parameters Contents Parameter names. Parameter names. they. A correlation matrix for the parameters. are printed on page six. standard error of the estimates. Eigenvalues for each level of the sort variables. These worksheets present the output in a form that can be used for reporting and further analyses. initial values.9 WinNonlin User’s Guide Secondary Parameters tab of the tabular output. a summary of the information in the ASCII output. CV%. The secondary parameter AUC is calculated from the model parameters. for each sort level. estimates. weighted sum of squares. Secondary Parameters If there are any secondary parameters. secondary parameters that depend on dose use the first dose in their computation. Note: Note that the AUC values may not be meaningful if the data were obtained after multiple dosing or IV dosing (due to time 0 extrapolation problems). sum of squared residuals (SSR). Summary of PK worksheets (continued) Item Variance-Covariance Matrix Summary Tablea Contents 9 A variance-covariance matrix for the parameters. estimate of residual standard deviation (S) and degrees of freedom (DF). and maximum number of iterations allowed. and CV% for each sort level. Title. The value of the partial derivatives for each parameter at each time point for each value of the sort variables. the correlation between observed Y and predicted Y. minimization method. Y. Values of the differential equations at each time in the data set. weighted corrected sum of squared observations (WCSS). 215 . residual. Secondary parameter name. model number. weighting scheme. Use the tabs at the bottom of the window to select the worksheet to use. for each sort level. X. Time and predicted Y for multiple time points. transformed Y. then X and Y will equal X(obs) and Y(obs). Diagnostics for each function in the model and for the total: corrected sum of squared observations (CSS). standard error of predicted Y. also includes CP. Select graph type using the tabs at the bottom of the window. the weighted correlation. If there are no statements to transform the data. For link models. also includes CP and Ce. weight. for each sort level. units. transformed X. weighted sum of squared residuals (WSSR). for each sort level. and two measures of goodness of fit: the Akaike Information Criterion (AIC) and Schwarz Bayesian Criterion (SBC).Modeling Output Table 9-2. standardized residuals. Note: Worksheets produced will depend on the analysis type and model settings. The sort variables. convergence criterion. standard error of the estimate. Chart output This PK analysis produces a chart window with five graphs for each level of the sort variable. predicted Y. Diagnostics Differential Equations Partial Derivatives Predicted Data Secondary Parameters User Settings a. These output worksheets are presented together in one workbook window. estimate. For indirect response models. Scatter should lie close to the 45 degree line. Weighted Predicted Y vs. a. then df(x. Weighted Residual Y: used to assess whether error distribution is appropriately modeled throughout the range of the data. b. based on parameters a. Used for assessing the model fit. Observed Y and Predicted Y: plots the predicted curves as a function of X. b. a. b. c)/db. Each derivative is evaluated at the final parameter estimates. Weighted Predicted Y: plots weighted predicted Y against observed Y. and c. X vs. Weighted Residual Y: used to assess whether error distribution is appropriately modeled across the range of the X variable. c) is the model as a function of x.9 WinNonlin User’s Guide X vs. 216 . df(x. Data taken at larger partial derivatives are more influential than those taken at smaller partial derivatives for the parameter of interest. c)/da. Partial Derivatives: plots the partial derivative of the model with respect to each parameter as a function of the x variable. b. b. df(x. with the Observed Y overlaid on the plot. a. Observed Y vs. If f(x. a. c)/dc are plotted versus x. One or more models can be saved together in an ASCII text file (*.TXT) or library (*. ASCII user models ASCII user models are custom models with or without differential equations.LIB). and used in Pharsight Model Objects and scripts. See also Gabrielsson and Weiner (2001). Individual models may be up to 24 KB in size. especially for models with differential equations.LIB). a library with two models might contain the following. and saved as text (*. An ASCII library cannot exceed 30 KB. Example user models appear in this chapter and the WinNonlin Examples Guide.DLL) file. • • ASCII user models are written using WinNonlin commands and programming language.TXT) or model library files (*. In addition. written in WinNonlin commands. WinNonlin supports two types of custom models: ASCII and compiled user models. FORTRAN. each model begins with a MODEL n statement identifying the model by number. Compiling a model can significantly speed model fitting. For example. Compiled user models are created and compiled in Fortran or C++ into a dynamic link library (*. Within an ASCII model file. and ends with an end-of-model (EOM) statement. 217 . Either type can be loaded via WinNonlin’s PK/PD/NCA Analysis Wizard.Chapter 10 User Models Creating custom models using WinNonlin command language. or C++ In addition to its library of standard PK and PK/PD models. the ASCII and compiled models in the WinNonlin \LIB and \USERCOMP subdirectories serve as examples and templates. copy the library (*.10 User’s Guide WinNonlin MODEL 1 <model statements> EOM MODEL 2 <model statements> EOM To load the second model into WinNonlin. To use a WinNonlin library model as a template. 218 . create a template with the appropriate number of parameters and equations using WinNonlin’s PK/PD/NCA Wizard as follows. Alternately. then enter the model number and the full file path as shown below. Details on model templates and structure appear under: • • • “Creating a new ASCII model” below “Model structure” on page 220 “ASCII model examples” on page 226 Creating a new ASCII model ASCII user models can be created in any text editor. including a WinNonlin text window or Notepad.LIB) file from the WinNonlin \LIB subdirectory and edit the copy as needed. choose to load a User Model and Existing ASCII Model in the first two screens of the wizard. you would use the PK/PD/NCA Analysis Wizard (see “Loading the model” on page 193). This information is entered in the User Model Parameters dialog of the Wizard as follows.LIB). open the library file in a text window. commands and syntax detailed in this chapter and under “Commands. and copy and paste the new model text into the existing file.LIB. 6. look at the FORTRAN or C++ code from which the user model . Review the settings. WinNonlin will create a new text window containing a template for the model. 4. The Selection Summary dialog appears. 7. The ASCII Model Selection dialog appears. To complete the model parameters settings: 1. 1. Several models can be saved to one . Note: To ascertain this information for compiled user models. Select User Model in the Model Types dialog and click Next. (See Table 10-1 on page 221. To do this. and the number and names of estimated and secondary parameters. Edit as needed using the structure.LIB file.User Models ASCII user models 10 To create a new ASCII model from a template: 1. Enter the model information as detailed under “User model parameters”. 2. 5. Select Create New ASCII Model and click Next. User model parameters In order to create a new ASCII model template or use a compiled user model. Click Back to change settings or Finish to accept them. WinNonlin’s PK/PD/NCA Analysis Wizard needs to know the number of algebraic and differential equations in the model. For ASCII models. Arrays and Functions” on page 557. The Model Parameters dialog appears. to identify the model within a library. To save the model.) 219 . The default model library file name is USERASCI. enter the model number. so long as each uses a unique model number. It is often convenient to save under a new name that describes the model. 3. Launch the PK/PD/NCA Analysis Wizard by clicking on the PK/PD/NCA Analysis Wizard tool bar button or choosing Tools>PK/PD/NCA Analysis Wizard from the WinNonlin menus. an integer between 1 and 32767. save it as an ASCII file (*. Click Next.DLL was compiled. Enter the primary (estimated) and secondary (optional) parameters in the grid. How many parameters and constants will be needed to define the model? 5. Does the data set contain all the necessary information. Model structure The following questions need to be considered in building custom models: 1. If the model includes differential equations. should be included within the model to reduce the amount of information that must be supplied to WinNonlin when using the model. Does the model change with values of the data or estimated parameters? 4. Enter the number of algebraic functions in the model. dosage—to make the model as general as possible. Parameter names can include up to 8 alphanumeric characters. 4. outlined in Table 10-1. If the model uses differential equations. 3. what are their initial values? Information that is fixed across modeling sessions. or will the model require new variables that are functions of the input data? 3. or a combination of the two? Examples of algebraic nonlinear models Examples of differential equations WinNonlin Y = exp ( – a *t ) Y = a ⋅ cos ( t ) + b ⋅ sin ( t ) dz ⁄ dt = – k 1 t dz ⁄ dt = k 1 w – k 2 y 2. For compiled models. Note: The maximum number of variables and operators in a model is 2000. check Include differential equations and enter the number of differential equations.. An ASCII model structures this information in blocks. a system of differential equations. such as the number and names of model parameters.10 User’s Guide 2. Is the model expressible as an algebraic nonlinear function.g. use variables to represent model quantities that may vary across uses—e. 220 . In contrast. enter parameters in the order in which they appear in the model. Sets initial conditions for the system of differential equations. or store it in a multi-model library. Either REMARK or REMA indicates that the rest of that paragraph contains a note and is to be ignored by WinNonlin. The other blocks may appear in any order. N. Each block must end with an END statement. a model requires a MODEL statement. call it in a script. The function number is required even if there is only one FUNCTION block. Marks the end of the model. and begin analysis. Useful for automated (scripted) analyses. At minimum. Statements defining the differential equations are included in this END block.e. Optional WinNonlin commands to load the model. Number each consecutively. A function block must be included for each set of algebraic equations to be fit to data.User Models ASCII user models 10 Table 10-1. Marks the beginning of the model. variables that will be used in more than one model block. It must begin with MODEL and end with EOM. Any block or command name can be represented by its first four characters or its full name. For example. Required. Required. i. The model number. FUNCTION block and EOM statement. The number must be an integer from 1 to 32767 and unique for each model in a library. Model structure Block: MODEL N Description Required. Optional statements to alter data values and/or create new variables before the model is fit. Required if differential equations are included. SECONDARY statements END EOM Optional statements to define parameters that are functions of the estimated parameters. is required to load the model via the PK/PD/NCA Wizard. alter model settings. Variables that are defined within another block persist only within that block. either FUNCTION 1 or FUNC 1 signifies that Function 1 is being defined.. 221 . TRANSFORM statements END COMMANDS statements END TEMPORARY statements END FUNCTION N statements END STARTING statements END DIFFERENTIAL Required if the model is a function of one or more differential equastatements tions. Optional commands to create global variables. the output will include both the original variable and the transformed values. Details appear under: • • • “WinNonlin variables” on page 231 “The WinNonlin programming language” on page 233 “Commands. prior to starting the estimation procedure. However. and EXP9. 222 . if it included: LOGY = log(Y) then the variable LOGY will not be included in the output worksheet. Independent variables can also be derived or transformed here. New variables created in the transform block will not be included. the blocks use WinNonlin programming statements and command language to define relationships between input data and parameter estimates. Arrays and Functions” on page 557 WinNonlin TRANSFORM This block of statements is called once for each observation being modeled. In general. See the user models in EXP3. EXP4.PMO. and the following statements are included: X8 = SQRT(X8) Y=log(Y) then the output worksheet will include the columns: X8 and Y (transformed values). and X8(Obs) and Y(Obs) (original values). See also “Weights in ASCII models” on page 243.10 User’s Guide Note: Comments can appear within or between any of the blocks. If this block transforms a variable from the input data set. for examples that use the TRANSFORM block. For example. it will override any weighting specified via the Data Variables dialog. Note: When WT is specified in a model file. using special variables that WinNonlin makes available to hold the model input and output.PMO. if a data set contains the variables X8 and Y. This block often defines the dependent variable (Y) or the weight (WT) as a function of other variables from the input data set. stored in the \EXAMPLES directory. Each individual comment line must begin with REMA (or REMARK).PMO. these items can be set using the WinNonlin interface. If a model will fit multiple functions that share one or more parameters.. LOWER and UPPER set initial estimates and bounds. i. and saved in a Pharsight model object. Instead of using a COMMAND block. to define model-related values such as NPAR (number of parameters). etc. When more than one function is defined (NFUN>1) it is important to define which observation data belong to each FUNCTION block. since the equations for both involve K10. the constants should be specified at run time using WinNonlin’s dosing interface. within the same MODEL using multiple FUNCTION blocks. the number of FUNCTION blocks. The variable K10 would be defined in a TEMPORARY block for use in both functions. or using the FNUMBER N command. Each constant must be initialized via either the CONSTANTS command or the WinNonlin dosing interface. 1 for FUNC 1. The NPARAMETERS command specifies the number of parameters to be estimated. This block cannot include WinNonlin programming statements. where N is the column number. that is.User Models ASCII user models 10 COMMANDS The COMMANDS block associates WinNonlin commands with a model. This variable is identified via WinNonlin’s Data Variables dialog. The input data set should contain all observations (Y) in one column. The NCONSTANTS command sets the number of (dosing) constants in the model. 2 for FUNC 2. A function variable in a separate column indicates which observations belong with which function. and must include the number of values indicated by NPARAM. one for urine. The equations for each compartment would appear in a separate FUNCTION block: one for plasma data. Some models allow a variable number of constants. For example: 223 .e. The NDERIVATIVES command tells WinNonlin the number of differential equations in the model. This block provides backward-compatibility with WinNonlin command files. the multiple dosing models in WinNonlin's library allow a variable number of doses per subject. its values must correspond to the FUNCTION numbers. when modeling both plasma and urine data for a one compartment IV bolus model it would be more efficient to model these simultaneously. NFUNCTIONS specifies the number of equation sets associated with observations. In such cases. For example. For example. it may be more efficient to model the functions simultaneously.e. i. INITIAL. The NSECONDARY command indicates the number of secondary parameters to be computed as functions of the estimated parameters.. variables that can be used by any block in the model. commands name the primary and secondary parameters.693/K10 END TEMPORARY Statements in the optional TEMPORARY block define global variables. 'K1E' PUNIT ‘1/hr’. units are not included in calculation but are carried along to the output. These names can then be used in the model definition statements.693/K10 END The PNAMES. 'VOLUME'.’1/hr’ NSEC 1 SNAMES 'CL' SUNIT ‘ml/hr’ END SECO CL =0. These names provide column labels in the output. numbers or underscores. ‘CONC’ NPARM 3 PNAMES 'K10'.10 User’s Guide WinNonlin COMMANDS NSEC 1 END SECO S(1)=0. 224 . For user models. Variables that are used only within a given block can be defined within that block. The PUNITS and SUNITS commands supply optional units for the primary and secondary parameters. but must begin with a letter. COMM DNAMES ‘SUBJECT’. ‘ml’. and may be used in programming statements. that is. ‘TIME’. and SNAMES. ‘FORM’. The DNAMES command may be used to define names for the columns in the input data. Variable names include up to eight (8) letters. e. F=Z(1)or F=Z(1)+3Z(2). If the initial values do not occur at X = 0. etc. starting values will equal the value of the dependent variable when X = 0. The reserved variable F should be assigned the function value as: • • an algebraic function. e. START Z(1)=C0 Z(2)=0 END DIFFERENTIAL This block defines a system of differential equations using the array: DZ(1). then X should also be assigned accordingly.e. Note that the initial values could be estimated as parameters.User Models ASCII user models 10 FUNCTION Each set of equations that will be fit to a body of data requires a FUNCTION block. NDERIVATIVES > 0). DZ(2). Most models include only one.g. In most instances. such as C0 below. F=(D/V)exp(-K10t). or a function of one or more differential equations defined in one or more DIFFERENTIAL blocks.. a STARTING block is required to set initial values for each.g. A DIFFERENTIAL block is required when NDERIVATIVES > 0.. Assign initial values to elements of the Z array as shown below. STARTING When a model includes differential equations (i.. FUNC 1 F=A*EXP(-ALPHA*T) END FUNC 2 F=Z(1) END This section can also define weights by setting WT equal to the weight. 225 . S(1) denotes the first secondary parameter. as C(0) = 0.5)/β 226 . SECO REMARK Note that D and AUC must be defined above S(1) = D/AUC END Assign secondary parameter values to the S array. use SNAMES as follows. implying that B = -A. COMM NSEC 1 SNAME ‘CL’ END SECO CL = D/AUC END ASCII model examples Example 1 To fit a set of data to the following model: C(t) = Aexp(-αt) + Bexp(-βt) A. This model will also estimate the half lives of α and β: -LN(0.5)/α and -LN(0. B is not estimated. though clearance is of greater interest. S(2) the second. A + B = 0). t denotes time and C(0) is constrained to be zero (i. a model may be defined micro constants..K20*Z(2) END SECONDARY In many instances. For example.e. α and β are parameters to be estimated. but in functions of them. Alternately. etc.10 User’s Guide WinNonlin DIFF DZ(1)=-K12*Z(1) DZ(2)=K12*Z(1) . interest is not in the model parameters per se. The SECONDARY block addresses this. α. as follows.5)/P(2) S(2)=-LOGE(*. S(1) and S(2) are the secondary parameters.5)/P(3) END EOM The FUNCTION . 227 . The SECONDARY .END block defines secondary parameters (functions of the model parameters). and β. descriptive parameter names and comments could make the model easier to read. as a function of X (which is time in this example) and three parameters A. The above model will work fine. F denotes the concentration predicted by the model. However.User Models ASCII user models 10 The following statements define the model and secondary parameters: MODEL 1 FUNCTION 1 F=P(1)*EXP(-P(2)*X)-P(1)*EXP(-P(3)*X) END SECONDARY S(1)=-LOGE(*. to be estimated.END block defines function one (the only function). 228 .5)/ALPHA BETA_HL=-LOGE(*. The PNAMES and SNAMES statements define the parameter and secondary parameter names that are used in the programming statements and the output. ‘BETA’ SNAMES ‘ALPHA_HL’.END block holds model commands. so they needn’t be set when loading the model to WinNonlin. ‘BETA_HL’ END FUNC 1 B=-A remark: Model is constrained so that C(0)=0 TIME=X E1=EXP(-ALPHA*TIME) E2=EXP(-BETA*TIME) F=A*E1 + B*E2 END remark: Secondary parameters SECO ALPHA_HL=-LOGE(*. The following statements would appear after the MODEL statement: COMMANDS NPAR 2 NSEC 2 END Example 2 Consider a model characterized by the following differential equations.10 User’s Guide WinNonlin MODEL 1 remark: Example problem COMMAND NPARM 3 NSEC 2 PNAMES ‘A’. It is useful to also store NSEC and NPAR with the model.5)/BETA END EOM The above models are functionally equivalent. the COMMAND . In the second model. ‘ALPHA’. DIFF DZ(1) = -K12 * Z(1) DZ(2) = K12 * Z(1) . ‘K20’. As in the previous example.K20 * Z(2) END EOM 229 .User Models ASCII user models 10 dZ 1 -------.= – K 12 ⋅ Z ( 1 ) dt dZ 2 -------.= K 12 ⋅ Z ( 1 ) – K 20 ⋅ Z ( 2 ) dt with initial conditions Z1(0) = D and Z2(0) = 0. define the initial values for the differential equations. MODEL 1 COMMAND NPARM 3 PNAMES ‘K12’. The parameters to be estimated are K12. In this case Z2 represents plasma concentration. FUNC 1 F=Z(2) END The next set of statements completes the model specification. ‘D’ END Next. K20 and D. START Z(1) = D Z(2) = 0 END A FUNCTION block is required to define the model associated with the observed data. first use a PNAMES statement to assign meaningful names to the parameters and facilitate model definition. 03 0.10 User’s Guide Example 3 This example shows how to use the predefined variables to alter a model based on the current observation. ‘B’.. The second is concentration. Suppose there are two sets of data to fit simultaneously..0 0. MODEL 1 COMM NPAR 5 PNAMES ‘A’.0 2.. 0 0 0 0.0 2..5 1..5 1. The model reads as follows. ‘LAG’ END FUNC 1 I = DTA(3) IF I = 1 THEN :First Set of Data T=X F = A*EXP(-ALPHA*T) + B*EXP(-BETA*T) ELSE :Second Set of Data T = X-LAG F = A*EXP(-ALPHA*T) + B*EXP(-BETA*T) ENDIF END EOM 230 . The third is an indicator variable.0 3.9 .02 2 2 2 2 1 1 1 WinNonlin The first column is time. ‘BETA’. Its value indicates to which set of data the observations belong.0 . 0. ‘ALPHA’. the statement: I = DTA(3) assigns I the value of the indicator variable. Since the third column contains the indicator values. Assume the following data: 0. use the DTA( ) variable. Both data sets have the same model and parameters except that the second set requires a time lag.7 0. To access its value during the model-fitting process. This allows “adaptive” models—that is.User Models WinNonlin variables 10 When I = 1 the function without the time lag is fit. using the FNUMBER command. the function with the time lag is fit. There is another way to fit these data. I = 2. in two FUNCTION blocks. ‘BETA’. Thus. The two functions can be defined separately. The first creates one summary table. P(2). listed below. and current values of the parameters to be estimated. DTA(1) = value of 1st variable (column) DTA(2) = value of 2nd variable (column) etc. When I is not 1. DTA(N) This array contains values of all columns from the input data file (in the same order) corresponding to the current observation. the second creates two. ‘ALPHA’. ‘LAG’ NFUNC 2 FNUM 3 END FUNC 1 T=X F = A*EXP(-ALPHA*T) + B*EXP(-BETA*T) END FUNC 2 T = X-LAG F = A*EXP(-ALPHA*T) + B*EXP(-BETA*T) END EOM These approaches produce the same parameter estimates but different output. one per function. WinNonlin variables WinNonlin provides access to input data and parameter estimates during the model-fitting process through special variables. These variables can also be accessed using DNAMES. MODEL 1 COMM NPAR 5 PNAMES ‘A’. The value of the function evaluated at the current observation. that is. models that can be altered based on the current observation or value of the parameter estimates. ‘B’. for a given observation. P(1). etc. F 231 . NVAR.. TEMP.g. PAUC. 232 . END. This array contains the current integrated values corresponding to a system of differential equations.. CARR.. or EOM) any WinNonlin command or data array (e. COMM. Reserved variable names Variable names must not begin with: • • any WinNonlin keyword (e. SECO. p. OBSN.g. This array contains the constants that have been input by the user (e. NPARM. NCAR. OBJF. Weight assigned to the current observation. These are created as global variables in the Temporary block or as local variables within any block. F. P( ) and Z( ). These variables can also be accessed using SNAMES. These values are computed by WinNonlin (except when starting values for the system of differential equations are defined by the user). NPAU.. WTARRAY(j). This array contains the current values of the differential equations evaluated at X.g.) If the model is a function of more than one independent variable. NOBS. STEA. dose). DTA. Values assigned to WT will override any weighting specified via the Data Variables dialog. or using DNAMES. DTAARRAY(i. TRAN. P. Z).10 User’s Guide X Value of the independent variable at which the functions or differential equations are to be evaluated. the independent variables can be accessed via the DTA( ) array. OBJFN. This value is identical to DTA(XNUM) (See the XNUM command. DTIM. DIFF. Variable names can include 1-8 letters. numbers and/or underscores and must start with a letter. FUNC. e. XARRAY(j). NSEC. NDER. NTOT. This array contains the current values of the parameters to be estimated. TIME(2) is invalid. X. These variables can also be accessed using PNAMES.g. WinNonlin P(N) CON(N) Z(N) DZ(N) S(N) WT Temporary variables Most models require other variables to define the model or parameters. STAR. This array contains the values of the estimated secondary parameters. 56. Subscripts are not allowed. YARRAY(j). CON. NCON.j). **. ELSE IF. IF I>4 THEN GOTO GREEN ELSE GOTO BLUE ENDIF Syntax Example 233 . -. Valid mathematical operators are +. Note that ** denotes exponentiation. IF expression THEN statements ELSE statements Where expression is a logical expression. and /. *. Expressions are evaluated using standard operator precedence.User Models The WinNonlin programming language 10 Aliases The commands PNAMES. secondary parameters. GOTO label Where label is the statement label of an executable statement. or ENDIF statement at the same IF level. It provides control statements. and DNAMES can define aliases for parameters. Control is transferred here via a GOTO statement. control is transferred to the next ELSE. Statements GOTO Action Syntax Remarks Example LABEL Action Syntax Example IF THEN ELSE Action Transfers processing to a labeled statement. if expression is false. There should not be any blank spaces between GO and TO. and columns from the data file. SNAMES. statements in the IF block are executed. constants. It is possible to define new variables for easier programming. The label must end with a colon (:). The WinNonlin programming language WinNonlin provides a flexible programming language for ASCII user models. GREEN: If expression is true. mathematical functions and loop control. GOTO GREEN Denotes a label. .. See IF . See IF . exp2 is a lower limit for the loop.e. exp1 is an expression whose value is the initial value of var. IF statements can be nested up to 10 deep with the default model size (2). and up to 30 for model sizes 3 or greater. DO var=exp1 TO exp2 [STEP exp3] Where var is a variable to be incremented during the looping. STEP exp3 is optional.THEN. If exp3 is positive. DO I=1 TO 3 DO J=1 TO 5 statements NEXT NEXT IF statements can be nested up to 10 deep with the default model size (2). Marks the beginning of an ELSE block. exp2 is an expression whose value is the limit for the loop. exp3 is an expression whose value is the incrementor for the loop. ELSE WinNonlin This statement is optional. var has the value that caused the loop to exit (i. Var is always initialized to exp1.. even if the loop is not executed (because the value of exp1 is beyond the limit placed by exp2). If omitted. An ELSE block consists of any executable statements between the ELSE statement and the next ENDIF statement at the same IF level. After executing the last round through the loop. through the NEXT statement that ends the loop. Remarks Example 234 .THEN. If exp3 is negative. NEXT Action Syntax Each block of statements corresponding to each IF or ELSEIF statement must end with an ENDIF.10 User’s Guide Remarks ELSE Action Syntax Remarks Example ENDIF Action Syntax Remarks Example DO . and up to 30 for model sizes 3 or greater. exp2 is an upper limit for the loop. Terminates an IF block. exp3 is assumed to be 1. Repeatedly executes the statements following the DO statement. it is beyond the limit set by exp2). Returns the minimum of X and Y. DMAX1 (X.G. DEXP (X) ALOG (X). Y) ABS (X). tan-1(X) tan-1(Y/X) Result Returns the value of var at the last execution. Truncates var to an integer value. Returns the absolute value of X.9) = 3 ex natural log of X: ln(X) ALOG10 (X). E. X). DLOG10 (X) LOG10 (X) log base 10 of X: log10 (X) X 235 . DATAN (X) ATAN2 (Y. DSQRT (X) SIN (X). DMIN1 (X. Y). Y) MIN (X. Operator AND OR GT LT EQ GE LE > < = >= <= or greater than less than equal to greater than or equal to less than or equal to Definition and Functions Function Name LAG(var) INT(var) EXP (X). DSIN (X) COS (X). Y). WT. or any temporary variable. DATAN2 (Y. DCOS (X) MAX (X. DLOG (X). Var can be X.: INT(3.User Models The WinNonlin programming language 10 Comparison operators Table 10-2. LOGE (X) SQRT (X). DABS (X) ATAN (X). Y. X) sine (X) cosine (X) Returns the maximum of X and Y. Each model requires two files: 236 . both EXP(X) and DEXP(X) will produce the same result.N) This function returns the wt. That is. For example. associated with the normit (NOR) and a sample size N. then p is set to 1/2N. WTNOR. namely ex. the model fitting process can be greatly accelerated by compiling the model into a 32-bit Windows dynamic link library (DLL) using either Compaq Visual FORTRAN 6.0. Bioassay functions Function name NORMIT (R. then p is set to 1-1/2N WTNORM (NOR. it returns the value NOR such that WinNonlin Note: If R=0. This function returns the normit of p. If R=N. where and LOGIT(P) LOGIT = Compiled user models For User Models defined in terms of several differential equations.0 or Microsoft Visual C++ 6.N) Result Let p denote R/N.10 User’s Guide Function names that are grouped together are interchangeable. Either of these f90 files can be used as a template. however. 2.DLL. Click OK. Open Visual Fortran. 2. The compiled model.C).0 To create a compiled user model in FORTRAN: 1. F_UEXP15. The files include: File F_UEXP6.DLL Compiled . The same blocks are included.DLL Purpose FORTRAN source file for a system of two differential equations with data on both compartments (same model as EXAMPLES\EXP6.F90 FORTRAN source file for a PK/PD link model (same model as EXAMPLES\EXP15.DLL is a reserved name in WinNonlin. that the project should not be named USRMOD. 5. Select Fortran Dynamic Link Library as the kind of product to create. Note.DLL for the above FORTRAN source file.User Models Compiled user models 10 1. Enter a project name at the top left. Example using Digital (now Compaq) Visual FORTRAN 6.PMO).DLL for the above FORTRAN source file.PMO ). This subroutine must be named USRMOD( ). Using FORTRAN Creating a source file The format of this file is very similar to that of WinNonlin ASCII user models. 237 .F90 F_UEXP6. 3. Choose File>New from the menus. The source code file contains one subroutine specifying the model. but implemented in FORTRAN or C++ instead of WinNonlin programming language. The source code file (*. 4. USRMOD. Two examples of the files required to build a DLL in FORTRAN have been installed to the \USERCOMP subdirectory of the WinNonlin installation directory.F90 or *. Compiled . F_UEXP15. *. Using C++ Creating a source file The format of this file is very similar to a WinNonlin ASCII model. Choose Build>Build mymodel. WinNonlin 12. Choose Build>Set Active Configuration… from the menus.0. An example of this file is provided for Microsoft® Visual C++ 6. The . This example uses modelname.10 User’s Guide 6. This minimizes the size of the file.F90 (located in the WinNonlin installation directory) in a text editor. 15. Select Fortran Free Format Source File. 13. Click OK to open a frame for the Fortran file. 7. 10.dll from the menus. To use one of the provided templates. 238 . Note: The source file stored on disk can be assigned any name. Select Win32 Release and click OK. with a C++ implementation. 9. 11.DLL is saved to the Release subdirectory in the current directory. Click OK. The same blocks are included. open USERCOMP\F_UEXP6. Arrays and Functions” on page 557 14. 8.F90 or USERCOMP\F_UEXP6.f90. See the following for model structure and syntax: • • • “Model structure” on page 220 “WinNonlin variables” on page 231 “Commands. Choose Project>Add to Project…>New from the Visual FORTRAN menus. but the subroutine must be named USRMOD( ). In the next dialog select Empty dll application. Click Finish.DLL. Enter a file name. for this example: D:\MYPROJECTS\MYMODEL\RELEASE\MYMODEL. Copy/paste the content into your new FORTRAN file and edit as needed. Enter the model code. Compiled .C C_UEXP6. 5. 239 . Press OK. 9.c. 4. run Example 15 in both the ASCII form and in the compiled form. 10. 6. In the next dialog select An empty dll project. Enter a file name. Note: It is important to use the . 3. Enter a project name. Two sets of files have been provided. Compiled .c file extension rather than the . These files include: File C_UEXP6. 8. Click Finish.cpp extension.DLL Purpose C source file for modeling example 6. Example using Microsoft Visual C++ 6.DLL file for modeling example 15. Click OK. Select Win32 Dynamic Link Library.DLL C_UEXP15. 7.0 To compile a user model: 1.User Models Compiled user models 10 The files required to build a DLL in Visual C++ have been installed to the \USERCOMP subdirectory. Note: To demonstrate the benefits of compiling models. Either can be used as a template.0. Select New from the File menu. C source file for modeling example 15. Select C++ Source File. Open Microsoft® Visual C++ v. 6.C C_UEXP15. 2. This example uses modelname. 11.DLL file for modeling example 6. On the Project menu select Add to Project…New. 3.10 User’s Guide 12.DLL) Note: Statements in the Visual C++ source file with // starting in the first column are comments.. Select from the Build menu Build mymodel. For each block. This is not essential. etc. Select from the Build menu Set Active Configuration…Win32 Release. Enter the code for the model. but it builds a smaller DLL file. A frame opens for the program. The DLL is built in the specified directory (D:\MYPROJECTS\MYMODEL\RELEASE\MYMODEL. other than the Command block. Use the templates provided (see table above). 13.dll. 2. 240 . .” with the parameters for the model as they would be stated in a Temporary block. All integers should be declared as INT and all real numbers should be declared as DOUBLE. 14. Replace “k12 = p[0]. include the corresponding block from the template file. but the subroutine must be called USRMOD( ). C++ makes no assumptions about types of variables. WinNonlin To alter this example for another model: 1.. The source file stored on disk can be assigned any name. i. programming statements may also be used to define weights as part of the model. WEIGHT = Yn. For example.Chapter 11 Weighting Weighted nonlinear regression WinNonlin provides flexibility in performing weighted nonlinear regression and using weighted regression. other than uniform weighting.e. Weighted least squares: weight by a power of the observed value of Y 2. through the WinNonlin user interface: 1. There are three ways to assign weights. and one or more of the Y values are less than or equal to zero.. 241 .e.5 instructs WinNonlin to weight the data by the square root of the reciprocal of observed Y. If n has a negative value. 1 ⁄ ( Y) . and setting n = -0. then the corresponding weights are set to zero. Iterative reweighting: weight by a power of the predicted value of Y 3. i. via the Model Options dialog (see “Model options” on page 203 for compartmental models and “Model options” on page 173 for NCA). WinNonlin weights each observation by the value of the dependent variable raised to the power of n. selecting this option. Reading weights from the data file: include weight as a column in the data For a user ASCII model. Weights are assigned after loading a model. Weighted least squares When using weighted least squares.. See “Scaling of weights” on page 242.11 User’s Guide The program scales the weights such that the sum of the weights for each function equals the number of observations with non-zero weights. if N is negative. selecting this option. For iteratively reweighted least squares the parameters change at each iteration. Scaling of weights When weights are read from the data file or when weighted least squares is used. The weights should be the reciprocal of the variance of the observations. ˆ 1 ⁄ ( Y) . the program will scale the weights such that the sum of the weights for each function equals the number of observations with non-zero weights (see “Scaling of weights”). As with Weighted least squares. the weights for the individual data values are scaled so that the sum of the weights for each function is equal to the number of data values (with nonzero weights). then the corresponding weights are set to zero. As with Weighted Least Squares.e. and one or more of the predicted Y values are less than or equal to zero. Reading weights from the data file It is also possible to have a variable in the data file that has as its values the weights to use. WinNonlin Iterative reweighting Iterative Reweighting redefines the weights for each observation to be Fn. iterative reweighting can be used to obtain maximum likelihood estimates (see the article by Jennrich and Moore in the References). For example.. and setting n = -0. where F is the predicted response.5 instructs WinNonlin to weight the data by the square root of reciprocal of the predicted value of Y. 242 . and therefore the predicted values and the weights change at each iteration. i. For certain types of models. Iterative reweighting differs from weighted least squares in that for weighted least squares the weights are fixed. Note: Scaling can be turned off by selecting the No Scaling of Weights option on the Weight tab of the Model Options dialog or by using the NOSCALE command in a command file.0123 0. Note that 0. it is possible to define the weights using programming statements at the same time the model is defined. WT is a special variable name in WinNonlin. Consider the following example: 243 . If the variable WT is defined with programming statements. (0. or 1. Each value of Y-2 is divided by the sum of the values in the third column.0051 0.0277778/.e.g.843/0. However.843 0.0278 0.0452 Weight assigned by WinNonlin 1. scaling of the weights provides increased numerical stability.819 0..0452254) * 3 = 1.Weighting Weights in ASCII models 11 The scaling of the weights has no effect on the model fitting as the weights of the observations are proportionally the same. this variable overrides any weighting specified in the WinNonlin Model Options dialog. and then multiplied by the number of observations.8426238..000 Suppose weighted least squares with the power -2 was specified (i. e. Consider the following example: X 1 10 100 Sum Y 6 9 14 Y-2 0.338 3.843 after rounding.338. Weights in ASCII models When writing a custom ASCII model.0278/0. weighting by 1/Y*Y).0051 is the same proportion as 1. The corresponding values are as shown in the third column. and 1/Y2 if X is less than or equal to 10. For example: MODEL 1 TEMP (additional statements here) END FUNC 1 (additional statements here) WT = (some function) END FUNC 2 (additional statements here) WT = (a different function) END (additional statements here) 244 .) Each function to be fit can use a different weighting scheme. the corresponding weights are also constant. (See “Iterative reweighting” on page 242. it is also possible to use programming statements to iteratively reweight the observations. As the values of Y are constant across iterations. However.A*EXP(-BETA*X) WT = 1/(F*F) END (additional statements here) This is equivalent to specifying Predicted to power n = -2. as shown in the following example: MODEL 1 FUNC 1 F = A*EXP(-ALPHA*X) .11 User’s Guide WinNonlin MODEL 1 TRANSFORM IF X > 10 THEN WT = 1/Y ELSE WT = 1/(Y*Y) ENDIF END (additional statements here) The above statements would set the weight of the observation equal to 1/Y if the value of X is greater than 10. based on concentration data from a single dose. This can be done by fitting the data to a compartmental model with some assumptions about the absorption rate of the drug. Nonparametric superposition In pharmacokinetics it is often desirable to predict the drug concentration in blood or plasma after multiple doses. Crossover design: performs nonparametric statistical tests. which does not assume any pharmacokinetic (PK) model.Chapter 12 The WinNonlin Toolbox Nonparametric Superposition. Semicompartmental modeling: estimates effect-site concentration given plasma concentration and a value for Ke0. The predictions are based upon an accumulation ratio computed from the terminal slope (Lambda Z). calculates confidence intervals for treatment median and median difference between treatments. and estimates relevance of direct. and period effects. Semicompartmental Modeling and Crossover Design The WinNonlin toolbox includes the following tools: • • • Nonparametric superposition: used to predict drug concentrations after multiple dosing at steady state. WinNonlin’s nonparametric superposition function is used to predict drug concentrations after multiple dosing at steady state. An alternative method is based on the principle of superposition. and is based on noncompartmental results describing single dose data. The feature allows predictions from simple (the same dose given in a constant inter245 . Numeric deconvolution: evaluating in vivo drug release and delivery based on data for a known drug input • Example are provided in the Examples Guide. residual. In order to predict the drug concentration resulting from multiple doses.…. and that linear pharmacokinetics apply. then C(t) = β exp(-λzt) for t > tn.12 User’s Guide val) or complicated dosing schedules. dosing. See “Performing nonparametric superposition” on page 248 for stepped instructions Data and assumptions Nonparametric Superposition assumes that each dose of a drug acts independently of every other dose. the time range included in the terminal phase is specified by the user. The former assumes that the effect of each dose can be separated from the effects of other doses. The results can be used to help design experiments or to predict outcomes of clinical trials when used in conjunction with the semicompartmental modeling function. Note: User-defined terminal phases apply to all sort keys. to characterize the drug absorption and elimination process. The required input data are the time. dosing schedules and doses are the same for all sort keys. linear pharmacokinetics. In addition. one may obtain the concentration C(t) at any time t through interpolation if t1 < t <tn. (i=1. If the absolute value of that line’s slope is λz and the intercept is ln(β). The slope λz and the intercept are estimated by least squares from the terminal phase. Two assumptions about the data are required: independence of each dose effect. and drug concentration. that is. assumes that changes in drug concentration will vary linearly with dose amount. (i=1.n). the terminal phase in the plot of log(C(t)) versus t is approximately a straight line. The drug concentration at any particular time during multiple dosing is then predicted by simply adding the concentration values as shown below. That is. WinNonlin Computation method Given the concentration data points C(ti) at times ti. one must have a complete characterization of the concentration-time profile after a single dose. so that a change in dose during the multiple dosing regimen can be accommodated.2. after a single dose D. If 246 . that the rate and extent of absorption and average systemic clearance are the same for each dosing interval.…. and linearity of the underlying pharmacokinetics. or extrapolation if t > tn. The latter. it is necessary to know C(ti) at sufficient time points ti.2. The extrapolation assumes a log-linear elimination process.n). The total predicted concentration at time t will be: Conc ( t ) = ∑ Cj ( t ) .e. t is relative to dose time) will be: Cn(t) = C1(t) + β exp[-λ(t+τ)] + β exp[-λ(t+2τ)] +…+ β exp[-λ(t+(n-1)τ)] = C1(t) + β exp[-λ(t+τ)]{1-exp[-λ(n-1)τ]} / [1-exp(-λτ)] As n→∞. The half life is ln(2)/λz.m If the same dose is given at constant dosing intervals τ. Then the concentration at time t after nth dose (i..2 .ti-1]/[ti. WinNonlin offers two methods: linear interpolation and log-interpolation. Let the concentration of the first dose be C1(t) for 0 < t < τ. The concentration due to dose Dj will be: Cj(t) = (Dj / D)∗C(t-τ j) where t is time since the first dose and C(t. then steady state can be reached after sufficient time intervals. WinNonlin assumes steady state is at ten times the half life. so τ is greater than tn.m. estimates from the best linear fit (based on adjusted R-square as in noncompartmental analysis) will be used. and each dose is administered after τ j time units from the first dose.τ j) = 0 for t ≤ τ j. For interpolation. j = 1. Linear interpolation is appropriate for log-transformed concentration data and is calculated by: C(t) = C(ti-1) + [C(ti)-C(ti-1)]*[t.… .…. and τ is sufficiently large that drug concentrations reflect the post-absorptive and post-distributive phase of the concentration-time profile. the steady state (ss) is reached: Css(t) = C1(t) + β exp[-λ(t+τ)]/[1-exp(-λτ)] To display the concentration curve at steady state. j j = 1 .ti-1].The WinNonlin Toolbox Nonparametric superposition 12 the user does not specify the terminal phase. ti-1 < t < ti 247 . Suppose there are m additional doses Dj. If units appear in the input worksheet. 6. 248 . drag one or more columns to the sort variables field. 4. Performing nonparametric superposition Note: All quantities in the Nonparametric Superposition dialog should be entered in the units of the input data set. Open the data set providing the single-dose concentrations. Time for First Dose: Drag the Time column to this field.12 User’s Guide Log-interpolation is appropriate for original concentration data. WinNonlin To perform nonparametric superposition: 1. A separate analysis is performed for each unique combination of sort variable values. Sort variables: If using sort variables to define individual data profiles. 3. Number of output data points: Enter the number of predicted values to output. those units are applied in nonparametric superposition. ti-1 < t < ti For additional reading see Appendix E of Gibaldi and Perrier (1982). 2. and is evaluated by: C(t) = exp{log(C(ti-1))+[log(C(ti))-log(C(ti-1))]*(t-ti-1)/(ti-ti-1)}. Concentration for First Dose: Drag the Concentration column to this field. Make sure the appropriate data set appears under Dataset. The Nonparametric Superposition dialog appears. and its variables appear under Variables. and appear in the output. 5. Choose Tools>Nonparametric Superposition from the WinNonlin menus. Variable Dosing: If the time between doses varies: a. enter doses and times for one cycle. Regular dosing: For dosing at a regular time interval: a. c. Generally. Log for untransformed data. and a chart window containing plots of predicted concentration over time for each sort level. Method for computations: Select linear or logarithmic interpolation and extrapolation. in time units matching those of the time variable in the data set. The Lambda Z worksheet contains the Lambda Z value and the half-life for each sort key. Enter the maintenance dose.The WinNonlin Toolbox Nonparametric superposition 12 7. b. The Concentrations worksheet contains times and predicted concentrations for each level of the sort variables. n. Define Terminal Phase: (Optional) To set the time range for Lambda Z calculations. c. the dosing interval. 10. If a regular dosing schedule is selected. 8. the output includes the times and concentrations within the supplied output range. check this check box and enter start and end times for the terminal elimination phase. d. Output Nonparametric Superposition generates a workbook containing predicted concentrations and Lambda Z values. Click Calculate. b. Enter the initial dose in the Initial Dose field of the Regular Dosing tab. 249 . select Linear for log-transformed concentration data. For a repeating dose regimen. 11. WinNonlin checks for duplicate time values in the data and prompts for sort variables if they have not been specified. The workbook contains two worksheets: Concentrations and Lambda Z. 9. this output represents times and concentrations between two doses at steady state. a Nonparametric Superposition Workbook and a Nonparametric Superposition Chart. Enter the value for Tau. Enter individual doses and dose times on the Variable Dosing tab. check Repeat every n time units and enter the repeat time. WinNonlin performs the calculations and generates two new windows. Enter time range for which to output predicted data. Select whether to display the chart at steady state or at a particular dose. If a variable dosing schedule is selected. Vozeh. The potential advantage of this approach is reducing the effect of model misspecification for Cp when the underlying PK model is unknown. See “Performing semicompartmental modeling” on page 252 for stepped instructions Data and assumptions Drug concentration in plasma Cp for each subject is measured at multiple time points after the drug is administrated. WinNonlin’s semicompartmental modeling estimates effect-site concentrations for given times and plasma concentrations and an appropriate value of keo. Computation method In the effect-site link model (Sheiner.12 User’s Guide WinNonlin Semicompartmental modeling Semicompartmental modeling was proposed by Kowalski and Karim (1995) for modeling the temporal aspects of the pharmacokinetic-pharmacodynamic relationship of drugs. The effect site concentration Ce is related to Cp by first-order disposition kinetics and can be obtained by solving the differential equation: 250 . This function should be used when a counterclockwise hysteresis is observed in the graph of effect versus plasma concentrations. This model was based on the effect-site link model of Sheiner. A scientist developing a PK/PD link model based upon simple IV bolus data can use this function to compute effect site concentrations from observed plasma concentrations following more complicated administration regimen without first modeling the data. Stanski. Effect-site link model is used and the value of the equilibration rate constant keo that accounts for the lag between the Cp and the Ce curves is required. The results can then be compared to the original data sets to determine if the model suitably describes pharmacodynamic action after the more complicated regimen. Vozeh. A piecewise log-linear model is assumed for Cp. Miller and Ham (1979) to estimate effect-site concentration Ce by using a piecewise linear model for plasma concentration Cp rather than specifying a PK model for Cp. Stanski. the time points need to be adequately sampled such that accurate estimation of the AUC by noncompartmental methods can be obtained. To minimize the bias in estimating Ce. Miller and Ham (1979)). The hysteresis loop collapses with the graph of effect versus effect-site concentrations. a hypothetical effect compartment was proposed to model the time lag between the PK and PD responses. t j-1 ). WinNonlin provides three methods.[ e –e ] k e0 – λ j λ 1 and C e1 are estimated by a linear regression. assuming Ce (0) =Cp(0) = 0. The ‘linear’ method uses linear piecewise PK model.The WinNonlin Toolbox Semicompartmental modeling 12 where keo is a known constant.λ j (t-t j-1)] j–1 here λ j = (ln C p . j j–1 j–1 Using the above two equations can lead to a recursive equation for Ce: C ej = C ej – 1 e – k e0 ( t j – t j – 1 ) λj –k ( t – t ) + ⎛ C Pj – 1 – -----. 251 . use a piecewise linear model for Cp: Cp (t) = C p + λ j (t-t j-1) j–1 t j-1 < t < t j where λ j = ( C p – C p )/( t j – t j-1 ) and C p = Cp ( tj – 1 ) . Here. the ‘log’ method uses log-linear piecewise PK model. one needs to know Cp as a function of time. In order to solve this equation.ln C p )/( t j . One can also assume a log-linear piecewise PK model for Cp Cp (t) = C p exp[ .⎞ [ 1 – e e0 j j – 1 ] + λ j ( t j – t j – 1 ) ⎝ ⎠ k e0 The initial condition is Cp(0) = Ce(0) =0. which is usually given by compartmental PK models. the default ‘log/linear’ method uses linear to Tmax and then log-linear after Tmax. The Effect field is not used for the calculation of Ce but when it is provided. the ECp and E-Ce plot will be plotted. and: j–1 j t j-1 < t < t j C ej = C ej – 1 e – k e0 ( t j – t j – 1 ) k e0 –λj ( tj – tj – 1 ) – k e0 ( t j – t j – 1 ) + C Pj – 1 ----------------. Choose Tools>Semicompartmental modeling from the WinNonlin menus. 6.12 User’s Guide Performing semicompartmental modeling To perform semicompartmental modeling: WinNonlin 1. A separate analysis is performed for each unique combination of sort variable values. at minimum. 3. Effect: Drag a column containing drug effect data to this field in order to include concentration-effect plots in the output. 8. 4. 7. or the default Log/linear for linear to Tmax and then log-linear after Tmax. The Semicompartmental Modeling dialog appears. 9. This column is not used in estimating effect site concentration. WinNonlin performs the computations and generates a new workbook and chart window. Make sure the appropriate data set appears under Dataset. drag one or more columns to the sort variables field. Click Calculate. Keo: Enter the value for the equilibration rate constant. 5. Open a data set containing plasma concentrations and times. 252 . Log for a log-linear piecewise PK model. Time: Drag the column containing times to this field. 2. Plasma Concentration: Drag the column containing blood concentrations here. and its columns appear under Variables. Sort variables: If using sort variables to define individual data profiles. Method for computations: Select Linear to use a linear piecewise PK model. Chart: The chart window contains plots of Cp versus Time and Ce versus Time. If Effect is included in the data set. Conc. 253 . Time. Ce (estimated effect site concentration). Note: Any units associated with the Time and Concentration columns in the input data are carried through to the semicompartmental modeling output. and Effect (if included). Examples are provided below. plots of Effect versus Cp and Effect versus Ce are also created.The WinNonlin Toolbox Semicompartmental modeling 12 Output Workbook: WinNonlin creates a new workbook with the following data: any sort variables. it calculates confi254 .12 User’s Guide WinNonlin When the data set has multiple sort levels. First. In many situations the assumptions of variance homogeneity and normality. WinNonlin’s Crossover Design tool performs statistical analysis of data arising from subjects when variance homogeneity and normality may not necessarily apply. may not be justified due to small sample size. a separate plot appears for each. This tool performs nonparametric statistical tests on variables such as Tmax and provides two categories of results. among other reasons. relied upon in parametric analyses. Crossover design The two-period crossover design is common for clinical trials in which each subject serves as his or her own control. The nonparametric methods proposed by Koch (1972) can be used to analyze such data. …. Suppose that n subjects are randomly assigned to the first sequence. i=1 and k=2 is treatment R. The data for one treatment must be listed first. in which subjects are given treatment T first and then treatment R following an adequate washout period. Finally. and period effects. The alternative is to place data for each treatment in a separate column. subject j (j = 1.The WinNonlin Toolbox Crossover design 12 dence intervals for each treatment median. Representation of the data Sequence (i) TR . Table 12-1.2. n+1. and column 5 gives the within-subject difference in Y between the two periods. for example i=1 and k=1 is treatment T.2). For “stacked” data.d2j)/2 are computed. Data and assumptions Consider a trial testing the effects of two treatments. then all the data for the other. it estimates the relevance of direct. See “Using the Crossover Design tool” on page 258 for stepped instructions. Similarly. with one or more additional columns flagging which data belong to which treatment. 255 . along with the median of these n*m points. residual. Treatment is implied by sequence i and period k.. TR. all measurements appear in a single column.. It then calculates the median difference between treatments and the corresponding confidence intervals. Crossover Design supports two data layouts: stacked in same column or separate columns.…n+m) and period k (k = 1.n. TR RT … RT Subject (j) 1 … n n+1 … n+m Yij1 y111 … y1n1 y2(n+1)1 … y2(n+m)1 Yij2 y112 … y1n2 y2(n+1)2 … y2(n+m)2 d2(n+m) d1n d2(n+1) Dij d11 dij = yij1 .yij2. Y represents the outcome of the trial (for example Cmax or Tmax) and yijk is the value observed on sequence i (i=1. m subjects are assigned to the RT sequence. Test (T) and Reference (R). The data are listed in columns 1 through 4 of Table 12-1 below.2). Then the n*m crossover differences (d1i . X(2). Let X(1). normal approximation is used: Xl = X(L). the exact probability for a binomial distribution is used: Xl = X(L). C. X(N) be the order statistic of samples X1. where U = where int(X) returns the largest integer less than or equal to X.….12 User’s Guide Descriptive analysis The median response and its confidence interval are calculated for Y of each treatment and the crossover difference between treatments. if N is even WinNonlin if N is odd The 100% confidence interval (CI) of Xm is defined as follows.2 ⎠ ⎝2 2 ⎛ N⎞ N – int ⎜ N – z 1 + p -------⎟ ------------.2 ⎠ ⎝2 2 Xu = X(U). where L = max ⎛ ⎜ i: ⎜ ⎝ ⎞ ⎛ N⎞ 2 – N < ( 1 – p ) ⁄ 2⎟ + 1 ∑ ⎝ i⎠ ⎟ ⎠ i=1 ⎞ ⎛ N⎞ 2 – N > ( 1 + p ) ⁄ 2⎟ – 1 ∑ ⎝ i⎠ ⎟ ⎠ i=1 U N N Xu = X(U). = X[(N+1)/2]. X2.…. where U = min ⎛ ⎜ i: ⎜ ⎝ The exact probability is ∑ ⎛ i⎞2 ⎝ ⎠ i=L N –N . 256 . • For N > 20. where L = ⎛ N⎞ int ⎜ N – z 1 + p -------⎟ + 1 ------------. • For N <= 20. XN. The median Xm is defined as: Xm = [X(N/2)+X(N/2+1)]/2. The test statistic is: n n+m T = min ( ∑ Rl l=1 . ….1) Similarly. hypothesis 2 can be tested using the Wilcoxon statistic on the difference D. ni=n for i=1. Hypothesis 4 can be tested using the bivariate Wilcoxon statistic on (Yij1.. n+m. and ni=m for i=2. R i2 -M]T. and 4. 2. …. 1980): [T . the statistic to be used for testing hypothesis 4 is: L = (n+m-1)[nU1T*S-1*U1 + mU2T*S-1*U2] Where Ui is a 2x1 vector: [ R i1 -M. l = 1. no sequence effect (or drug residual effect).. y1nk. no treatment and no sequence effect.The WinNonlin Toolbox Crossover design 12 Hypothesis testing Four hypotheses are of interest: 1. where j=1. y2(n+1)k. M = (n+m+1)/2 257 . ni. If R(Sl) is the rank of Sij in the whole sample. Hypothesis 1 above can be tested using the Wilcoxon statistic on the sum S.n(n+m+1)/2] / nm ( n + m + 1 ) ⁄ 12 ~N(0. The statistics are in the form described above.. 3. ∑ l = n+1 Rl ) The p-value is evaluated by using normal approximation (Conover.…. let Rijk equal: rank { yijk: y11k. Thus. y2(n+m)k } The average rank for each sequence is ni R ik = ∑ Rijk ⁄ ni j=1 . no period effect given no sequence effect. no treatment effect given no sequence effect. hypothesis 3 can be tested using the Wilcoxon statistic on the crossover difference C. Yij2). For each period k. Note: See the WinNonlin Examples Guide for examples of crossover design. 6. Each unique combination of sort variable values for each subject and treatment represents one data profile. or each treatment placed in a separate column. Separate columns: 258 .12 User’s Guide S is the 2x2 covariance matrix: WinNonlin S = Σi Σj ( R ij1 – M ) 2 ( R ij1 – M ) ( R ij2 – M ) ( R ij2 – M ) 2 ( R ij1 – M ) ( R ij2 – M ) and L ~ chi-square(2). An example of each appears below. 2. drag one or more columns to the sort variables field. Subject: Drag the variable representing subject identifiers from the Variables list box to the Subject field. Make sure that the data set to use is open and active. To use crossover design: 1. 4. Treatment data: Select whether the treatment data are in separate columns or stacked in one column. Choose Tools>Crossover Design from the WinNonlin menus to open the Crossover Design dialog. so the p-value can be evaluated. Using the Crossover Design tool Crossover design supports two data formats: data for both treatments stacked in one column. 5. 3. treatment and sequence to define individual data profiles. Sequence: Drag the variable representing treatment sequence to this field. Sort variables: If using sort variables in addition to subject IDs. b. If response data for both treatments are stacked in a single column: a. Stacked in one column: 8. Treatment: Drag the column the identifies each treatment to this field. 259 . Click Calculate to perform the analysis.The WinNonlin Toolbox Crossover design 12 7. Response: Drag the column containing response data to this field. If data for each treatment appears in a separate column. drag the column containing response data for each treatment to the Treatment A and Treatment B fields as shown above. 9. The second worksheet. The output workbooks display units in the median. • 260 “Deconvolution methodology” . treatment and period. lower. and upper value parameters.12 User’s Guide Output Crossover Design creates a new workbook with two worksheets. as well as treatment and residual simultaneously. lists the test statistic and p-value associated with each of four tests. The first sheet lists the confidence intervals (at 80%. Deconvolution calculations and stepped usage instructions appear under the following headings. and 95%) for the treatment medians in the crossover data and the treatment difference median. 90%. called Effects. Numeric deconvolution Deconvolution evaluates in vivo drug release and delivery based on data for a known drug input. WinNonlin Note: Units associated with the Response variable in the input data set units are carried through to the crossover design output. It includes tests for an effect of sequence. integration. gastro-intestinal release) or a composite form. Perhaps the most common application of deconvolution is in the evaluation of drug release and drug absorption from orally administered drug formulations. Linearity assumptions The methodology is based on linear system analysis with linearity defined in the general sense of the linear superposition principle. differentiation. convolution. deconvolution. Depending upon the type of reference input information available. all kinetic models defined in terms of linear combinations of linear mathematical operators (e. but code to process deconvolution output may need modification to account for enhancements to the workbook output. However. See “Numeric deconvolution” on page 260 for instructions on performing deconvolution. Deconvolution methodology Deconvolution is used to evaluate in vivo drug release and delivery when data from a known drug input are available. gastro-intestinal release is evaluated if the reference is an oral solution (oral bolus input). including transdermal release and delivery.. It considers other types of delivery. It is well recognized that classical linear compartmental kinetic models exhibit superposition linearity due to their origin in linear differential equations. Scripts written prior to WinNonlin 5. the bioavailability is evaluated if the reference input is a vascular drug input.1 established several enhancements to numeric deconvolution.The WinNonlin Toolbox Numeric deconvolution 12 • Note: “Using deconvolution” on page 269 WinNonlin version 5.) constitute linear systems adhering to the superposition principle. The Deconvolution feature is not limited in scope to drug delivery via the oral route. It also allows the user to direct the analysis to investigate issues of special interest through different parameter settings and program input. etc.g. the drug transport evaluated will be either a simple in vivo drug release (e. Deconvolution provides automatic calculation of the drug input. The scope and generality of the linear system approach thus ranges far beyond linear com261 . Similarly. In this case. drug delivery from implanted devices. typically consisting of an in vivo release followed by a drug delivery to the general systemic circulation. etc.g.1 should operate in later versions. This assumed property is called time-invariance. in the present case the expected number of drug molecules to be found at sampling site B at time t (NB(t)) is equal to Ng(t). Imagine repeating this one molecule experiment an infinite number of times and observing the fraction of times that the molecule is at B at time t. That fraction represents the probability that a molecule is at point B given that it entered point A at time 0. Denote this probability by the function g(t). but now at time tA 262 . a bolus injection. To fully appreciate the power and generality of the linear system approach. Convolution/deconvolution . The possible presence of the molecule at B at time t is a random variable due to the stochastic transport principles involved. but is otherwise independent of the entry time. Consider again the simultaneous entry of N molecules. For example. The result is that the probability of being at point B at time t depends on the elapsed time since entry at A.. In addition the probability of being at B at time t is the same and equals g(t) for all of the molecules. Thus the probability of being at B for a molecule that enters A at time tA is g(t-tA). Let B be a sampling point (a given sampling space or volume) anywhere in the body (e. Let it be assumed that there is no significant interaction between the drug molecules. a blood sampling) that can be reached by the drug molecule entering at A. consider the entry of a single drug molecule at point A at time 0. it is best to consider the kinetics of drug transport in a non-parametric manner and simply use stochastic transport principles in the analysis. For example. it is best to depart from the typical.g.stochastic background WinNonlin The convolution/deconvolution principles applied to evaluate drug release and drug input (absorption) can be explained in terms of point A to point B stochastic transport principles. Now let's further assume that the processes influencing transport from A to B are constant over time. Next. Then the transport to sampling site B is both random and independent. To cut through all the complexity and objectively deal with the present problems. consider a simultaneous entry of a very large number (N) of drug molecules at time 0. It is this combination of independence and equal probability that leads to the superposition property that in its most simple form reveals itself in terms of dose-proportionality.12 User’s Guide partmental models. physically structured modeling view of pharmacokinetics. e. such that the probability of a drug molecule's possible arrival at sampling site B is not significantly affected by any other drug molecule. It can be seen from this that the concentration of drug at the sampling site B at the arbitrary time t is proportional to the dose (N is proportional to the dose and g(t) does not depend on the dose due to the independence in the transport).g. superposition.e. the expected number of molecules at B at time t is the product of N and the probability of being at B at time t.e. It is instead a random quantity tA distributed according to some probability density function h(t).... a molecule can't get to B before it enters A. As argued before.e. i.e.The WinNonlin Toolbox Numeric deconvolution 12 instead of 0. i. As a result the quantity g(t — tA)h(tA) is nonzero only over the range from 0 to t and the equation becomes g(t) = ∫0 g ( t – tA )h ( tA ) dtA t Suppose again that N molecules enter at point A but this time they enter at random times distributed according to h(t). Combining this property with those of independent and equal probabilities. For typical applications drug input is restricted to non-negative times so that h(t) = 0 for t < 0. results in an expected number of molecules at B at time t given by NB(t) = Ng(t-tA). t Pr ( t A ≥ t ) = ∫ h ( u ) du 0 The probability that such a molecule is at point B at time t is the average or expected value of g(t-tA) where the average is taken over the possible values of tA according to g(t) = ∫–∞ g ( t – tA )h ( tA ) dtA ∞ Causality dictates that g(t) must be 0 for any negative times. i. i. This is equivalent to saying that the expected rate of entry at A is given by N’A (t) = Nh (t).. Suppose that the actual time at which the molecule enters point A is unknown. N B ( t ) = Ng ( t ) = N ∫ g ( t – t A )h ( t A ) dt A = 0 t ∫0 g ( t – tA )N'A ( tA ) dtA t Converting from number of molecules to mass units: MB ( t ) = ∫0 g ( t – tA )f ( tA ) dtA 263 t . c(t). The piecewise linear component is parameterized in terms of a sum of hattype wavelet basis functions.a. Let Vs denote the volume of the sampling space (point B).12 User’s Guide where MB(t) is the mass of drug at point B at time t and f(t) is the rate of entry in mass per unit time into point A at time t. f(t). Tj + 1 – Tj 264 Tj < t < Tj + 1 Tj + 1 < t < Tj + 2 .. of drug at B is WinNonlin c(t) = ∫0 f ( t – u )cδ ( u ) du ≡ f ( t )*cδ ( t ) cδ ( t ) = g ( t ) ⁄ Vs t where “*” is used to denote the convolution operation and The above equation is the key convolution equation that forms the basis for the evaluation of the drug input rate. The function cδ (t) is denoted the unit impulse response (a. cδ is simply equal to the probability that a molecule entering point A at t = 0 is present in the sampling space at time t divided by the volume of that sample space. The former provides substantial flexibility whereas the latter provides smoothness. characteristic response or disposition function). Then the concentration. The process of determining the input function f(t) is called deconvolution because it is required to deconvolve the convolution integral in order to extract the input function that is embedded in the convolution integral. The function representation WinNonlin models the input function as a piecewise linear “precursor” function fp(t) convolved with an exponential “dispersion” function fd(t). Tj + 1 – Tj t – Tj h j ( t ) = --------------------.. The unit impulse response function (cδ ) provides the exact linkage between drug level response c(t) and the input rate function f(t).k. hj(t): fp ( t ) = ∑ xj hj ( t ) xj ≥ 0 t – Tj h j ( t ) = --------------------. The dispersion function provides the smoothing of the input function that is expected from the stochastic transport principles governing the transport of the drug molecules from the site of release to subsequent entry to and mixing in the general systemic circulation. 265 . This point is simply included to create enough capacity for drug input prior to the first sampling. The dispersion function is normalized to have a total integral (t = 0 to ∞ ) equal one. if any. is the convolution of the precursor function and the dispersion function: –t ⁄ δ f ( t ) = f p ( t )*f d ( t ) ≡ ∫ f p ( t – u )f d ( u ) du 0 t The general convolution form of the input function above is consistent with stochastic as well as deterministic transport principles. f(t). with the exception of the very first support point that is used to define the lag-time. such as stomach emptying. one support point is injected halfway between the lag-time and the first observation. c(t). Furthermore. The drug level profile. absorption window. resulting from the above input function is. The extra support point is well suited to accommodate an initial “burst” release commonly encountered for many formulations.The WinNonlin Toolbox Numeric deconvolution 12 hj ( t ) = 0 . for the input function. finite duration drug releases to be considered together with other factors that result in discontinuous delivery. pH changes. The wavelet support points (Tj) are constrained to coincide with the observation times. The input function. which explains the scaling with the δ parameter. etc. The hat-type wavelet representation enables discontinuous. otherwise where Tj are the wavelet support points. according to the linear disposition assumption. Having just one support point prior to the first observation would limit this capacity. The dispersion function fd(t) is defined as an exponential function: fd ( t ) = e ----------δ where δ is denoted the dispersion or smoothing parameter. given by: c ( t ) = f p ( t )*f d ( t )*c d ( t ) where “*” is used to denote the convolution operation. Consider. Third. including WinNonlin’s approach. Thus.12 User’s Guide Deconvolution through convolution methodology WinNonlin deconvolution uses the basic principle of deconvolution through convolution (DTC) to determine the input function. fd(t). Due to the stochastic transport principles involved. The deconvolution method implemented in WinNonlin is novel in the way the regularization (smoothing) is implemented. Decreasing the value of the dispersion function parameter δ results in a decreasing degree of smoothing of the input function. the convolution operation that leads to the drug level response c(t). 266 WinNonlin . a convolution operation acts like a “washout of details”. The WinNonlin deconvolution method instead introduces the regularization directly into the input function through a convolution operation with the dispersion function. The more modern DTC methods. that is smoother than the precursor function. The objective function may be based solely on weighted or unweighted residual values (observed–calculated drug levels). f(t). the drug level at time t is made up of a mixture of drug molecules that started their journey to the sampling site at different times and took different lengths of time to arrive there.g. integral of squared second derivative). Regularization methods in some other deconvolution methods are done through a penalty function approach that involves a measurement of the smoothness of the predicted drug level curve (e. a smoothing due to the mixing operation inherent in the convolution operation. the drug level response at time t depends on a mixture of prior input. consider both the residuals and other properties such as smoothness of the total fitted curve in the definition of the objective function. The purely residual-based DTC methods ignore any behavior of the calculated drug level response between the observations. The finer details (higher frequencies) are attenuated relative to the more slowly varying components (low frequency components). DTC methods differ basically in the way the input function is specified and the way the objective function is defined. In essence. for example. WinNonlin allows the user to control the smoothing through the use of the smoothing parameter δ. The DTC method is an iterative procedure consisting of three steps. First. the convolution of the precursor function with the dispersion function results in an input function. i.e. the input function is adjusted by changing its parameter values. Second. the new input function is convolved with cδ (t) to produce a calculated drug level response. It is exactly this mixing in the convolution operation that provides the smoothing. Thus. The three steps are repeated until the objective function is optimized. The convolution operation acts essentially as a low pass filter with respect to the filtering of the input information. the agreement between the observed data and the calculated drug level data is quantitatively evaluated according to some objective function. The WinNonlin Toolbox Numeric deconvolution 12 Similarly, larger values of δ provide more smoothing. As δ approaches 0, the dispersion function becomes equal to the so-called Dirac delta “function”, resulting in no change in the precursor function. The smoothing of the input function, f(t), provided by the dispersion function, fd(t), is carried forward to the drug level response in the subsequent convolution operation with the unit impulse response function (cδ ). Smoothing and fitting flexibility are inversely related. Too little smoothing (too small a δ value) will result in too much fitting flexibility that results in a “fitting to the error in the data.” In the most extreme case, the result is an exact fitting to the data. Conversely, too much smoothing (too large a δ value) results in too little flexibility so that the calculated response curve becomes too “stiff” or “stretched out” to follow the underlying true drug level response. Cross validation principles WinNonlin's deconvolution function determines the optimal smoothing without actually measuring the degree of smoothing. The degree of smoothing is not quantified but is controlled through the dispersion function to optimize the “consistency” between the data and the estimated drug level curve. Consistency is defined here according to the cross validation principles. Let rj denote the differences between the predicted and observed concentration at the j-th observation time when that observation is excluded from the data set. The optimal cross validation principle applied is defined as the condition that leads to the minimal predicted residual sum of squares (PRESS) value, where PRESS is defined as: m PRESS = ∑ rj j=1 2 For a given value of the smoothing parameter δ, PRESS is a quadratic function of the wavelet scaling parameters x. Thus, with the non-negativity constraint x > 0, the minimization of PRESS for a given δ value is a quadratic programming problem that has a unique solution that can be determined numerically. Let PRESS (δ) denote such a solution. The optimal smoothing is then determined by finding the value of the smoothing parameter δ that minimizes PRESS (δ). This is a one variable optimization problem with an embedded quadratic-programming problem. 267 12 User’s Guide User control over execution WinNonlin WinNonlin's deconvolution function permits the user to override the automatic smoothing by manual setting of the smoothing parameter. The user may also specify that no smoothing should be performed. In this case the input rate function consists of the precursor function alone. Besides controlling the degree of smoothing, the user may also influence the initial behavior of the estimated input time course. In particular the user may choose to constrain the initial input rate to 0 (f(0) = 0) and/or constrain the initial change in the input rate to 0 (f'(0) = 0). By default WinNonlin does not constrain either initial condition. Leaving f(0) unconstrained permits better characterization of formulations with rapid initial “burst release,” e.g., extended release dosage forms with an immediate release shell. This is done by optionally introducing a bolus component (or “integral boundary condition”) for the precursor function so the input function becomes: f(t) = (fp(t) + xδ(t)) * fd(t) = fp(t) * fd(t) + xδ fd(t) where the fp(t) * fd(t) is defined as before. The difference here is the superposition of the extra term xδ fd(t) that represents a particularly smooth component of the input. The magnitude of this component is determined by the scaling parameter xδ , which is determined in the same way as the other wavelet scaling parameters previously described. The estimation procedure is not constrained with respect to the relative magnitude of the two terms of the composite input function given above. Accordingly, the input can “collapse” to the “bolus component” xδ fd(t) and thus accommodate the simple first order input commonly experienced when dealing with drug solutions or rapid release dosage forms. A drug suspension in which a significant fraction of the drug may exist in solution should be well described by the composite input function option given above. The same may be the case for dual release formulations designed to rapidly release a portion of the drug initially and then release the remaining drug in a prolonged fashion. The prolonged release component will probably be more erratic and would be described better by the more flexible wavelet-based component fp(t) * fd(t) of the above dual input function. Constraining the initial rate of change to 0 (f'(0) = 0) introduces an initial lag in the increase of the input rate that is more continuous in behavior than the usual abrupt lag time. This constraint is obtained by constraining the initial value of the precursor function to 0 (fp(0) = 0). When such a constrained pre268 The WinNonlin Toolbox Numeric deconvolution 12 cursor is convolved with the dispersion function, the resulting input function has the desired constraint. The extent of drug input The extent of drug input in the Deconvolution module is presented in two ways. First, the amount of drug input is calculated as: t Amount input = ∫ f ( t ) dt 0 Second, the extent of drug input is given in terms of fraction input. If test doses are not supplied, the fraction is defined in a non-extrapolated way as the fraction of drug input at time t relative to the amount input at the last sample time, tend: t t end Fraction input = ∫ f ( t ) dt ⁄ ∫ f ( t ) dt 0 0 The above fraction input will by definition have a value of 1 at the last observation time, tend. This value should not be confused with the fraction input relative to either the dose or the total amount absorbed from a reference dosage form. If dosing is entered, as detailed under “Deconvolution dosing” on page 272, the fraction input is relative to the dose amount. Using deconvolution Numeric deconvolution is used to evaluate in vivo drug release and delivery when data from a known drug input are available. WinNonlin’s Deconvolution tool can estimate the cumulative amount and fraction absorbed over time for individual subjects, given PK profile data and dose per subject. To use numeric deconvolution: 1. Obtain the unit impulse response function (UIR). For example, fit time and concentration data in WinNonlin using PK model 1, 8, or 18 (for N=1, 2, or 3, respectively), and adjust to reflect a single unit dose. The UIR must take the form: 269 12 User’s Guide WinNonlin N cδ ( t ) = ∑ Aj e j=1 –αi t Since the UIR assumed the response from 1 dose unit, for model 1 (one compartment), take V (volume) from the 1-compartment IV model output and let A = 1/V. Make units corrections if needed (see below). For model 8 (two compartment), take the A and B from the model 8 output and divide by the Stripping Dose to get A1 and A2 for the UIR. Again, perform the units conversion if needed. Use model 18 (three compartment) similarly model 8. If the dose units differ from the concentration mass units in the deconvolution input dataset, take care in entering the A values: A values must use the units of the concentration data divided by the units of the dose used to determine the UIR. (See the example immediately below.) Alternately, use the Column Properties dialog to convert the concentration data to use the same mass units as the dose, or convert the dose amounts to use the concentration mass units. For example, suppose the oral concentrations use ng/mL, and a 1 mg dose was used in determining the unit impulse response. In the one-compartment case, A = 1/V, so the units for 'A’ in the model 1 output equal the inverse of the volume units. ‘A’ values must be converted to ng/mL/mg before they are used in the Deconvolution tool, i.e., A = 10^6/V. 2. Once the UIR is established and units adjustments are complete, open a data set containing individual PK profiles in a WinNonlin workbook. 3. Choose Tools>IVIVC>Deconvolution (IVIVC license) or Tools>Deconvolution (no IVIVC license) from the WinNonlin menus. 4. Define the variable mappings, dosing and model settings as detailed under “Deconvolution variables” below, “Deconvolution dosing” on page 272 and “Deconvolution settings” on page 273. 5. Click Calculate. Deconvolution generates a new chart and workbook, discussed under “Deconvolution output” below. 270 The WinNonlin Toolbox Numeric deconvolution 12 Deconvolution variables Numeric deconvolution requires a data set containing time-concentration data, with sort variables identifying individual profiles. To select variables for numeric deconvolution: 1. Open the Variables tab of the Deconvolution dialog (Tools>IVIVC>Deconvolution (IVIVC license) or Tools>Deconvolution (no IVIVC license)). 2. Dataset: If needed, select the data set containing individual PK profiles. 3. Sort variables: If one or more variables contain values that differentiate individual data profiles, drag those columns to the Sort Variables field. 4. Time: Drag the variable representing time (or other independent variable) to this field. 5. Concentration: Drag the dependent variable to this field. 6. Use the Save Settings and/or Load Settings buttons to save or load all settings in the Deconvolution dialog via an external file. 7. Proceed to “Deconvolution dosing” below. 271 12 User’s Guide Deconvolution dosing Numeric deconvolution supports input of test dose information directly to the Deconvolution dialog, so that the fraction input can be computed relative to the administered dose. Dose information is optional; without it, deconvolution will run correctly, with the fraction input approaching a final value of 1. Dose time is assumed to be zero for all sort levels. If a dose value is entered for any sort level, values are required for all sort levels (all or none). WinNonlin To define dosing for numeric deconvolution: 1. Open the Dosing tab of the Deconvolution dialog (Tools>IVIVC>Deconvolution (IVIVC license) or Tools>Deconvolution (no IVIVC license)). 2. Dose units: (Optional) Enter units for all doses here. The same units must apply to the UIR and the test (oral) doses entered in this dialog. Dose units are applied only if the Time and Concentration columns in the data set have units. See Table 3-2 and Table 3-3 on page 54 for accepted units abbreviations. 3. Dose: Enter the dose amount for each sort level in the input data. 4. Use the Save Settings and/or Load Settings buttons to save or load all settings in the Deconvolution dialog via an external file. 5. Proceed to “Deconvolution settings” below. 272 The WinNonlin Toolbox Numeric deconvolution 12 Deconvolution settings Numeric deconvolution in WinNonlin assumes a unit impulse response function (UIR) of the form: N cδ ( t ) = ∑ Aj e j=1 –αi t The analysis can use the observed times from the input data set, or in vivo times from an entirely separate workbook. To define model settings for numeric deconvolution: 1. Open the Settings tab of the Deconvolution dialog (Tools>IVIVC>Deconvolution (IVIVC license) or Tools>Deconvolution (no IVIVC license)). 2. Number of exponential terms in the unit impulse response: Select the number of exponential terms (N in the equation above, with 1 < N < 9). 3. A and alpha: For each profile, enter (or copy/paste) values for each exponential term in the unit impulse response function, e.g., values obtained by fitting WinNonlin PK model 1, 8, or 18 (for N=1, 2, or 3, respectively). 273 12 User’s Guide CAUTION: The A and alpha values must use the units shown in the A or alpha column WinNonlin headers (if any), to match the units used in the input data and dosing. In particular, A values must match the units of the concentration data. This may require conversion of A values if dose and concentration units differ. See “Using deconvolution” on page 269 for further discussion. 4. Smoothing: Select one of the following choices. • • • Automatic allows WinNonlin to find the optimal value for the dispersion parameter delta. None disables smoothing that comes from the dispersion function. If selected, the input function is the piecewise linear precursor function. User-Specified allows manual entry of the value of the dispersion parameter delta as the Smoothing Parameter, where delta > 0. Increasing delta increases the amount of smoothing. 5. Initial Rate is Zero: Check this option to constrain the estimated input rate to be zero at the initial time (lag time). 6. Initial Change in Rate is Zero: Check this option to constrain the derivative of the estimated input rate to be zero at the initial time (lag time) (requires initial rate of zero for automatic or user-specified smoothing). 7. Output data points: at the lower left area of the dialog, select the number and times for output data points by selecting from among the following options. a. Output from 0 to last time point: WinNonlin will generate output at even intervals from time 0 to the time of the final input data point for each profile, inclusive. Enter the total number of data points (<=1001) below. b. Output from X to Y: WinNonlin will generate output at even intervals from time X to time Y, inclusive. Enter the total number of data points (<=1001) below. c. Use observed times from input workbook: WinNonlin will generate output at each time point in the input data set for each profile. d. Use times from a workbook column: Select the workbook and time column. The worksheet containing the time values must be open in a WinNonlin workbook window, and can contain no more than 1001 times. 8. Click Calculate. Deconvolution generates a new chart and workbook, discussed under “Deconvolution output” below. 274 The WinNonlin Toolbox Numeric deconvolution 12 Deconvolution output WinNonlin numeric deconvolution generated workbook and chart output. The workbook output includes one workbook with the following six worksheets. Table 12-2. Deconvolution workbook output Worksheet Values Content Time, input rate, cumulative amount (Cumul_Amt, using the dose units) and fraction input (Cumul_Amt / test dose or, if no test doses are given, then fraction input approaches 1) for each profile The smoothing parameter delta and absorption lag time for each profile Values for the UIR exponential terms for each profile Parameters Exponential Terms Settings History Dosing Fitted Values Predicted data for each profile Input settings for smoothing, number of output points and start and end times for output History of operations on the workbook. See “History worksheet” on page 37. If user-specified test doses were entered in the Deconvolution dosing dialog, a Dosing worksheet is included in the output, listing the doses and dose units. Deconvolution generates one chart window with three charts for each profile: • • • Fitted Curve: observed time-concentration data vs. the predicted curve Input Rate: rate of drug input vs. time Cumulative Input: cumulative drug input vs. time 275 12 User’s Guide WinNonlin 276 Chapter 13 The IVIVC Toolkit In vivo-in vitro correlation and validation tools for formulation development Users who have a WinNonlin In Vivo-In Vitro Correlation IVIVC Toolkit License can access additional tools for evaluating correlations between in vitro drug properties, such as dissolution, and in vivo responses, such as fraction absorbed, in order to predict PK responses such as maximum concentration or area under the curve. WinNonlin IVIVC features are detailed under the following headings: • “Numeric deconvolution” on page 260: Users with or without an IVIVC Toolkit License can use deconvolution to evaluate drug absorption from time-concentration data, based on the unit impulse response. “Wagner-Nelson and Loo-Riegelman methods” on page 278: Estimate drug absorption assuming a one-compartment (Wagner-Nelson) or twocompartment (Loo-Riegelman) PK model. “Convolution” on page 293: Estimate or verify PK profiles based on the unit impulse response function and expected absorption profile. “Levy plots” on page 300: Comparing predicted and observed concentrations over time as a measure of correlation model fit. “Sigmoidal/dissolution models” on page 547: Sigmoidal models, including Hill, Weibull, Double Weibull and Makoid-Banakar, for modeling dissolution. “The IVIVC Wizard” on page 302: Structured environment for building, executing, saving and reloading common IVIVC work flows. • • • • • See the WinNonlin Examples Guide for examples demonstrating common IVIVC inquiries. 277 13 User’s Guide WinNonlin Wagner-Nelson and Loo-Riegelman methods Users with a WinNonlin IVIVC Toolkit License can access Wagner-Nelson and Loo-Riegelman methods to estimate fraction of drug absorbed. Wagner-Nelson inputs and calculations The Wagner-Nelson method estimates the fraction of drug absorbed over time, relative to the total amount to be absorbed, following the method described in Gibaldi and Perrier (1975) pages 130 to 133. It uses as a basis AUC values computed for each time point in your time-concentration data. Note: WinNonlin Wagner-Nelson computations assume single-dose PK data with a concentration value of 0 at dose time. If no concentration value exists at dose time, WinNonlin will insert a concentration value of 0. You can choose to calculate AUC using either the linear, linear/log, or linear up/log down trapezoidal rule. (See “Calculation methods for AUC” on page 177 and “AUC calculation and interpolation formulas” on page 183 for definitions and calculation methods.) Note that the two linear non-compartmental analysis (NCA) methods are equivalent, as no interpolation is used in the Wagner-Nelson method.) Similarly, you choose the method for computing Lambda Z: best fit, userspecified range, or user-specified value. If you specify a Lambda Z value, you may also enter an intercept or have WinNonlin compute it using the last positive concentration and associated time value: intercept = Lambda Z * tlast + ln(Clast) If you do not enter a value for AUCINF, it will be computed as for WinNonlin NCA: AUCINF = AUClast + Clast / Lambda Z You choose whether to use the observed or predicted value for Clast, where Clast_pred = exp(intercept - Lambda Z(tlast).) The cumulative amount absorbed at time t, normalized by the central compartment volume V1, is computed as: Cumul_Amt_Abs_V (t) = C(t) + Lambda Z * AUC(t) 278 The IVIVC Toolkit Wagner-Nelson and Loo-Riegelman methods 13 Therefore, Cumul_Amt_Abs_V at t = infinity is: Cumul_Amt_Abs_V(inf) = Lambda Z * AUCINF The relative fraction absorbed at time t is then: Rel_Fraction_Abs(t) = Cumul_Amt_Abs_V(t) / Cumul_Amt_Abs_V(inf) The percent remaining at time t is computed as: Percent remaining (t) = 100 * (1 - Rel_Fraction_Abs(t) ) Data points with a missing value for either time or the concentration will be excluded from the analysis and will not appear in the output. Loo-Riegelman inputs and calculations Like Wagner-Nelson, the Loo-Riegelman method estimates the fraction of drug absorbed as a function of time, relative to the total amount to be absorbed, following the method described in Gibaldi and Perrier (1975) pages 136 to 142. It computes AUC values for each time point in your time-concentration data set, as in Wagner-Nelson. For Loo-Riegelman, you must provide estimates for K10, K12, and K21, as may be obtained by running WinNonlin PK model 7 on separate IV data. Note: WinNonlin Loo-Riegelman computations assume single-dose PK data with a concentration value of 0 at dose time. If no concentration value exists at dose time, WinNonlin will insert a concentration value of 0. You may supply a value for AUCINF or allow WinNonlin to compute it as for non-compartmental analysis: AUCINF = AUClast + Clast / Lambda Z You can choose to use the observed or predicted value for Clast, where Clast_pred = exp(intercept - Lambda Z(tlast). As in the Wagner-Nelson method, you choose the method for computing Lambda Z: best fit, user-specified range, or user-specified value. You may specify an intercept for the last case. If the intercept is not specified, WinNonlin will compute it using the last positive concentration and associated time value: intercept = Lambda Z * tlast + ln(Clast) 279 13 User’s Guide Note that the Loo-Riegelman method uses Lambda Z only to compute AUCINF; it uses the intercept only to compute Clast_pred. The cumulative amount absorbed at time t, normalized by the central compartment volume V1, is: Cumul_Amt_Abs_V(t) = Conc(t) + K10 * AUC(t) + Amt_Peripheral_V(t) The amount in the peripheral compartment at time t, normalized by V1 is computed sequentially as: Amt_Peripheral_V(t) = (Prior_Amt_Peripheral_V) * exp(-K21*δt) + ( K12 * Prior_Conc / K21) * (1 - exp(-K21*δt) + ( K12 * δt * δConc / 2 ) See Gibaldi and Perrier (1975) pages 138-139 for the derivation of this equation. Cumul_Amt_Abs_V at time infinity is: Cumul_Amt_Abs_V(inf) = K10 * AUCINF The fraction absorbed at time t is computed as: Rel_Fraction_Abs(t) = Cumul_Amt_Abs_V(t) / Cumul_Amt_Abs_V(inf) The percent remaining at time t is computed as: Percent remaining (t) = 100 * (1 - Rel_Fraction_Abs(t) ) Data points with a missing value for either time or the concentration will be excluded from the analysis and will not appear in the output. Using the Wagner-Nelson or Loo-Riegelman method To use the Wagner-Nelson or Loo-Riegelman method: WinNonlin 1. Open the data set containing individual PK profiles in a WinNonlin workbook. 2. For deconvolution using a one-compartment PK model: choose Tools>IVIVC Tools>Wagner-Nelson from the WinNonlin menus. 280 Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280. X Variable: Drag the independent variable (generally. representing the newly-loaded model. 3. Run the analysis by choosing Model>Start Modeling from the WinNonlin menus. 4. For deconvolution using a two-compartment PK model: choose Tools>IVIVC Tools>Loo-Riegelman from the WinNonlin menus. time) to this field. 281 .The IVIVC Toolkit Wagner-Nelson and Loo-Riegelman methods 13 3. 6. 5. Y Variable: Drag the dependent variable (concentration) to this field. Define the variable mappings. loaded and refreshed as for any WinNonlin model. A model (text) window will appear. 2. model properties and options as detailed under the following headings: • • • • • • • • “Data variables” on page 281 “Dosing regimen” on page 282 “Loo-Riegelman model parameters” on page 283 (Loo-Riegelman deconvolution only) “Lambda Z ranges” on page 284 “Output options” on page 287 “Weighting” on page 288 “Title” on page 290 “Settings” on page 291 5. Sort Variables: Drag the variables needed to identify individual profiles to the Sort Variables list. 4. Choose Model>Data Variables from the WinNonlin menus. Data variables To define the data variables to use for the Wagner-Nelson or Loo-Riegelman method: 1. using the options on the File menu. The model and settings may be saved. skip to “Loo-Riegelman model parameters” on page 283 for LooRiegelman or “Lambda Z ranges” on page 284 for Wagner-Nelson. Dosing regimen If the data were loaded from PKS. WinNonlin 7. Carry Alongs: Drag variables to be included in the output workbook (but not used in calculations) to this list. Value: For each profile. In that case. and set the data variables (Model>Data Variables). 2. Click OK and proceed to set the Dosing regimen. Use the Save button to save the dosing information for later re-use. enter the dose time. WinNonlin will insert a concentration value of 0 at the specified dose time. or use the Load button to load previously-saved dosing information from a file. Choose Model>Dosing Regimen from the WinNonlin menus.13 User’s Guide 6. Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280. 4. 3. 282 . Note: If no concentration value exists at dose time for a given profile. To define the dosing: 1. the dosing should also have been loaded. enter initial parameter estimates for each parameter. use the buttons at the bottom of the dialog to load the parameter 283 . Alternately. and set the data variables (Model>Data Variables). To set up initial model parameters for the Loo-Riegelman method: 1. Parameter estimates may be obtained by running WinNonlin PK model 7 on the data prior to using the Loo-Riegelman tool. Choose Model>Model Parameters from the WinNonlin menus. skip to the Lambda Z ranges. 3. For Loo-Riegelman models. Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280. proceed to set the Loo-Riegelman model parameters. 2. 6. Click OK.The IVIVC Toolkit Wagner-Nelson and Loo-Riegelman methods 13 5. For Wagner-Nelson. Loo-Riegelman model parameters The Loo-Riegelman method requires initial estimates for the model parameters describing the two-compartment model that fits the IV data. Value: For each profile. For each profile. select Lambda Z and AUC settings as follows. Choose whether to enter Lambda Z information using a plot of each profile or by entering tabular data. WinNonlin 4. 4.13 User’s Guide values from a PK output workbook obtained by running WinNonlin PK model 7 on the data prior to this analysis. 3. Use the arrow buttons directly below the plot to move between profiles in the Plot view. Click OK and proceed to set the Lambda Z ranges. using the Graph and Data buttons at the top right. and set the data variables (Model>Data Variables). 2. Lambda Z ranges To set Lambda Z ranges for the Wagner-Nelson or Loo-Riegelman method: 1. Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280. 284 . Choose Model>Lambda z Ranges from the WinNonlin menus. where Clast is the last observed concentration. where Clast is the last predicted concentration. Note that for the Loo-Riegelman method. Area Under the Curve: Select one of the following options for calculation of area under the curve to time = infinity: Use AUCINF_pred: WinNonlin will calculate the area under the curve as: AUCINF = AUClast + Clast / Lambda Z. If AUCINF is a user-specified value. User specified AUCINF: Enter a value for the area under the curve. Note: If observed data does not extend significantly into the terminal phase (i. the Lambda Z options apply only for the AUCINF computation. resulting in scaling errors in the computation of fraction absorbed.e. 6. absorption phase must be completed) then use of calculated AUCINF may significantly underestimate the actual AUCINF. Lambda Z Calculation Method: Select one of the following options: 285 . then the other Lambda Z settings are unnecessary for the Loo-Riegelman method.The IVIVC Toolkit Wagner-Nelson and Loo-Riegelman methods 13 5. Use AUCINF_obs: WinNonlin will calculate the area under the curve as: AUCINF = AUClast + Clast / Lambda Z. click Apply to save the settings and move to another tab of the Model Properties dialog. Select the first data point to be included: Left-click on a data point or enter a time value under Start Time in the Data view. Intercept and K values entered must be greater than 0. Repeat to reverse an exclusion. When done. c.) Enter a value for Lambda Z and.13 User’s Guide Best fit: If this option is selected in the Plot view. for the intercept. These exclusions apply only to Lambda Z calculations. User Time Range: Define the terminal elimination phase as follows: a. If using the Plot view. To save the settings and close the dialog. hold down the Ctrl key and left-click on that point in the Plot view or enter the time value under Exclusions (or in the next column to the right. Proceed to set the Output options. the Lambda Z table in the output will give Predicted and Residual values for all Time and Conc values. WinNonlin will determine the Lambda Z range as detailed under “Lambda Z or slope range selection” on page 162. Select the last data point to be included: Hold down the Shift key while left-clicking on a data point or enter a time value under End Time. Exclusions (optional): To exclude a data point from Lambda Z calculations. If no intercept value is entered. select the axis scale for the graphical representation below the lower left corner of the plot: Linear or Log Linear. User Specified Lz and Optional Intercept: (Available for AUC calculated as AUC_pred only. 8. or Start Time. WinNonlin will compute it as: intercept = Lambda Z * tlast + ln(Clast). 286 . Since no specific points will be used in the computation of Lambda Z. End Time and Lambda Z are empty in the Data view. optionally. the excluded data points will be included in computation of AUC’s. click OK. Lambda_z. WinNonlin 7. d. b. for multiple exclusions) for that profile in the Data view. Note: Any AUC. 2. Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280. with content as selected in the following options. 287 . After the analysis is run.The IVIVC Toolkit Wagner-Nelson and Loo-Riegelman methods 13 Output options To define the output options for the Wagner-Nelson or Loo-Riegelman method: 1. Note: Set default output options via Tools>Options in the WinNonlin menus. These worksheets present the same information as the text output. Choose Model>Model Options from the WinNonlin menus and open the Output Options tab. Workbook: Check this option to generate an output workbook. The workbook contains the worksheets summarized in Table 13-1. but in tabular form. and set the data variables (Model>Data Variables). one or more new windows is generated. 3. Check the options to apply from among the following choices. including time. Loo-Riegelman Lambda Z Fit User Settings History Chart: If workbook output is selected. amount in the peripheral compartment. Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280. cumulative AUC and cumulative amount and relative fraction absorbed. Loo-Riegelman method only. Units are those of the input data. User defined settings History of actions done in and to the workbook. Not included for Loo-Riegelman with userspecified AUCINF. If Lambda Z is user-specified. and relative fraction absorbed. 288 . “History worksheet” on page 37. Details for fitting Lambda Z. including time. Page Breaks: Check this box to include page breaks in the ASCII text output. Time. containing the same information as the workbook output. Click OK and proceed to set the Weighting. Summary of output worksheets Sheet name Wagner-Nelson Contents Wagner-Nelson method only. 4. Modeling errors are also reported in this file.13 User’s Guide . concentration and cumulative AUC. To define the weighting scheme to use in fitting Lambda Z: 1. the latter plot is excluded. amount absorbed. and set the data variables (Model>Data Variables). check this item to generate a Chart window with two plots per profile: a linear plot of Relative_Fraction_Absorbed vs. WinNonlin Table 13-1. Profile estimates for each level of the sort variable. Weighting The weighting options for the Loo-Riegelman and Wagner-Nelson methods apply to the calculation of Lambda Z (see “Lambda Z ranges” on page 284). Profile estimates for each level of the sort variable. Text: This option provides an ASCII text version of the analysis results. and a plot of the Lambda Z fit (excluded for Loo-Riegelman with user-specified AUCINF). concentration. Units are those of the input data. 2. or use the Pre-selected scheme button to select the most common weighted least square options: 1/Y and 1/Y2. Note that it is not the weights themselves that are important. if this were the case. 2. 289 . Common schemes are detailed below. Select the desired weighting scheme. Select the Pre-selected Scheme radio button. concentrations between 0 and 1 would have negative weights. rather it is the relative proportions of the weights. To use uniform weighting: 1. and therefore could not be included. However. When weighting by 1/Y and using the Linear/Log trapezoidal rule.The IVIVC Toolkit Wagner-Nelson and Loo-Riegelman methods 13 2. Choose Model>Model Options from the WinNonlin menus and open the Weight tab. To use weighted least squares: 1. approximately equal to 1/Yhat2. it already uses weighting implicitly. See “Weighting” on page 241 for discussion of weighting schemes. rather than 1/Y. one could argue that the weights should be 1/LogY. 3. Select Observed to the Power n button and enter the value of n. CAUTION: When a log-linear fit is done. Select Uniform Weighting from the pull-down list. 4. and at the top of each chart in chart output. and set the data variables (Model>Data Variables). Click OK and proceed to set the Title. Choose Model>Model Options from the WinNonlin menus and open the Title tab. the title will not display even if text is entered below. 290 . Enter the title text below. Make sure that Title Text is checked. To define and include the title: 1. Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280.13 User’s Guide WinNonlin 3. Title If included. 2. The title will be centered on the page. It may include up to 5 lines of text. the title is displayed at the top of each printed page in ASCII text output. To include a line break use Shift + Enter. on the User Settings worksheet in the workbook output. 3. If it is not. 2.The IVIVC Toolkit Wagner-Nelson and Loo-Riegelman methods 13 5. Click OK and proceed to the IVIVC Settings tab. Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280. and set the data variables (Model>Data Variables). Choose Model>Model Options from the WinNonlin menus and open the Settings tab. 291 . Settings To set the AUC calculation method for Wagner-Nelson or Loo-Riegelman: 1. The Linear/Log Trapezoidal. Linear Up/Log Down Method (method 3). and the logarithmic trapezoidal rule is used any time that the concentration data is decreasing. Similarly. and the Linear Trapezoidal (with Linear/Log Interpolation) methods all apply the same exceptions in area calculation and interpolation: If a Y value (concentration. since no interpolation is done. Equivalent to Linear Trapezoidal rule. Linear Trapezoidal rule (Linear Interpolation). 292 .13 User’s Guide 3. applied to each pair of consecutive points in the data set that have non-missing values. and sums up these areas. given under “AUC calculation and interpolation formulas” on page 183. If Cmax is not unique. the program defaults to the linear trapezoidal or interpolation rule. Four methods are available. Note: 4. then the first maximum is used. or after C0 for bolus (if C0 > Cmax). WinNonlin • • • Note: Methods 1 and 4 are the same for Wagner-Nelson and Loo-Riegelman. All methods reduce to the log trapezoidal rule and/or linear trapezoidal rule. Calculation Method: Select the method for computing area under the curve. Uses the log trapezoidal rule (given under “AUC calculation and interpolation formulas” on page 183) after Cmax. since interpolation is not used in Loo-Riegelman and Wagner-Nelson methods. but differ with regard to when these rules are applied. Click OK. • Linear/Log Trapezoidal Method (method 1). Linear Trapezoidal Method with Linear/Log Interpolation (method 4). The method chosen applies to all AUCINF computations. the Linear Up/Log Down. The linear trapezoidal rule is used any time that the concentration data is increasing. rate or effect) is less than or equal to zero. if adjacent Y values are equal to each other. the program defaults to the linear trapezoidal or interpolation rule for that point. and linear trapezoidal otherwise. This method (method 2) uses the linear trapezoidal rule. F ( t ) ⋅ g )(t) = ∫0 f ( t – u )g ( u ) du t where f(t) and g(t) are constrained to be either polyexponentials or piecewise polynomials (splines). You can choose to convolve either this spline function. For a spline function. the A's and alpha's. if cumulative input data is fit with linear splines. and will use these polynomial splines in the convolution. e.The IVIVC Toolkit Convolution 13 Convolution The WinNonlin Convolution tool supports analytic evaluation of the convolution integral.. Splines must be piecewise constant or linear splines. To specify a polyexponential.g. m = 0. the Convolution tool will fit the data with the requested degree of polynomial splines. or the derivative of this spline function. so the piecewise constant function is convolved with the user's other specified function. you must supply a dataset to be fit with the splines.e. you must supply the polyexponential coefficients. then the derivatives are piecewise constant. Polyexponentials must take the form: N f(t) = ∑ Ai e –( αi ⋅ t ) i=1 where N <= 9 and t >= 0. otherwise where N is restricted to be 1 or 2. 2. and use the derivatives in the convolution. for cumulative input data (so the derivative is input rate data). tε ( t j. they must take the form: N i–1 f(t) = ∑ bij ( t – tj ) i=1 . For the first case.. 293 . e.. the Convolution tool will find the spline with the requested degree of polynomial splines. For example. j = 1. For the derivative of a spline.e.g. i. differentiate the splines to obtain polynomial splines of one degree lower. in any combination. t j + 1 ). i.. …. for input rate data. Open the data set(s) to be convolved in a WinNonlin workbook. with the constants being the slopes mi. Select the appropriate function type and settings for the Input function and Unit Impulse Response (UIR) function. as for input rate data.9 sets of coefficient (“A’s”) and exponent (“alphas”) values per subject. The derivative of a linear spline function is a piecewise constant function. • • 294 .ti). If two polyexponentials are to be convolved. ti+1). For input rate data.ti). and set the degree of polynomial splines (Linear or Constant). This option is not available for cumulative input data since the derivative will be zero. where mi is the slope (yi+1 . Select Use this fitted function to fit the data with the piecewise polynomial. 2. • Splines: Convolution will find the polynomial coefficients. Polyexponential: Optionally. you may select a dataset containing 1 . To use convolution: WinNonlin 1. For t ε (ti. y = yi + mi (t . either could represent the input function or unit impulse response function. 3. Select the workbook containing time concentration data to be fit under Required Dataset. Select Use derivative of the fitted function to fit the data with the derivative of the piecewise polynomial (first of two polynomials only). The Convolution Wizard appears. each piecewise constant spline will have the value that is the average of the y-values of the endpoints. See the above discussion for detail on function types and data requirements.13 User’s Guide Linear splines are the lines that connect the (x. none of the alpha values for the first polyexponential can be the same as any of the alpha values for the second polyexponential.yi) / (ti+1 .y) points. Note that 'linear splines' are required for cumulative input data. Choose Tools>IVIVC Tools>Convolution from the WinNonlin menus. The first dialog defines the functions f(t) and g(t) described above. as for cumulative input data. Note that f(t) and g(t) are interchangeable. to proceed to “Convolution settings”. depending on the function types selected in the Convolution Function dialog. described under “Convolution” on page 293. 2) WinNonlin will use the cross-product of any unmatched sort keys. if you selected two polyexponentials without optional input data sets. 5. 295 . Note: When both functions use an input data set: 1) the data will be joined on any sort keys that have identical names in both data sets. Click Next to proceed to “Data variables” or. Use the Save or Load button to save or load convolution settings to/from an external file. Data variables The options in this dialog vary. The example below shows the setup for one polynomial and one polyexponential with an optional input data set.The IVIVC Toolkit Convolution 13 4. they are not used in calculations. above. Sort Variables: Drag variables needed to identify individual profiles to this field. 2. See the above note about sort keys. 4. Carry Alongs: Drag data columns to include in the output to this field. Cumulative Input: Drag the dependent variable to this field. 3. To set up a polyexponential function that uses a dataset for exponentials: 1. These columns will be carried as-is to the results. Time: Drag the independent variable to this field. 296 . Choose the number of coefficient/exponent pairs under Number of exponential terms. Sort Variables: Drag the variables needed to identify individual profiles to this field. 2. Note that the data set may contain no more than one time record for each profile (unique combination of Sort variable values).13 User’s Guide To set up a polynomial function (splines): WinNonlin 1. See the note about sort keys. depending on the function types selected in the Convolution Function dialog. respectively. 4. inclusive. Select the time option: • Use default times: (not available for two polyexponentials) Output will include time points from 0 (zero) to one of the following: – For two polynomials. twice the last time value in the data set • Output from X to Y: Output will spread the number of data points evenly across times from X to Y. To set the number and times of output data points (optional): 1. as must all Alphas. These columns will be carried as-is to the results. Terms Variables: Drag each coefficient and exponent pair to the A and Alpha fields. in the same row. they are not used in calculations. The example below shows the setup for one polynomial and one polyexponential without an optional input data set. described under “Convolution” on page 293. Click Next to proceed to “Convolution settings”. 297 . Convolution settings The options in this dialog vary. All A’s must use the same units. 2.The IVIVC Toolkit Convolution 13 3. Carry Alongs: Drag data columns to include in the output to this field. 5. the last time value in one of the data sets plus the last time value in the other dataset – For one polynomial. Enter the total number of output data points under Number of output data points. Choose the number of coefficient/exponent pairs to enter under Maximum number of exponential terms. The output is described under “Convolution output”. The units can be blank (no units). Use the abbreviations detailed in Table 3-2 on page 54. Optionally. their column units will apply. Convolution output Convolution generates workbook and chart output. Units in the output are described under “Convolution units” on page 299.13 User’s Guide WinNonlin To set up a polyexponential function that did not use a dataset for exponentials: 1. 3. Note that alpha units must equal 1/(Time units). enter units for the coefficients (A) and exponents (alphas) to the right. 4. The workbook contains the worksheets listed in Table 13-2. If the A’s and alphas are drawn from a workbook. 298 . 2. A’s and Alphas: Enter values for each coefficient and exponent pair together in the same row. The chart output includes one plot of the convolved data over time for each level of the sort variables. Click Calculate to run the convolution. then the convolved data has units: ted function" Time units * Input Rate units * Y units The two Time columns must use the same units (if any). Table 13-3. with option "use provided units for the A's.If the Time and Cumulative Input columns have units.The IVIVC Toolkit Convolution 13 Table 13-2. Convolution workbook output Worksheet name Convolution Settings History Content Two columns: Time and the convolved data points for each level of the sort variables Settings and input workbook names for the convolution Record of operations on the workbook. and if the alpha units are 1 over a time unit.columns have units. If alpha and A units are provided. and you have nential. The two Time columns must use the same units (if any). then the convolved data has units: tive of this fitted funcCumulative Input units * Y units tion" Polyexponential and polyexponential All alpha units must be the same. then the convolved data have units: derivative of this fitted Cumulative Input units * A units function" Splines and splines. “History worksheet” on page 37. If all input with option "use this fit.) * A units (2nd poly. Time units * Input Rate units * A units Splines and polyexpo. Note that the alpha units must equal 1/(Time units).If the Time and Input Rate columns have units.columns have units. as follows. Convolution units The units in convolution output differ with the functions used. then the convolved data has units: Time units from alpha units * A units (1st poly. and you have pronential. then the convolved data have the following this fitted function" units. with option "use deriva. all A units must also be the same. Convolution units Functions Units in output Splines and polyexpo.) 299 . with option "use vided units for the A's. If all input Splines and splines. drag the variable identifying different drug formulations from the Variables list to the Formulation field. WinNonlin will use the cross-product of sort values to perform the plotting and modeling. 4. drag the variable containing sample times from the Variables list to the Independent field. 6. using linear interpolation to calculate data points as needed. drag the variable containing sample data from the Variables list to the Values field. Matched time pairs will be sampled from the data within the range from zero to the maximum observed dissolution or deconvolved absorption value. 7. The two data sets may contain the same and/or different sort variables. WinNonlin will include only those sort values that exist in both data sets. WinNonlin will group data by formulation within each Levy Chart. as well as a line at unity. Dataset: Select the appropriate data sets... must match in the two data sets. if any. To create one or more Levy plots: 1. Formulation (optional): For each data set. and the correlation (R2) value and the slope for the model line. one containing in-vivo and another containing in-vitro data. Each 300 .13 User’s Guide WinNonlin Levy plots Users with a WinNonlin IVIVC Toolkit License can create Levy plots to compare in-vivo absorption versus in-vitro dissolution data visually. 5. 3. For nonmatching sort variables. The plots include a linear fit of the data points with intersection at origin. WinNonlin uses these data sets to plot in-vivo versus in-vitro times at user-specified Y values (typically. Values: For each data set. For sort variables that exist in both data sets (identical column name). Sort Variables (optional): Drag any variables required to identify separate plots from the Variables list to the Sort Variables field for one or both data sets. 2. Choose Tools>IVIVC Tools>Levy Plots. one each for the in-vivo (vertical axis) and in-vitro (horizontal axis) profiles. Levy plots require two separate workbooks. Units for these variables. Independent: For each data set. from the WinNonlin menus to open the Levy Plots dialog. whichever is smaller. fraction absorbed or dissolved). Open the in-vivo and in-vitro data sets in separate WinNonlin workbooks. in the corresponding Dataset fields. as time-versus-time plots at matched values for absorption and dissolution. use Ctrl+Enter to insert line breaks. Carry Alongs (optional): For each data set. Data for formulation values that appear in only one data set will not appear in the chart or workbook output (for a given profile. Plot labels: If desired. 30. 9. The same selections apply across all sort levels. To enter more than one title line. The maximum Y value shown will be lesser of the maximum observed fraction dissolved and the calculated fraction absorbed. Show datapoint labels: Check this box to display the numeric time values with each data point in the charts as (T(in-vitro). 301 . then at the specified interval up to the end of the sample data range.. T(in-vivo)). Plot at these values: Select this radio button and enter a list of Y values to plot interpolated times at those values.02) rather than percents (e. WinNonlin will use only those formulation values that appear in both data sets. edit the chart title and axis labels in the appropriate fields at the bottom of the dialog box. no output will appear at all. in the units of the Y variables. Values to Match: WinNonlin will plot in-vivo versus in-vitro times corresponding to the Y values selected here. WinNonlin will multiply the values by 100 before plotting. For example. if sort keys are used). etc. The first value will appear at Y=0. Datapoints are fractional: Check this box if the data are expressed as fractions (e. b. 10. all other values are computed using simple linear interpolation. d. These variables are not used in the plots. 2). Plot Values at every N: Select this radio button to plot interpolated time values at regular intervals across the sample data. up to the end of the input data. drag any additional variables to include in the output workbook from the Variables list to the Carry Alongs field. If none of the formulation values match between data sets for any sort key. 10. Note: The summary line on the Levy plot is fit to all formulations combined. 20. a.. Enter the interval to the right. WinNonlin will use the data. 0. c.g. 8.The IVIVC Toolkit Levy plots 13 formulation will appear in a different color with a corresponding entry in the plot legend.g. Note: When a Y value to be plotted exists in the data set. WinNonlin will not plot any points that would be extrapolated from one of the data sets. an interval value of 10 would generate plot values at Y = 0. The Wizard supports setup and running of all or a subset of operations required to create an IVIVC. Use the Save Settings button to save entries in the dialog to an external file for later use. including: sort variables. and Carrys Along variables (if included). See “History worksheet” on page 37. Output The Levy Plot tool outputs one workbook and one plot window. The workbook contains the following worksheets: Table 13-4. and for each unique combination of sort variable values: X. Matched_Y. The plot window contains one Levy plot for each level of the sort variables. including estimates and statistics for the Yintercept (INT) and slope. Output from fitting model 502. IV or IR formulation. It provides workflow support for complete.IVC). 12. Use the Load Settings button to load values from such a file. Predicted data based on linear regression. Levy workbook output Worksheet Plotted Values Content Values used in the plots. including: sort variables (if included) and for each unique combination of sort variable values: Formulation (if included). Vivo_Time. When all settings are complete. History of operations on the workbook. if any. validate it and use it to make predictions.13 User’s Guide 11. click the Create Chart button to generate the Levy plots and accompanying workbook. Any set of operations set up in the Wizard can be saved and reloaded as an IVIVC project (*. two-stage IVIVC development. including interpolated and original data values (where interpolation wasn’t necessary). Vitro_Time. WinNonlin Predicted Values Parameters History The IVIVC Wizard The IVIVC Wizard provides an organized environment in which to build and manipulate in-vitro in-vivo correlation (IVIVC) models using the underlying functionality described under “The IVIVC Toolkit” on page 277. Y and Carry Along variables (if included). including the 302 . plus a feature to automatically estimate the unit impulse response (UIR) for one or more profiles given an oral. • “In-vitro tab” on page 306 303 . from the WinNonlin menus to launch the IVIVC Wizard.g. and another with PK profiles. e. either individual or average. set your monitor resolution to a minimum setting of 1024 x 768. To use the IVIVC Wizard: 1. Open the necessary data set(s) in WinNonlin workbook windows. 2. Note: To maximize IVIVC Wizard operation. Choose Tools>IVIVC>IVIVC Wizard.. with one profile per formulation. See the WinNonlin Examples Guide for examples demonstrating a full IVIVC workflow. You must have a valid WinNonlin license and a valid WinNonlin IVIVC Toolkit License installed in order to access this menu. one workbook for dissolution data.. 3. Note that in-vitro and in-vivo data must appear in separate workbooks. Individual functions are run from the separate tabs of the IVIVC Wizard as detailed in the following sections.The IVIVC Toolkit The IVIVC Wizard 13 related data sets. Close any models that are open in WinNonlin. minimize the main WinNonlin window using the third button from the right end of the WinNonlin title bar.13 User’s Guide • • • • Note: “In-vivo tab” on page 307 “Correlation tab” on page 310 “Prediction tab” on page 313 “Options” on page 315 WinNonlin Output generated from within the IVIVC wizard should be refreshed by saving. Process is not ready to run. Use the buttons next to the items to view related data or settings. Process is ready to run. but not up to date due to changes to the data in WinNonlin or the settings in the IVIVC Wizard. Note that there is no requirement to complete all items. the IVIVC Wizard remains in the Windows foreground so long as it is open and WinNonlin is not minimized. To minimize the IVIVC Wizard (and leave it open). Project Status The right hand pane of the wizard shows the status of steps that may be completed in the process of creating. Project status color codes Color Green Yellow Red Meaning Process has been run successfully and is up to date. Table 13-5. Each item appears next to a square that can take the following colors and meanings. The Dependencies dialog will not work for IVIVC Wizard operations and output. 304 . one or more required settings or data items is not available. Minimizing the IVIVC Wizard To avoid potential issues with other WinNonlin dialogs. testing and using and IVIVC. reloading and re-running the IVIVC project. Choose Tools>IVIVC>IVIVC Wizard. all the WinNonlin objects associated with the IVIVC project will be saved to the same directory. It will not be possible to load a saved . button at the top right corner of the wizard. charts) together.ivc file. 305 . Click Yes to proceed. Note: Use PKS>Save or PKS>Save As to save an IVIVC project and. To load a saved IVIVC project: 1. text. Click Open to load the project into the IVIVC Wizard. The WinNonlin objects associated with the IVIVC project can be easily recognized because they have had the IVIVC project name prefix associated with their filenames. Locate and select the settings (*.ivc file if any of the associated files are not present in the same directory. 2. Set up the desired project settings in the IVIVC Wizard either manually or by loading a previously-saved project. To save an IVIVC project to your hard drive: 1.The IVIVC Toolkit The IVIVC Wizard 13 Saving and loading IVIVC projects You may save any collection of settings.. with or without the file extension (*. from the WinNonlin menus. Click Save. Click the Load Project button at the top right corner of the wizard. 3. output.. navigate to an appropriate directory and enter any Windows-compatible filename.ivc + workbooks. to the PKS. 4. input data and output in the IVIVC Wizard as an IVIVC project.IVC) file...IVC). Note: Once a user saves an . Keep the wizard open. If any windows are open in WinNonlin. be sure to move all files (. In the Save Project dialog. optionally. Note that any open windows must be closed in WinNonlin before an IVIVC project can be loaded. Click the Save Project. 4. If the project files must be moved to another location. 2. the Wizard will warn you that those objects will be closed. 3. Run any analyses for which you would like to save data as part of the project. dissolution and formulation identifiers. Variables: Drag column names from the Variables list to the appropriate field on the right to identify those containing time. Sheet: Select the worksheet containing the percent dissolution data.13 User’s Guide In-vitro tab The IVIVC Wizard. Dissolution data partitioning Drag and drop a formulation to one or more of the following to identify: 306 . from among those open in WinNonlin. as fraction dissolved over time. test and reference (target) formulations. and to fit a dissolution model to smooth the data if desired. WinNonlin Dissolution data selection Dataset: Select the workbook containing your dissolution data. In-Vitro tab includes settings to identify your dissolution data. See “Model parameters” on page 197 for instructions. optionally based on observed Y or predicted Y (Yhat) values. Click Fit Dissolution Data to run the model or linear interpolation and generate output. 307 . Select one of the IVIVC models described under “Sigmoidal/dissolution models” on page 547. Finf. Alternately. (See “Prediction tab” on page 313. Depending on the model. to use linear interpolation on the raw data instead of modeling. For example. In-Vivo tab includes settings to identify your time-concentration or absorption data. See also “Weight” on page 205.2. Use this tab to fit a unit impulse response (UIR) function to reference concentration data. and/or to deconvolve an existing UIR and concentration data in order to estimate fraction of drug absorbed over time.1. (See “Correlation tab” on page 310.The IVIVC Toolkit The IVIVC Wizard 13 Internal validation (required): Formulation(s) to be used in building the IVIVC model(s) and for validation on the Correlation tab. In-vivo tab The IVIVC Wizard.0 to prevent incomplete measurements from biasing other parameter estimates.3.1. formulations and dosages. can be fixed to 1. then set the following. the asymptotic fraction dissolved. you may wish to fix certain parameters to known values. via the Predictions tab. check Use raw data above the models.) Dissolution model Optionally. in order to provide smoothing you may fit the dissolution data to a sigmoidal dissolution model. Target batch (optional): Formulation to be used as the comparator in predictions. Weighting: Select how to weight data during modeling.) External validation (optional): Formulation(s) to be used in model validation. Model parameters: 1. The reference formulation must use all oral. IV bolus or IV infusion input. 308 .13 User’s Guide WinNonlin PK data selection Dataset: Select the workbook containing your time-concentration or timeabsorption data from among those open in WinNonlin. Variables: Drag column names from the Variables list to the appropriate field on the right to identify those identifying individual profiles (Sort variables). Formulation information Reference formulation: Select the identifier for the formulation to be used in computing the UIR (and also in building IVIVC correlation models). IV bolus or IV infusion. Reference data type: Identify whether the reference input was oral. and drug formulation identifiers (Formulation). The IVIVC Wizard does not support combinations of inputs for UIR calculation. Sheet: Select the worksheet containing the data. concentration (Values) or fraction absorbed (Absorption). time (Independent). You can paste in the values that you prefer to use. The IVIVC Wizard will use the patient and formulation sort keys specified to determine individual profiles to deconvolve. check this box to include a lag time parameter in the UIR model. This assumes that absorption is truly first order and fits a model that is n-compartment polyexponential. If you have already determined the UIR and (average) Fraction Absorbed. It will generate estimates for fraction absorbed at 200 time points. It will fit a polyexponential function to each profile and choose the one with the best fit based on Akaike Information Criterion or Weighted SSR. check this box to generate the polyexponential describing the IV bolus PK.pwo. the UIR will simply be a polyexponential model of the reference data. respectively. the Wizard can compute UIR’s for all subjects.The IVIVC Toolkit The IVIVC Wizard 13 Dose units: Optionally. Unit impulse response Based on the reference formulation you select. using the doses entered into the Wizard and the UIR’s just generated. If this option is not checked. in the units identified above. In UIR. (See “AIC” on page 483 and “WRSS” on page 500. the A* and alpha* columns on the Transposed Final Parameters sheet are editable. Then. click Deconvolve to perform numeric deconvolution. Strip Ka: For an oral reference formulation. Enter the dosage amount for each formulation in the table below. The absorption is mathematically separated from the decay. you must still follow the above procedure to generate the output workbooks in their appropriate format. Then check Data already deconvolved and averaged to make two of the output workbooks editable. enter the dose units. For an infusion. In the Fa-avg. as chosen under Model Selection. enter the infusion time for each formulation.pwo workbook.) Include time lag: For an oral reference formulation. 309 . to compute fraction absorbed over time. Click Generate UIR when all settings are ready to fit the unit impulse response function(s) and generate related charts and workbook output. convolved with first-order absorption. the Time and Mean columns are editible. the Wizard can estimate AUC’s and maximum concentrations based on the model predictions.13 User’s Guide Correlation tab The IVIVC Wizard. WinNonlin 310 . Correlation tab supports creation and validation of an IVIVC correlation model for dissolution data defined as “internal” on the InVitro tab. then run noncompartmental analysis on that predicted PK data and the observed in-vivo data. The correlation will be built using those formulations identified as “internal” (or “reference”) on the In-Vitro and In-Vivo tabs. It will fit the selected correlation model to absorption data using the dissolution profiles from the In-Vitro tab as inputs to the correlation function. it will convolve predicted fractions absorbed and UIR’s from the other tabs. To validate the correlation. and absorption data set up or estimated via the In-Vivo tab. See “Invitro tab” on page 306 and “In-vivo tab” on page 307. To do so. and compare to actual values from the in-vivo data. Click Generate Plots to create the plots.) Average: Select whether to compute averages as the mean or geometric mean. or click the Edit User Model button to write your own in the WinNonlin modeling language. where Clast is the last observed concentration (AUCINF_obs). Validation AUC: Select which computation of area under the curve to use in the validation: from dose time through the last positive concentration value (AUClast). where Clast is the last predicted concentration (AUCINF). enter the time after which the parameter Diss should stop increasing. Calculation method: You can choose to calculate AUC using either the linear. linear/log. or linear up/log down trapezoidal rule.The IVIVC Toolkit The IVIVC Wizard 13 Correlation model Select the correlation model to use to fit fraction absorbed (Fabs).) Tcutoff (optional): If using a library model. Click Run Validation to generate a table of prediction errors for AUC and Cmax: %_PE = 100* (Predicted-Observed) / Observed. 311 . from dose time through the final observation time (AUCall). Fabs vs Fdiss: Fraction absorbed and fraction dissolved over time. Model parameters: See “Model parameters” on page 197. Plots Check the plots to include in the output: Tvitro vs Tvivo: Levy plot of in-vitro versus in-vivo times for a given fraction absorbed or dissolved. (See “Calculation methods for AUC” on page 177 and “AUC calculation and interpolation formulas” on page 183 for definitions and calculation methods. Click Build Correlation to run the IVIVC model. AUClast + Clast/ Lambda Z. (See “ASCII user model for correlations” on page 312. or AUClast + Clast/Lambda Z. which is called T. Edit the statements to the right to reflect your model. Use the syntax of the WinNonlin modeling language detailed under “User Models” on page 217. WinNonlin IVINTERP() .13 User’s Guide For “Avg Internal. and X is then scaled or shifted by model parameters to yield an in vitro time. ASCII user model for correlations A custom model for an IVIVC correlation (see “Correlation tab” on page 310) needs to include only the parameter definitions and model equations. 312 . DOSEVIVO .This function is used to interpolate dissolution data. Enter the number of model parameters in the topmost field near the left side of this dialog. The wizard will translate this variable into the DTA array that is part of the WNL modeling language. No other statements are required. The first and last values in the dissolution data are used when T is respectively before or after the time range of the dissolution data.This variable contains the dose amount for each formulation.” the average is taken over the absolute values of %_PE for the formulations specified as Internal. and the following keywords that are specific to IVIVC. A common usage is for X to be the in vivo time. and enter its initial value below. IVINTERP is then used to interpolate the dissolution data to find the dissolution at time T. The IVIVC Toolkit The IVIVC Wizard 13 Prediction tab The IVIVC Wizard. 313 . Prediction tab generates a table of prediction errors for AUC and Cmax by comparing predicted data for test formulations on the Prediction tab to the target formulation identified on the In-Vitro tab. 13 User’s Guide WinNonlin Dissolution data selection Dataset: Select the workbook containing your dissolution data. enter the dose units for the formulations. as fraction dissolved over time. Variables: Drag column names from the Variables list to the appropriate field on the right to identify those containing time. in the units identified above. fraction dissolved and formulation identifiers. Formulation information Dose units: Optionally. Enter the dosage amount for each formulation in the table below. 314 . Sheet: Select the worksheet containing the percent dissolution data. from among those open in WinNonlin. the asymptotic fraction dissolved. then set the following. in order to provide smoothing you may fit the dissolution data to a sigmoidal dissolution model. comparing test formulations to the target formulation identified on the In-Vivo tab. 315 . can be fixed to 1. Options tab includes settings that control how operations on the other tabs will behave. to use linear interpolation on the raw data instead of modeling.0 to prevent incomplete measurements from biasing other parameter estimates. For example. Options The IVIVC Wizard. Click Fit Dissolution Data to run the model or linear interpolation and generate output. optionally based on observed Y or predicted Y (Yhat) values. Finf. check Use raw data above the models. Model parameters: 1.The IVIVC Toolkit The IVIVC Wizard 13 Dissolution model Optionally. Prediction Click Predict PK to generate the table of prediction errors for AUC and Cmax. Weighting: Select how to weight data during modeling. Depending on the model. Select one of the IVIVC models described under “Sigmoidal/dissolution models” on page 547. See also “Weight” on page 205.1. See “Model parameters” on page 197 for instructions.2. you may wish to fix certain parameters to known values.1.3. Alternately. IVIVC Wizard Output Objects File Name “Dissolution Out” Description Process Simulated dissolu. See also “Hidden windows” on page 18. Output objects Select whether to hide specific output windows generated by the IVIVC Wizard in the WinNonlin window.Fit Dissolution Data tion data points.13 User’s Guide WinNonlin Modeling runtime options Enter the number of predicted values to be generated by modeling operations initiated in the IVIVC Wizard and the maximum allowed number of iterations during model fitting. using fitted model selected on “InVitro” tab Object Types Workbook Text Chart 316 . The IVIVC Toolkit The IVIVC Wizard 13 ”Deconvolution Out” Deconvolution out.Deconvolve put – fraction absorbed vs. time for each profile Unit Impulse Generate UIR Response (UIR) modeling output Average fraction absorbed by formulation Fraction absorbed data going into the correlation model fitting Fraction dissolved data going into the correlation model fitting Correlation modeling output Input vivo data to correlation model simulation for validation step Input vitro data to correlation model simulation for validation step Simulated correlation output using “internal” dissolution profiles Validation convolution output – individual profiles of UIRs convolved with simulated correlation output Deconvolve Workbook Chart “UIR” “Fa-avg” Workbook Text Chart Workbook “Corr Model Vivo In” Build Correlation Workbook “Corr Model Vitro In” Build Correlation Workbook “Corr Model Out” Build Correlation “Corr Model Sim Vivo Validate Correlation Workbook Text Chart Workbook “Corr Model Sim Vitro” Validate Correlation Workbook “Val Corr Sim Out” Validate Correlation Workbook Chart “Val Conv Out” Validate Correlation Workbook Chart 317 . with PE computations. using fitted model selected on “Prediction” tab “Pred Corr Sim Input vivo data to Vivo” correlation model simulation for prediction step “Pred Corr Sim Input vitro data to Vitro” correlation model simulation for prediction step “Val PK NCA In” Validate Correlation Workbook WinNonlin Validate Correlation Validate Correlation Workbook Workbook Generate Plots Chart Generate Plots Chart Validate Correlation Workbook Validate Correlation Workbook Predict PK Workbook Text Chart Predict PK Workbook Predict PK Workbook 318 .goes into NCA “Val Baseline NCA NCA results from Out” in vivo dataset “Val PK NCA Out” NCA results from [Val PK NCA In.pwo] “Levy Plot Out” Levy plot using [Fa-avg] and modeled dissolution data “Corr Plot Out” Correlation plot using [Fa-avg] and modeled dissolution data “Avg PE Stats Out” Average absolute prediction error statistics for ‘internal’ formulations “Val Stats Out” NCA outputs averaged by formulation for observed and predicted data.13 User’s Guide Validation convolution output sampled at in vivo times. “Pred Dissolution Simulated dissoluOut” tion data points. sampled at in vivo times. and (optionally) test in vitro data. 2. Predict PK Workbook Chart Predict PK Workbook Predict PK Workbook Predict PK Predict PK Workbook Workbook Chart Predict PK Workbook Using the IVIVC Wizard with PKS Initiating a New IVIVC Project in an existing PKS Scenario Note: Use these steps when you have an existing PKS Scenario that you want to apply an IVIVC analysis to. Proceed to use the IVIVC Wizard. Load the Scenario. in vitro. Save per instructions below. 319 . Create relevant workbooks for in vivo. with PE computations.The IVIVC Toolkit Using the IVIVC Wizard with PKS 13 “Pred Corr Sim Out” Simulated fraction absorbed output from correlation model “Pred Target NCA” NCA results from in vivo dataset for ‘Target’ formulation “Pred PK NCA In” Predicted concentration profiles. 1. for input to NCA “Pred PK NCA NCA results for Out” predicted profiles “Pred Conv Out” Prediction convolution output – individual profiles of UIRs convolved with simulated correlation output “Pred Stats Out” NCA outputs averaged by formulation for target and predicted data. 3. Proceed to use the IVIVC Wizard. Load the Scenario. 3. Then save the project to a clean directory. 4. Save the IVIVC project to disk. open the IVIVC project in the IVIVC Wizard. Load the Scenario. This way. 4. 2. 5. 5. Open the IVIVC Wizard. 1. 1. 1.ivc file) into WNL to open them.ivc) from the disk location it was extracted to. Open the IVIVC Wizard. text. Saving an IVIVC Project to PKS Note: Use these steps to save an IVIVC project to a PKS Scenario. Proceed to use the IVIVC Wizard.ivc) from “Other Documents. Load the IVIVC project file (filename . Open all IVIVC project files (workbook.” 3. Merging an Existing IVIVC Project into PKS Note: Use these steps when you have an existing IVIVC project on your local disk and would like to merge the work into an existing PKS Scenario. 2. TIP: Before beginning the merging instructions. Extract the IVIVC project file (filename . Save per instructions below.13 User’s Guide Loading a IVIVC Project from PKS WinNonlin Note: Use these steps to load an IVIVC project that has been properly saved to a PKS scenario.ivc) from its disk location. in step 2 you can use the file explorer to drag and drop all files (except the . Save per instructions below. and chart objects) from their disk locations. Open the IVIVC project file (filename . 320 . 5. 321 .The IVIVC Toolkit Using the IVIVC Wizard with PKS 13 2.ivc file that you just saved. Close the IVIVC Wizard. Add the *. Click on the PKS > Save or PKS > Save As to save the PKS scenario. Click on PKS > Other Documents. 4. 3. 13 User’s Guide WinNonlin 322 . and regression models The Linear Mixed Effects function (LinMix) performs analyses using linear mixed effects models. “LinMix Wizard” on page 350 provides stepped instructions. LinMix is discussed in the following sections.Chapter 14 Linear Mixed Effects Modeling ANOVA. “The linear mixed effects model” on page 326 discusses LinMix general theory in more detail. • • • • An example The following example is used as a framework for discussing the functions and computations of the Linear Mixed Effects Modeling wizard. It can analyze regression and covariance models. ANCOVA. • “An example”: Since linear mixed effects modeling is most easily introduced using an example. “LinMix output” on page 361 and “Computational details” on page 366 cover LinMix calculations and output parameters. each subject 323 . Therefore. as well as classical ANOVA modeling terms. Because responses are fairly heterogeneous among people. and can calculate both sequential and partial tests. which should increase the power of the study. this chapter starts by creating a statistical model for a common experimental design that utilizes several error terms. “Comparing LinMix to ANOVA and SAS’s PROC MIXED” on page 372 explains similarities and differences among three analysis packages. It is a statistical analysis system for analysis of variance for crossover and parallel studies. labeled A and B. An investigator wants to estimate the difference between two treatments for blood pressure. the investigator decided that each study subject will be his own control. including unbalanced designs. Hence. note that this crossover study has two periods. 2) q = treatment index (1. The purpose of the study is to estimate τ1 . This mixing of the regression model and the random effects is known as a mixed effect model. half the subjects will receive A first. Note that in model (1). εijkm is assumed to be normally distributed with mean 0 and variance σ2 > 0. and the other half will receive B first. For the analysis. 1 and 2. it is desirable to treat subject as a random effect. The model will be of the form: yijkm = μ + πi + δj + τq + Sk(j) + εijkm where: i = period index (1 or 2) j = sequence index (AB or BA) k = subject within sequence group index (1 … nj) m = observation within subject index (1. Therefore. Study subjects are assumed to be randomly selected from the population at large. 2) μ = intercept πi = effect due to period δj = effect due to sequence grouping τq = effect due to treatment. the mean is: E{yijkm} = μ + πi + δj + τq (2) (1) WinNonlin 324 . and two sequence groups AB and BA.τ2 in all people with similar inclusion criteria as the study subjects. the purpose of the study Sk(j) = effect due to subject εijkm = random variation Typically. The mixed effect model allows one to model the mean of the response as well as its variance. assume that subject effect is normally distributed with mean 0 and variance Var(Sk(j)) ≥ 0. Since the treatment order might affect the outcome.14 User’s Guide will be exposed to treatments A and B. 63 9. (3) E{yijkm}is known as the fixed effects of the model.6 9. Table 14-1.23 9.Linear Mixed Effects Modeling An example 14 and the variance is given by: Var{yijkm} = Var{Sk(j)}+ σ2. Data for the example Subject 1 2 3 4 5 6 7 8 9 10 11 12 7 8 9 10 11 Sequence AB AB AB AB AB AB BA BA BA BA BA BA BA BA BA BA BA Period 1 1 1 1 1 1 2 2 2 2 2 2 1 1 1 1 1 Treatment A A A A A A A A A A A A B B B B B Response 9.02 7. This is appropriate since the fixed effects model the mean of the response. The data are shown in Table 14-1.67 325 .24 11.43 10.13 9. The LinMix wizard is used to analyze these data.15 9. since these terms represent random variables whose variances are to be estimated. while the random effects model the variance of the response.70 10.2 7. while Sk(j) + εijkm constitutes the random effects. A model without the fixed effects is known as a random effect model. The fixed and random effects of the model are entered on different pages of the wizard.1 9.31 9. A model with both is a mixed effect model.91 8.88 9.63 9.82 11. A model with only the residual variance is known as a fixed effect model. ” The linear mixed effects model WinNonlin’s Linear Mixed Effects Modeling feature is based on the general linear mixed effects model.31 WinNonlin LinMix can fit linear models to Gaussian data. Independent variables can be continuous or discrete.14 User’s Guide Table 14-1.34 8.28 11. This model provides a context for “The linear mixed effects model. The mean and variance structures can be simple or complex.73 9. which can be represented in matrix notation as: y = Xβ + Zγ + ε 326 .72 11.77 8. and no distributional assumption is made about them. Data for the example (continued) Subject 12 1 2 3 4 5 6 Sequence BA AB AB AB AB AB AB Period 1 2 2 2 2 2 2 Treatment B B B B B B B Response 11.91 8. . the fixed effects dialog specifies the model terms for constructing X. then R is the residual variance—i. i. normally distributed with mean 0 and variance G. Using the assumptions for the linear mixed effects model. normally distributed with mean 0 and variance R. and ε is a random vector.Linear Mixed Effects Modeling The linear mixed effects model 14 where: X is a matrix of known constants and the design matrix for β.e. R = σ2I. If the model doesn’t include a repeated specification.. the model includes one or more random effects. one can show that the variance of y is: V = ZGZ + R If Z ≠ 0 . Listing all the parameters in a vector. γ is a random unobserved vector (the vector of random-effects parameters). In the LinMix wizard. The fixed effects model is shown in equation 2 under “An example” on page 323. then the LinMix model reduces to the ANOVA model. If the model includes a repeated specification. the random effects tabs in the Variance Structure dialog specifies the model terms for constructing Z and the structure of G.e. the G matrix is determined by the variance structures specified for the random models. then the R matrix is determined by the variance structure specified for the repeated model. Z is a matrix of known constants. and the repeated tab specifies the structure of R. It is instructive to see how a specific example fits into this framework. LinMix breaks the model into the following areas: • • • Note: Fixed effects Variance structure Least squares means • • Contrasts Estimates In the linear mixed effects model. if R = σ2I and Z = 0. β is a vector of unknown (fixed effect) parameters. one obtains: T 327 . the design matrix for γ. γ and ε are uncorrelated. δ1. below.14 User’s Guide βT = [μ. The Z matrix will consist of 1’s and 0’s. The parameter to be estimated for G is Var{Sk(j)}. LinMix takes this sum of terms and constructs the X matrix. π1. The rules for translating terms to columns of the X matrix depend on the variable types. 328 . The LinMix wizard references this column as the intercept. j) will be 1 when observation i is from subject j. A user specifies a model by giving a sum of terms. the X matrix is composed entirely of 1’s and 0’s. regressor/covariate or classification. To exclude it. if A and B are variables in the data set: A + B + A*B. The collection can be directly entered into γ. τ2] For each element of β. A summary of the wizard pages and the part of the model they generate is listed below. Element (i. For example. there is one Sk(j). The variance of γ is of the form G ≡ Var(γ) = Var{Sk(j)} In(1)+n(2) where Ib represents a b×b identity matrix. The way this is done depends upon the nature of the variables A and B. Wizard Page Fixed Effects Specifies Tabs on page Specifies X matrix Variance Structure Specifies Z matrix Random tabs (1 or more) Repeated tab (exactly 1) Creates Generates β Generates G = var(γ) Generates R = var(ε) WinNonlin Construction of the X matrix A term is a variable or a product of variables. π2. the first column of X contains all ones. The No Intercept option By default. The details will be discussed later. and 0 otherwise. δ2. This is demonstrated using two classification variables. check No Intercept in the Fixed Effects dialog of the wizard. In this model. The exact method of constructing it is discussed under “Construction of the X matrix”. For each subject. The residual variance is R = Var{εijkm} = σ2 I2[n(1)+n(2)] The parameter to be estimated for R is σ2. τ1. Drug and Form. there is a corresponding column in the X matrix. The product of a classification variable with a continuous variable is obtained by multiplying the continuous value by each column of the indicator variables in each row. The product operations described in the preceding paragraph can be succinctly defined using the Kronecker productA. m and j=1. The product of two continuous variables is the ordinary product within rows. 0 0 0 0 1 Value Name capsule drops syrup tablet capsule 0 1 2 3 0 Form Value 48 34 27 23 45 120 124 123 172 200 Age Weight Let A = (aij) be a m×n matrix and let B = (bij) be a p×q matrix. The product of Drug and Age is: [0 1 0] ⊗ [45] = [ 0 45 0]. When a single classification variable is specified for a model. Age and Weight. Then the Kronecker product A ⊗ B = (aij B) is an mp × nq matrix expressible as a partitioned matrix with aij B as the (i. Table 14-2. The product of Drug and Form is: [0 1 0] ⊗ [1 0 0 0] = [0 0 0 0 1 0 0 0 0 0 0 0] The result is the fifth row of Table 14-5.…. The association between levels and codes is determined by the numerical order for converted numerical variables and by the character collating sequence for converted character variables. an indicator variable is placed in the X matrix for each level of the variable. Now consider the fifth row of Table 14-4. A term containing one continuous variable produces a single column in the X matrix. Table 14-3 shows the product of Drug and Age. ….n. Variables in sample data set Drug Name DrugA DrugA DrugA DrugA DrugB A. in Table 14-2. For example. i=1. 329 . Table 14-4 shows two classification and indicator variables. An integer value is associated with every level of a classification variable. Table 14-3 demonstrates this for the variable Drug. The model specification allows products of variables. consider the fifth row of Table 14-3.Linear Mixed Effects Modeling The linear mixed effects model 14 and two continuous variables. For example. j)th partition. Table 14-5 gives their product. Variables in sample data set (continued) Drug Name DrugB DrugB DrugB DrugC DrugC DrugC DrugC 1 1 1 2 2 2 2 Value Name drops syrup tablet capsule drops syrup tablet 1 2 3 0 1 2 3 Form Value 26 41 32 38 17 32 23 150 120 130 190 125 175 150 Age Weight WinNonlin Table 14-3. Indicator variables and a product of a classification variable with a continuous variable Drug Name DrugA DrugA DrugA DrugA DrugB DrugB DrugB DrugB DrugC DrugC DrugC DrugC 0 0 0 0 1 1 1 1 2 2 2 2 Value 1 1 1 1 0 0 0 0 0 0 0 0 Indicator variables DrugA DrugB 0 0 0 0 1 1 1 1 0 0 0 0 DrugC 0 0 0 0 0 0 0 0 1 1 1 1 48 34 27 23 45 26 41 32 38 17 32 23 Age DrugA 48 34 27 23 0 0 0 0 0 0 0 0 Drug*Age DrugB 0 0 0 0 45 26 41 32 0 0 0 0 DrugC 0 0 0 0 0 0 0 0 38 17 32 23 330 .14 User’s Guide Table 14-2. capsule drops syrup tablet 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 0 0 0 1 0 0 0 1 Name Val. DrugA DrugB DrugC capsule 0 capsule 0 Table 14-5. Product of the two classification variables Drug DrugA DrugA DrugA DrugA DrugB DrugB DrugB DrugB DrugC DrugC Form capsule drops syrup tablet capsule drops syrup tablet capsule drops 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 Indicator 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 331 . Two classification variables and their indicator variables Drug DrugA 0 DrugA 0 DrugA 0 DrugA 0 DrugB 1 DrugB 1 DrugB 1 DrugB 1 DrugC 2 DrugC 2 DrugC 2 DrugC 2 1 1 1 1 0 0 0 0 0 0 0 0 Indicators 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 Form Name drops syrup tablet drops syrup tablet drops syrup tablet capsule 0 1 2 3 1 2 3 1 2 3 1 0 0 0 1 0 0 0 1 0 0 0 Indicators Val.Linear Mixed Effects Modeling The linear mixed effects model 14 Table 14-4. In the context of 332 .τ3. such as μ + τ1. can be generated by choosing appropriate values of lj. τ1 . as an example. the model where the expected value of the response of the jth subject on the ith treatment is: μ + τi (4) where μ is a common parameter and μ+τi is the expected value of the ith treatment. The expression: (5) Σljβj j with the lj constant is called a linear combination of the elements of β. and τ2. The X matrix has a column of ones in the first position followed by three treatment indicators. The range of the subscript j starts with 0 if the intercept is in the model and starts with 1 otherwise. Linear combinations of β are entered in LinMix through the wizard’s Estimates and Contrasts pages. A linear combination of β is said to be estimable if it can be expressed as a linear combination of expected responses. Most common functions of the parameters. Suppose there are three treatments. The β vector is: β = [β0 β1 β2 β3] = [μ τ1 τ2 τ3] The first form is for the general representation of a model.14 User’s Guide Table 14-5. The second form is specific to model (4) above. Product of the two classification variables (continued) Drug DrugC DrugC Form syrup tablet 0 0 0 0 0 0 0 0 0 0 Indicator 0 0 0 0 0 0 0 0 0 0 1 0 0 1 WinNonlin Linear combinations of elements of β Most of the inference done by LinMix is done with respect to linear combinations of elements of β. Consider. Determining estimability numerically The X matrix for the model in equation (4) has three distinct rows.Linear Mixed Effects Modeling The linear mixed effects model 14 model 4. Also shown in Table 14-6 is a vector (call it z) that is a basis of the space orthogonal to the rows of X. Allow for some rounding error in the computation of z. A linear combination of effects. and tol is the Estimability Tolerance.τ3 is a contrast of the τ’s and thus is estimable. The linear combination 5 in vector notation is lTb and is estimable if lTΖ = 0. where the sum of the coefficients is zero. In general. the number of rows in the basis of the space orthogonal to the rows of X is equal the number of linear dependencies among the columns of X. c2. For example.< tol T l ∞ Z Z T (8) where || • ||∞ is the largest absolute value of the vector. so μ + τ1 is estimable. The linear combination is estimable if: l Z ------------------------. and these are shown in Table 14-6. is called a contrast. a linear combination of the elements of β is estimable if it can be generated by choosing c1.3. A rearrangement of expression (6) is: (c1 + c2 + c3)μ + c1τ1 + c2τ2 + c3τ3.2. and c3=0 generates μ + τ1. lj = cj for j=1. and c3. and only if. If there is more than one z. The linear combination τ1 . l1 + l2 + l3 = 0. Also. Let l be a vector whose elements are the lj of expression (5). any linear combination involving only τs is estimable if. Note that c1=1. The linear combination τ2 is not a contrast of the τ’s and thus is not estimable. then (8) must be satisfied for each z. 333 . (7) (6) The form in (7) makes some things more obvious. c1(μ + τ1) + c2(μ + τ2) + c3(μ + τ3). c2=0. regress row 10 on row 5. j = 1. so this will be used as an example. 2. Notice is given in the text output when the substitution is made. The regression coefficient is 0. and it is the user's responsibility to check the coefficients to see if they are reasonable. Consider the model: yijk = μ + αi + βj + (αβ)ij + εijk (9) where μ is the over-all mean. (Rows 6 through 8 are all zeros and are ignored.14 User’s Guide WinNonlin Table 14-6. (Row 4 is all zeros and is ignored. so there is no need to compute residuals. 2. The canonical form of the QR factorization of X is displayed in rows 0 through 8 of Table 14-7. Now calculate residuals: (row 9) .e. Within the αβ columns. αi is the effect of the i-th level of a factor A. In the μ column. Within the α columns. LinMix tries to be helpful. not just the α columns. The basis of the space orthogonal to the rows of X Elements of β Distinct rows of X μ 1 1 1 Basis of orthogonal space -1 τ1 1 0 0 1 τ2 0 1 0 1 τ3 0 0 1 1 Substitution of estimable functions for non-estimable functions It is not always easy for a user to determine if a linear combination is estimable or not.. i = 1.α2 are shown in row 9. The process is a sequence of regressions followed by residual computations. σ2). The result is shown in row 10. This operation goes across the entire matrix. and if a user attempts to estimate an non-estimable function. j is the effect of the j-th level of a factor B. At this point. and the εijk are independently distributed N(0. and the next operation applies to row 10.).) The regression coefficient is 1. The most common attempt to estimate a non-estimable function of the parameters is probably trying to estimate a contrast of main effects in a model with interaction. (αβ)ij is the effect of the interaction of the i-th level of A with the j-th level of B. The numbers in the β columns of row 10 are zero. LinMix will substitute a “nearby” function that is estimable. This section is to describe how the substitution is made. i. row 9 is already 0 so no operation is done.1× (row 3). The coefficients to generate α1 . regress row 9 on row 3. so skip to the αβ columns. row 10 is a deviation of row 9 from 334 . 5 -1 0 β1 0. Put the coefficients of a linear combination in a work row.5 0 -1 0 0 0..e. Subtract the final values in the work row from the original coefficients to obtain coefficients of an estimable linear combination.25 0. This section provides a general description of the process just demonstrated on a specific model and linear combination of elements of β.5 0 0. do a least squares fit) the work row on the rows in the diagonal block of the QR factorization. The QR factorization of X scaled to canonical form μ 0 1 2 3 4 5 6 7 8 9 10 0 0 1 0 1 -1 0 -1 0 0 0 0 0 0 0 -0.5 0 0 1 β2 0.5 -0.5 0.5 -0. Overwrite the work row with these residuals.25 0.5 0 1 0 0 0 0 0. Subtract the deviation (row 10) from the original (row 9) to get the estimable function displayed in the last row. do the following: 1.5 0 0 -1 0 (αβ)11 (αβ)12 (αβ)21 (αβ)22 0.5 Final result 0 335 . Calculate the residuals from this fit across the entire matrix.5 0 0.5 0 0. 2. Table 14-7. Regress (i.25 -0. The result is the expected value of the difference between the marginal A means.5 0. from left to right.25 -0. Partition the QR factorization of X vertically and horizontally corresponding to the terms in X.5 1 α2 0.Linear Mixed Effects Modeling The linear mixed effects model 14 an estimable function.5 0 -0. 3.5 0 -1 0 0.5 0 1 0. This is likely what the user had in mind.5 1 α1 0. For each vertical partition.5 0 -0.5 0 -0. X1. Hence β3 would be set to zero. Output parameterization Suppose X is n×p. To solve this issue. In addition. If X has rank p. Suppose column j is in the span of columns 0. consider a simple one-way ANOVA model. 336 . …. j-1. If rank(X) < p. denoted rank(X). When this happens. then each parameter can be uniquely estimated. In this case. X0 and X1 are clearly linearly independent. as is X2 linearly independent of X0 and X1. But X3 is not linearly independent of X0. its parameter estimate is set to 0. since X3 = X0 . then there are infinitely many solutions.14 User’s Guide WinNonlin Fixed effects The specification of the fixed effects (as defined in “The linear mixed effects model” on page 326) can contain both classification variables and continuous regressors. it seems sensible to not use the column in the model fitting. one must impose additional constraints on the estimator of β. The design matrix is X0 1 1 1 1 1 1 X1 1 0 0 1 0 0 X2 0 1 0 0 1 0 X3 0 0 1 0 0 1 X0 is the intercept column. and X2. The degrees of freedom for an effect is the number of columns remaining after this process. the model can include interaction and nested terms. In this example. X1. treatment effect has 2 degrees of freedom. As an example. where n is the number of observations and p is the number of columns. X2. with three levels.X1 -X2. Then column j provides no additional information about the response y beyond that of the columns that come before. See “Predicted values and residuals” on page 364 for computation details. 1. the number typically assigned in ANOVA. and X3 correspond to the three levels of the treatment effect. Transformations are applied to the response before model fitting. Notice that X2 = X0 . Note that all estimates using log will be the same estimates found using ln. β4. then the vector is called singular. Table 14-8. β2. then center the data. or log transformation (log10). the design matrix X would be expanded into that presented in Table 14-8. The vector of fixed effects corresponding to that design matrix would be β = (β0. β6). If the norm of the vector after GS divided by the norm of the original vector is less than δ. Perform a Gram-Schmidt orthogonalization process. Hence set β2 = 0. Using that model. This implicitly assumes that the error structure is multiplicative with lognormal distribution. X4 = X0 .Linear Mixed Effects Modeling Fixed effects 14 Singularity tolerance If intercept is in the model. only scaled by ln 10. and hence set β4 = 0 and β6 = 0. Design matrix expanded for fixed effects model Sequence + Period + Treatment Intercept X0 1 1 1 1 1 1 1 1 1 1 1 1 Sequence AB [X1] BA [X2] 0 0 0 0 0 0 1 [X3] 1 1 1 1 1 1 Period 2 [X4] 0 0 0 0 0 0 1 1 1 1 1 1 Treatment A [X5] B [X6] 0 0 0 0 0 0 337 . taking the logarithm often stabilizes the variance. β3. Transformation of the response The transformation pull-down menu provides three options: No transformation. β5. β1.X3 and X6 = X0 . since log(x) = ln(x)/ln(10). Similarly. If the standard deviation is proportional to the mean.X1.X5. ln transformation (loge). resulting in better fit. Crossover example continued The fixed effects are given in equation (5). Only 1 repeated effect may be specified. G. The user may have 0. 338 . or more random effects specified. Design matrix expanded for fixed effects model Sequence + Period + Treatment (continued) Intercept X0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 Sequence AB [X1] BA [X2] 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 1 [X3] 0 0 0 0 0 0 1 1 1 1 1 1 0 0 0 0 0 0 Period 2 [X4] 1 1 1 1 1 1 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 Treatment A [X5] B [X6] 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 WinNonlin Variance structure The LinMix Variance Structure page specifies the Z. The Random effects specify Z and the corresponding elements of G = Var(γ). The Repeated effect specifies the R = Var(ε).14 User’s Guide Table 14-8. 1. and R matrices defined in “The linear mixed effects model” on page 326. 2. then subject within treatment group.. use the group variable. …. If no repeated effect is specified. Put the model term in the Repeated Specification box. where N denotes the number of observations. one may assume that the data are sorted by treatment group. then R = σ2 ΙN is used. Specifying the repeated effect is optional. ng. An example will make this easier to understand. The variance Σ is specified using the Type pull-down menu. Time(Subject)). 339 . so the correlation structure will differ among T1. To specify a particular repeated effect. T2. including unstructured. and ΙN is the N × N identity matrix. Several variance structures are possible. and further suppose that the data are sorted according to the variance blocking variable. To model heterogeneity of variances. and T3. Time*Subject). and no-diagonal factor analytic.g. Note that a model term can be a variable name. Suppose 5 subjects are randomly assigned to each of 3 treatment groups. 3.g. heterogeneous compound symmetry. 1.Linear Mixed Effects Modeling Variance structure 14 Repeated effect The repeated effect is used to model a correlation structure on the residuals. Suppose further that this correlation is affected by the treatment itself. T2. and T3. Suppose the variance blocking model term has b levels. autoregressive. If a group variable has levels g = 1. Without loss of generality. Group will accept any model term. then the variance for observations within group g will be Σg. The autoregressive is a first-order autoregressive model.. 2. The variance blocking model term creates a block diagonal R matrix. call the treatment groups T1. Then R would have the form R = Ιb ⊗ Σ where Σ is the variance structure specified. one must have a classification model term that uniquely identifies each individual observation. an interaction term (e. See the Help file for the details of the variance structures. All variables used on this page of the LinMix wizard must be classification variables. Suppose further that each subject is measured in periods t = 0. and that serial correlations among the measurements are likely. The effect is to create additional parameters of the same variance structure. or a nested term (e. It produces a variance structure that looks like WinNonlin R1 0 0 R = 0 R2 0 0 0 R3 where each element of R is a 15×15 block. Each Rg has the same form. 340 . Because the variance blocking variable is specified. the form of each Rg is I5 ⊗ Σg. The output will consist of six parameters: σ2 and ρ for each of the three treatment groups. Other variance types are also possible. Within each subject. Often compound symmetry will effectively mimic autoregressive in cases where autoregressive models fail to converge. the variance structured specified is 1 ρg ρg ρg 2 2 3 σg 2 ρg 1 3 2 ρg ρg 2 ρg ρg 1 ρg ρg ρg ρg 1 This structure is the autoregressive variance type. I5 is used because there are five subjects within each treatment group.14 User’s Guide First consider the effect of the group. It is not possible to delete all random effects. constructed in the same way that columns are produced in the X matrix.. it expands the Z matrix by taking the Kronecker product of the Variance Blocking Variables with the columns produced by the intercept check box and the Random Effects Model terms. then it will be ignored. Model terms entered into the Random Effects Model produce columns in the Z matrix. To delete a previously specified random effect. Mathematically. it is possible to specify up to 10 random effects. all 1’s) into the Z matrix. click the Delete Random button.Linear Mixed Effects Modeling Variance structure 14 Random effects Unlike the (single) repeated effect. similarly to Repeated Effects. 341 . however. The Variance Blocking Variables has the effect of inducing independence among the levels of the Variance Blocking Variables. The variance matrix appropriate for that type is sized according to the number of columns produced in that random effect. It is possible to put more than one model term in the Random Effects Model. The random intercept check box puts an intercept column (i. if no entries are made in the random effect. Additional random effects can be added by clicking the Add Random button. but each random page will correspond to a single variance type. The Group model term is used to model heterogeneity of the variances among the levels of the Group model term.e. while R2 has two levels. This example models the variance using the compound symmetry variance type. γ3. build the resulting Z and G matrices. put the columns of each variable in the Z matrix to get the following. Suppose further that R1 has three levels. To illustrate this. The hypothetical data are shown as follows. R1 A 1 0 0 1 0 0 R2 C 0 0 1 0 0 1 B 0 1 0 0 1 0 Y 1 1 1 0 0 0 Z 0 0 0 1 1 1 The random effects can be placed in a vector γ = (γ1.14 User’s Guide Multiple random effects vs. and C. multiple effects on one random effect Suppose one has two random effect variables. R1 and R2 share the same variance matrix G where: 342 . γ5). R1 A B C A B C R2 Y Y Y Z Z Z WinNonlin Specifying a random effect of R1+R2 will produce different results than specifying R1 on Random 1 and R2 on Random 2 pages. A. R1 and R2. γ4. Y and Z. γ2. B. For R1+R2 on Random 1 page. the columns in the Z matrix and the corresponding G is: R2 1 1 1 0 0 0 Y Z 0 0 0 1 1 1 In this case: ⎡G 0 ⎤ G=⎢ 1 ⎥ ⎣ 0 G2 ⎦ 343 .Linear Mixed Effects Modeling Variance structure 14 Now. the Z matrix columns are: R1 B 0 1 0 0 A 1 0 0 1 C 0 0 1 0 A 1 0 0 1 0 0 R1 B 0 1 0 0 1 0 C 0 0 1 0 0 1 For R2. and R2 is placed on Random 2 page. consider the case where R1 is placed on Random 1 page. For R1. μ is the over-all mean. The defining 344 . These are not comparable because of the different coefficients of β. xij is the value of the covariable for the jth individual in the ith treatment. Thus. Least squares means are given in the context of a defining term. αi is the intercept for the ith treatment. The respective expected values of the averages of all values on level 1 of A and the averages of all values on level 2 of A are μ+(0.4β1+0. or if there are unbalanced data. sample means cannot be used to compare levels of A because they contain different functions of β1 and β2. that represent the expected values of sample means in balanced data. one can estimate α1 + 10β and α2 + 10β where the overall mean of the covariable is used in each linear combination. and εij is a random error with zero expected value. the subsample means are not estimates that can be validly compared. If there are covariables with unequal means for the different levels. Instead. Instead. and εijk is a random error with zero expected value. in the presence of unbalanced data. Suppose there are two treatments.4β2) +α1 and μ+(0. βj is the effect of the jth level of factor B. Suppose there are six observations for the combinations where i = j and four observations for the combinations where i ≠ j. The model is: yijk = μ + αi + βj + εijk where yijk is the observed response on the kth individual on the ith level of factor A and the jth level of factor B. statistics that make proper adjustments for unequal covariable means and unbalanced data in the linear mixed effects model. The respective expected values of the sample means are α1 + 5β and α2 + 15β. Now consider a 2 × 2 factorial design. β is the slope with respect to x. αi is the effect of the ith level of factor A. though the process can be repeated for different defining terms for the same model. This section describes least squares means. Consider a completely randomized design with a covariable.6β2)+α2. The preceding examples constructed linear combinations of parameters. the average of the x1j is 5.14 User’s Guide WinNonlin Least squares means Sometimes one would like to see the mean response for each level of a classification variable or for each combination of levels for two or more classification variables. The model is: yij = αi + xij β + εij where yij is the observed response on the jth individual on the ith treatment. This is the idea behind least squares means.6β1+0. one compares the linear combinations μ + (β1 + β2)/2 + α1 and μ + (β1 + β2)/2 + α2. and the average of the x2j is 15. The results for DrugB and DrugC are also shown in Table 14-9. so average values are used. The first coefficient is 1 for the implied intercept. 345 . No new principles would be illustrated by finding the coefficients for the Form least squares means. three sets of coefficients are created. Treatment is the defining term in the first example.25 ] = [0. It is possible that some least squares means will not be estimable.25 0. the average of a four factor indicator variable with balanced data. but not in the defining term. Form is not in the defining term. For example.Linear Mixed Effects Modeling Least squares means 14 term must contain only classification variables and it must be one of the terms in the model. For all variables in the model. the average of Age. The next four coefficients are all 0. and 0. The next twelve elements are: [ 1 0 0 ] ⊗ [0. LinMix automatically generates the coefficients lj of expression (5) and processes them almost as it would process the coefficients specified in an estimate statement. The coefficients for the Drug*Form least squares means would be like representative rows of X except that Age would be replaced by average of Age.25 0. the indicator variables associated with DrugA. the defining term is Drug.25 0. suppose the fixed portion of the model is: Drug + Form + Age + Drug * Form. The average value of a numeric variable (covariable) is the average for all cases used in the model fitting.17. average values of the variables are the coefficients. assume the average of each indicator variable is 1/k.25 0 0 0 0 0 0 0 0] The final result is shown in the DrugA column of Table 14-9. A set of coefficients are created for each of all combinations of levels of the classification variables in the defining term. For a classification variable with k levels. The value 1/k would be the actual average if the data were balanced. If some terms in the model are products. This chapter describes generation of linear combinations of elements of β that represent least squares means. the products are formed using the same rules used for constructing rows of the X matrix as described in the section on Fixed Effects. Build the coefficients associated the first level of Drug. The values of all variables in the model have now been defined. above. To get means for each level of Drug.25 0. and factor A is the defining term in the second example.25.25 0. The next coefficient is 32. The next three coefficients are 1. 0. DrugA. Since Drug has three levels. When the user requests least squares means.25 0. Contrasts facilitate the testing of linear com346 .25 0.25 0.25 0.14 User’s Guide WinNonlin Table 14-9.25 0.25 0.17 0.25 0.25 Contrasts A contrast is a linear combination of the parameters associated with a term where the coefficients sum to 0.25 0.25 32.25 0 0 0 0 0 0 0 0 DrugB 1 0 1 0 0.25 0.25 0.17 0 0 0 0 0.25 32. Coefficients of least squares means lj Column of X Intercept Drug DrugA DrugB DrugC Form capsule drops syrup tablet Age Drug*Form DrugA*capsule DrugA*drops DrugA*syrup DrugA*tablet DrugB*capsule DrugB*drops DrugB*syrup DrugB*tablet DrugC*capsule DrugC*drops DrugC*syrup DrugC*tablet βj β0 β1 β2 β3 β4 β5 β6 β7 β8 β9 β10 β11 β12 β13 β14 β15 β16 β17 β18 β19 β20 DrugA 1 1 0 0 0.25 0.25 0.25 0.25 0.25 32.17 0 0 0 0 0 0 0 0 0.25 0.25 0 0 0 0 DrugC 1 0 0 1 0.25 0.25 0.25 0.25 0. both of which have one column. Note that this number must be no more than one less than the number of distinct levels for the model term of interest. Low Dose. enter the coefficients as follows. Using the first approach. Both of these are contrasts since the sum of the coefficients is 0. This requirement is enforced in the wizard.e. in a single test with a predetermined level of the test. To see the difference. Contrasts are tested using the Wald (1943) test statistic: –1 T –1 Tˆ ˆ T ˆ ( Lβ ) ( L( X VX ) L ) ( Lβ ) ˆ ˆ where β and V are estimators of β and V.. This approach will produce two independent tests. One could write this hypothesis as: H0: μPlacebo .(μLow + μMedium + μHigh) = 0. This is true generally: contrasts are invariant under changes in scaling. i. The Wald statistic is compared to an F distribution with rank(L) numerator degrees of freedom and denominator degrees of freedom estimated by the program or supplied by the user.Linear Mixed Effects Modeling Contrasts 14 binations of parameters.(μLow + μMedium + μHigh)/3 = 0 or equivalently: H0: 3μPlacebo . and High Dose. consider a completely randomized design with four treatment groups: Placebo. Note that both contrasts make the same statement. A second approach would be to put the same term in the neighboring contrast. L must be a matrix such that each row is in the row space of X. each of which is at the specified level. Contrast #1 Joint tests 347 Title . For example. Medium Dose. The user may wish to test the hypothesis that the average of the dosed groups is the same as the average of the placebo group. compare Placebo to Low Dose and Placebo to Medium Dose. This tests both hypotheses jointly. Joint versus single contrasts Joint contrasts are constructed when the number of columns for contrast is changed from the default value of 1 to a number larger than 1. H0: – 1 1 0 0 β = –1 0 1 0 0 0 Now compare this with putting the second contrast in its own contrast. testing the following hypothesis. H01: [-1 1 0 0] β = [0] H02: [-1 0 1 0] β = [0] Note that this method inflates the overall Type I error rate to approximately 2α. Contrast #1 Title Effect Contrast #1 #2 #3 #4 Low vs. Nonestimability of contrasts If a contrast is not estimable. at the possible expense of some power. Placebo Treatment Treatment Placebo Low Dose Medium Dose High Dose -1 1 0 0 Contrast #2 Medium vs.14 User’s Guide Effect Contrast #1 #2 #3 #4 Treatment Treatment Placebo Low Dose Medium Dose High Dose WinNonlin -1 1 0 0 -1 0 1 0 This will produce one test. See “Substitution of estimable functions for nonestimable functions” on page 334 for the details. then an estimable function will be substituted for the nonestimable function. Placebo Treatment Treatment Placebo Low Dose Medium Dose High Dose -1 0 1 0 This produces two tests. 348 . whereas the joint test maintains the overall Type I error rate to α. The degrees of freedom must greater than 0.Linear Mixed Effects Modeling Estimates 14 Degrees of freedom The degrees of freedom check box allows the user to control the denominator degrees of freedom in the F approximation. 349 . If the contrast entered is estimable. Note: For purely fixed effects model. If the contrasts are not estimable. Satterthwaite and residual degrees of freedom yield the same number. The Show Coefficients check box will produce extra output. including the coefficients used to construct the tolerance. and enter an appropriate number. see the section “Satterthwaite’s Approximation for Degrees of Freedom”. The confidence intervals are marginal. multiple model terms may be included in a single estimate. then it will repeat the contrasts entered. As such.95 will yield a very narrow confidence interval with coverage just less than 1%. To override the default degrees of freedom.e. Other options To control the estimability tolerance. All other options act similarly to the corresponding option in the Contrasts. it will enter the estimable function used instead of the nonestimable function. check the box. Additionally. Check the box for residual degrees of freedom. Estimates The estimates page produces estimates output instead of contrasts. The numerator degrees of freedom is always calculated as rank(L). Unlike contrasts. If the check box is not checked. then the default denominator degrees of freedom will be used. i. enter an appropriate tolerance in the Estimability Tolerance text box (see “Substitution of estimable functions for non-estimable functions” on page 334). the coefficients need not sum to zero. the Satterthwaite degrees of freedom. For details of the Satterthwaite approximation. The confidence level must be between 0 and 100. estimates do not support joint estimates. Note that it is entered as a percent: Entering 0.. by clicking the Save Model button. LinMix limits and constraints Cell data may not include question marks (?) or either single (') or double (") quotation marks. 2. Note: To reuse the model last run in LinMix (during the same WinNonlin session).LML) for future re-use. ANOVA command files (*.. click the Load. Refer to “An example” on page 323 for an in-depth discussion of linear mixed effects modeling in the context of an example study. 6. as described under “Fixed effects” below. (Optional) At any point in the LinMix wizard. To set up a linear mixed effects model: 1.ANV) can also be loaded into LinMix so long as bioequivalence is not included. Some steps are optional. Choose Tools>Linear Mixed Effects Wizard from the WinNonlin menus or click the Linear Mixed Effects Wizard tool bar button. Click Calculate to start the calculations and close the LinMix wizard. Open the worksheet to be analyzed. depending which calculations are included. button and open the LinMix Command file (*. 4. save the model to a LinMix Command file (*. Set up any other optional calculations and options as described under: – “Variance structure” on page 353 – “Contrasts” on page 356 – “Estimates” on page 358 – “Least squares means” on page 359 – “LinMix general options” on page 360 5. Define the model in the Fixed Effects dialog.14 User’s Guide WinNonlin LinMix Wizard The general steps to create a linear mixed effects model are outlined below. 3. See the Examples Guide for further linear mixed effects modeling and bioequivalence examples.LML). 350 . To reload a saved model.. click the Previous Model button. or ASCII) can have single or double quotation marks—but not both.). Table 14-10.000 10. Titles (in Contrasts.535 128 128 5 lines 351 . . question marks (?) or semicolons (. variable names may not have single or double quotes in them.Linear Mixed Effects Modeling LinMix Wizard 14 Variable names must begin with a letter. The following are maximum counts for LinMix variables and other objects.. They cannot include spaces or the following operational symbols: +. =.000 1. Estimates. They also cannot include parentheses (). LinMix limits Item model terms factors in a model term sort keys dependent variables covariate/regressor variables variables in the data set random and repeated statements variables contrast statements estimate statements combined length of all variance parameter names (total characters) combined length of all level names (total characters) levels per variable path and file name (total characters) number of records passed from WinNonlin to other programs characters per data cell column name length (characters) titles (e. After the first character. Finally.000 256 65.g. valid characters include numbers and underscore (my_file_2). /. output title or chart title) Maximum 30 10 16 128 255 256 10 256 100 100 10.*. Regressor variables or covariates are continuous independent variables. When a weight variable is specified. Classification variables or factors are categorical independent variables. Sort variables define subsets of data to be processed separately. treatment. Note: Parentheses in the model definition represent nesting of model terms. It can also be used in the model. drug concentration. or b. the analysis will be performed for the missing level and MISSING will be printed as the current value of the sort variable. drag variable names and click on the operators to be used in the model.g. In the box labeled Model Specification. each row of data is multiplied by the square root of the corresponding weight.g. e. see “The linear mixed effects model” on page 326. enter the fixed effects model: a. e. If the sort variable has missing values. that is. 2. a separate analysis will be done for each level of the sort variables.. and gender. In the most common situation.g. using an example to illustrate. A weight variable can be included if weights for each record are included in a separate data column. It must have been declared as a regressor/covariate before it can be used as a weight variable. the data are averages and weights are sample sizes associated with the averages. See also “Weighting” on page 241.. Hence: Seq+Subject(Seq)+Period+Form 352 • . type the model. formulation. For discussion of the model. The Weight Variable cannot be a classification variable. The weights are used to compensate for observations having different variances. temperature or body weight. e. Drag and drop variables to categorize those needed in the LinMix model: • • • • Dependent variables contain the values to which the model will be fit..14 User’s Guide Fixed effects To define the model: WinNonlin 1. Weight variable values should be proportional to the reciprocals of the variances. See “Construction of the X matrix” on page 328 for discussion. Note that this will change the parameterization of the classification variables. it is possible to save the model to a LinMix Command file (*. 4.) It provides a variety of covariance structures.Linear Mixed Effects Modeling LinMix Wizard 14 is a valid use of parentheses and indicates that Subject is nested within Seq. Select the No Intercept option to eliminate the intercept in the model. Variance structure The Variance Structure dialog follows the Fixed effects dialog in the LinMix wizard. Mixed linear models contain both fixed effects and random effect parameters. (See “LinMix Wizard” on page 350. Drug + Disease + (Drug*Disease) is not a valid use of parentheses in the model definition. (n = 1 if the group option is not used) t = number of time points in the repeated context b = dimension parameter: for some variance structures. The table uses the following notation: n = number of groups. do not select a log transformation from the pull-down list. Select the transformation from the pull-down list for Dependent Variables Transformation. Check the Dependent Variables Transformation check box if the dependent variable values should be log-transformed before analysis. CAUTION: If the dependent variable data in the workbook are already log transformed. The variances of the random-effects parameters become the covariance parameters for this model. Click Next to move to the Variance structure or Calculate to run the analysis. Accept the default setting for Fixed Effects Confidence Level or enter the confidence interval for calculations of the fixed effects model. 6. as this will transform the data a second time. by clicking the Save Model button. the num353 . The choices for variance structures are shown in Table 14-11.LML) for future re-use. 7. See “Variance structure” on page 338 for discussion of variance in linear mixed effects models. At any time in the LinMix wizard. 5. 3. or “Repeated measurements” on page 356 to set covariance structures for repeated measurements on subjects. 354 .14 User’s Guide ber of bands. Variance structure types Description Variance Components Number of Parameters ith. n×t n × b if b < t . jth element WinNonlin n σ k 1 ( i = j ) and i corresponds to kth effect 2 Unstructured Banded Unstructured (b) if Toeplitz Banded Toeplitz Compound Symmetry Heterogeneous Compound Symmetry No-Diagonal Factor Analytic n × t × (t + 1) ⁄ 2 n × b ⁄ 2 ( 2t – b + 1 ) b < t . 0 otherwise Table 14-11. for others. same as No-Diagonal Factor Analytic Autoregressive (1) Heterogeneous Autoregressive (1) n×2 n × (t + 1) σ ρ 2 i–j i–j σi σj ρ See “Random coefficients” below to specify traditional variance components and/or to specify random coefficients. same as Unstructured. same as Toeplitz n×2 n × (t + 1) n × t(t + 1) ⁄ 2 σi–j 2 σ ij σ ij ( i – j < n ) σi–j + 11( 2 +1 i – j < n) σ1 + σ 1 ( i = j ) σ i σ j [ ρ1 ( i ≠ j ) + 1 ( i = j ) ] min ( i. j. otherwise. otherwise. otherwise. the number of factors 1(expression) = 1 if expression is true. n ) ∑ k=1 λ ik λ jk Banded No-Diagonal Factor Analytic (b) n × b ⁄ 2 ( 2t – b + 1 ) if b < t . for instance. Subject produces a block diagonal structure with identical blocks.” on page 354. The options are explained in Table 14-11. Random Intercept: Check this box to indicate that the intercept is random. It can be used to specify traditional variance components and/or to specify random coefficients. This is most commonly employed when a subject is specified in the Variance Blocking field. All observations having the same level of the group effect have the same covariance parameters. Note: The same variable cannot be used in both the fixed–effects specification and the random effects specification unless it is used differently—as part of a product. Group: (Optional) This variable defines an effect specifying heterogeneity in the covariance structure. Each new level of the group effect produces a new set of covariance parameters with the same structure as the original group.Linear Mixed Effects Modeling LinMix Wizard 14 Random coefficients The Random tab of the Variance structure dialog provides a means to incorporate random effects in a model. The Repeated tab is used to specify covariance structures for repeated measurements on subjects. 4. To specify random effects: 1. The variable must be a classification model term. 2. Thus. This Subject field identifies the subjects in a data set. The model can be built from classification variables and/or regressors and the operators at the top of the dialog. “Variance structure types. or nested variables) should not appear in both. This variable must be a classification model term. The default is off-no random intercept. Complete independence is assumed among subjects. The same term (single variables. 5. 355 . Variance Blocking Variables (Subject): (Optional) Drag the appropriate variable to this field. Random Effects Model: Enter the model by typing it or by dragging the variable names and operators to be used in the model from the appropriate boxes to the Random Effects Model box using the mouse. The random effects can be either classification or continuous. 3. Use parentheses to indicate nesting. Other random models (tabs) can be added by clicking the Add Random button. Type: Select the type of covariance structure. products. It defines an effect specifying heterogeneity in the covariance structure. The repeated effect must contain only classification variables. 356 . Contrasts provide a mechanism for creating custom hypothesis tests. Group: (Optional) This must be a classification variable with no regressors or operators. 3. R 2 is assumed to be equal to σ I .” on page 354. Complete independence is assumed among subjects. Syntax rules for the repeated specification are the same as those for the fixed– effects model. All observations having the same level of the group effect have the same covariance parameters.) Further discussion of contrasts is provided in “Contrasts” on page 346. Contrasts The Contrasts dialog is next in the LinMix wizard after the Variance structure. This Subject field identifies the subjects in a data set. 5.14 User’s Guide Repeated measurements The Repeated tab of the Variance structure dialog provides the means to specify the R matrix in the mixed model. Note: Interactions can be used as contrasts only if the interaction is a model term. Contrasts can be computed only for classification variables. (See “Fixed effects” on page 352 to select classification variables. Click Next to proceed to the Contrasts or Calculate to run the analysis. Each new level of the group effect produces a new set of covariance parameters with the same structure as the original group. Variance Blocking Variables (Subject): (Optional) This must be a classification variable. Subject produces a block diagonal structure with identical blocks. the same is true for nested terms. 2. If no Repeated statement is specified. 4. “Variance structure types. To specify the repeated variance structure: WinNonlin 1. Type: Select the type of covariance structure from the options explained in Table 14-11. Repeated Specification: enter the model by typing or by dragging the classification variable names and operators from the appropriate boxes to this field. the grid automatically expands to allow for Contrast #2. The Estimability tolerance indicates the closeness of the zero vector before invoking the algorithm to make it estimable. d. Drag the appropriate Model term to the cell labeled Effect. Each contrast is tested independently of other contrasts. This shows the actual coefficients used in the output. Once contrast #1 has been specified. If the contrast is not estimable. 3. Note: If all values of the dependent variable are missing or excluded for a given level of the effect variable. If the algorithm doesn’t seem to be using appropriate choices for the degrees of freedom. If a column contains both numeric and alphabetic data. They are specific to that contrast. the settings for the contrast can be specified. 357 . b. check User-specified [denominator] degrees of freedom and enter an integer for df greater than 1. the column cannot be used as an effect variable when building contrasts.Linear Mixed Effects Modeling LinMix Wizard 14 To specify contrasts: 1. Note: The coefficients for single contrasts must sum to zero. Specify the Settings for the contrast entered. Enter the appropriate contrast coefficients in the empty cells associated with Contrasts. 2. Once Contrast #1 is entered. These columns are tested jointly. but they may not be both. Effect variables may be either numeric or alphabetic. c. a nearby estimable function will be used instead. a. Enter a descriptive label for the contrast in the Title cell. 4. The Show Coefficients of Contrasts check box ensures that the contrast is estimable. up to a maximum of 100. These settings are specific to the active contrast. Number of columns for contrast: check this box to test multiple linearly independent combinations simultaneously. This grid will allow additional contrasts. then this level of the effect variable will not appear in the Contrast vector. 3. Set the Confidence Level for confidence interval calculations. Repeat as necessary for other estimates.14 User’s Guide 5. To enter estimates: WinNonlin 1. Enter the appropriate coefficients in the cells associated with the new term. 8. (Optional) To create compound estimates. 4. Drag the appropriate Model term to the cell labeled Effect. then the term Intercept is included with the model terms. Note: The marginal confidence interval is generated for each estimate. Enter a descriptive label for the estimate in the cell labeled Title. Enter the appropriate coefficients in the empty cells associated with Effects. The grid automatically expands up to the maximum number of estimates. The grid expands to display every unique level of that term. If the Intercept term is used as an effect for the estimate. it works like a regressor. Note: Coefficients for estimates are not bound by the same limits as are those for Contrasts. 2. drag another model term below the existing estimate. Estimates The LinMix Estimates dialog is similar to the Contrasts dialog—except that multiple effects can be used in each estimate statement. See also “Estimates” on page 349 and “The linear mixed effects model” on page 326. 5. Specify the Settings for the Estimate: a. Interaction terms and nested terms can be used if they appear in the model. Click Next to proceed to the Estimates. 7. 358 . If the fixed–effects model includes an intercept (the default). or Calculate to run the analysis. 6. The grid expands to include the new term and its levels. Repeat as necessary for other terms within any one estimate. only one number will be entered in the grid. Drag each model term for which the least square means are to be calculated up to the Least Squares Means box. Checking Show Coefficients of Estimates displays in the output the actual coefficients used and. Least Squares Means can be computed for any classification model term. If a data set is balanced. c. That is. If the algorithm doesn’t seem to be using appropriate choices for the degrees of freedom. select the User-specified [denominator] degrees of freedom check box and enter an integer for df greater than 1. The settings selected are applied to all terms. If the test is not estimable. Least squares means Least squares means are generalized least-squares means of the fixed effects. Click Next to proceed to the Least squares means. 3. these are the means predicted by the ANOVA model. estimable function. a. To specify Least Squares Means: 1. 9. They are estimates of what the mean values would have been had the data been balanced. 359 . (Optional) Select the Settings for the Least Squares Means. Checking Show Coefficients of LSM’s ensures that the contrast is estimable. If the algorithm doesn’t seem to be using appropriate choices for the degrees of freedom. this option will apply a nearby. or Calculate to run the LinMix model. See “Fixed effects” on page 336 to set classification variables. The Least Squares Means dialog follows the Estimates dialog in the LinMix wizard. The Estimability tolerance indicates the closeness of the zero vector before invoking the algorithm to make it estimable. check User-specified [denominator] degrees of freedom and enter an integer for df greater than 1. d. the least squares means will be identical to the raw (observed) means.Linear Mixed Effects Modeling LinMix Wizard 14 b. See also “Least squares means” on page 344. estimable function. c. applies a nearby.) 2. no change is made. If the hypothesis is estimable. (See “LinMix Wizard” on page 350. if the estimate is not estimable. Set the Confidence Level for confidence interval calculations. b. The linear mixed effects model terms are provided in the Model Classifiable Terms field. The columns in X and Z are eliminated if their norm ≤ this number. Singularity Tolerance: (Optional) Enter the level. This is the number of iterations in the Newton fitting algorithm. The Compute Pairwise differences of LSM’s check box provides additional output with the differences of confidence intervals and LSM’s. 3. Click Next to proceed to the LinMix general options or Calculate to run the LinMix analysis.14 User’s Guide d. and final variance matrix in the ASCII output. press ENTER to move to the next line. 5. 7. Title: Enter a title for the ASCII output (optional). e. This function tests η 1 = η 2 . 8. If these are not specified. See “LinMix output” on page 361. select the Initial Variance Parameters button to edit a grid of the parameters. reduced data matrix. maximum iterations: (Optional) Enter the number of maximum iterations. use the Save Model button. Hessian. 6. 2. 4.LML) for later re-use. the application uses the method of moments estimates. 4. When entering a multi-line title. (Optional) Select Intermediate Calculations to include the design matrix. Satterthwaite computes the df base on χ approximation to distribution of variance. 360 . (Optional) If the model has random effects. LinMix general options To specify LinMix general options: WinNonlin 1. discussed under “Convergence criterion” on page 369. To save the model to a LinMix command file (*. asymptotic covariance matrix of variance parameters. Degrees of freedom calculation form: (Optional) Residual is the same as in a 2 purely fixed effects model. The Estimability tolerance indicates the closeness of the zero vector before invoking the algorithm to make it estimable. 9. Click OK. Check ASCII output to include text output in addition to workbook output. The title can include up to 5 lines. Convergence Criterion: (Optional) Specify the desired option. Tests of fixed effects. standard error. LinMix output LinMix produces the following text and/or workbook output. Linear mixed effects text This is the optional ASCII output selected in the LinMix general options. For details on ASCII and workbook output. and p-value Estimate. including all output as well as analysis settings and any errors that occur. Depends upon the order that model terms are written. F-statistic. Depending which model options have been set up. workbook may contain the items listed in Table 14-12. Click Calculate to run the LinMix analysis. in sequence. standard error. See also “Computational details” on page 366. Linear mixed effects worksheets Item Diagnostics Sequential Tests Partial Tests Final Fixed Parameters Final Variance Parameters Parameter Key Contrasts Estimates Contents Model-fitting information Tests of fixed effect for significant improvement in fit. Linear mixed effects workbook The Linear Mixed Effects Workbook contains summary tables of the output. Table 14-12. and confidence interval 361 . but each term is tested last Estimates. p-value. or Back to return to the Least squares means and other previous settings. and confidence intervals Estimates and units of the final variance parameters Information about the variance structure Hypothesis. It contains a complete summary of the analysis. see “LinMix output” on page 361. Specific output depends on the options set in the LinMix Wizard (see “LinMix Wizard” on page 350). t-statistic.Linear Mixed Effects Modeling LinMix output 14 10. pvalue. df. Details for specific output follow below. t-statistic. standard error. Then it is tested whether the third model term should enter. standard error and residuals Information about model settings Initial values for variance parameters Data on the variable estimation in each iteration WinNonlin Tests of hypotheses The tests of hypotheses are the mixed model analog to the analysis of variance in the fixed model case. This is computed using a QR factorization of the XY matrix. 362 . Unlike sequential tests. p-value. units. This is computed by modifying the basis created by the QR factorization to yield a basis that more closely resembles that found in balanced data.. It does this by factoring out of each model term the contribution attributable to the remaining model terms. Linear mixed effects worksheets (continued) Item Least Squares Means Residuals User Settings Initial Variance Parameters Iteration History Contents Least squares means. etc. observed and predicted values. The tests are performed by finding appropriate L matrices and testing the hypotheses H0: Lβ = 0 for each model term. t-statistic. given that the first term is in the model. Then it is tested whether the second model term should enter the model. given that the first two terms. First it is tested whether the first model term should enter the model.14 User’s Guide Table 14-12. The QR factorization is segmented to match the number of columns that each model term contributes to the X matrix. partial tests are invariant under the order in which model terms are listed in the Fixed Effects dialog. df. and confidence interval Dependent variable(s). Partial tests Partial tests test each model term given every other model term. Sequential tests The sequential tests test each model term sequentially. until all model terms are exhausted. some terms in ANOVA may be contaminated with undesired effects. For the sequential ANOVA.. but the sums of squares are correlated and do not add up to the model sums of squares. N is the number of observations used. the sums of squares are statistically independent.e. s = rank(X) + dim(θ)).e. the ANOVA represents a partitioning of the model sum of squares.e. The mixed effects tests have similar properties.Linear Mixed Effects Modeling LinMix output 14 Discussion For fixed effects models.. The partial ANOVA is designed to eliminate the contamination problem.. However. s = rank(X) + dim(θ) ). Also. i. Akaike’s Information Criterion (AIC) The Linear Mixed Effects module uses the smaller-is-better form of Akaike’s Information Criterion: AIC = –2LR + 2s where LR is the restricted log-likelihood function evaluated at final fixed ˆ ˆ parameter estimates β and the final variance parameter estimates θ . and s is the rank of the fixed effects design matrix X plus the number of parameters in θ (i. the sum of the individual sums of squares is equal to the model sum of squares. 363 . Schwarz’s Bayesian Criterion (SBC) The Linear Mixed Effects module uses the smaller-is-better form of Schwarz’s Bayesian Criterion: SBC = – 2L R + s log ( N – r ) where LR is the restricted log-likelihood function evaluated at the final estiˆ ˆ mates β and θ . r is the rank of the fixed effects design matrix X. and s is the rank of the fixed effects design matrix X plus the number of parameters in θ (i. certain properties can be stated for the two types of ANOVA. the user should supply the desired points in the input data set with missing dependent variables. Predicted values are estimated by using the rows of the fixed effects design matrix X as the L matrices for estimation. then the predicted values are also transformed. Fixed effects parameter estimates In the Linear Mixed Effects module. There is one eigenvalue per variance parameter. Note that if the user requested a log-transform. and therefore the predicted values ˆ ˆ ˆ for the input data are equal to y = Xβ . Tˆ T ˆ Var ( y ) = X ( X V X ) X –1 WinNonlin The standard errors are the square root of the diagonal of the matrix above. T-type confidence intervals for the predicted values are also computed. the residuals are the transformed values minus the predicted values. Predicted values and residuals In the Linear Mixed Effects module. The confidence level is given in Conf_Level. The residuals are then the observed values minus the predicted values. Positive eigenvalues indicate that the final parameter estimates are found at a maximum. To obtain predicted values for data points other than the observed values. The variance of the prediction is as follows. the final fixed effects parameter estiˆ mates β are computed by solving ˆ T00 β = y0 364 . The Lower_CI and Upper_CI are the confidence bounds on the predicted response. predicted values for the dependent variables are computed at all of the input data points. or in the case of transforms.14 User’s Guide Hessian eigenvalues The eigenvalues are formed from the Hessian of the restricted log likelihood. where β are the final fixed parameters estimates. This is controlled on the Fixed Effects page of the wizard. the variance parameter addresses. This provides the opportunity to inspect what combination is actually used to see if it represents the intentions. Coefficients If Display Coefficients is checked on any of the page of the LinMix wizard. then the estimable replacement will be displayed. or group. etc. two indices are appended to the variance parameter names in order to clarify which random or repeated model. and β are the final fixed parameters estimates. if the residual variance 365 Type . The first index indicates the random model or repeated model to which the parameter applies. An index of 1 indicates the model on the Random 1 tab in the user interface. If the Group option is not used in the model or if there is only one random or repeated model. Source Random if the parameter is from a random model Repeated if the parameter is from a repeated specification Assumed if the variance parameter is the residual Variance type specified by the user. the Parameter column of this grid contains the variance parameter names assigned by LinMix. then zero iterations will be displayed. The highest index indicates the repeated model if there was one. Iteration history This worksheet holds the fitting algorithm results. then the coefficients will be the same as those entered. When there are multiple solutions. the coefficients of the parameters used to construct the linear combination will be displayed. then the correˆ sponding element of β is set to zero.Linear Mixed Effects Modeling LinMix output 14 where T00 and y0 are elements of the reduced data matrix as defined in the ˆ section on Restricted Maximum Likelihood Estimation. If the initial estimates are the REML estimates. then indices are not used in the parameter names. If the linear combination is estimable. The second index is the group index indicating that the parameters with the same group index are for the same group level. or Identity. If the linear combination is not estimable. Parameter key If the model included random or repeated effects. a convention is followed such that if any column of T00 is a linear combination of preceding columns. Otherwise. Let θ be a vector consisting of the variance parameters in G and R. if there was a subject Group term for this variance structure. the data matrix [X | Z | y] is reduced by a QR factorization to an upper triangular matrix. Let [R0 | R1 | R2] be an orthogonal matrix such that the multiplication with [X | Z | y] results in an upper triangular matrix: R0 T T 00 T 01 y 0 0 T 11 y r 0 0 ye T R1 [ X Z y ] = R2 366 T . Hence for models that do not contain any fixed effects. if there was a group WinNonlin Computational details Restricted maximum likelihood For linear mixed effects models that contain random effects.14 User’s Guide Bands Subject_Term Group_Term Group_Level Number of bands or factors if appropriate for the variance type Subject term for this variance structure. The linear mixed effects model described under “The linear mixed effects model” on page 326 is: y = Xβ + Zγ + ε. restricted maximum likelihood estimation (REML) maximizes a likelihood that is only a function of the variance parameters θ and the observations y. if there was a group Level of the group that this parameter goes with. the first step in model analysis is determining the maximum likelihood estimates of the variance parameters associated with the random effects. In contrast. REML would be the same as ML. and not a function of the fixed effects parameters. V = Variance(y) = ZGZT + R. The full maximum likelihood procedure (ML) would simultaneously estimate both the fixed effects parameters β and the variance parameters θ by maximizing the likelihood of the observations y with respect to these parameters. This is accomplished in LinMix using Restricted Maximum Likelihood (REML) estimation. To obtain the restricted likelihood function in a form that is computationally efficient to solve. yr. For more information on REML. yr)) = –½ log (|Vr |) – ½ yr T Vr –1 yr log(L(θ.y e ) ) = -. [ y e1 y e2 … y en ] . see Corbeil and Searle (1976). and N = number of observations used. ye is treated as a multivariate residual. the negative of the log of the restricted likelihood function is the sum of the separate negative log-likelihood functions: log(L(θ.∑ y ei V y ei 2 2 i=1 n where V is the dispersion matrix for the multivariate normal distribution. ye)) = –ne/2 log(θe) – ½ (yeTye / θe) Vr = Variance(yr) θe = residual variance ne = residual degrees of freedom. and ignores y0. in order to increase the speed of the program and then the log-likelihood can be computed by the equivalent form: n 1 T –1 –1 log ( L ( θ . For some balanced problems. Since ye has a distribution that depends only on the residual variance. ye)) – (N–rank(X))/2 log(2π) where: log(L(θ.log ( V ) – -. ye)) = log(L(θ. yr)) + log(L(θ. and since yr and ye are independent.Linear Mixed Effects Modeling Computational details 14 REML estimation maximizes the likelihood of θ based on yr and ye of the reduced data matrix. 367 . or when it is unable to continue. the objective function used in the algorithm is the negative of the restricted log-likelihood without the constant term: objective function = – log(L(θ.14 User’s Guide Newton’s Algorithm for maximization of restricted likelihood As a general review. ye)) Newton’s algorithm is modified by adding a diagonal matrix to the Hessian when determining the direction vector for Newton’s. In the Linear Mixed Effects module. and Wright. a modification of Newton’s algorithm is used to minimize the negative restricted log-likelihood function with respect to the variance parameters θ. Since the variances can be determined through the types specified for G and R (e.. Since a constant term does not affect the minimization. the first and second derivatives of the variance can WinNonlin 368 . The algorithm needs a starting point. Another modification to Newton’s algorithm is that. and then at each iteration. where H–1 is the inverse of the Hessian of the objective function.). Newton’s algorithm is an iterative algorithm used for finding the minimum of an objective function. a line search is done as described in Fletcher (1980) to find an approximate minimum along the direction vector. once the direction vector d = – ( Hi + Di )–1 gi has been determined. Note from the discussion on REML estimation that the restricted log-likelihood is a function of the variance matrix Vr and the residual variance. it finds the next point using: θi+1 = θi – H–1(θi) g(θi). The algorithm stops when a convergence criterion is met. Unstructured. yr)) – log(L(θ. etc. Variance Components. which are in turn functions of the parameters θ.g. Murray. hence yielding the variance parameters that maximize the restricted likelihood. so that the diagonal matrix plus the Hessian is positive definite. and g is the gradient of the objective function. ensuring a direction of descent: θi+1 = θi – ( Hi + Di )–1 gi The appropriate Di matrix is found by a modified Cholesky factorization as described in Gill. The gradient and Hessian of the objective function with respect to θ are therefore functions of first and second derivatives of the variances with respect to θ. indicating that the model may be over-specified. the initial values are determined by the method of moments. and the system of equations will be solved to obtain initial parameter estimates. 1. 0. It may be in a trough. or on a flat surface. More often. Unstructured. General Options page. indicating successful convergence at a local—and hopefully global—minimum. LinMix will determine initial values. If either method fails. Compound Symmetry. then LinMix will calculate a “sample” variance by solving for random effects and taking their sums of squares and cross-products. and Toeplitz). The default value is 1×10–10.. via the Initial Variance Parameters… button. The algorithm converged. • 369 . The user can supply these in the LinMix Wizard. where ι is the convergence criterion specified on the General Options page of the wizard. Convergence criterion The convergence criterion used in the Linear Mixed Effects module is that the algorithm has converged if: gT (θ) H–1(θ) g(θ) < ι.01 for correlations. Sometimes the method of moments will yield estimates of θ that are REML estimates. but not at a local minimum. The possible convergence outcomes from Newton’s algorithm are: • Newton's algorithm converged with a final Hessian that is positive definite. The output is suspect. LinMix will use values that should yield a reasonable variance matrix.g. In this case the program will not need to iterate. and hence the gradient and Hessian are evaluated in closed-form in the optimization algorithm. Starting values Newton’s algorithm requires initial values for the variance parameter θ. For variance structures that are linear in θ (Variance Components.0 for variance components and standard deviations. at a saddle point.Linear Mixed Effects Modeling Computational details 14 be evaluated in closed-form. Newton's algorithm converged with a modified Hessian. e. If the variance structure is nonlinear. The sample variance will then be equated with the parametric variance. Intermediate variance matrix is not positive definite. the Wald statistic is: T ˆ –1 –1 T –1 ˆ T ˆ ( Lβ ) [ L( X V X ) L ] ( Lβ ) (11) ˆ where β and V are estimators of β and V.14 User’s Guide • • • Initial variance matrix is not positive definite. In common variance component models with balanced data. LinMix uses a method due to Allen (2001) for finding an F distribution to approximate the distribution of the Wald statistic. V) (10) where X is a matrix of fixed known constants. and V is the variance matrix dependent on other unknown parameters. Their methodology applies to variance component models with unbalanced data. This indicates an invalid starting point for the algorithm. L must be a matrix such that each row is in the row space of X. the Wald statistic has an exact F-distribution. WinNonlin Satterthwaite’s approximation for degrees of freedom The linear model used in linear mixed effects modeling is: Y ~ N(Xb. Using the chi-squared approximation results in a test procedure that is too liberal. The Wald statistic is asymptotically distributed under the null hypothesis as a chi-squared random variable with q = rank(L) degrees of freedom. b is an unknown vector of constants. Failed to converge in allocated number of iterations. Wald (1943) developed a method for testing hypotheses in a rather general class of models. ˆ 370 . The algorithm cannot continue. Giesbrecht and Burns (1985) suggested a procedure to determine denominator degrees of freedom when there is one numerator degree of freedom. To test the hypothesis H0: Lb = 0 in (10). Allen derived an extension of their technique to the general mixed model. Linear Mixed Effects Modeling Computational details 14 A one degree of freedom (df) test is synonymous with L having one row in (11).∼ χ . νu). Assuming that the u-th term is distributed F(1.= -Var –1 T –1 ν L(X V X) L The approach is to use (13) to solve for ν. When L has a single row. Find an orthogonal matrix R and a diagonal matrix Λ such that: T ˆ –1 –1 T T L ( X V X ) L = RΛR T –1 –1 T (13) The Wald statistic (11) can be re-expressed as: q ˆ ˆ W = ( R Lβ ) Λ (R Lβ ) = T –1 T T ∑ u=1 ˆ ( ru L β ) ------------------ 2 (14) λu where ru is the u-th column of R.⁄ -------------------------------------T –1 –1 T T –1 –1 T ˆ –1 –1 T L(X V X) L L(X V X) L L(X V X) L (12) Consider the right side of (12).-------–1 T –1 ν L( X V X) L 2 T –1 –1 T is true. the expected value of W is: 371 . Each term in (14) is similar in form to (12). The objective is to find ν such that the denominator is approximately distributed as a chi-squared variable with ν df. and λu is the u-th diagonal of Λ. If: ˆ L(X V X) L ν -------------------------------------.= -------------------------------------. the Wald statistic can be re-expressed as: T ˆ –1 –1 T ˆ 2 ˆ 2 L(X V X) L (Lβ) (Lβ) -------------------------------------. The distribution of the numerator is approximated by a chi-squared random variable with 1 df. then: ˆ L(X V X) L 2 -------------------------------------. ν). ANOVA Uses only Residual. and between LinMix and PROC MIXED in SAS. The partial tests in LinMix are not equivalent to As SAS’s PROC GLM Type III the Type III method in SAS though they coincide in most situations. LinMix and ANOVA Table 14-13. Equating expected values and solving for ν. by default. If a variable is not estimable. User can also select Residual. its expected value is qν/(ν-2). one obtains: ν = 2 E{W} / (E{W} – q) Comparing LinMix to ANOVA and SAS’s PROC MIXED This section outlines differences between LinMix and the previous WinNonlin ANOVA module. Prints “Not estimable” Not estimable parameters 372 . See “Partial tests” on page 362.’ LinMix provides the option to write out 0 instead. LinMix outputs ‘Not estimable. LinMIx compared to WinNonlin ANOVA Feature Degrees of freedom Partial tests of model effects LinMix Uses Satterthwaite approximation for all degrees of freedom.14 User’s Guide q WinNonlin E( W) = u–1 ------------∑ νu – 2 νu Assuming W is distributed F(q. Saturated variance structure If the random statement saturates the variance. where g is the gradient and H is the Hessian of the negative log-likelihood. LinMix sets the residual variance to zero. too many parameters in the model). The default convergence criterion options in SAS are ‘convh’ and ‘relative’. To make LinMix and SAS® consistent. put them on the same random tab. random. set the ‘noprofile’ option on the PROC MIXED statement. on separate tabs. In LinMix. In addition. SAS behaves similarly. This is how it is done in LinMix. as defined in Corbeil and Searle (1976). except that it does not follow this rule in the case of Variance Components. and the likelihood calculation method in LinMix also differs from SAS.Linear Mixed Effects Modeling Comparing LinMix to ANOVA and SAS’s PROC MIXED 14 LinMix and PROC MIXED Table 14-14. and set some parameters to zero. by default. PROC MIXED PROC MIXED chooses different defaults based on the model. Multiple factors on the same random command REML By default. Regardless of variance structure: If random effects are correlated.e. set the SAS ‘ddfm=satterth’ option. LinMIx compared to SAS PROC MIXED Feature Degrees of freedom LinMix Uses Satterthwaite approximation for all degrees of freedom. so LinMix and SAS will differ for models with multiple terms on a random statement using the VC structure.. This often occurs when the variance is over-specified (i. In particular. and hence will produce different final parameter estimates. Convergence cri. effects associated with the matrix columns from a single random statement are assumed to be correlated per the variance specified. Akaike and LinMix uses the smaller-is-better form of Schwarz’s criteria AIC and SBC in order to match WinNonlin and WinNonMix. The SAS residual needs to be added to the other SAS variance parameter estimates in order to match the LinMix output. SAS allows the final Hessian be non positive definite. A userspecified variance structure takes priority over that supplied by LinMix. REML (Restricted Maximum Likelihood) estimation. 373 . SAS uses the bigger-is-better form. User can also select Residual. SAS sets the residual to some other number. To be consistent with LInMix. SAS modifies the REML by ‘profiling out’ residual variance.The convergence criterion is gT H–1g < tolteria erance with H positive definite. LinMix –log(likelihood) may not equal twice SAS ‘–2 Res Log Like’ if noprofile is not set. does maximum likelihood estimation on the residual vector. AIC and SBC are functions of the likelihood. The LinMix criterion is equivalent to SAS with the ‘convh’ and ‘absolute’ options set. If independent. Modified likelihood LinMix always tries to keep the final Hessian (and hence the variance matrix) positive definite. and repeated statements. It will determine when the variance is over-specified. the residual is not estimable. LinMix outputs SAS writes the value of 0. LinMIx compared to SAS PROC MIXED (continued) Feature LinMix PROC MIXED WinNonlin Keeping VC posi.For VC structure. but will not nent. where Dii is the original diagonal of XTX.com. it outputs 0. but requires data to be sorted and considers each new regressor value to indicate a new subject or group. LinMix uses the standard deviations for the SAS prints the variances in the output for final parameters. SAS doesn’t require an explicit repeated out any time vari. and the parameter estimates are flagged as biased. See “Partial tests” on page 362. force individual variances > 0.parameters. Classification variables are treated similarly. terms of standard deviations. Repeated withLinMix requires an explicit repeated model. Partial tests of model effects The partial tests in LinMix are not equivalent to the Type III method in SAS though they coincide in most situations. For details. and prints the standard devia. 374 . Not estimable parameters ARH. The i-th column of X is deemed linearly dependent on the proceeding columns if dii2 < (singularity tolerance)×Dii. The Type III results can change depending on how one specifies the model. consult the SAS manual or contact support@pharsight. Group as classifi. ‘Not estimable. Singularity tolerance Define dii as the i-th diagonal element of the QR factorization of X. If intercept is present. SAS allows regressors.LinMix requires variables used for the subcation variable ject and group options to be declared as classification variables. LinMix keeps the final If SAS estimates a negative variance compotive variance matrix positive definite. then the parameter associated with the column is set to 0. and Dii the sum of squares of the i-th column of the QR factorization. CSH variance structures If a variable is not estimable. It instead creates an implicit time field able would need to be added. the first row of Dii is omitted. though the model is expressed in tions in the output. a time variable model. If the absolute diagonal element |dii| < (singularity tolerance)×Dii.To run some SAS examples.’ LinMix provides the option to write out 0 instead. Define dii as the diagonal element of the XTX matrix during the sweeping process. and assumes data are sorted appropriately.14 User’s Guide Table 14-14. bioequivalence comparisons rely on a criterion. 375 . Further details about WinNonlin bioequivalence are provided under the following headings. • • • • • • “Introduction to bioequivalence” on page 375 “The model and data” on page 378 “Average bioequivalence” on page 379 “Individual and population bioequivalence” on page 387 “Bioequivalence Wizard” on page 395 “Bioequivalence output” on page 403 Introduction to bioequivalence Basic terms Bioequivalence is said to exist when a test formulation has a bioavailability that is similar to that of the reference. bioequivalence involves comparison between test and reference drug products. a predetermined bioequivalence limit. individual. and population bioequivalence WinNonlin includes a Bioequivalence Wizard that calculates average or population/individual bioequivalence. Defined as relative bioavailability. and Population. where the test and reference products can vary. and calculation of a confidence interval for that criterion. Although bioavailability and bioequivalence are closely related. depending upon the comparison to be performed.Chapter 15 Bioequivalence Calculating average. There are three types of bioequivalence: Average. Individual. the 1992 guidance also provided specific recommendations for (1) logarithmic transformations of pharmacokinetic data. usually 80125% for the ratio of the product averages. Population and individual bioequivalence include comparisons of both the averages and variances of the study measure. Population bioequivalence assesses the total variability of the measure in the population. sometimes referred to as relative bioavailability. population and individual bioequivalence. such as area under the curve (AUC) and peak concentration (Cmax) be based on a test procedure termed the two one-sided tests procedure to determine whether the average values for pharmacokinetic parameters were comparable for test and reference formulations. Individual bioequivalence assesses within-subject variability for the test and reference products as well as the subject-by-formulation interaction. The objective of a bioequivalence study is to determine whether bioequivalence exists between a test formulation and reference formulation and to hence identify whether the two formulations are pharmaceutical alternatives that can be used interchangeably to achieve the same effect. (2) methods to evaluate sequence effects. the 1992 guidance recommended that statistical analysis for pharmacokinetic parameters. The US FDA Guidance for Industry (January 2001) Statistical Approaches to Establishing Bioequivalence recommends that average bioequivalence be supplemented by two new approaches. To establish bioequivalence. indicates that the two formulations are therapeutically equivalent and will provide the same therapeutic effect. and (3) methods to evaluate outlier data. In addition to specifying this general approach. with random assignment to the two possible sequences of drug product administration.15 User’s Guide The July 1992 FDA guidance on Statistical Procedures for Bioequivalence Studies Using a Standard Two-Treatment Crossover Design recommended that a standard in vivo bioequivalence study design be based on the administration of either single or multiple doses of the test and reference products to healthy subjects on separate occasions. the Food and Drug Administration (FDA) usually does not require a new drug 376 WinNonlin . This approach is termed average bioequivalence and involves the calculation of a 90% confidence interval for the ratio of the averages of the test and reference products. For example. Further. to get approval to market a generic drug product. the calculated confidence interval should fall within a bioequivalence limit. Bioequivalence studies are important because establishing bioequivalence with an already approved drug product is a cost-efficient way of obtaining drug approval. Bioequivalence Bioequivalence between two formulations of a drug product. A procedure for establishing average bioequivalence was recommended by the FDA in the 1992 guidance on “Statistical Procedures for Bioequivalence Studies Using a Standard Two-Treatment Crossover Design. Drug pre377 . Three different types of bioequivalence have been established by the FDA. new routes of administration of a drug product. usually calculated at the 90% confidence level. should fall within a predetermined acceptability limit. The individual bioequivalence approach assesses within-subject variability for the test and reference products as well as the subject-by-formulation interaction. and to determine the confidence interval for the difference. population and individual bioequivalence approaches include comparisons of both the averages and variances of the study measure. All types of bioequivalence comparisons are based on the calculation of a criterion (or point estimate). The FDA also recommended an equivalent approach using two. population bioequivalence. with random assignment to the possible sequences of drug administration. These added approaches are needed because average bioequivalence uses only a comparison of averages. The concepts of population and individual bioequivalence are related to two types of drug interchangeability—prescribability and switchability. The guidance then recommended that a statistical analysis of pharmacokinetic parameters such as AUC and Cmax be done on a log-transformed scale to determine the difference in the average values of these parameters between the test and reference data. and individual bioequivalence. Both the confidence interval approach and the two one-sided t-tests are described further below. and not the variances. one-sided t-tests. on the calculation of a confidence interval for the criterion. and on a predetermined acceptability limit for the confidence interval.” The FDA recommended administration of either single or multiple doses of the test and reference formulations to subjects. These are average bioequivalence. the confidence interval. To establish bioequivalence. of the bioequivalence measures. population and individual bioequivalence. The population bioequivalence approach assesses the total variability of the study measure in the population. usually within 20% of the reference average. In contrast. and for a drug product that has changed after approval. Bioequivalence studies are also useful for testing new formulations of a drug product. The January 2001 guidance “Statistical Approaches to Establishing Bioequivalence” recommends that average bioequivalence be supplemented by two new approaches.Bioequivalence Introduction to bioequivalence 15 application submission if the generic drug company can provide evidence of bioequivalence between the generic drug product and the innovator product. 15 User’s Guide scribability refers to when a physician prescribes a drug product to a patient for the first time. Variance structures The variance structures available in the Bioequivalence Wizard are the same as those available in the Linear Mixed Effects Modeling wizard. See “The linear mixed effects model” on page 326. Drug switchability is usually assessed by individual bioequivalence. For more details on the models required for different kinds of bioequivalence. See “Variance structure” on page 353.). the Bioequivalence Wizard is based on a mixed effects model. valid characters include numbers and underscore (my_file_2). They cannot include spaces or the following operational symbols: +. =. /. 378 . After the first character. This program does not impute missing values. variable names may not have single or double quotes in them. If the data have many missing observations. They also cannot include parentheses (). Drug prescribability is usually assessed by population bioequivalence. Variable name limits for use in WinNonlin bioequivalence analyses. consider imputing estimated values to produce complete records per subject. question marks (?) or semicolons (. and must choose from a number of bioequivalent drug products.*. WinNonlin The model and data Linear model Like the Linear Mixed Effects Modeling wizard. see the US FDA Guidance for Industry (August 1999). Finally. . If a subject has a missing observation. Drug switchability refers to when a physician switches a patient from one drug product to a bioequivalent product. such as when switching from an innovator drug product to a generic substitute within the same patient. the program assumes complete data for each subject. that subject is not included in the analysis. variable (column) names must begin with a letter. Missing data For population and individual bioequivalence. Replicated crossover designs should be used for individual bioequivalence studies. Period 1 T R Period 2 R T Sequence 1 Sequence 2 Recommended models for average bioequivalence The recommended model to be used in average bioequivalence studies depends on the type of study design that was used – parallel. nonreplicated crossover. Replicated designs involve multiple doses. for the two formulations T and R. then the study is considered a replicated design. For example. In nonreplicated designs. nonreplicated crossover. In the Bioequivalence Wizard. whereas in a crossover design each subject receives different formulations in different time periods. subjects receive only one dose of the test formulation and only one dose of the reference formulation. 379 . Period 1 T R Period 2 R T Period 3 T R Sequence 1 Sequence 2 An example of a nonreplicated crossover design is the standard 2x2 crossover design described in the 1992 FDA guidance and in the table below. Crossover designs are further broken down into replicated and nonreplicated designs. or replicated crossover. a 2×3 crossover design as described in the table below is a replicated crossover design.Bioequivalence Average bioequivalence 15 Average bioequivalence Study designs The most common designs for bioequivalence studies are replicated crossover. and parallel. if all formulations have at least one subject given more than one dose of that formulation. The FDA recommends that a bioequivalence study should use a crossover design unless a parallel or other design can be demonstrated to be more appropriate for valid scientific reasons. In a parallel design. each subject receives only one formulation in randomized fashion. and can be used for average or population bioequivalence analysis. the correct standard errors will be computed for sequence means and tests of sequence effects. 380 . so the default error model ε ~ N(0.15 User’s Guide Replicated crossover designs For replicated crossover designs. the default model used in the Bioequivalence Wizard is the example model given for replicated designs in Appendix E of the FDA guidance: Fixed effects model is: • DependentVariable = Intercept + Sequence + Formulation + Period WinNonlin Random effects model is: • • • Treatment Variance Blocking Variable set to Subject Type set to Banded No-Diagonal Factor Analytic with 2 factors Repeated specification is: • • • • Period Variance Blocking Variables set to Subject Group set to Treatment Type set to Variance Components Rather than using the implicit repeated model as in Appendix E. but using maximum likelihood instead of method of moments to estimate inter-subject variance. the time variable is made explicit as Period. This is equivalent to the classical analysis method. σ2 I) will be used. Using Subject as a random effect this way. Fixed effects model is: • DependentVariable = Intercept + Sequence + Formulation + Period Random effects model is the nested term: • Subject(Sequence) There is no repeated specification. the default model is as follows. Nonreplicated crossover designs For nonreplicated crossover designs. F-statistic and p-value. which contain the degrees of freedom (DF).1 ) Note that. Sum of Squares (SS).Bioequivalence Average bioequivalence 15 Using a fixed effect model. in addition to using the default model. Var(Subj(Seq)) is the intersubject (between subject) variance. for this default model. Note: In each case. Mean Squares (MS). When this default model is used for a standard 2x2 crossover design. WinNonlin creates two additional tabs in the output called Sequential SS and Partial SS. the user may supplement or modify the model to suit the analysis needs of the data set in consideration. then the Final Variance Parameters tab will contain two additional parameters: Intersubject CV = sqrt( exp(Var(Subj(Seq)) . Fixed effects model is: • Formulation There is no random model and no repeated specification. the default model is as follows. the first step is the computation of the least squares means (LSM) and standard errors of the test and reference formulations and the standard error of the difference of the test and reference least squares means. These 381 . one must construct pseudo-F tests by hand to accomplish the same task. for each of the model terms. the data is also ln-transformed. Note that the F-statistic and p-value for the Sequence term are using the correct error term since Subject(Sequence) is a random effect. Parallel designs For parallel designs. and Residual_Variance is the intrasubject (within subject) variance. Least squares means In determining the bioequivalence of a test formulation and a reference formulation. These tables are also included in the text output. If. so the residual error term is included in the model.1 ) Intrasubject CV = sqrt( exp(Residual_Variance) . For ln-transform or data already ln-transformed. DiffSE = standard error of the difference in LSM (computed by LinMix). RatioSE = standard error of the ratio of the least squares means. WinNonlin The geometric LSM are computed for transformed data. TestLSM = test least squares mean (computed by LinMix). For non-transformed. fractionToDetect = (user-specified percent of reference to detect) / 100. See also “Least squares means” on page 344.15 User’s Guide quantities are computed by the same process that is used for the Linear Mixed Effects module. Ratio(%Ref) = 100 (TestLSM / RefLSM) For ln-transform or data already ln-transformed. To simplify the notation for this section. the ratio is obtained on arithmetic scale by exponentiating. Difference = TestLSM – RefLSM The ratio calculation depends on data transformation. let: • • • • • RefLSM = reference least squares mean (computed by LinMix). RefGeoLSM = 10RefLSM TestGeoLSM = 10TestLSM The difference is the test LSM minus the reference LSM. the ratio is Ratio(%Ref) = 100 · 10Difference 382 . RefGeoLSM = exp(RefLSM) TestGeoLSM = exp(TestLSM) For log10-transform or data already log10-transformed. Ratio(%Ref) = 100exp(Difference) Similarly for log10-transform or data already log10-transformed. lower = Difference – t (1-alpha/2). Chow and Liu (2000). CI_Lower = 100 (1+lower/RefLSM) = 100 ⎛ 1 + ----------------------------------------------------. first the following values are computed using the students-t distribution.⋅ RatioSE⎞ ⎝ RefLSM – t ( 1 – α ⁄ 2 ) . for example. See. and for the confidence level that the user gave on the bioequivalence tab if that value is different than 80. CI_Lower = 100 · exp(lower) CI_Upper = 100 · exp(upper) For log10-transform or data already log10-transformed. in which case the Bioequivalence module will print “Not Estimable. df RefLSM⎠ = where the approximation RatioSE = DiffSE / RefLSM is used.⋅ --------------------⎞ ⎝ TestLSM 100 ⎛ ---------------------------------------------------. For ln-transform or data already ln-transformed. Similarly. page 86. It is possible that there will not be convergence.df · DiffSE These values are then transformed if necessary to be on the arithmetic scale.Bioequivalence Average bioequivalence 15 Classical and Westlake confidence intervals Output from the Bioequivalence module includes the classical confidence intervals and Westlake confidence intervals for confidence levels equal to 80. To compute the classical confidence intervals.df · DiffSE upper = Difference + t (1-alpha/2). or 95. Computing these intervals is an iterative procedure for which the NelderMead simplex algorithm is used.” 383 . 90. df ⎠ ( TestLSM – RefLSM ) DiffSE RefLSM – t ( 1 – α ⁄ 2 ) . and translated to percentages. 90. 95. CI_Upper = 100 (1 + upper / RefLSM) The procedure for computing the Westlake confidence intervals is given in many statistical references. where alpha = (100 – confidenceLevel) / 100. CI_Lower = 100 · 10lower CI_Upper = 100 · 10upper For no transform. 05. the first test is done similarly using log10 instead of ln. average bioinequivalence has been shown. CI_Upper) is contained within LowerBound and UpperBound. Two one-sided T-tests For ln-transform or data already ln-transformed. Otherwise. these options are entered on the bioequivalence options tab of the Bioequivalence wizard.0–fractionToDetect)) = 100 (1 / (1 – fractionToDetect)) UpperBound (no transform) = 100 + (percent of reference to detect) If the interval (CI_Lower. average bioequivalence has been shown. less than a 5% chance of rejecting H0 when it was actually true. the CI_Lower and CI_Upper for the user-specified value of the confidence level are compared to the following lower and upper bounds.15 User’s Guide The bioequivalence conclusion that appears in the text output depends on the user-specified values for the confidence level and for the percent of reference to detect. If the p-value is <0. CI_Upper) is completely outside the interval (LowerBound. If the interval (CI_Lower. UpperBound). To conclude whether bioequivalence has been achieved. LowerBound = 100 – (percent of reference to detect) UpperBound (ln-transforms) = 100 exp( –ln(1 – fractionToDetect) ) = 100 (1 / (1 – fractionToDetect)) UpperBound (log10-transforms)=100 · 10^( –log10(1. the first t-test is a left-tail test of the hypotheses: H0: Difference < ln(1 – fractionToDetect) (bioinequivalence for left test) H1: Difference ≥ ln(1 – fractionToDetect) (bioequivalence for left test) The test statistic for performing this test is: t1 = ( (TestLSM – RefLSM) – ln(1 – fractionToDetect) ) / DiffSE The p-value is determined using the t-distribution for this t-value and the degrees of freedom. then the user can reject H0 at the 5% level. the module has failed to show bioequivalence or bioinequivalence. For log10-transform or data already log10-transformed. Note that the upper bound for log10 or ln-transforms or for data already transformed is adjusted so that the bounds are symmetric on a logarithmic scale. WinNonlin 384 . i.e. the second test is a right-tail test of the hypotheses: H0: Difference > – ln(1 – fractionToDetect) (bioinequivalence for right test) H1: Difference £ – ln(1 – fractionToDetect) (bioequivalence for right test) The test statistic for performing this test is: t2 = ( (TestLSM – RefLSM) + ln(1 – fractionToDetect) ) / DiffSE For log10-transform or data already log10-transformed. However for log10 or ln-transforms. the second test is: H0: Ratio > 1 + fractionToDetect H1: Ratio £ 1 + fractionToDetect The test statistic for performing this test is: t2 = ( (TestLSM / RefLSM) – (1 + fractionToDetect) ) / RatioSE where the approximation RatioSE = DiffSE / RefLSM is used. the first test is (where Ratio here refers to Ratio(%Ref) / 100 ): H0: Ratio < 1 – fractionToDetect H1: Ratio Š 1 – fractionToDetect The test statistic for performing this test is: t1 = [ (TestLSM / RefLSM) – (1 – fractionToDetect) ] / RatioSE where the approximation RatioSE = DiffSE / RefLSM is used. the second test is done similarly using log10 instead of ln.8) = – ln(1. if the percent of reference to detect is 20%. For example. the right-tail test is Prob(>125%). the maximum of these p-values. then the lefttail test is Pr(<80%). since ln(0. the test will be symmetric on a logarithmic scale.25). the p-value for the second test above. and total of these p-values. The second t-test is a right-tail test that is a symmetric test to the first. but for ln-transformed data.Bioequivalence Average bioequivalence 15 For data with no transformation. The output for the two one-sided t-tests includes the p-value for the first test described above. For data with no transformation. The two one-sided t-tests procedure is 385 . For ln-transform or data already ln-transformed. μR < θL vs. Power = 1 – (probability of a type II error) = probability of rejecting H0 when H1 is true. or page 99 of Chow and Liu (2000). if the classical (1-2×alpha) ×100% confidence interval for difference or ratio is within LowerBound and UpperBound.15 User’s Guide operationally equivalent to the classical confidence interval approach. Briefly.μR > θL H02: μT . That is. Anderson-Hauck test See Anderson and Hauck (1983). and data already ln-transformed. these limits are entered on the bioequivalence options tab. HA1: μT . the hypotheses being tested are: H0: RefLSM = TestLSM H1: RefLSM ≠ TestLSM For the no-transform case. The Anderson-Hauck test statistic is tAH given by WinNonlin YT – YR – ( θL + θU ) ⁄ 2 t AH = -----------------------------------------------------Diff SE where Diff SE is the standard error of the difference in means. the Anderson-Hauck test is based on the hypotheses: H01: μT .μR > θU vs. this changes to: 386 . Rejection of both null hypotheses implies bioequivalence. Power Power is the post-hoc probability of detecting a difference greater than or equal to a specified percentage of the reference least squares mean.μR < θU where θL and θU are the lower and upper Anderson-Hauck limits. Under the null hypothesis. the power is the probability of rejecting H0 given: |TestLSM – RefLSM | ≥ fractionToDetect × RefLSM For ln-transform. HA2: μT . then both the H0’s given above for the two tests are also rejected at the alpha level by the two one-sided t-tests. In general. In the bioequivalence calculations. this test statistic has a noncentral t-distribution. Bioequivalence Individual and population bioequivalence 15 |TestLSM – RefLSM | ≥ – ln (1 – fractionToDetect). For the sake of illustration.2×RefLSM) = Pr (Difference/DiffSE > t (1-alpha/2).df – 0. |TestLSM – RefLSM| = 0. assume fractionToDetect = 0.df given |Difference| = 0. replicated crossover designs are 387 . So: Power = Pr (rejecting H0 at the alpha level given |Difference| = 0.2×RefLSM) Let: t1 = t (1-alpha/2).2×RefLSM) Note that tstat is T-distributed.2×RefLSM. within-formulation variation. and similarly for log10-transform and data already log10-transformed. and no transform was done on the data. Let p1 be the p-value associated with t1 (the area in the tail).2×RefLSM) = Pr (|Difference/DiffSE| > t (1-alpha/2). note that p2 may be negligible).df – 0.2×RefLSM) + Pr (Difference/DiffSE < –t (1-alpha/2).2×RefLSM / DiffSE tstat = (Difference – 0.2×RefLSM) + Pr(tstat < t2 given |Difference| = 0. RefLSM>0.2. Also the maximum probability of rejecting H0 occurs in the case of equality. Because individual bioequivalence relies on estimates of within-subject.df given |Difference| = 0. and let p2 be the p-value associated with t2 (the area in the tail.2×RefLSM) / DiffSE Then: Power = Pr(tstat > t1 given |Difference| = 0.2×RefLSM / DiffSE t2 = – t (1-alpha/2). Then: Power = 1 – (p1 – p2) Individual and population bioequivalence Introduction Individual and population bioequivalence are used to assess switchability and prescribability. df given |Difference| = 0. < θ P if σ R > σ P 1 2 -.< θ P if σ R ≤ σ P 1 2 -. the sum of within.σ P 2 where σP defaults to 0.E ( Y jR – Y j'R ) 2 E ( Y jT – Y j'R ) – E ( Y jR – Y j'R ) -----------------------------------------------------------------------.PercentReference) + εP) /σP2. Designs that can be appropriately analyzed include. σR2 is computed by the program and is the total variance of the reference formulation. In the wizard.. The criteria take the linearized form 2 2 2 2 η P1 = E ( μ T – μ R ) + σ T – ( 1 + θ P ) σ R < 0 388 2 2 2 if σ R > σ P . Bioequivalence criterion The FDA population bioequivalence (PBE) criteria are E ( Y jT – Y j'R ) – E ( Y jR – Y j'R ) -----------------------------------------------------------------------. but are not limited to.20. Sequence Period Design WinNonlin 2 4 4 4 2 3 2 6 2 6 4 4 2 3 5 3 3 3 4 4 TRTR/RTRT TRTR/RTRT/TRRT/RTTR TT/RR/TR/RT TRT/RTR/TRR/RTT TRRTT/RTTRR TRR/RTR/RRT RTR/TRT TRR/RTT/TRT/RTR/TTR/RRT TRRR/RTTT TTRR/RRTT/TRRT/RTTR/TRRR/RTTT The algorithm works for balanced and unbalanced data with an equal number of periods in each sequence and one measurement per subject in each period. This algorithm was developed by Francis Hsuan.15 User’s Guide required. The default value for PercentReference is 0.and between-subject variance. σP is called the Total SD standard.2 and θP = (ln(1 . i.e. < θ I 1 2 -.05. For constant scaling. and the default value for εI is 0.Bioequivalence Individual and population bioequivalence 15 η P2 = E ( μ T – μ R ) + σ T – σ R – θ P σ P < 0 The FDA individual bioequivalence (IBE) criteria are 2 2 2 2 if σ R ≤ σ P E ( Y jT – Y jR ) – E ( Y jR – Y j'R ) ----------------------------------------------------------------------. The method of calculating that upper confidence bound follows. use one of the following. The confidence interval is set on the Bioequivalence Options page. The default value for Percent Reference is 0.E ( Y jR – Y j'R ) 2 E ( Y jT – Y jR ) – E ( Y jR – Y j'R ) ----------------------------------------------------------------------. For mixed scaling. 389 . use ηP2 or ηI2. σWR2 is computed by the program and is the within-subject variance of the reference formulation.σ I 2 2 2 2 2 if σ WR > σ I if σ WR ≤ σ I where σI defaults to 0. use ηP1 or ηI1. In the wizard. The IBE criteria take the linearized form η I1 = E ( μ T – μ R ) + σ D + σ WT – ( 1 + θ I ) σ WR < 0 and 2 * 2 2 if σ WR > σ I η I2 = E ( μ T – μ R ) + σ D + σ WT – σ WR – θ I σ I < 0 where σ D will be defined below.20. * 2 * 2 2 2 if σ WR ≤ σ I For reference scaling. Population Individual If σR > σP If σR ≤ σP If σWR > σI If σWR ≤ σI use ηP1 use ηP2 use ηI1 use ηI2 If the upper confidence bound on the appropriate η is less than 0. then the product is bioequivalent in the chosen sense.2 and θI = (ln(1 – PercentReference) + εI) /σI2.< θ I 1 2 -. σΙ is called the within-subject SD standard. σ TR} ˜ are defined as follows: 2 2 390 . Let piT:= sum( Z iT ) and piR:= sum( Z iR ) be the number of occasions T and R ˜ respectively be˜assigned to the subjects in sequence i. ˜ ˜ ˜ and d i : = s T u iT – s R u iR .Y ijk't') follow the model –1 –1 –1 –1 ˜ ˜ Σ i = σ WT diag ( Z iT ) + σ WR diag ( Z iR ) + σ TT Z iT Z' iT ˜ ˜ + σ RR Z iR Z' iR + σ TR { Z iT Z' iR + Z iR Z' iT } ˜ ˜ ˜ ˜ ˜ ˜ ˜ ˜ 2 2 (1b) where the parameters σ = {σ WT . μR are the population mean responses of Y with treatments T and R respectively. Assume the mean responses of Y ij . be the number of subjects in sequence i. u iT : = p iT Z iT . σ WR . ˜ ˜ u iI : = ( u iT – u iR ) .σ RR . Let Z iT and Z iR be the ˜ design vector for treatment T and R respectively in sequence i.σ TT . ˜ ˜ u iR : = p iR Z iR .15 User’s Guide Details of computations Let Yijkt be the response of subject j in sequence i at time t with treatment k. The components ˜ of Y are arranged in the ascending order of time. μ i follow a linear model ˜ WinNonlin ˜ μ i = Z iT μ T + Z iR μ R ˜ ˜ ˜ (1a) where μT. and Y ij' be a vector of responses for subject j in sequence i. ˜ ˜ ˜ Assume that the covariances (Y ijkt.˜ni > 1. and Xi are design/model matrices. Y R) . For PBE and IBE investigations. corr (Y T.Y R') . 2 2 σ WT = σ T – σ TT is the intra-subject variance ( Y T – Y T' ) ⁄ 2 .Y T') . 1 > ρ RR > 0 . corr (Y T. corr (Y R. but not identical. It satisfies the equation 391 . it is useful to define additional parameters: 2 2 2 2 σ D = σ TT + σ RR – 2σ TR σ iT σ iR σ iI *2 *2 *2 * (2a) 2 = ( σ T – ( 1 – p iT )σ WT ) = ( σ R – ( 1 – p iR )σ WR ) * –1 2 –1 2 2 –1 2 2 –1 (2b) (2c) (2d) = ( σ D + p iT σ WT + p iR σ WR ) Except for (2a). 2 ρ TR = σ TR ⁄ ( σ T σ R ) = intra-subject coeff. 1 > ρ TT > 0 . all the quantities above are non-negative when they exist.Bioequivalence Individual and population bioequivalence 15 σ T = σ TT + σ WT and σ R = σ RR + σ WR are var ( Y ijTt ) and var ( Y ijRt ) . σ RR = ρ RR σ R and σ TR = ρ TR σ T σ R are intra-subject covariances. The quantity is similar. 1 > ρ TR > – 1 . 2 ρ RR = σ RR ⁄ σ R = intra-subject correlation coeff. σ TT = ρ TT σ T . 2 2 2 2 ρ TT = σ TT ⁄ σ T = intra-subject correlation coeff. σ WR = σ R – σ RR is the intra-subject variance ( Y R – Y R' ) ⁄ 2 . to the “subject-by-formulation” variance component in the mixed model discussed in the FDA guidance Section III. and Vi is the withinsequence sample covariance matrix. TRT/RTR. S iI = u' iI V i u iI ˜ ˜ ˜ ˜ ˜ ˜ S iWT = ( 1 – p iT ) ( tr { diag ( u iT )V i } – u' iT V i u iT ) ˜ ˜ ˜ S iWR = ( 1 – p iR ) ( tr { diag ( u iR )V i } – u' iR V i u iR ) ˜ ˜ ˜ where Y i• is the sample mean of the ith sequence. It yields identical results to those described in the FDA guidance Appendices F & H when the design is TRTR/RTRT.15 User’s Guide WinNonlin var ( Y ijT – Y ijR ) = σ D + σ WT + σ WR * 2 * 2 2 In general. Given the ith sequence. This method for PBE/IBE is based on the multivariate model. page 205). It can be shown that –1 –1 –1 –1 ˆ –1 Δ ∼ N (μ T – μ R.∑ n i d' i ∑ d i) ˜ ˜ i i (4a) S iT ∼ σ iT ( n i – 1 ) χ S iR ∼ σ iR ( n i – 1 ) χ S iI ∼ σ iI ( n i – 1 ) χ 2 –1 *2 *2 *2 –1 2 ni – 1 (4b) (4c) (4d) (4e) –1 2 ni – 1 –1 2 ni – 1 ( p iT – 1 ) ( n i – 1 ) S iWT ∼ σ WT ( p iT – 1 ) ( n i – 1 ) χ –1 2 392 . or TxRR/xRTT/RTxx/TRxx/xTRR/RxTT (Table 5. S iR = u' iR V i u iR .7 of Jones and Kenward. let ˆ Δ = ∑ d'i Yi• ˜ i S iT = u' iT V i u iT . This method is applicable to a variety of higher-order two-treatment crossover designs including TR/RT/TT/RR (the Balaam Design). * 2 and the quantity σ D is identical to σ D whenever the former is nonnegative. The FDA mixed model is a special case of the multivariate model above. this σ D may be negative. while FDA’s σ D is always non-negative. chosen to yield the method of moments estimates of the η’s. one may define unbiased moment estimators for the PBE criteria: 393 . SiI. and that the four statistics Δ.Bioequivalence Individual and population bioequivalence 15 S iWR ∼ σ WR ( p iR – 1 ) ( n i – 1 ) χ 2 –1 –1 2 ( p iT – 1 ) ( n i – 1 ) (4f) Furthermore. SiR} (for PBE) are statistically independent from {Δ} and {SiWT. SiWR}. Let αi’s be sets of normalized weights. SiWR (for IBE) are statistically independent. SiWT. it can be shown that {SiT. Then define the estimators of the components of the linearized criterion by Using the above notation. H q = n q E q ⁄ χ . H and U statistics: 2 ˆ ˆ ˆ ( Δ ) ) 1 ⁄ 2⎞ ⎛Δ +t H0 = . and U q = Uq for all other q. 394 * 2 * . P4. I1 and I2. that indicates bioequivalence. 2 Uq = ( Hq – Eq ) for all q where the degrees of freedom nq are computed using Satterthwaite’s approximation. * 1⁄2 ˆ Hη = η + ( ∑ Uq ) where U q = ( 1 + θ P ) U q for q = P3.95.nq for q = P1. If Hq < 0.05. I3a and I3b. Hq > 0 fails to show bioequivalence.95. P4. Then.n 0 ( var ⎝ ⎠ H q = n q E q ⁄ χ . Hsuan and Holder (2000). This can be generalized to compute the following sets of nq. P2. nq 2 2 for q = P3. the 95% upper bound for each η is .15 User’s Guide WinNonlin ˆ2 ˆ η P1 = Δ + E P1 + E P2 – ( 1 + θ P )E P3 – ( 1 + θ P )E P4 ˆ2 2 ˆ η P2 = Δ + E P1 + E P2 – E P3 – E P4 – θ P σ P and for the IBE criteria: ˆ2 η I1 = Δ + E I1 + E I2 – E I3a ˆ2 2 η I2 = Δ + E I1 + E I2 – E I3b – θ I σ I The FDA Guidance specifies a procedure to construct a 95% upper bound for η based on the TRTR/RTRT design using Howe’s approximation I and a modification proposed by Hyslop. To reload the most recently run model. 3. 2. For a discussion of the analysis types. as follows. To create a new analysis. The Bioequivalence Wizard provides options to use prior model settings (optional). click the Load button and open a LinMix/bioequivalence command (*. 5. • • 4. see “Average bioequivalence” on page 379 and “Individual and population bioequivalence” on page 387. Click Calculate to run the model. Start the Bioequivalence Wizard by choosing Tools>Bioequivalence Wizard or by clicking on the Bioequivalence Wizard tool bar button. • To load a previously-saved bioequivalence model. Open the data set to be analyzed. click the Previous Model button. 395 .Bioequivalence Bioequivalence Wizard 15 Bioequivalence Wizard To set up and run a bioequivalence analysis: 1.LML) file containing bioequivalence settings. For average bioequivalence. select the study type at the top left. proceed to step 4 below. Type of Bioequivalence: Select individual/population or average bioequivalence using the radio buttons at the upper right. regressor(s)/covariate(s). 9. to a command (*. 6. only Formulation and its reference value need to be specified. from the pull down list labeled Reference Value. If WinNonlin has not already selected the appropriate variables. and sort variable(s) must be specified. Period. Sequence. against which to test for bioequivalence.15 User’s Guide Based on the analysis and study types. For Average bioequivalence other classification variables can be added by dragging them from the Variable Collection to the Classification 396 . and Formulation fields. Select the reference formulation. WinNonlin will try to match the variable names from the active data set with the Subject. Note: For Average bioequivalence with a parallel study type. At any point in the wizard. and Formulation. Average bioequivalence analysis settings fall under the following headings: • • • • • “Fixed effects for average bioequivalence” on page 396 “Random effects” on page 398 “Repeated variance” on page 399 “Bioequivalence options” on page 400 “Bioequivalence general options” on page 401 Fixed effects for average bioequivalence The dependent variable(s). select the variables that identify the Subject.LML) file for later re-use. Sequence. WinNonlin 8. Click Next to proceed with the analysis: • • “Using average bioequivalence” on page 396 “Using population/individual bioequivalence” on page 401 Using average bioequivalence See the Examples Guide for an example using average bioequivalence. Period. including variable names. 7. the Save button can be used to save the analysis settings. Regressor variables or covariates are continuous independent variables. Drag and drop variables to categorize those needed in the LinMix model: • • • • Dependent variables contain the values to which the model will be fit. See also “Weighting” on page 241. Sort variables define subsets of data to be processed separately. For Average bioequivalence the Model Specification field automatically displays an appropriate fixed effects model for study type. formulation. see “The model and data” on page 378. The Weight Variable cannot be a classification variable. The weights are used to compensate for observations having different variances. In the most common situation.LML). When a weight variable is specified. If the sort variable has missing values.g. It must have been declared as a regressor/covariate before it can be used as a weight variable. See “Fixed effects” on page 352. Weight variable values should be proportional to the reciprocals of the variances. It can also be used in the model. the analysis will be performed for the missing level and MISSING will be printed as the sort variable value. e.Bioequivalence Bioequivalence Wizard 15 Variable field. To reuse the most recently run model.. the data are averages and weights are sample sizes associated with the averages. To define the fixed effects model: 1. a separate analysis will be done for each level of the sort variables. each row of data is multiplied by the square root of the corresponding weight. click the Previous Model button.e. Select the weight variable if weights for each record are included in a separate column.. • 397 . e. Classification variables or factors are categorical independent variables. e. temperature or body weight. i.. The Bioequivalence Wizard Fixed Effects dialog works very much like the LinMix Fixed Effects dialog.g. Edit the model as needed. and gender.g. click the Load Model button. Note: To reuse a model in a command file (*. drug concentration.. treatment. For discussion of the model. 2. Fixed Effects Confidence Level: Accept the default setting [95] or enter the confidence level for the parameter estimates. select Already Ln-transformed or Already Log10-transformed. select a log transformation from the pull down list for Dependent Variables Transformation. If the input data are already transformed. or Next to specify Random effects or Repeated variance. Random effects Use this tab of the Bioequivalence Wizard to set traditional variance components and/or to random coefficients.) 1. else see “Repeated variance” on page 399. Complete 398 .15 User’s Guide WinNonlin 2. Note: The Bioequivalence Wizard Variance Structure dialog operates exactly as it does in the LinMix wizard. The confidence level for the bioequivalence test is specified later. It identifies the subjects in a data set. 3. (See also “Variance structure” on page 353. To transform the dependent variable values before analysis. Click Calculate to run the model with default settings. Classification variables and regressors: Drag and drop column names (or type them) to as needed for the random effects model(s). 4. Variance Blocking Variables (Subject): (Optional) This variable must be a classification model term. Click the Add Random button to create multiple variance models. See also “Repeated measurements” on page 356.Bioequivalence Bioequivalence Wizard 15 independence is assumed among subjects. Thus. Group variable: (Optional) This variable must be a classification model term. It defines an effect specifying heterogeneity in the covariance structure. Click Calculate to run the analysis or Next to set Bioequivalence options. Subject produces a block diagonal structure with identical blocks. The repeated effect must contain only classification variables. If no Repeated statement is specified. Repeated variance Select the Repeated tab of the Bioequivalence Wizard to specify the R matrix in the mixed model. 4. 3. This is most commonly employed when a subject is specified in the Variance Blocking field. 5. R is assumed to be 2 equal to σ I . 7. 6. All observations having the same level of the group effect have the same covariance parameters. Check Random Intercept to indicate that the intercept is random. Each new level of the group effect produces a new set of covariance parameters with the same structure as the original group. if needed. select the type of covariance structure here. 399 . See “Variance structure” on page 353 for details on each type. Type: If a random effects model is specified. The default is off—no random intercept. Click Calculate to run the analysis or Next to set Bioequivalence options. Type: If a Repeated Specification has been included. Each new level of the group effect produces a new set of covariance parameters with the same structure as the original group. 1. Bioequivalence options This dialog of the Bioequivalence Wizard sets the range for bioequivalence. Repeated Specification: Enter the model by typing it or by dragging the classification variable names and clicking on the operators. 3. Complete independence is assumed across subjects. 2.15 User’s Guide To specify the repeated variance settings: WinNonlin 1.25 for log-transformed data. Group variable: (Optional) This must be a classification variable with no regressors or operators. Subject produces a block diagonal structure with identical blocks. This Subject field identifies subjects in the data set. The default model follows the FDA guidance. Thus. To set the bioequivalence options: 1. • • • • Confidence Level (90%) Percent of Reference to Detect (20%) Anderson-Hauck Lower Limit (0. select the type of covariance structure. All observations having the same level of the group effect have the same covariance parameters. Variance Blocking Variables (Subject): (Optional) This variable must be a classification model term. See “Average bioequivalence” on page 379.8) Anderson-Hauck Upper Limit (1. 5.2 for not transformed) 400 . “Variance structure” on page 353 for types. It defines an effect specifying heterogeneity in the covariance structure. Syntax rules for the repeated specification are the same as those for the fixed–effects model terms. The following options are pre-set to meet FDA requirements. 4. Note: The default Repeated model depends upon whether the crossover study is replicated or not. an optional ASCII text window). See the Examples Guide for an example. Bioequivalence general options This dialog of the Bioequivalence Wizard sets the following operations: • • • • • • ASCII output can be selected as an additional output form. 401 .Bioequivalence Bioequivalence Wizard 15 2. Launch the Bioequivalence Wizard and set up the variables for the analysis as described under “Bioequivalence Wizard” on page 395. See “Individual and population bioequivalence” on page 387 for a discussion of theory. To specify population/individual bioequivalence information: 1. A title can be assigned as a header for the ASCII output. and Intermediate Calculations. Convergence Criterion. The maximum number of iterations can be set. Using population/individual bioequivalence Following is an overview of the basic steps to perform population/individual bioequivalence in the Bioequivalence Wizard. up to a maximum of five lines. When all the appropriate settings have been made. Press Ctrl+Enter to move to the next line. Note: Population/Individual bioequivalence can only use replicated crossover study data. How output that is not estimable is to be represented may be set. The average bioequivalence calculations begin and the output appears in a new workbook (and. click the Calculate button to run the model. Click Calculate to run the analysis or Next for Bioequivalence general options. For each period each subject must have one measurement. Each sequence must contain the same number of periods. The kind of degrees of freedom can be stipulated. The advanced user options include: – Initial estimates for the Variance Parameters – Change defaults for Singularity Tolerance. or Next to proceed to the Fixed effects for population/individual bioequivalence. To complete the model: 1. Click Calculate to run the model with the default settings.. button to reload the last-run model. Drag the appropriate term(s). The Bioequivalence Wizard Fixed Effects dialog for population/individual bioequivalence works very much like that for LinMix. For Population bioequivalence. or Next to proceed to the Bioequivalence options for population/individual bioequivalence.15 User’s Guide 2. from the Variable Collection to the Dependent Variables field. Bioequivalence options for population/individual bioequivalence To set the bioequivalence options: 1. and fixed effects model are disabled and preset to FDA guidelines. The following options are pre-set to meet FDA standards. e.. WinNonlin Fixed effects for population/individual bioequivalence The Fixed effects model is prebuilt following FDA guidelines. select Already Lntransformed or Already Log10-transformed.2) and Epsilon (0. 2.g. • • • 402 Confidence Level (95%) Percent of Reference to Detect (20%) Population Limits: Total SD Standard (0.. Click Calculate to run the model. See “Fixed effects” on page 352. See “Individual and population bioequivalence” on page 387. Note: Use the Load Model button to load a previously-saved model. If the input data are already transformed. classification variables. 3. (Optional) Select a log transformation from the pull down list for Dependent Variables Transformation. Select Ln(x) or Log(x) to transform the values before analysis. concentration. Appropriate terms for study type are specified automatically as classification variables. regressors.02) . Use the Previous Model. SBC. F-tests and p-values for partial test Sort variables. Hypothesis degrees of freedom. Units. mean squares. Denominator degrees of freedom. etc. To start calculating Population/Individual bioequivalence. and Hessian eigenvalues Sort Variable(s). Units. Denominator degrees of freedom. click Calculate. p-value. mean squares.05) 2. Hypothesis degrees of freedom. residual variance. F statistic. confidence intervals. Dependent Variable(s). number of observations used. Average bioequivalence can produce text output as an option. Variance parameter names assigned by WinNonlin 403 Sequential Tests Sequential SS Partial Tests Partial SS Final Fixed Parameters Final Variance Parameters Final estimates of the variance parameters Parameter Key . Average bioequivalence output worksheets Worksheet Average Bioequivalence Ratio Tests Diagnostics Contents Output from the bioequivalence analysis Ratios of reference to test mean computed for each subject Number of observations. Bioequivalence output The Bioequivalence Wizard generates a new bioequivalence workbook. AIC. Average bioequivalence Average bioequivalence generates a workbook containing following output. residual degrees of freedom. and p-value Only for 2x2 crossover designs using the default model: ANOVA sums of squares. Hypothesis. degrees of freedom. dependent variable(s). Table 15-1. t-statistic.2) and Epsilon (0. and p-value Only for 2x2 crossover designs using the default model: ANOVA sums of squares. effect level. F statistic. F-tests and p-values for sequential test Sort Variable(s). Dependent Variable(s).Bioequivalence Bioequivalence output 15 • Individual Limits: Within Subject SD Standard (0. Population/individual bioequivalence always generates text output as well. Hypothesis. residual sums of squares. parameter estimate. units. restricted log likelihood. degrees of freedom. The output also contains the Average Bioequivalence Statistics. Average bioequivalence output worksheets (continued) Worksheet Least Squares Means LSM Differences Residuals User Settings Initial Variance Parameters Iteration History Contents Sort Variable(s). Population/individual output worksheets Worksheet Population/Individual Bioequivalence Ratio Tests User Settings Contents Output from the bioequivalence analysis Ratios of reference to test mean User-specified settings Depending which settings were specified in the wizard. Obj_function. Units. The Linear Mixed Effects module. and the least squares means for the test formulation Difference in least squares means Sort Variable(s). the output also contains the ratios table. Dependent Variable(s). Therefore the output for average bioequivalence includes all the output obtained by running LinMix. Units. described under “Ratio tables” on page 406. and degrees of freedom for the difference. For crossover designs. computes these quantities. and column for each variance parameter WinNonlin To determine average bioequivalence of a test and reference formulation. the output may not include all of these worksheets. These are described in detail in “Introduction to bioequivalence” on page 375 and “Average bioequivalence” on page 379.15 User’s Guide Table 15-1. LinMix. the first step is computation of the least squares means of the test and reference formulations and their standard errors. and information on residual effects vs. Population/Individual bioequivalence Table 15-2. 404 . as described in “LinMix output” on page 361. Dependent Variable(s). and computation of the standard error of the difference between test and reference least squares means. predicted and observed effects User-specified settings Parameter and estimates for variance parameters Iteration. if the user specified a percent of reference to detect of 20%. Note: The FDA guidance suggests that both the ratio test and the tests listed below need to be passed to indicate bioequivalence. the Within Subject SD Standard. the Total SD Standard. This is compared with the percent of reference to detect that was specified by the user. Population/Individual Bioequivalence Calculations Statistic Const_Pop_eta Ref_Pop_eta Mixed_Pop_eta Const_Indiv_eta Ref_Indiv_eta Mixed_Indiv_eta Indication Test statistic for population bioequivalence with constant scaling Test statistic for population bioequivalence with reference scaling Test statistic for population bioequivalence with mixed scaling Test statistic for individual bioequivalence with constant scaling Test statistic for individual bioequivalence with reference scaling Test statistic for individual bioequivalence with mixed scaling 405 . For example. to determine if bioequivalence has been shown. SigmaR . to determine whether mixed scaling for population bioequivalence will use the constant-scaling values or the reference-scaling values. The population/individual bioequivalence statistics are described further in “Introduction to bioequivalence” on page 375.value that is compared with sigmaP. SigmaWR . 125%) to show bioequivalence for the ratio test. They include: Difference (Delta) = difference in sample means between the test and reference formulations. to determine whether mixed scaling for individual bioequivalence will use the constant-scaling values or the reference-scaling values.value that is compared with sigmaI. Each of the following is given. An upper confidence bound < 0 indicates bioequivalence. then Ratio(%Ref) needs to be in the interval (80%. with its upper confidence bound and bioequivalence conclusion. Ratio(%Ref) = ratio of the test to reference means expressed as a percent. Table 15-3. and also specified a ln-transform.Bioequivalence Bioequivalence output 15 Errors and warnings in the computations appear in the ASCII output as well as the application log or data history. Let ref be the corresponding value for the reference formulation. then let test be the mean of the values of the dependent variable measured after administration of the test formulation. Ratio(%Ref) = 100×exp(test – ref) = 100×exp(test)/exp(ref) 406 . for the cases of no transform. and similarly for ref.15 User’s Guide For crossover designs. For each subject. This is in response to the FDA guidance “Measures for each individual should be displayed in parallel for the formulations tested. For data that was specified to be ‘already ln-transformed.’ these values are back-transformed to be in terms of the original data: Difference = exp(test) – exp(ref). If multiple test or reference values exist for the same subject. In particular. and similarly for ref. except for the cases of ‘ln-transform’ and ‘log10-transform’ where the geometric mean should be used: ‘ln-transform’: WinNonlin ⎛ ⎞ test = exp ⎜ ∑ ln ( X i ) ⁄ N⎟ ⎝ i ⎠ N –1 ‘log10-transform’: test = 10 ∑ log i 10 X i where Xi are the measurements after administration of the test formulation. Ratio(%Ref) = 100 × test / ref. or log10-transform. for the case in which only one dependent variable measurement is made. for each BE measure the ratio of the individual geometric mean of the T product to the individual geometric mean of the R product should be tabulated side by side for each subject. the ratios table contains: Difference = test – ref. ln-transform.” Let test be the value of the dependent variable measured after administration of the test formulation for one subject. the output also contains the ratio tables. Ratio tables For any bioequivalence study done on a crossover design. below. the output also includes for each test formulation a table of differences and ratios of the original data computed individually for each subject. Bioequivalence Bioequivalence output 15 Similarly. 407 . for data that was specified to be ‘already log10-transformed’: Difference = 10test – 10ref Ratio(%Ref) = 100×10test – ref = 100×10test/10ref Note: For ‘already transformed’ input data. if the mean is used for test or ref. then this is equal to the geometric mean. and the antilog of test or ref is taken above. 15 User’s Guide WinNonlin 408 . and doses WinNonlin simulations can use library models or user defined models to calculate response values at specified time points. This capability is especially useful. data variables. or determining optimal sampling times. as detailed under “Modeling” on page 193. except that no Y variable is specified. and the Simulate check box. a set of time (X variable) values. to illustrate. Simulations can use any compiled or ASCII model. Running a simulation of a model can quickly uncover errors in the model specification. Simulation of a compiled library model This example uses a compiled WinNonlin library model to illustrate the use of simulation to generate multiple-dose concentrations based on parameters from fitting single-dose data. and dosing constants (for some models). parameter values for that model. and an example of simulation using a user-defined ASCII model. Simulations are also useful when writing custom models. to explore the model shape and behavior. a model. in simulating steady state responses from single dose data. must be checked. Simulating responses Simulation of output values requires a set of time or other X-variable values. in the Data Variables dialog. and dosing are set up as for model fitting. 409 . See the Examples Guide for an example using simulation to evaluate competing sampling schedules. The model. An example is provided below. for example. including a user model. Note that a compiled or ASCII model could be used. comparing drug concentrations for different dosing regimens or for different values of model parameters (such as absorption or elimination rates).Chapter 16 Simulation Generating response values based on a model. 18 in cell A2. and drag down to Row 201. incrementing by 0. an automatic record of the data history. Enter time values in the first column. In the resulting dialog. enter 0 in the first row. or double-click on the column header. highlight cell A2 and position the cursor over the lower right hand corner of the cell until the cursor changes into a small black cross. Model 11 from WinNonlin's compiled pharmacokinetic library. 2. Select File>New from the WinNonlin menus. and =A1+0.18 hours. the process of selecting the model (a compiled model or an ASCII model) is exactly the same as it would be when modeling. 410 . for data. Setting up the model The kinetics will be modeled by a two-compartment open model with firstorder input. To enter all of the times quickly. Use Data>Column Properties in the menus. the second. 3. A new workbook appears with two worksheets: the first. 4. 5. To create the time value data: WinNonlin 1. Hold the left mouse button down. Next. Press the Enter key. When simulating. to name Column A: Time. in increments of 0. 6.16 User’s Guide Creating the input data Simulation requires a data set with X values defining the time points over which to simulate Y variable values. This example uses a data set with times ranging from 0 to 36 hours. The formula will be copied. select the Workbook radio button and click OK.18 in each row. 411 . Choose Model>Data Variables from the menus. To set model variables: 1. Choose Tools>PK/PD/NCA Analysis Wizard from the menus or click the PK/PD/NCA Analysis Wizard tool bar button. Click Next. 3. Check the Simulate Data check box.Simulation Simulation of a compiled library model 16 To select the PK model: 1. Select Pharmacokinetic in the Model Types dialog. click Finish to load the model. Drag Time to the X Variable box. The time variable must be identified. 2. In the final dialog. 2. Data variables The Simulate Data check box in the Data Variables dialog selects simulation rather than model fitting. and confirm that the Compiled check box is selected. 4. Select Model 11. Click Next. 3. Click Apply (leave the dialog box open). Model parameters The simulation requires parameter values. to calculate output. K21 = 1. Select Model>Model Parameters from the menus. Click OK. 2. 412 .16 User’s Guide WinNonlin 4.2. as constants. K01 = 5. K10 = 0.5. To enter parameter values: 1. 3. Enter the following parameter values into the grid as shown below: V_F = 10. K12 = 1. To enter the dosing information: 1.Simulation Simulation of a compiled library model 16 Dosing regimen This example will dose 150 mg every 24 hours. Enter mg in the Current Units field. 4. Dosing Constant Number of doses Dose #1 Time of dose #1 Dose #2 Time of dose #2 Value 2 150 0 150 24 3. Enter the values shown in the table below. 2. Click OK. 413 . Open the Dosing Regimen tab. and uncheck the Transpose Final Parameter Table check box. The simulated response data appear in the Summary Table worksheet of the PK Workbook. Select the PK Settings tab. Click the Start Modeling button or choose Model>Start Modeling from the menus. Click OK when finished. Running the simulation To start the simulation: 1. 414 .16 User’s Guide Model options To specify model options: WinNonlin 1. Click the Model Options tool bar button or choose Model>Model Options from the menus. and in the PK Chart output. 2. keeping the total daily dose of 150. For example. 415 . but this time administering 75 mg every 12 hours. Only the Dosing Regimen need be edited. and Dosing Regimen. Lower values indicate better estimates. Note: The secondary parameter estimates in simulation output are based on the first dose. WinNonlin remembers the values for the Data Variables. Model Parameters. Comparing dosing schemes Simulations provide an easy means of modifying a scenario to see the impact on concentrations or effects. re-run the prior simulation. After the previous simulation (“Simulation of a compiled library model” on page 409).Simulation Comparing dosing schemes 16 The Variance Inflation Factor (VIF) is useful in determining optimal sampling schedule for parameter estimation. Select Model>Dosing Regimen from the menus. 12 and 24. Choose Model>Start Modeling to re-run the simulation. 2.16 User’s Guide Dosing regimen For this new simulation. use doses of 75 mg at times 0. 416 . 3. To enter dosing information: WinNonlin 1. Enter the new doses and click OK. users can devise script files. to customize WinNonlin and run a number of WinNonlin operations in batch mode. Using the scripting language. This section includes: • • • • • • “Terminology” on page 417: object-based programming terminology “Writing and editing a script” on page 418 “Executing a script” on page 420 “Scripting object properties and methods” on page 421: a quick reference guide to properties and methods for each WinNonlin object and engine “Script file components” on page 426 “Processing multiple profiles in script files” on page 428 See “Scripting Commands” on page 579 for details on each scripting method and property. which are in essence computer programs.Chapter 17 Scripting WinNonlin’s object-oriented scripting language The WinNonlin scripting language is an object-based programming language for automating WinNonlin tasks and analyses. Terminology Object-oriented programming terms used in WinNonlin scripting follow.com to learn about WinNonlin automation software and services. Note: E-mail sales@pharsight. 417 . See the WinNonlin Examples Guide for example scripts. LINMIX.PSC) is an ASCII file that contains WinNonlin scripting code. For example. They can be opened in this version of WinNonlin. and WinNonlin itself. TABLE. The WinNonlin engines are: PK. each type of “child window” that appears in the main WinNonlin parent window is an object.WSC. Script files are ASCII text files. Every object has properties that define how the object should appear and interact.WDO. There are five types of objects in WinNonlin: Data objects. A method is a set of computer code that performs functionality on a specific object or engine. The Method RUN causes an associated PK Model (Engine) to run the model. 418 . A Pharsight script file (*. Scripting terminology Objects Object-oriented programming languages deal with objects. In short. ANOVAa. Engines Method Property Script File a. the Method LOAD causes an associated data file (an Object) to be loaded. PLOT. A third property would define whether or not the named data file contains data headers (column names) as the first row. Note: Scripts written with early versions of WinNonlin have the extension *. a property of an output data object could contain a file name to identify that object. ANOVA is available for backward compatibility with older versions of WinNonlin. it invokes an action. but cannot be saved back to that file type. Graph objects.BIOEQ.17 User’s Guide WinNonlin Table 17-1. Only one Engine may be in use at any one time. In WinNonlin.PSC file extension. there could be many Objects of each type and several Engines. Properties also set values for objects or engines. DESC. For example. Within a script. Older script files will run but may require minor modifications to file types. MERGE and TOOLBOX. TRANSFORM. Another property could be the file type to be used when loading the object. New script files use the . Text objects. Writing and editing a script This section discusses the steps in writing and editing a script. LinMix or Bioequivalence Wizards provide newer and more powerful options than ANOVA. output must be saved to an appropriate file type—a .PWO for a workbook rather than a . They have the default extension of .PSC. In short. properties are attributes or characteristics of an object. For example. Engines execute tasks in WinNonlin. Model objects. Case does not matter for any of the Properties or Methods. 8. 6. Select the radio button for Text. Comments may be added to the Script file by adding lines to the script that begin with REMA. 4. O). From the File menu. The File Save dialog appears. select Save As from the File menu (Alt+F.PSC) in the List Files of Type box. Select Script Files (*. When done. Click Save.Scripting Writing and editing a script 17 To write a script: 1. When the script is complete.PSC) from the pull-down menu in the Save as type box. The New dialog appears. click Save. or click the Open button on the tool bar. Double click on the name of the script file to be edited. 6. 7. 2. Select Script Files (*. The File Open dialog appears.PSC. In the File name box. or REMARK. Select the appropriate drive and directory if necessary. Select Open from the File menu (Alt+F. 5. A text window appears. 9. 3. 5. Click OK. 4. click the text line where changes are needed. type the desired file name. 419 . To edit a script: 1. Type the content of the script. using the file extension . 3. N). then type in the changes. 2. A). The script file will appear in a text window. Select the appropriate drive and directory. All script files in the selected directory will be listed. In the text window. select New (Alt+F. Select Run. 4. no WinNonlin windows may be open. When executing from the Windows command line. Select the appropriate drive and directory and double click on the script file name or click on the script file name and click Open. When running a script from within WinNonlin.17 User’s Guide WinNonlin Executing a script Once a script is created. Script files may be executed either from within WinNonlin or from a Windows command line. 420 . type WINNONLN. 3. select Run Script (Alt+F. followed by the script file name. Close all WinNonlin windows. Within WinNonlin To execute a script while in WinNonlin: 1. The script file executes and the output appears in a WinNonlin output window. again including the full file path.PSC 5. 2. A Windows dialog will appear. R). From the File menu. 2. or WINNONLN. including the script. Exit WinNonlin if it is already running. 3.EXE along with the appropriate path. In the Open box. The Run Script dialog appears. the following steps are used to run it. Click OK. For example: C:\WINNONLN\WINNONLN C:\WINNONLN\EXAMPLES\EXMPL_1. Click the Windows Start button. WinNonlin must be closed. From the command line To execute a script from the command line in Windows NT or Windows 98/2000: 1. Text. Table 17-2. If no errors occur during a script file execution. A text window will appear showing the contents of that log file.EXE. A quick reference table of applicable properties and methods is given below for each Object. a log file is created. Model and Graph objects. the file extension . Objects There are five WinNonlin objects: the WinNonlin. Scripting object properties and methods This section lists all properties and methods that are applicable to each WinNonlin object and engine. Refer to this Quick Reference Guide to see which properties and methods are used for each operation.Scripting Scripting object properties and methods 17 Note: When a script file cannot be successfully executed because of an error or errors in the file. To do so. The Pharsight application log will contain appropriate entries for actions taken during the script file execution (exclusions. etc.). From file manager or explorer To execute scripts from Windows File Manager or from Explorer: Script files can be run simply by double clicking on their names in Windows Explorer (Windows 98/2000 or Windows NT). Detail on each command is provided under “Scripting Commands” on page 579. See the WinNonlin Examples Guide for examples of their usage. Data. a log file will not be created. WinNonlin object properties and methods Arrange Cascade CloseAll DeleteFiles ExportAll FileMask Load MsgBox PrintAll PrintData PrintGraph PrintModel Save TileHorizontal TileVertical Unload WorkingDir WindowState 421 .PSC must be associated with WINNONLN. 17 User’s Guide Table 17-2. Text object properties and methods Filename FileType Load Print Save Unload WindowState Table 17-5. WinNonlin object properties and methods (continued) Halt PrintText WinNonlin Table 17-3. Model object properties and methods DoseUnit Filename FileType 422 LinkFile Load ParmFile Print Save Unload . Data object properties and methods Append AppendFilename AppendFileType AppendSheet AutoFileOptions CaseSensitive Column ConseqDelim Copy Cut Delimiter EndCol EndRow Exclude Filename FileType Find Font FontSize Headers Include Load Missing MultiGrid Operation Paste Print Remove RemoveCol RemoveRow RenameCol Replace Save SearchArea Sort SortVar( ) SortVars StartCol StartRow Tab Tolerance Unload Unit WindowState With Table 17-4. Model object properties and methods (continued) LamzFile PartFile WindowState Table 17-6. Descriptive statistics engine properties and methods Confidence InData OutData Percentiles SortVars SummaryVar( ) 423 . PK engine properties and methods OutData OutGraph OutText Run RunWordExport WordExportFilename Table 17-8. Graph object properties and methods Filename FileType FootnoteFont FootnoteFontSize LegendFont LegendFontSize Load LoadTemplate MoveFirst MoveLast MoveNext MovePrevious MultiGraph Print PrintMulti PrintMultiAcross PrintMultiDown Save SaveAll SaveAllPrefix SaveTemplate SortLevel TitleFont TitleFontSize Uniform Unload WindowState XLabelsFont XLabelsFontSize XTitleFont XTitleFontSize XUnits YLabelsFont YLabelsFontSize YTitleFont YTitleFontSize YUnits Engines The tables below provide a quick reference to the properties and methods applicable to each WinNonlin engine.Scripting Scripting object properties and methods 17 Table 17-5. Table 17-7. See “Scripting Commands” on page 579 for details on individual properties and methods. ) GroupVars( ) InData( ) Legend( ) NumData OutGraph Run SameErrCol( ) ShowUnits SortVar( . Table 17-10. ANOVA engine properties and methodsa Filename FileType OutData OutText Run a. ) JoinGroupVars JoinIDVar( . The former ANOVA functions are now handled by the Linear Mixed Effects Modeling engine. ) SortVars( ) Title UpErr( ) XTitle XUnits XVar( ) YTitle YUnits YVar Table 17-11. ) JoinCrossVars JoinGroupVar( . Table engine properties and methods AggregateVar CrossVar( ) CrossVars DefinitionFile Filename GroupVar( ) GroupVarBreak( ) GroupVars 424 IDVars InData( ) JoinCrossVar( . Scripts using the ANOVA engine will continue to work. ) JoinIDVars OutData PageFooter PageHeader RegVar( ) RegVars Run Stats TableNum .17 User’s Guide Table 17-8. Descriptive statistics engine properties and methods (continued) Interval Run SortVar( ) SummaryVars Weight WinNonlin Table 17-9. but it is best that new scripts use the LinMix engine. Plot engine properties and methods AutoSort DnErr( ) ErrType GroupVar( . Transform engine properties and methods DestArea DestCol Extra Function InData Run SourceCol Table 17-13.function assignments Toolbox. Merge engine properties and methods CarryData IncludeVar( . ) IncludeVars( ) a. Toolbox engine properties and methodsa Toolbox. ) SortVars Table 17-14. Table engine properties and methods (continued) IDVar( ) NumData Table 17-12. LinMix and Bioequivalence engine properties and methods Filename FileType OutData OutText Run Table 17-15.Function=Crossover Applicable properties and methods InData OutData Run Response SortVar( ) SortVars ShowUnits Subject TreatmentA TreatmentB Stacked 425 .Scripting Scripting object properties and methods 17 Table 17-11. Must be equal to 2. InData( ) NumDataa OutData Run SortVar( . Function can be assigned one of four values (above) to select the analysis tool.2. operations are specified via files that are prepared in advance.Function=Superposition Toolbox.1. other. more complex. Toolbox engine properties and methodsa (continued) Toolbox.1. 426 .Function=Deconvolution a.17 User’s Guide Table 17-15. Valid properties are listed for each Function property assignment. but the deconvolution output may differ due to enhancements made in version 5.function assignments Toolbox. Script file components While some operations are specified in Script files completely through the use of the scripting language commands described in this chapter. Note: Deconvolution commands from scripts written prior to WinNonlin version 5. This section describes how each engine is implemented in the scripting language.Function=Semicompartmental Applicable properties and methods ConcCol Ecol InData Keo OutData ConcCol DoseUnit InData InitialDose MaintenanceDose Npoints OutData OutGraph AlphaTerms() Aterms() Bolus ConcCol DoseUnit Delta InputLag OutGraph Run SortVar( ) SortVars TimeCol Run SortVar( ) SortVars Tau TerminalLower TerminalPhase TerminalUpper TimeCol TimeRepeat Npoints NumTerms Run Smoothing SortVar( ) SortVars TimeCol WinNonlin Toolbox.1 will function properly in WinNonlin 5. The set of properties available to the Toolbox engine depends on the value assigned to the Function property. The Load and Save operations on this Data Object then rely on a PKS file pointed to by the Data Object.LML).CMD) from prior versions of WinNonlin can also be loaded. that data file will not be available for any other processing because it will not have an alias assigned to it. that data file will not be available for any other processing. for each sort level. PKS Access to data in the Pharsight Knowledgebase Server (PKS) is possible using scripting in WinNonlin. Compartmental models are specified in scripts via Pharsight Model Objects (*. it will be necessary to load it separately. and linear regression can be included in scripts using a linear mixed effects command file (*.Scripting Script file components 17 Table 17-16. that data file is not available for any other processing because it does not have an alias assigned to it. and individual bioequivalence models. Charts are specified in scripts via scripting language commands. The file format is XML and is defined by an embedded DTD. To use Comtal analysisa mand files with different Lambda Z ranges.PMO). Descriptive statistics LinMIx Merging data sets Noncompartmen Noncompartmental models are specified in scripts via Pharsight Model Objects (*. Any combination of data from two data sets can be merged and saved as one data set. Scripting the PKS is done by adding an additional FileType property to the WinNonlin Data Object. ANCOVA. etc. for each sort level. it will be necessary to load it separately. To use Command files with different model parameters. see “Processing multiple profiles in script files” on page 428. Script file operations and methods of implementation Engine Bioequivalence Charts Compartmental models Method of implementation Bioequivalence analysis. 427 .ANV) from previous versions of WinNonlin can also be loaded.ANV). To use that data set for any other purpose within the script. Note that although a workbook will be loaded and used for the modeling as a result of either the Command file or Pharsight Model Object.CMD) from prior versions of WinNonlin can also be loaded. see “Processing multiple profiles in script files” on page 428. To use that data set for another purpose. The merging of data sets are specified in Scripts via scripting language commands that use the merge engine properties and methods. including ANOVA. The PKS file is the same format as the files generated by the Study Definition and Study Mappings dialogs used in the creation of a study in the PKS. including average. Descriptive Statistics are specified in Scripts via scripting language commands that address the descriptive statistics engine. Command files (*. Note: ANOVA command files (*. Linear mixed effects modeling. it will be necessary to load it separately. can be specified in scripts using a bioequivalence command file (*. Note that although a data grid will be loaded and used for the modeling as a result of loading the ANOVA command file (*. The user specifies which columns are summary variables and a new output data grid is generated.. Command files (*.PMO). population.. To use that data set for any other purpose within the script.LML). Note that although a workbook will be loaded and used for the modeling as a result of either the Command file or Pharsight Model Object. etc. Also. It is possible to overwrite existing files in this way. or. with a separate variable for each of the Final Parameters in the transposed output. A script can call any existing table definition file. a. this option simply determines which format will appear in the worksheet entitled Final Parameters. see “Chart templates” on page 129 WinNonlin Transformations Data transformations are specified in Scripts via scripting language commands that access properties and methods of the transform engine. See “Transpose final parameter table” on page 206.PMO) . When writing scripts that will use the NCA Data output. Processing multiple profiles in script files Two options are available to process compartmental models and noncompartmental analyses in the Scripting Language: • 428 Loading a Pharsight model object (*. Note on saving files CAUTION: When saving files with the Scripting Language. Scripting language commands can specify: The Input data set(s) Table template number Variable mappings Summary Statistics (all or none) Page Break options Using only the Scripting Language commands loses some flexibility. WinNonlin does not check to see if that file name already exists. A table definition file contains all of the information needed to create the table except the input data set(s) to be used. For more information on table definition files. The NCA Data output can appear with the variables (columns) Parameter and Estimate (in addition to the sort and carry-along variables). or using a table definition file (*.TDF). default formatting is used. it is important to be aware of the form in which this outpwill appear.17 User’s Guide Table 17-16. but it is not possible to select a subset. Script file operations and methods of implementation (continued) Engine Tables Method of implementation Tables may be specified in Script files in two different ways: through the use of scripting language commands. Both transposed a. Summary statistics can be included. The Final Parameters worksheet will be transposed if NCATRANS is included in the NCA specification or if the user has set up “transposed output” as the WinNonlin default. PARMFILE for specifying parameter values – . The information contained in the files specified by these properties is expected to be in a very specific format. to load a command file. however.DOSEFILE for specifying dosing information – . Set MYMODEL = Model MYMODEL.OutData= “MYOUTDATA” REMA Text and graphics output objects may also be named at this time Pk. Command files. however. These ASCII files can be created and saved within WinNonlin or a text editor.Filetype= “ASCII” MYMODEL.PartFile= “Partinfo.PARTFILE for specifying partial area information The following example script files shows how a command file could be loaded. and the methods used. store only one set of values to be used for all sort levels.CMD) from an earlier version of WinNonlin When using a Pharsight model object.Dosefile= “Doseinfo.cmd” MYMODEL. It is possible. partial areas or lambda_z range for each sort level automatically.LAMZFILE for specifying lambda_z range information – . then load ASCII files that contain unique information for each sort level. The next section discusses the ASCII file formats that are expected when using the following methods with Model Objects. parameter values. – .LINKFILE for specifying PK/PD models that require two sets of parameter values – .Load MYMODEL. different values may be used for dosing. 429 .File Name= “Profiles.dat” Pk.Scripting Processing multiple profiles in script files 17 • Loading a Command file (*.dat” MYMODEL.Run Note: File references within scripts can use absolute or relative file paths. Note: These files may be created using a text editor or from within WinNonlin. 3rd 4th nProfiles nRows. UNITS for preferred parameter units in non-compartmental analysis. • • • • • • LAMZFILE names a file containing lambda_z range information (start and end times for computation of Lambda Z in noncompartmental analysis). PARTFILE for partial area information (start and end times for computation of partial areas under the curve). WinNonlin The general structure for all of the ASCII files is given.17 User’s Guide Special file formats This section discusses the required ASCII file formats when using the following methods with an ASCII user model or Model Object. General structure Order in file 1st 2nd Contents “CheckSum or Identifier” nType Comments Must be included on the first line of file. LINKFILE for PK/PD models that require two sets of parameter values. nRows & nCols are integers that indicate how many rows and columns of data are in the file. nType is between 1 to 5 as follows: 1 lamdbaz range information 2 dosing information 3 model parameter values 4 link parameter values 5 partial areas nProfiles is a value < 32767 that indicates how many profiles or subsets of values are included. DOSEFILE for dosing information. using the Save buttons on the appropriate tabs of the Model Properties or Model Options dialogs. nCols 430 . followed by annotated examples for several. PARMFILE for initial parameter values and bounds (optional). n = 4. of profiles). When ntype = 2 or 3 or 4. Model used: NCA Model 200. nProfiles = 3. the number of columns equals 2 + the maximum number of exclusions. Data values of each profile Note: When ntype = 1 or 5. 3. the number of rows equals nProfiles (the number of profiles). as the number of profiles (Subjects) used is 3. Set(s) of values for each nRow.Scripting Processing multiple profiles in script files 17 Order in file 5th(nColns+4)t h record Remaining records Contents n. if any. etc. respectively. so nCols = 4. Char Comments For each column. of constants) or NPAR (no. as the number of decimal places. nRow = nProfiles (no. Table 17-17. Char = N as the data type is numeric for the 1st column. When nType is 1. is 4. so nRows = 3. Lambda Z file structure Record 1st 2nd 3rd 4th Contents CheckSum or Required Identifier 1 3 3. Lambda Z range information (nType=1) Data Grid with Lambda Z Range Information for three profiles. 2. Subject 1 2 3 Start time 3 3 3 End time 18 18 18 Exclusions 4 3 5 ASCII File with Lambda_z Range information of the above three profiles. as 1 represents lambda_z range information. N . 4 nType = 1. 431 Comments 5th 4. When nType is 1. these values indicate the number of decimal places and data type (“A” for strings or “N” for numeric) of the data. Data values of column 1. nRow = NCON (no. of parameters). Numeric data in the 1st column (Start Time). Invalid exclusions in the 3rd column. is 4. Numeric data in the 3rd column (Exclusions). is 4. Numeric data in the 1st column (Start Time). 2nd profile. 1st profile.17 User’s Guide Table 17-17. Numeric data in the 4th column (Exclusions). 2nd profile. WinNonlin Note: Values that are in the excluded part of the data grid but are not valid exclusions have a value of 0. 1st profile. N 3 18 4 5 3 18 3 0 3 18 -1e150 0 Comments n = 4. Char = N as the data type is numeric for the 2nd column. (See Note below). if any. Numeric data in the 2nd column (EndTime). n = 4. Invalid exclusions in the 4th column. 3rd profile. 2nd profile. Invalid exclusions in the 4th column. Lambda Z file structure (continued) Record 6th 7th 8th 9th 10th 11th 12th 13th 14th 15th 16th 17th 18th 19th Contents 4. Dosing information (nType = 2) Data Grid with Dosing Information for three profiles. Numeric data in the 1st column (Start Time). Model used: NCA Model 200. Numeric data in the 2nd column (EndTime). Numeric data in the 2nd column (EndTime). as the number of decimal places. as the number of decimal places. 1st profile. 1st profile. This does place the restriction that Time=0 may not be an excluded data point. if any. Char = N as the data type is numeric for the 3rd column. (See Note below). Subject 1 1 Constant Dose Time of Last Dose Value 100 0 432 . 3rd profile. N 4. 3rd profile. Numeric data in the 3rd column (Exclusions). (See Note below). 3rd profile. 2nd profile. nProfiles = 3. as the number of decimal places. Null data in the 3rd column. 1 nType = 2. so nCols = 1. Numeric data in the 1st column (Value). Table 17-18. N 100 0 0 100 0 0 100 0 0 a. the number of columns equals 1. 3rd profile.Scripting Processing multiple profiles in script files 17 Subject 2 2 3 3 Constant Dose Time of Last Dose Dose Time of Last Dose Value 100 0 100 0 ASCII File with dosing information of the above three profiles. Null data in the 3rd column. 433 . Char = N as the data type is numeric for the column. Can be used to save Identifier dosing units. the number of rows equals NCON (the number of constants). the format for the first line becomes V32. Dosing file structure Record 1st 2nd 3rd 4th Contents Comments CheckSum or Was not used before WinNonlin version 3.2. 1st profile. The commas must be included. is 4.DAT files. sdoseunit. 1st profile. 2 3 3. 1st profile. 2nd profile. 2nd profile. See note belowa. so nRows = 3. ]. (See Note below on nRows). Null data in the 2nd column. When nType is 2. Numeric data in the 1st column (Value). as the number of profiles (Subjects) used is 3. Null data in the 2nd column. It is possible to save dosing units and dose normalization information to ASCII *. even without the units [hence it could be V32. n = 4. To do so. Null data in the 3rd column. When nType is 2. Numeric data in the 1st column (Value). if any. sdosenormalization. 3rd profile. 3rd profile. 2nd profile. as 2 represents dosing information. Null data in the 2nd column. 5th 6th 7th 8th 9th 10th 11th 12th 13th 14th 4.. DoseFile method with NCA models. so nRows = 3. Model parameter file structure Record 1st 2nd 3rd 4th Contents CheckSum or Identifier 3 2 3. Model used: PK Model 3 Subject 1 1 1 2 2 2 Parameter Volume_F K01 K10 Volume_F K01 K10 Initial value 0. When steady state is being calculated. as the number of decimal places. as the number of decimal places. as 3 represents parameter values. these values should be set to 0. is 4. When nType is 3. N 434 . n = 4. (See Note below on nCol). so nCols = 3. the number of columns equals 3. Table 17-19. When nType is 3. n = 4. N 4. if any.5 Lower 0 0 0 0 0 0 Upper 0 0 0 0 0 0 WinNonlin ASCII File with parameter values of the above two profiles.15 33 1.17 User’s Guide Note: When using the Model. Model parameter values (nType = 3) Data grid with parameter values on two profiles. Char = N as the data type is numeric for the 1st column (Initial Value). Char = N as the data type is numeric for the 2nd column (Lower Limit). nRows is equal to NCON+2 to accommodate the Tau values. If steady state is not being calculated. if any. 3 Unused nType = 3. as the number of profiles (Subjects) used is 2. is 4.2 20 2 0. Comments 5th 6th 4. the number of rows equals NPAR (the number of parameters). nProfiles = 2. nRows is equal to NCON+1 in order to accommodate the Time of Last Dose values. 2nd profile. Null data of the K10 parameter in the 2nd column (Lower Limit). 2nd profile. 1st profile.Scripting Processing multiple profiles in script files 17 Table 17-19. Numeric data of the K01 parameter in the 1st column (Initial Value). Null data of the K01 parameter in the 2nd column (Lower Limit). 1st profile. Null data of the Volume/F parameter in the 3rd column (Upper Limit). Null data of the K01 parameter in the 2nd column. Model parameter file structure (continued) Record 7th 8th 9th 10th 11th 12th 13th 14th 15th 16th 17th 18th 19th 20th 21st 22nd 23rd 24th 25th Contents 4.2 0 0 20 0 0 2 0 0 . Null data of the K10 parameter in the 3rd column. Numeric data of the K01 parameter in the 1st column. 1st profile. Numeric data of the Volume/F parameter in the 1st column (Initial Value).5 0 0 Comments n = 4. 2nd profile. Char = N as the data type is numeric for the 3rd column (Upper Limit). Null data of the Volume/F parameter in the 3rd column (Upper Limit). 2nd profile. Null data of the K01 parameter in the 3rd column (Upper Limit). is 4. Null data of the K10 parameter in the 3rd column (Upper Limit). 435 . 1st profile. Null data of the Volume/F parameter in the 2nd column (Lower Limit). 1st profile. 1st profile. Numeric data of the K10 parameter in the 1st column (Initial Value). 2nd profile. 2nd profile.15 0 0 33 0 0 1. 1st profile. Numeric data of the Volume/F parameter in the 1st column (Initial Value). 2nd profile. Null data of the K10 parameter in the 2nd column. N 0. 1st profile. 1st profile. as the number of decimal places. Null data of the Volume/F parameter in the 2nd column (Lower Limit). Numeric data of the K10 parameter in the 1st column. 2nd profile. Null data of the K01 parameter in the 3rd column. if any. 2nd profile. so nRows = 3. so nCols = 3. is 4. N 436 . if any. When nType is 4. Data grids with link parameter values on two profiles. the number of rows equals NPAR (the number of parameters). When nType is 4. as the number of profiles (Subjects) used is 2.5 Lower 0 0 0 0 0 0 Upper 0 0 0 0 0 0 WinNonlin ASCII File with parameter values of the above two profiles. as the number of decimal places. (See Note below on nCol). 3 nType = 4. nCol3 = Upper Limit. but must still be part of the file. See note belowa. n = 4. as 4 represents Link parameter values. the Linkfile is included. the number of columns equals 3.15 33 1. Table 17-20. nCol2 = Lower Limit. Char = N as the data type is numeric for the 1st column (Initial Value). Link PK parameter values Record 1st 2nd 3rd 4th Contents Comments CheckSum or Was not used before WinNonlin version 3. 4 2 3.17 User’s Guide Note: The three columns: nCol1 = Initial Value. are not always used.2. Can be used to Identifier save PK concentration units. Link parameter values (nType = 4) Since a PK/PD model actually requires two sets of parameter values. nProfiles = 2. Models used: PK Model 3/PD Model 105 The following data grid is applied to the PK Model: Subject 1 1 1 2 2 2 Parameter Volume_F K01 K10 Volume_F K01 K10 Initial Value 0. 5th 4.2 20 2 0. N 4. Numeric data of the Volume/F parameter in the 1st column (Initial Value). The comma must be included. 1st profile. 1st profile. The following data grid is applied to the PD Model: Subject 1 1 1 2 2 2 Parameter Emax Ece50 Gamma Emax Ece50 Gamma Initial value 5 6 7 1 2 3 Lower 0 0 0 0 0 0 Upper 0 0 0 0 0 0 ASCII File with parameter values of the above two profiles. n = 4. is 4. Link PK parameter values (continued) Record 6th 7th 8th 9th 10th Contents 4. Null data of the Volume/F parameter in the 2nd column (Lower Limit). as the number of decimal places. The remainder of the file structure is the same as the file shown in Example 3.Scripting Processing multiple profiles in script files 17 Table 17-20. It is possible to save PK concentration units for the PK tab of PK/PD models to ASCII .PKconcentrationunit. 1st profile. is 4. N 0. Table 17-21. if any. Char = N as the data type is numeric for the 2nd column (Lower Limit). Null data of the Volume/F parameter in the 3rd column (Upper Limit). the format for the first line becomes V32. a.2 0 0 Comments n = 4. PD parameter values Record 1st Contents CheckSum or Identifier Unused Comments 437 .DAT files. as the number of decimal places. To do so.). if any. Char = N as the data type is numeric for the 3rd column (Upper Limit). even without the units (hence it could be V32. Null data. 1st profile. as 4 represents Link parameter values. 1st profile. 2nd profile. Null data. Null data. N 5 0 0 6 0 0 7 0 0 1 0 0 2 0 438 . if any. Numeric data. N 4. 2nd profile. so nCols = 3. WinNonlin 5th 6th 7th 8th 9th 10th 11th 12th 13th 14th 15th 16th 17th 18th 19th 20th 21st 4. nProfiles = 2. Null data. Emax parameter in the 3rd column (Upper Limit). Gamma parameter in the 3rd column. as the number of decimal places. if any. Ece50 parameter in the 3rd column. (See Note below on nCol. the number of rows equals NPAR (the number of parameters). 2nd profile. is 4. n = 4. Ece50 parameter in the 2nd column. Numeric data. N 4.) n = 4. if any. 1st profile. 3 Comments nType = 4. Null data. When nType is 4. Null data. the number of columns equals 3. Emax parameter in the 2nd column (Lower Limit). PD parameter values (continued) Record 2nd 3rd 4th Contents 4 2 3. as the number of decimal places. Numeric data. Numeric data. so nRows = 3.17 User’s Guide Table 17-21. Null data. Char = N as the data type is numeric for the 1st column (Initial Value). 1st profile. Gamma parameter in the 1st column. 1st profile. 1st profile. Emax parameter in the 1st column (Initial Value). Ece50 parameter in the 2nd column. Null data. Ece50 parameter in the 1st column. as the number of decimal places. Emax parameter in the 3rd column(Upper Limit). 1st profile. Char = N as the data type is numeric for the 3rd column (Upper Limit). 2nd profile. Char = N as the data type is numeric for the 2nd column (Lower Limit). 1st profile. When nType is 4. Emax parameter in the 2nd column (Lower Limit). Emax parameter in the 1st column (Initial Value). as the number of profiles (Subjects) used is 2. is 4. Null data. 2nd profile. Numeric data. Ece50 parameter in the 1st column. n = 4. Gamma parameter in the 2nd column. is 4. 1st profile. 2 nType = 5. Null data. When nType is 5. if any. N . are not always used. nProfiles = 3.25 . Char = N as the data type is numeric for the 1st column (AUC1 Lower). Numeric data. Model used: NCA Model 200 Subject 1 2 3 Min Time . Partial areas (nType = 5) Data grid with partial area information for three profiles. is 4. as 5 represents Partial Areas. Note: The three columns: nCol1 = Initial Value. 2nd profile. 2nd profile. but must still be part of the file. the number of rows equals nProfiles. 439 Comments 5th 4. as the number of decimal places. so nCols = 2. so nRows = 3. n = 4. PD parameter values (continued) Record 22nd 23rd 24th 25th Contents 0 3 0 0 Comments Null data. 2nd profile. Gamma parameter in the 3rd column. the number of columns equals the maximum number of partial areas multiplied by 2. When nType is 5. Partial areas file format Record 1st 2nd 3rd 4th Contents CheckSum or Unused Identifier 5 3 3.Scripting Processing multiple profiles in script files 17 Table 17-21. 2nd profile. Null data. Gamma parameter in the 1st column. nCol2 = Lower Limit. Gamma parameter in the 2nd column. nCol3 = Upper Limit.25 Max Time 48 48 48 AUC1 Lower 5 8 2 AUC1 Upper 6 12 3 ASCII File with partial areas information of the above three profiles. Ece50 parameter in the 3rd column. Table 17-22.25 . as the number of profiles (Subjects) used is 3. Scripting the Pharsight Knowledgebase Server The Pharsight Knowledgebase Server (PKS). WinNonlin Note: Values that are in the partial area portion of the data grid. See the WinNonlin Examples Guide for examples of scripts to run WinNonlin/ PKS operations. can be accessed by scripting from WinNonlin. Numeric data of the column AUC1 Upper of the 1st profile. Numeric data of the column AUC1 Upper of the 3rd profile. Numeric data of the column AUC1 Lower of the 2nd profile.17 User’s Guide Table 17-22. is 4. The workflow rules associated with the manipulation of PKS data via Scripting can be encapsulated based on the method being invoked (i. . Char = N as the data type is numeric for the 2nd column (AUC1 Upper). This PKS file is the same format as the files generated by the Study Definition and Study Mappings dialogs used in the creation of a study in the PKS.e. Load Note: 440 Rules follow the user interface. but are not valid partial areas. See “PKS file format” on page 441 for details. N 5 6 8 12 2 3 Comments n = 4. Numeric data of the column AUC1 Upper of the 2nd profile. The Load and Save operations on this Data Object will rely on a PKS file pointed to by the Data Object. if any. Load or Save). and the files stored as studies and scenarios in it. Partial areas file format (continued) Record 6th 7th 8th 9th 10th 11th 12th Contents 4. Numeric data of the column AUC1 Lower of the 1st profile. Scripting the PKS is done by adding an additional FileType property to the Data Object. The file format is XML and is defined by an embedded DTD. as the number of decimal places. Numeric data of the column AUC1 Lower of the 3rd profile. should be given lower and upper values of 0. then only a study will be loaded. If a PKS session has been created via loading a study only and no other Objects (i. then the Save Method will create a study in the PKS based on the information in the PKS file pointed to by the File Name Property of the Data Object. Save Note: A more involved set of rules based on the state of the application. If a PKS session has been created via loading a study and scenario or a study only and other Objects (i.e. workbooks) are present in WinNonlin. It is saved as an XML file with the following format. If the PKS file pointed to by the Data Object has a study name and a scenario name. If the PKS file pointed to by the Data Object has only a study name. 2. <?xml version="1. charts. then the Save Method will update the Study (if it changed) and Save the scenario to a new version or an entirely new scenario. texts. workbooks) are present in WinNonlin. no connection has been made to the PKS).e.Scripting Scripting the Pharsight Knowledgebase Server 17 1. charts. 2. then the Save Method will update the Study only.Mappings?)> <!ATTLIST PWRImport UserID CDATA #IMPLIED Password CDATA #IMPLIED Server CDATA #IMPLIED 441 . 3. Scripting will not load previous versions of a scenario. PKS file format The PKS file contains XML that describes load and/or save operations to and/ or from the PKS.0"?> <!DOCTYPE PWRImport [ <!ELEMENT PWRImport (StudyInformation?. texts. A new version will be created if and only if the scenario name being saved is the same as the scenario that was loaded. then both the study and latest version of the scenario will be loaded. Scenario? . If no PKS session has been created (i. 1.e. Mapping*)> <!ATTLIST Mappings Type (SAMPLE|DOSE|BOTH) #IMPLIED > <!ELEMENT Sources EMPTY> <!ATTLIST Sources Observation CDATA #IMPLIED Dose CDATA #IMPLIED Demographics CDATA #IMPLIED > 442 WinNonlin .17 User’s Guide Port CDATA #IMPLIED Root CDATA #IMPLIED > <!ELEMENT StudyInformation EMPTY> <!ATTLIST StudyInformation BlindingType CDATA #IMPLIED Compound CDATA #IMPLIED Description CDATA #IMPLIED Design CDATA #IMPLIED EndTime CDATA #IMPLIED Indication CDATA #IMPLIED Name CDATA #IMPLIED Portfolio CDATA #IMPLIED Project CDATA #IMPLIED StartTime CDATA #IMPLIED State CDATA #IMPLIED Type CDATA #IMPLIED Reason CDATA #IMPLIED > <!ELEMENT Scenario EMPTY> <!ATTLIST Scenario Name CDATA #IMPLIED Reason CDATA #IMPLIED GetLatest (TRUE|FALSE) #IMPLIED > <!ELEMENT Mappings (Sources?. Scripting Scripting the Pharsight Knowledgebase Server 17 <!ELEMENT Mapping (FromFields*. FieldOption*)> <!ATTLIST Mapping FromField CDATA #IMPLIED ToField CDATA #IMPLIED FieldType (Subject_ID|Subject_Data|Observation|DCP_Description|Actual_Clock_Time|Relative_Nomina l_Time|Sample_Number|Relative_Actual_Time|Relative_Ac tual_End_Time|Treatment|Period|Phase|Undefined|Dose) #IMPLIED TreatAsDCP (TRUE|FALSE) #IMPLIED > <!ELEMENT FromFields EMPTY> <!ATTLIST FromFields Name CDATA #IMPLIED > <!ELEMENT FieldOption EMPTY> <!ATTLIST FieldOption OptionName (Unit|Sample_Description|Sample_Matrix|Analytical_Metho d|Status|LLOQ|ULOQ) #IMPLIED IsColumn (TRUE|FALSE) #IMPLIED Value CDATA #IMPLIED > ] > 443 . 17 User’s Guide WinNonlin 444 . These functions and operations are covered in the Pharsight Knowledgebase Server’s documentation. This chapter covers WinNonlin/PKS interactions under the following headings. scripts and results among the members of a drug development team.Chapter 18 Using the Pharsight Knowledgebase Server Data access in a secure environment The Pharsight Knowledgebase Server (PKS) is a database system for secure storage of study data and analyses.S. The combination also supports compliance with the U. • • • • • • • • • • “The PKS information structure” on page 446 “Dosing” on page 448 “Accessing the PKS from WinNonlin” on page 449 “Creating a study” on page 452 “Loading a study or scenario” on page 463 “Creating scenarios” on page 465 “Optimizing PKS/WNL performance” on page 467 “Change tracking and audit information” on page 470 “Dosing information” on page 472 “Appending or updating a study” on page 474 445 . which provides additional operations. The combination of the PKS and WinNonlin Enterprise provides an effective means to share source data. This chapter covers the use of WinNonlin as a PKS client: loading PKS data to WinNonlin and saving data from WinNonlin as PKS studies and scenarios. Food and Drug Administrations’s 21 CFR Part 11 regulation. The PKS can also be accessed by way of a web interface. and an XML API. models. The PKS provides a secure data management system with complete auditing capabilities. . a study must include at least one column containing subject identifiers. each data collection point can have multiple observations. baseline covariates. subject-specific information such as body weight or gender. last name). Analysis settings and output are added to the study as scenarios. such as analysis code files. measurements of blood pressure. heart rate. A study may include (but is not required to have) multiple columns of subject data—time-invariant. a sampling variable is used to differentiate the samples. A collection of fields taken together can be used for subject identification (e. PKS saves observations and related data as a PKS study. Observation data may include records of concentration. including model settings and results. observation times and subject identifiers. Non-WinNonlin documents. concurrent measurements for any subject (e. A study starts with a collection of raw data that must include longitudinal observations. etc. can be loaded into the PKS and associated with a scenario using WinNonlin’s “Other documents” feature. chart. Those objects fit into specific niches in the PKS data structure. first name. multiple assays performed on a given sample or multiple sam- 446 . i. In this case. As analyses are performed on these data.g. can be added as scenario(s) within the study. As the PKS manuals explain. First. reports and graphics that are related to the study. If the observation data include multiple. the analyses.18 User’s Guide • • • “PKS libraries and object shares” on page 475 “PKS Information” on page 479 “WinNonlin and PKS differences” on page 480 WinNonlin See also “Scripting the Pharsight Knowledgebase Server” on page 440.e. The PKS information structure WinNonlin is an object (text. A study may include time-dependent observation and dosing data. model.. A PKS study requires specific fields.. and may include dosing and subject demographics.g. data) based application. Studies The basic unit of information in the PKS is a study. Project. the data set must include a variable that differentiates the measurements (e. a model with model parameters. a study must include relative nominal time. This metadata provides ways to search and filter the data within the database. 2nd and 3rd samples). If only the Relative Nominal Time is defined in the data set. although it is possible to create further scenarios from the study (more children) or from an existing scenario (grandchildren. 3 for the 1st. This is in contrast to relative actual time. a duplicate column will be created for the actual times. are children of) a particular data set (study). Portfolio. the relative nominal end time must be identified as well. and/or statistical tests to be applied to a particular data set for fitting or analysis. Indication. with values for each row (as placeholders). Note: To create a study that does include longitudinal observation data.Using the Pharsight Knowledgebase Server The PKS information structure 18 ples during a given visit). Compound. initial parameters. See “Creating a study” on page 452 and “Loading a study or scenario” on page 463. etc. Scenarios are naturally associated with (i. Study Type. Time fields Because a study generally includes time-dependent observation data. the combination of data.e. 447 . Scenario When WinNonlin analysis is performed on a PKS study. indicating the times at which observations (and doses. See also “Creating scenarios” on page 465.). Study metadata When a study is created. Scenarios are roughly comparable to the WinNonlin concept of a workspace in that it is possible to save. dosing. if included) were scheduled in the protocol. as well as any derived output from those tests or analysis. model. The metadata are saved with the study. in one place and under one name.) If the data uses infusion dosing.. information about the study can be entered in the Study Definition dialog (see below). 2. and (optional) results data can then be saved in the PKS as a scenario of the original study. and the analytical output. (Relative time is time elapsed since the start of the study.g. A scenario can include a structural model. related work— the data source. This dialog includes information about the study: Study Name.. period or phase. in the study. include a dummy time column. etc. other model settings. representing the times that observations or doses really happened. Study Design. values 1. 18 User’s Guide Other documents The PKS can store documents that are not related to WinNonlin. The columns representing subject identifiers and nominal time must be named exactly as the corresponding columns are in the workbook containing the observation data. The dosing information must appear in a different worksheet from that containing the study observation data. and its variables are assigned as part of mapping the study variables (see “Creating a study” on page 452). Dose data can be added to a study by one of three methods. though it is not required. The worksheet must contain. 448 . Dosing is generally saved as part of a PKS study. Users can select which dose columns to use for a given scenario. SAS or NONMEM code files. If multiple scenarios are included in a study. The dosing worksheet can contain columns for more than one analysis scenario. from a separate worksheet (not the worksheet containing observation data) Append dosing after study creation. as detailed under “Dosing information” on page 472. That worksheet is selected as the dose source during study creation. such as study reports. subject identifiers. Dosing worksheet It is possible to include dosing data when creating a study in the PKS. and relative nominal dose times. Sets of such documents can be associated with scenarios. at minimum. or a graphic created by saving a WinNonlin chart to a JPEG file. WinNonlin Dosing Dosing information is a special case of study data. See “Dosing information” on page 472 for details. using the WinNonlin Dosing Regimen dialog. each may use different dose data. It must contain the sort variables required to distinguish profiles. See “Adding non-WinNonlin documents to a scenario” on page 466 for instructions. from a dosing worksheet Dosing dialog: enter dosing as part of a modeling scenario. dose amounts. The PKS will use those columns to match observation data to dose data for given subject(s). • • • Dosing worksheet: load dosing during study creation. then dosing data can be entered via the WinNonlin Dosing Regimen or Dosing Data Variable dialog. and subsequently saved to a scenario. Dosing dialog When a PKS study or scenario with no dosing data is loaded in WinNonlin. Dosing entered for a simulation. The menu items include: Table 18-1. and a model is selected. resides with the scenario information. The PKS will use the subject and time columns to match dose data to records for specific subjects at specific times. model parameters and dosing information. The WinNonlin dosing dialog can still be used for simulations and for user models. at minimum. The worksheet must contain. if any Saves changes to the study. Thereafter. any changes to the dosing then must be done on the Dosing worksheet rather than through the WinNonlin Dosing Regimen. and loads a PKS study or scenario Saves changes to the study and creates a new version of the current scenario. and creates a new scenario Opens the mapping wizard to set up a new PKS study from the current workbook 449 . if any. PKS menu Menu item Load Save Save As Create Study Used when Always A PKS study is loaded A PKS study and scenario are loaded No PKS session is open and a workbook is loaded Function Closes the current PKS session. dose amounts. subject identifiers.Using the Pharsight Knowledgebase Server Accessing the PKS from WinNonlin 18 Append dosing To append dosing data to a PKS study. and relative nominal dose times. Accessing the PKS from WinNonlin Access to the Pharsight Knowledgebase Server occurs via the PKS menu in WinNonlin Enterprise. Use the PKS Append/Update feature to map the new data columns to PKS study fields. close all WinNonlin windows. then open a workbook containing the dose information in a single worksheet. When the scenario is saved to the PKS with model. the dosing is transferred to a Dosing worksheet inside the study workbook. if any. and does not become the study dosing data. any changes are documented for audit trail purposes. PKS menu (continued) Menu item Append/Update Study Dosing Add columns Used when No PKS session is open and a workbook is loaded Function Opens the mapping wizard. and other items to be used as standards across many studies and analyses Checks whether any currently-loaded library or share items are up to date with the most recent version in PKS Displays the PKS study identifier and scenario identifier. so that data columns from the current workbook can be added to an existing PKS study WinNonlin A PKS study or scenario is Sets dosing column((s) to be used in the loaded current model or analysis (scenario) A PKS study is loaded Loads additional study data columns from PKS into WinNonlin Opens the WinNonlin Document Manager. share link can be saved with a scenario Loads or creates a PKS system object. While the link between WinNonlin and the PKS is not constant. WinNonlin knows whether data came from the PKS. storing a binary file containing code. If data are then saved back to PKS (whether to an existing study or scenario or to a new one). to save or load non-WinNonlin documents to or from a PKS scenario Loads a non-WinNonlin document (“other document”) from any PKS scenario for use in the current analysis. if one is loaded Other documents Always Share Always Libraries Always Status A library or share item is loaded A PKS study is open Information PKS sessions The Enterprise edition of WinNonlin has a concept of a PKS session.18 User’s Guide Table 18-1. and tracks all changes. 450 . A PKS session begins when WinNonlin retrieves and uses data from the PKS. parameter names. For example.Using the Pharsight Knowledgebase Server Accessing the PKS from WinNonlin 18 Connecting to the PKS The PKS configuration must be set up before WinNonlin can contact the PKS. 5.. This will be used to select a specific PKS database instance when logging in. 3. and root directory for the PKS database under Root. port and root. choose PKS>Load. Enter a name describing the PKS database under PKS Alias. 4. The configuration settings are accessed from the PKS Logon dialog. enter the port number under Port. used to run examples from PKS Examples Guide. Select any available item in the PKS menu to open the PKS Logon dialog. This information need be entered once only. To configure the PKS connection: 1. Enter the PKS server machine name under Server. Click the Configure button to edit the PKS setup. For example. many sites have one production database and one example database. Similarly.. as more than one PKS database may be available on the local network. Click OK. 451 . Note: The system administrator who set up the PKS database should provide the values for the server. WinNonlin will save it. 2. and must contain the same data collection point identifiers. Choose PKS>Create Study from the WinNonlin menus. columns for subject identifiers. phase. etc. Dosing must be in a separate worksheet from observations.18 User’s Guide Logging in User name and password Each time data are loaded to or from the PKS. who must set up an account for each user to access the PKS (see the PKS User’s Guide for details). open those as well. Demographic data must be time-invariant and subject-specific. and indicators of period. A PKS study stores raw data to be modeled and analyzed in one or more scenarios. and can be loaded from the same worksheet as the observations.e. See “Connecting to the PKS” on page 451. To create a study: 1. Note: See “Dosing” on page 448 for methods to add dosing information to a study. at minimum. i. WinNonlin Creating a study Studies are the fundamental unit of PKS organization. subject identifiers. 452 . set up the PKS configuration. 2. Data requirements: Observation data must contain. See “Studies” on page 446 for further discussion. the user must log in and provide a password. sample values and nominal (scheduled) times. If logging in for the first time. time fields.. If dosing and/or demographics data are to be loaded. Open sample data in a WinNonlin workbook. The log in user name and password are provided by your PKS system administrator. WinNonlin will contact the PKS and open the Study definition dialog. These are not necessarily the same as those used to log onto you Windows network. e. You must have WRITE system access in order to create studies.Using the Pharsight Knowledgebase Server Creating a study 18 3. 4.) These attributes are called study metadata. Enter the user name and password supplied by your PKS administrator.g. etc. Select the desired PKS database under PKS Alias and click OK.: 05/05/2002 453 . year. must be unique within the PKS database Data entry Typed Format 50 character max Study start time Dates of first and last days in and end time the study Typed Month day. Table 18-2. (Required fields are marked with an asterisk in the Study Definition dialog. e. Study metadata fields Attribute Study namea Description Used to identify and locate study in PKS. 2002 or mm/dd/yyyy. Study definition Creating a study requires specification of the following information. indication. They are useful in database searches for studies on a given compound.: May 5. See your database administrator for local rules for metadata values.g. . Required field. Study Creation Wizard 454 .g.18 User’s Guide Table 18-2. Click OK to proceed to the Study Creation Wizard. parallel or crossover Typed or selected 100 character from list of prior values max Typed or selected 50 character max from list of prior values Typed or selected 30 character max from list of prior values Typed or selected 30 character max from list of prior values Typed or selected 30 character max from list of prior values Main drug compound(s) under Typed or selected 50 character max development from list of prior values Whether the study will be locked against changes or not Selected Normal or locked a. 1. Study metadata fields (continued) Attribute Description Portfolio Project Indication Blinding typea Study typea Study designa Compound Status Description Text describing the study. use Ctrl+Enter for line breaks Data entry Typed Format 2000 character max WinNonlin Development portfolio to which Typed or selected 50 character max the study belongs from list of prior values Development project to which the study belongs Primary indication(s) for the study blinding level for the study type of study Experimental structure. e. Note: The Save button saves all study creation settings. 2. or to assign units to a column: 455 . Click Next to proceed to the Subject identifiers. Subject ID attributes If units already exist for a given column in the WinNonlin worksheet. Select a variable under PKS Fields and click the Attributes button to set units for the subject ID variable(s). those units are loaded to the PKS. including file paths/names and variable mappings. The Load button reloads the settings from that file. Click Next to map Subject data. 3. 3. Select dosing and demographics worksheets (optional). Drag the variable(s) that identify individual subjects to the PK Fields box. as an XML file with a *.Using the Pharsight Knowledgebase Server Creating a study 18 1. Note: If dosing and/or demographics worksheets are used. 2. Select the worksheet for the observation data under Observation Worksheet. named identically. if needed. To view them. as detailed under “Subject data attributes” below. will appear in the Subject Identifiers list. only columns that appear in all worksheets. Subject identifiers 1.PKS file extension. Drag columns that contain subject demographics (baseline covariates) and other time-invariant. mapping subject-specific data as observations slows performance. To set one unit for an entire column. controlled list. 4. but if a unit is not defined in the PKS. 456 . Click the Hide button to return to the Subject identifiers. In particular. CAUTION: Improper variable mappings can slow PKS operations substantially. subject-specific columns to the PKS Fields box. drag that column to the Unit field. To enter a new name for a column.18 User’s Guide 1. as detailed under “Subject data attributes” below. 3. Time units must come from a fixed. See also “Optimizing PKS/WNL performance” on page 467 and “Performance Tips” in the PKS System Administrator’s Guide 2. Observations include one record for every data collection point. to be used in the PKS study. Subject data WinNonlin 1. The name may include up to 50 characters. rather than one record per subject. Other units can be any string. 2. unit conversions will not be available for it. Select the variable under PKS Fields and click the Attributes button to set units. select that column under PKS Fields and click the Rename button. Click Next to specify variables related to Time. and must follow WinNonlin column name limits. To load units for each record from a column in the data set. 3. check Static and enter or select the unit under Unit. Other units can be any string. Set other time variables (optional). drag that column to the Unit field. All time variables must exist with identical names in both the observations worksheet and the dosing worksheet. expressed relative to the start of the study. 2. or to assign units to a column: 1. See your system administrator for a list of allowed units. 3. or check Static and enter the unit under Time Unit. Relative Nominal Time: Drag the variable containing scheduled times. controlled list. Time 1. To set one unit for an entire column. check Static and enter or select the unit under Unit. Time unit: Either choose a column containing the units for each record. unit conversions will not be available for it. Click the Hide button to exit the Attributes dialog and return to the Subject data dialog. 457 . Time units must come from a fixed. 3. to this field. 4. To view them.Using the Pharsight Knowledgebase Server Creating a study 18 Subject data attributes If units already exist for a given column in the WinNonlin worksheet. See “Time fields” on page 447 for a discussion of the PKS time variables. but if a unit is not defined in the PKS. 2. if included. those units are loaded to the PKS. To load units for each record from a column in the data set. Click Next to map Sample data. Note that infusion dosing requires Relative Actual End Time. 3. 4. Click Next to set Dosing. Sample attributes All attributes are optional. If a column includes multiple. drag the variable that differentiates measurements to Sample Number.18 User’s Guide Sample data WinNonlin 1. Drag each column containing sample data to the PKS Fields box. 2. concurrent measurements for any given subject. Select a variable under PKS Fields and click the Attributes button to edit the Sample attributes. 458 . or missing due to dropout or non-adherence. – Analytical Method: assay or technique used to obtain the sample data. check Static and enter or select the value. e. To load attribute values for each record from a column in the data set. – Status: indicates data status. – Unit: units of measurement. or a unique value for each observation record (from a column in the data set). below a limit of quantification (BQL). Click the Hide button to exit the Attributes dialog and return to the Sample data.) 459 . drag that column to the appropriate field. 2. – LLOQ: lower limit of quantification for the assay. Dosing 1. the Map Dosing Data screen appears. as needed.. (See “Dosing attributes” below.” The list of statuses is defined by your PKS administrator.g. 3. A number value. A number value. To set one attribute value for an entire column. Select a dose variable under PKS Fields and click the Attributes button to set units and other optional attributes.Using the Pharsight Knowledgebase Server Creating a study 18 Each attribute can take one value for an entire column. Valid data are “finite. 1. Drag the column(s) containing dose amounts to the PKS Fields box. 2. If a dosing worksheet is used. – ULOQ: upper limit of quantification for the assay. 3. – Sample Description: further information about the sample. Each can take one value for an entire column. missing due to dropout or nonadherence.. or a unique value for each observation record (from a column in the data set). 2. – Status: indicates dose status. Valid data are “finite. Click Next to continue to the Data Collection Point (DCP) screen. Dosing attributes WinNonlin All attributes are optional. To load attribute values for each record from a column in the data set. – Dose Description: further text information about the sample. 1. 3. – State: further text information about the sample.” The list of statuses is defined by your PKS administrator. – Unit: units of measurement. 460 .g. check Static and enter or select the value. To set one attribute value for an entire column. e. – Analytical Method: assay or technique used to obtain the sample data. drag that column to the appropriate field. Click the Hide button to exit the Attributes dialog and return to the Dosing.18 User’s Guide 4. phase (optional) and other study descriptors to the appropriate boxes. 2. 3. an indicator variable noting whether a given record contains a measurement of plasma concentration or pain score. Click Next to set Treatment variables. for example.Using the Pharsight Knowledgebase Server Creating a study 18 Data Collection Point 1. Drag variables representing the study period (required for crossover designs only). DCP Description: (Data Collection Point description) use this field for variables that distinguish different measurements stored a single observation column. Treatment 461 . Click OK. 3. When all necessary data has been entered. Use Save to save mappings to a re-loadable file. Route: variable identifying route (oral.18 User’s Guide 1. and the Back button to make any adjustments. 462 . Fasted/fed: variable identifying whether subject ate before dosing. f. etc. 2. Once all data are successfully loaded to the PKS. a confirmation dialog will appear. c. The Study creation dialog appears. Regimen: variable identifying dose regimen. Treatment Code: variable identifying the treatment. Click Finish when the variable mappings are complete. 1. uncheck Reload Study. Study creation This dialog is the final step in study creation. State: additional descriptive field. click Next to save the PKS study. IV.) WinNonlin Note: The study name cannot be changed in this dialog. 3. Use the Preview button to review the variable mappings. IM. g. e.) for each dose. (See “Creating a study” on page 452. 2. Formulation: variable identifying drug formulation for each dose. Treatment description: additional descriptive field. causing study data to be saved to PKS. b. d. If the study does not need to remain open in WinNonlin for modeling or other further work. Drag variables to the following (optional) to add treatment descriptors per your company’s standards: a. 4. Use the Preferences button to organize the study tree by study name. the preferences select whether all scenarios will be displayed (“All”). 7. that is. Click OK. 3. PKS will close them before loading any study. Log into PKS as detailed under “Connecting to the PKS” on page 451. from the WinNonlin menus. The list is determined by your user access rights. Un-check the Read Only check box if any alterations or additions are to be made to the existing study or scenario (as opposed to creating a new scenario). 5. 463 . 2. WinNonlin will contact the PKS database and load a tree view of available studies and scenarios. Use the Search button to search by study attributes. or only the most-recent (“Latest”). and/or portfolio.Using the Pharsight Knowledgebase Server Loading a study or scenario 18 Loading a study or scenario To load a PKS study into WinNonlin: 1. In addition. if needed. Do not leave any windows open. See “Scenario search”. Note: A red scenario icon indicates that at least one derived output object is out of sync with its source. Choose PKS>Load. assigned by your PKS administrator. A green scenario icon means that the study data have changed since the scenario last loaded the study data. Loading the study as Read Only improves performance significantly. indication. the final leaf on each scenario branch in the study tree.. compound. 6.. Launch WinNonlin. compound and indication. scenario name and/or value of study metadata. Select dose data Select the columns to display in the dosing worksheet and click OK to load the data into WinNonlin. including portfolio. Select observations WinNonlin Drag and drop columns to include in the observation worksheet when the study is loaded into WinNonlin. and click OK. Scenario search The Search button available while Loading a study or scenario allows a search of available studies by study name. to select data columns to appear in observation and dose worksheets.18 User’s Guide 8. Select the study or scenario and click OK to load the scenario or. 464 . for studies. 3. choose PKS>Save from the menus. Load the model or analysis. If analysis results will be saved in the scenario.Using the Pharsight Knowledgebase Server Creating scenarios 18 Creating scenarios Each scenario contains analysis settings and results for one analysis. 4. it can be added with the scenario. creating a new version of the scenario. To save analysis results to PKS: 1. including any related settings. run the model or analysis. Once the analysis is complete. Each scenario can be updated. If the PKS study does not include dose data. 465 . Load the PKS study to analyze in WinNonlin as detailed under “Loading a study or scenario” on page 463. 2. A scenario can be derived from an existing study or from another scenario. in WinNonlin. . Click OK. The Scenario Field Selection dialog appears. Enter a scenario name and reason for the study (scenario) update. Adding non-WinNonlin documents to a scenario Non-WinNonlin documents can be saved to (or loaded from) a PKS with a scenario. Log in to PKS if required. choose PKS>Save from the WinNonlin menus. If any output windows were not saved and named. To save the updated scenario. among others. Microsoft Word reports created with the PKS Reporter. 466 . and viewed in an appropriate software program. saved to disk. Enter a scenario description (optional). Adding data columns to a scenario After a scenario is created. data sets from other analysis programs. or graphics created from WinNonlin charts.18 User’s Guide 5. See also “PKS share” on page 475 and “Libraries” on page 476. Any file can be attached to a scenario. but may not be viewable until extracted from the PKS. from the WinNonlin menus.. The new scenario will be saved with the original study. 7. Such documents may include graphics files. WinNonlin Drag study data columns from the left side to the Selected Fields box to add those data columns to the Observations worksheet for the scenario. PKS can display JPG files in the Web Administrator interface. you can add columns from the study to the scenario by loading the scenario and choosing PKS>Add Columns. a dialog will appear requesting names. 6. A final “Save complete” dialog will confirm scenario creation. Other documents can be attached. Click OK. The PKS Client for Word and the PKS Client for Excel can view XL and Word documents. Both are required. Optimizing PKS/WNL performance A number of issues impact the performance of operations between WinNonlin and the PKS. To commit modifications of the document manager list to the scenario in the PKS. 7. 467 . If the study data are loaded as read-only–allowing the user to analyze them in WinNonlin but not change data values in the source data–the loading can be done much more quickly. Read-only access When loading data. as part of the current scenario. Use the Add button to locate and save a file to the PKS. See “Loading a study or scenario” on page 463. Choose PKS>Other Documents from the WinNonlin menus to open the Document Manager. 4. 5. choose PKS>Save in the menus. 3. 2. WinNonlin and the PKS must do a significant amount of work to track any changes to the study data. Select a document in the list and click the Extract button to save it to a hard drive or network location. Select a document and click the Delete button to remove it from the scenario. Load the scenario. 6. Click OK when finished.Using the Pharsight Knowledgebase Server Optimizing PKS/WNL performance 18 To save or load a non-WinNonlin document as part of a PKS scenario: 1. or (4) Observation. the observation data are checked for any changes. WinNonlin and the PKS track any changes to the data. Appropriate field mappings Study data fields must be mapped to an appropriate data type when initially loaded from the WinNonlin client into the PKS. (2) Subject data. All scenarios are. Thus. In creating a study or scenario and loading/reloading.18 User’s Guide Note: The read-only check box on the PKS Load dialog applies to the study data and not the scenario. Subject data (unlimited number of fields) are data related to a subject that does not change over time. Subject identifiers (up to five fields) uniquely identify a subject. Make sure data that varies over time are mapped as observation data while demographic data are mapped as subject data. WinNonlin 468 . If baseline covariates are mapped as observations rather than subject data. While they can be changed (corrected). This data may also be called baseline covariates or demographics. the more they must be checked when loaded and the slower the loading is. read-only. while others (observation data) vary. Actual time differs from nominal time when subject adherence to the protocol schedule is imperfect. by their very nature. Changes are captured in an audit. the more that data are mapped as observation data. they are not normally modified in the course of a study and its analysis. Time data identify the nominal and actual time points for the observed data. Observation data (unlimited) include sample or dose data that change over time for a given subject. Some data types (such as subject identifiers and subject data) are constant across time. Nominal times are the dose and observation times designated in the study protocol. (3) Time (actual. Only new versions of scenarios can be created. When studies are loaded. the loading/reloading process becomes much slower because those data cells must be checked for changes. and combinations). The basic data types are (1) Subject identifier. nominal. Users must have system level rights and right to the original study in order to create a new scenario. This allows WinNonlin to load the study data in a more condensed format if read-only study data access is selected. 5.e. See “Loading a study or scenario” on page 463.Using the Pharsight Knowledgebase Server Optimizing PKS/WNL performance 18 Note: Mapping fields in the creation of a study is particularly important. 2.e.. Enter the password and click OK. Give the object a name (i. Load the study from the PKS into WinNonlin. 8.e. Save the base scenario via PKS>Save. create a base scenario that contains all the columns that will be used for a set of scenarios on a study. Click OK. Enter a reason (i. Selecting the correct data types can ensure performance is optimized. 7. and selecting Object Name. create a text object describing the base scenario (File>New then Text). Scenario Description). Example: 1. Modeling and other analysis can be done and saved as new scenarios using PKS>Save As. Click Save. All subsequent scenarios should be derived from this base scenario. The logon dialog appears. Creation of a base scenario for a study Loading a scenario is faster than loading a study. 469 . Base Scenario). to save time on subsequent operations. Now all work can be done by loading the base scenario with read-only access to the study. 3.. Once the study is loaded.. Name the Text object by right clicking on it. 6. this will be much faster than loading the source study. File>Close All (it may take some time to close all since a bit of memory is being freed up). An Update Reason dialog is displayed. It is not possible to change the data types in a study once it has been created. Highlight the Text object in the list box and press F2. 4. Enter a scenario name for the base scenario (i. Base Scenario). Therefore. the Audit Information dialog appears. and the new data are saved back to the PKS. PKS user name and change reason. Audit information If the study data have changed. are stored and made available via the PKS web client’s audit trail.18 User’s Guide Note: Make sure that all dosing data are loaded into the study. A change reason must be entered for every changed value in study data. fill in a reason for each item. Note: If 100 records or fewer have changed. the system will display the following dialogs to document all changes. WinNonlin Change tracking and audit information If any changes are made to a PKS study or scenario. Dosing data are considered part of the study and cannot be added after studies are loaded in readonly mode. A reason for change is required for every alteration to study data. including date. 470 . If more than 100 records have changed. Enter information for all changes. and for each scenario update. See your system administrator to inquire about standard reason text or click the Lookup button for a list of company-standard values. This reason will be recorded for every change that was made. All change records. the Audit Information dialog presents only one Reason field. change made. Use the Copy and Paste All buttons to speed the process. Worksheets with fewer cells will save individual data values to PKS. rather than saving individual data values to the databse.Using the Pharsight Knowledgebase Server Change tracking and audit information 18 Update reason When saving a new scenario or scenario version. This will speed loading of those data objects to and from PKS. Charts may be saved in JPG or Windows metafile (WMF) format. you will see the Update Reason dialog. instead of as Pharsight chart objects. which provides a means to save worksheets and charts in binary formats. rather than saving individual data values to the databse. update reason and (optional) scenario description. Check Native format for large worksheets to save all worksheets larger than a set limit as binary objects. created by your system administrators. Enter the minimum worksheet size to be saved in binary format under # cells and click Apply. Check items to save in binary format. Use Ctrl+Enter to insert line breaks. Worksheets may be saved as binary files. The Format button opens the PKS Save Native Format dialog. but make data values unavailable to query and other tools within the PKS Web Client. 471 . Enter the scenario name (for new scenarios). Use the Lookup button to select a reason for change from a standard list. 18 User’s Guide WinNonlin Once the Audit Information dialog is complete. Select an object name. Name objects WinNonlin will prompt you to provide names for any unnamed objects in the scenario. Clicking Yes adds a new scenario with the current WinNonlin settings and objects. PKS asks whether to create a base scenario with the updated data and any other changes that have been made in WinNonlin. or as part of a modeling scenario (via the WinNonlin Dosing 472 . then click on it again to edit the name. Dosing information Dose data are added to PKS studies either through a separate dose worksheet at study creation. click OK to enter the update reason(s). If no scenario was loaded. 473 . Changing the dose regimen When the dosing data exists on the Dosing worksheet. and one that records the data history (History). Dose_B. The Dosing worksheet can contain dose information for one or more analysis scenarios. See “Dosing” on page 448 for discussion. suppose that a PKS study contains time-concentration data for a crossover trial comparing IV bolus and oral drug delivery. After loading a study or scenario that contains dose information. Dose times and amounts for the two formulations appear in four columns of the Dosing worksheet: • • • • IV_Time IV_Amount Oral_Time Oral_Amount. To choose dosing columns to apply in the current scenario: 1.Using the Pharsight Knowledgebase Server Dosing information 18 Regimen dialog. See “Dosing worksheet” on page 448 for data requirements. after loading the model). Non-compartmental analysis will be performed for each. it is possible to indicate which fields apply to the current scenario. For example. two scenarios are required: • • Oral scenario: uses NCA model 200 and dose columns Oral_Time and Oral_Amount IV scenario: uses NCA model 201 and dose columns IV_Time and IV_Amount The selection of dose columns for each scenario is made as follows. etc. Load a study or scenario containing multiple dose regimens as described above.] as follows. Since the IV and Oral formulations require different NCA models. and applied to different scenarios by including different data columns [Dose_A. Thus different dosing schedules can saved in a study or scenario. and WinNonlin can load only one model at a time. WinNonlin contains a workbook with three worksheets: observation data (Observations). one containing dosing data (Dosing). or existing data can be updated. from the menus. Select PKS>Dosing. 6. The Study Creation Wizard appears. Open the data to be added in a WinNonlin workbook window. The column(s) containing the subject identifiers must be identified. or in which to update data. additional data can be added to the study. (See “Logging in” on page 452 for details. Note: Append/Update is a convenient way to add dosing information to the study or scenario. 7. 2. The Dosing Field Selection dialog appears. Click OK. Click OK. 3. Choose PKS>Append/Update Study.18 User’s Guide 1... To append/update a study: 1. select the study to which to append data. WinNonlin Appending or updating a study After a study is created in the PKS. 3. Drag the dosing fields for the current analysis to the Dosing Columns box. This method can also be used to add subject data to the study. 8. to map the variables in the data set to the selected study. In the Pharsight Knowledgebase Server Load dialog. from a workbook. 4. Drag variables from the workbook and drop them on or under the corresponding study variables. 474 . Close any PKS study or scenario that is open in WinNonlin. add them beneath the existing study variables. such as extra subject data or other observations. Click OK.) 5. Log in. Variables listed in the Non-Dosing Columns box will not be used in the current scenario. 2. If there are new data variables. For example. When the scenario is loaded in WinNonlin. Whenever there are significant differences between the actions in WinNonlin and those in PKS. the documentation systems point them out. To load files from the Pharsight Knowledgebase Server: 1. It is also possible to load data into WinNonlin then specify units or convert units—then save them back to the PKS. as described under “Logging in” on page 452. any PKS scenario can share data objects that are part of another scenario. the share associations are saved with the new scenario. data. In the Pharsight Knowledgebase Server Load dialog.Using the Pharsight Knowledgebase Server PKS libraries and object shares 18 Note: The PKS documentation set covers a number of other operations that affect studies and scenarios in the PKS. PKS libraries and object shares PKS provides two features for sharing models. • • “PKS share” on page 475 “Libraries” on page 476 PKS share A new analysis in WinNonlin can share non-Pharsight analysis and data files contained in any PKS scenario. and maintain a link to that object so that updates are shared. Details appear under the following headings. Log in to the PKS. When an analysis that uses shared files is saved to PKS. First. it is possible to specify or convert units in the PKS without loading the data in WinNonlin. 475 . 2. the PKS library feature supports creation of system-level objects that can be referenced by any scenario. choose the study and scenario containing the file(s) to be shared. code and other data objects across studies and scenarios. Second. any shared files may be updated to reflect changes in their source files as described under “Library and share status” on page 478. 3. Choose PKS>Share from the WinNonlin menus. Select items to load. presenting a list of non-Pharsight files that can be referenced in the current analysis. 476 . Consequently. Libraries Note: Each application has it’s own folder/container for storing the binary files. 6. Click OK.18 User’s Guide WinNonlin 4. 5. applications cannot see other applications’ library systems. Hold down the Ctrl key to select multiple items. Click OK to load the items. The Select Share Item(s) dialog will appear. PKS Client may add a library item and it will not be visible by the WinNonlin application. For example. NONMEM or SAS code. Hold down the Ctrl key to make multiple selections. as described under “Connecting to the PKS” on page 451. Library items are not associated with any PKS study or scenario. 477 . and load it back again. Click OK. Select the library(ies). 2. using the Library function. and other files as Libraries in the PKS. The library function is intended to store a small number of files containing code. 1. and other items that will be used as standards across many studies and analyses. The library functions include the following: • • “Loading a library from the PKS” “Adding or updating a library in the PKS” on page 477 Loading a library from the PKS Use WinNonlin to load standardized parameter names. Choose PKS>Libraries>Load from the WinNonlin menus. 3. 4. NONMEM or SAS code. Adding or updating a library in the PKS Use this feature to save standardized parameter names.Using the Pharsight Knowledgebase Server PKS libraries and object shares 18 WinNonlin can save any type of binary file as a PKS system level object. Log in to the PKS. and other files stored as Libraries in the PKS. parameter names. The Select Library Item(s) dialog will appear. Choose PKS>Libraries>Add/Update from the WinNonlin menus. With PKS share and/or library files loaded in WinNonlin. Open the file in the WinNonlin using File>Open. The Select Library Item(s) dialog appears. 478 . Library and share status WinNonlin tracks whether PKS share items and Libraries are current with the latest version in the PKS. The Share and Library Status dialog appears. Items that are out-of-date can be updated as follows. Select the item(s) to be loaded to the PKS and click OK. Make any desired changes.18 User’s Guide To save a file to the PKS as a library or update an existing library: WinNonlin 1. or load an existing PKS library as described under “Loading a library from the PKS” on page 477.. following any standards set within your company. 7. Hold down the Ctrl key to select multiple items. Click OK. 2. 6. Click Yes. 3. enter the Library Description and the Reason for Update. to confirm that you wish to add PKS data. You may save the file to a local disk to save interim changes before uploading them to the PKS. 8. Log in to the PKS. A dialog will appear. as described under “Connecting to the PKS” on page 451. Click OK after the save operation is complete. choose PKS>Status. In the PKS SAS Save dialog. Items that have newer versions available in the PKS appear in red. 5.. To synchronize shared files and/or libraries with the latest versions in the PKS: 1. 4. from the menus. All dates and times. its data cannot be updated in PKS. including time stamps for study updates in PKS. • • Server Information: time zone of the PKS server. choose PKS>Information from the menus. – Description: description of the most-recent version of the study.Using the Pharsight Knowledgebase Server PKS Information 18 2. – User selected read only: if true. Study information: – Locked: if the study is locked. With a PKS study or scenario loaded in WinNonlin. will use this time zone. no changes may be saved to PKS. – Name: study name in PKS. entered by the user when the study was last saved or updated. The PKS Information dialog appears. – Type: study type. – Internal ID: study identifier used in PKS. WinNonlin will load the latest version from the PKS. the study is loaded as read-only. Check the items to be synchronized with the latest PKS version and click OK. To view summary information for the currently-loaded study: 1. It shows the following information. 479 . PKS Information Information about a study and/or scenario loaded into WinNonlin is summarized in the PKS Information dialog. etc.. • • • Scenario Name in PKS Scenario Description: entered when the scenario was last saved in PKS. they are never required. While units are useful in WinNonlin. there are some differences that are important to note. However. Current user rights: PKS access rights of the PKS user account used to load the study into WinNonlin. because each is also designed to work with other tools as well. Conversions are not available for custom units in WinNonlin. PKS includes all of WinNonlin’s default units. WinNonlin WinNonlin and PKS differences WinNonlin is closely integrated with the Pharsight Knowledgebase Server. By default. custom units can be entered and carried through operations.18 User’s Guide – Design: study design. with conversion factors to define relationships to existing units. • • 480 . See the PKS user documentation for a description of various levels of user access. This is not required in WinNonlin. Units can be handled at the cellular level in the PKS. (The units would be stored in an adjacent column. This level determines whether the user can save changes to this study in the PKS. – State: state of the study in the PKS. Relative Nominal Time. WinNonlin performs calculations and conversions using only its internal set of units. Units • • Units are required for some fields in the PKS. units are associated with an entire column. each record in a given column can use different units. That is. PKS requires that all time-dependent data (Relative Actual Time. The PKS provides the means perform this conversion. Thus there are situations when valid data from a PKS data set cannot load correctly in WinNonlin until conversions are performed to make units uniform within each column. New units can be added and defined in the PKS.) In WinNonlin. for dosing and observations) use the same time unit. but they do not affect calculations. – BlindingType: study blinding scheme. The PKS can handle multiple drugs within one study. except that workspaces require all child windows to have filenames prior to being saved. If a data set does not include time or subject data. dummy columns must be created. WinNonlin analyzes only one drug at a time. PKS requires the child windows to have “object names.Using the Pharsight Knowledgebase Server WinNonlin and PKS differences 18 Studies and scenarios • Saving scenarios to PKS is analogous to saving a workspace in WinNonlin.” Time and subject data must be present to save and import data from WinNonlin to the PKS. • • 481 . 18 User’s Guide WinNonlin 482 . j = 1 …. the zero time intercept associated with the Alpha phase. A A In Modeling. P is the number of parameters. available for entering clock times from research data. this time may differ from the nominal time. Italics indicate terms that are defined elsewhere in the glossary. actual time The time that a study dose or observation actually happened. accumulation index actual clock time AIC 483 . When comparing several models for a given data set. AIC = N log (WRSS) + 2P for modeling in WinNonlin. A measure of goodness of fit based on maximum likelihood. 1 -------------------------------–( λz × τ ) [1 – e ] In PKS. In Deconvolution A and alpha refer to the parameters of a polyexponential unit impulse response function of the form: UIR(t) = sum[A(j) * exp(-alpha(j) * t)]. Appropriate only for comparing models that use the same weighting scheme. Not used in any WinNonlin calculations. N is the number of observations with positive weight. N. Akaike Information Criterion (AIC). the model associated with the smallest AIC is regarded as giving the best fit. a placeholder.Appendix A Glossary WinNonlin and PKS terminology The following terms and definitions pertain specifically to objects and operations in WinNonlin and its operations with Pharsight Knowledgebase Server. WRSS is the weighted residual sum of squares. and S is the estimate of the residual standard deviation. If C(t) denotes the concentration of analyte at time t.ANV) containing all information required to recall and reproduce a given ANOVA model. AndersonHauck pval AndersonTest statistic applicable for testing equality of means in a bioequivalence Hauck statistic study.0 and later. The test statistic t is: 1 X E – X S – -. nE and nS are the group sample sizes. Sometimes denoted ALPHA_HL.( A + B ) 2 t = ---------------------------------------------S 1 ⁄ nE + 1 ⁄ nS where the X's are the respective sample means. and variance components for linear mixed-effects modeling calculations. This is derived from a noncentral t distribution. These versions can open and run and AVNOVA command files and call them in scripts. time curve. AIC is defined as: AIC = – 2 L R + 2S where L R is the restricted likelihood evaluated at converged estimates. the macro rate constant associated with the distribution phase. ANOVA command file (*. N. Area under a concentration of analyte vs. then: b a b AUC AUC = ∫ c ( t ) dt a AUCall 484 AUC computed from time zero to the time of the last Y value. Alpha In modeling. The p-value of the Anderson-Hauck test statistic. j = 1 …. In Deconvolution. Alpha half-life The half life associated with the macro constant Alpha. but cannot save to that format.A WinNonlin User’s Guide ˆ ˆ In LinMix. A and alpha refer to the parameters of a polyexponential unit impulse response function of the form: UIR(t) = sum[A(j) * exp(-alpha(j) * t)]. and s is the rank of the fixed-effects design matrix plus the number of variance parameters.LML) in WinNonlin versions 4. Replaced with the LinMix command file (*.anv) Legacy ANOVA command file (*. A and B are the lower and upper limits. . for use in scripting. AUMClast = AUMC t last 0 B B balanced design The zero time intercept associated with the beta phase. the design is no longer balanced. each with two levels. individual. The half life associated with the macro constant Beta. An experimental design in which the number of observations is the same for every combination of the experimental effects. is balanced if each of the four combinations of formulation by period has the same number of observations. AUMCINF = AUMC ∞ 0 AUMClast Area under the moment curve computed to the last observation.Glossary B AUCINF AUClast AUMC A AUC from time zero extrapolated to infinity. “Macro” rate constant associated with the elimination phase. a crossover design with effects for formulation and period. AUC computed from time zero to the time of the last positive Y value. If a subject drops out of one sequence. Average bioequivalence states that aver485 Beta Beta half-life bioequivalenc e . Also called BETA_HL. and population. There are three types of bioequivalence: average. Area under the moment curve: b AUMC = a b ∫ c ( t )t dt a AUMCINF Area under the moment curve when the time concentration curve is extrapolated to infinity. For example. Bioequivalence is said to exist when a test formulation has a bioavailability that is similar to that of the reference. and is similar to population bioequivalence. box and whiskers plot The box and whiskers plot is designed to show the distribution of values.5 times the H-spread [the distance between the hinges] from the hinges. carry along variables Cavg 486 . The box marks the 25th and 75th percentiles. The whiskers end at the Max and Min points unless there are outliers beyond 1. Profiles with fewer than five data points plot the data points directly with no box. every parameter must be given lower and upper bounds. it is recommended that Lower and Upper bounds be used routinely. The lower and upper values limit the range of values the parameters can take on during model fitting. BQL Below the lower limit of quantification (LLOQ) for a given assay. and the median between them. Although bounds are not always required. user-defined bounds are preferred. This can be very valuable if the parameter values become unrealistic or the model will not converge. C C The zero–time intercept associated with the absorption phase for an extravascular model or the zero time intercept associated with the gamma phase for a 3 compartment IV bolus model.A WinNonlin User’s Guide age bioavailability for test formulation is the same as for the reference formulation. The WinNonlin default bounds are 0 for one bound and 10 times the initial estimate for the other bound. bounds Upper and lower constraints on the initial parameter estimates for compartmental modeling. Individual bioequivalence is meant to assess switchability of products. If they are used. For models with parameters that may be either positive or negative. For steady-state data: computed as AUC(0-τ) divided by τ. Population bioequivalence is meant to assess equivalence in prescribability and takes into account both differences in mean bioavailability and in variability of bioavailability. Variables that are not required for the current analysis. but will be copied and included in the output data set. magnitude of each dose. Renal Clearance. For extravascular models the fraction of dose absorbed cannot be estimated. The concept is used in PKS to group study objects that pertain to the same molecule.PMO). For steady-state data: an estimate of the total body clearance. smoking status). CL= Dose/AUC In a LinMix model. classification variables or covariates are independent variables that employ values from a discrete set (e. but not save *.g. A command file can contain all information required to reproduce a compartmental or noncompartmental model.Glossary C CL classification variables A Total body clearance. Legacy command file (*. COND is the square root of the ratio of the largest to the smallest eigenvalue of the matrix of partial derivatives. Condition number of the matrix of partial derivatives. Clss/F Cmax Cmin The peak or maximum concentration. For steady-state data: the minimum concentration between 0 and τ (corresponding to Tmin).CMD files. and dosing times. gender. Time expressed as a date and/or time. It may include one or more formulations. therefore Clearance for these models is actually Clearance/F where F is the fraction of dose absorbed: Dose = --------------τ AUC 0 clock time CLR Clss. Replaced by Pharsight Model Objects (*. rather than continuous values. WinNonlin can open and run. Each library model requires specific constants.CMD) from an earlier version of WinNonlin. A drug molecule under study. rather than relative to the first dose or the start of a study. Dosing information such as number of doses. See also LinMix command file and table definition file. See also regressors. command file compound condition number (COND) constant 487 . ” See also independent variable. blood pressure. In this way. suppose that effect Trt has three levels: Placebo. 488 . For example. such as age.5. or body weight. some subject characteristic of interest. Concentration that achieves 50% of predicted maximum effect in an Emax model.” or “time of sample. 10mg=-0. It is not required that data be available at every intersection of the factors.. construct the contrast: Placebo=1. This normally includes the designated dose and the time of the dose. possibly unknown.5. It may also include the dose cycle.5=0. Convergence is achieved when the relative change in the residual objective function is less than the convergence criteria. This compares the average of the 10mg and 20mg groups versus placebo. linear combinations of levels of an effect. Covariates are continuous measurements. be written as a sum of exponentials. the scheduled times for observations.” “period. A variable such as “drug concentration” is usually a dependent variable because it depends on factors like “formulation. The convergence criterion determines when WinNonlin can stop with a successful fit. See also nested factors.A WinNonlin User’s Guide contrasts In LinMix.5-0. To compare the two dose groups against placebo. such that the sum of the coefficients is zero. In a LinMix crossed-classification model every level of every factor could be used in combination with every level of every other factor. dosing regimens E E EC50 Denotes Effect. the factors “cross” each other. This method can be used to determine initial estimates required as a prelude to nonlinear least-squares regression. Generally. 20mg=-0. etc. depending on the particular model's requirements. 10mg. and 20mg. Independent variables in a LinMix model. Note that 1-0. to the independent variables. convergence criterion covariates crossed factors curve stripping A graphical method for estimating parameters associated with models that can D dependent variable Dependent variables have some functional relationship. The schedule of dosing. TABLE.covariance matrix. For steady-state data: % fluctuation is computed as 100*(Cmax-Cmin)/Cavg. For compartmental modeling. Generally associated with explanatory variables. F Fabs Fdiss first-order Fraction absorbed over time. Very large values of the condition number may be indicative of instability in the model fitting process. E0 Emax engine Effect at time 0. This method can be used to determine initial estimates required as a prelude to nonlinear least-squares regression. The point of dose input into a structural (PK/PD) model. The eigenvalues included in WinNonlin's modeling output are associated with the variance . DESC. where x is the eigenvector. WinNonlin engines are: PK. PLOT. or quantitative. LINMIX. Maximum effect. BIOEQUIVALENCE. Eigenvalues and their associated eigenvectors can be thought of as building blocks for matrices. such that Ax = λx for some vector x. 489 fixed effects % fluctuation formulation function variable . as in standard linear regression. as in a standard linear model. TRANSFORM. A column in the data set that indicates which observations belong with which model functions. for fitting multiple functions to one data set. estimated from in-vivo time-concentration data. ANOVA. Fraction dissolved over time in-vitro. Used to denote rate constants when the transfer rates are proportional to the amount in the source compartment. exponential A graphical method for estimating parameters associated with models that can curve stripping be written as a sum of exponentials. WinNonlin prints a condition number that is equal to the square root of the ratio of the largest to the smallest eigenvalue. as in traditional ANOVA. Required for simultaneous link models. Scripting language engines execute tasks in WinNonlin using methods. MERGE. where Cmin and Cmax were obtained between 0 and Tau. The variables can be either qualitative.Glossary F eigenvalue A An eigenvalue of matrix A is a number λ. Gamma halflife GaussNewton algorithm geometric mean group variables H harmonic mean The harmonic mean is the reciprocal of the arithmetic mean of the reciprocals of the Ys (Y > 0). Note that group variables may appear as a column in the body of the table.covariance matrix of the parameters. Levenberg Modification: A model fitting algorithm that performs well on a wide class of problems. This requires that each Y > 0. They are printed only when their value changes and at page breaks. Multiple group variables may be used. The half-life associated with the macro constant Gamma. A model fitting algorithm. Hartley Modification: This method is not as robust as others but is very fast. The geometric mean is the nth root of the product of the n Y's: GM = n Y Y Y …Y . This method requires the estimated variance . or as an information line above the column titles. 1 2 3 n In the Chart Wizard. however the Levenberg modification suppresses the magnitude of the change in parameter values from iteration to iteration to a reasonable amount. first-order model. Modifications of the Gauss-Newton algorithm include the Hartley and Levenberg modifications. group variables allow for heterogeneity of variances while retaining the same mean model. including ill-conditioned data sets. a group variable breaks the data into separate plot lines. This method is the default. 1 HM = ⎛ -⎝n ∑ ---⎞ Y⎠ 1 –1 490 . For the Table Wizard. Sometimes denoted GAMMA_HL.A WinNonlin User’s Guide G Gamma The terminal elimination rate associated with a three compartment. group variables are much like ID variables. with the exception that only unique values of group variables are printed. as selected using the Tools>Options menu item. In LinMix. When summary statistics are selected these variables are not considered. R. and its related uses in formulation development. Note that multiple ID variables may be used. Interaction is the degree to which the effect of one factor differs with varying values of another factor. Nonlinear algorithms require derivatives of the models with respect to the parameters. Indirect Pharmacodynamic Response Model. increment for partial derivatives independent variable indirect response models The independent variables are controlled by the experimenter (e. the value of the response when all effects are zero. of the response may be inhibited or stimulated.001. of a response may be either inhibited or stimulated.g. to a drug may be produced by an indirect mechanism. formulation) or do not depend on other factors. For LinMix ANOVA and regression models.Glossary I A I ID variables Used by the Table Wizard to identify data in adjacent “Data” variable columns. interaction intercept IPR iterative reweighting IVIVC 491 . The measured response. A weighting scheme where weights associated with the individual observations are functions of the predicted values. kout.≅ -------------------------------------------------------dP ( 1 ) Δ where Δ = Increment times the parameter value. The default is 0. Interaction terms would be included in a LinMix or Bioequivalence model using the “*” symbol: Treatment*Period. See also dependent variable. then Treatment and Period are said to interact. and that this may be a convenient way to include units.. For example. a model describing the relationship between dissolution in-vitro and absorption in-vivo. kin. Factors controlling the input or production. if a treatment has a different effect in one period than it does in another period. For example: dF F(P(1) + Δ) – F(P(1)) -------------. In-vitro in-vivo correlation. The program estimates these derivatives using a forward difference. or determinants of loss. A WinNonlin User’s Guide J join variables In the Tables Wizard, when merging data from two workbooks into one table, joins identify corresponding variables to be used in merging profiles from the two data sets, e.g., “Subj” and “Subject.” K K half-life K01 The half-life associated with the rate constant K. Sometimes denoted K_HL. The rate at which the drug enters the central compartment from outside the system. It is sometimes denoted as Ka. The rate at which the drug leaves the system from the central compartment. The elimination rate. The half-life associated with the rate constant K10. Often denoted K10_HL. For first-order rate constants; the rate from Compartment i to Compartment j. Zero-order input or infusion rate constant. Michaelis constant for models involving nonlinear kinetics. K10 K10 half-life Kij K0 KM L lag time (LT) Lag time is a time delay between drug administration and the onset of absorption. An example of such a process is the stomach-emptying time, following oral administration of a drug that is not absorbed from the stomach. When a drug is administered by bolus intravenous injection there is no lag phase. phase. This is estimated via linear regression of time vs. log concentration. Lambda Z (λz) First order rate constant associated with the terminal (log-linear) elimination least squares fitting least squares means Minimizes the sum of the squares of the deviations between the observed and predicted values of the dependent variable. Least squares means are estimates of what the mean values would have been had the data been balanced (see balanced design). That is, least squares means are the means predicted by the fixed-effects model. If a data set is balanced, the least square means will be identical to the raw (observed) means. 492 Glossary M linear kinetics linear regression A A model in which the transfer rates are first order. A model relating an independent variable to dependent variables such that the unknown parameters enter into the model linearly. In the simplest case, a straight line is fit to the data. The intercept and slope are determined by least squares. The slope and intercept are called model coefficients, or parameters. File (*.LML) storing all information needed to reproduce a LinMix analysis, often used in scripting. See “Linear Mixed Effects Modeling” on page 323. Lower limit of quantification, the concentration whose lower bound (of variability in the assay) includes zero. Analyte may be detectable below this concentration even though it is not quantifiable. LinMix command file LLOQ M macro rate constants Compartmental models involving first order kinetics can be written as a sum of exponentials. The associated parameters are called macro rate constants and are functions of the underlying micro rate constants. Maximum value. The arithmetic average. Arithmetic average of the logarithms of the Ys. The variance is normally a function of weighted residual SS/df, or the mean square. For certain types of problems, when computing the variance the mean square needs to be set to a fixed value. It is possible to use a specified value other than the residual mean square in the estimates of the variances and standard deviations. This value replaces the residual sum of squares in the estimates of variances and standard deviations. max mean mean of the logs mean square method In the scripting language, a method is a set of computer code that performs functionality on an object or engine. For example, the method 'LOAD' causes an associated data file (an object) to be loaded. The method 'RUN' causes an associated PK Model (engine) to run the model. A type of model often employed for models involving nonlinear kinetics. The V = V max × C ⁄ ( K m + C ) Michaelis-Menten equation is: 493 MichaelisMenten A WinNonlin User’s Guide micro rate constants min MRT Transfer rates associated with a compartmental model. See also macro rate constants. Minimum value. Mean Residence Time is the average amount of time a particle remains in a compartment or system, equal to AUMC / AUC. Mean Residence Time when the drug concentration profile is extrapolated to infinity. This can be estimated from single dose data or steady state data. Mean Residence Time when the drug concentration profile is not extrapolated to infinity, but rather is based on values up to and including the last measured concentration: MRTlast = AUMClast / AUClast MRTINF MRTlast N N Nmiss Nelder-Mead simplex algorithm In modeling output, the number of observations with non-missing data. In modeling output, the number of observations with missing data. A model fitting algorithm that does not require the estimated variance—covariance matrix as part of its algorithm. It often performs very well on ill-conditioned data sets (data sets that do not contain enough information to precisely estimate all of the parameters in the model). When used to fit a system of differential equations, it can be very slow unless the model has been compiled. nested factor are unrelated to one another and are nested within a level of another factor. For example, in a crossover study, Subjects are assigned to Treatment Sequences. This nested factor would be included in a LinMix or bioequivalence model using parentheses: Subject(Sequence). Subjects are not crossed with Sequences, because the subjects within one sequence are unrelated to the subjects within another sequence. See also crossed factors. nested factors In a nested classification, unlike a crossed classification, the levels of the Nobs nominal time The number of observations. The time specified in the protocol for a dose or observation—which may differ from the actual time that a dose or observation was taken. Noncompartmental, or model independent analysis describes the concentration-time curve by estimating noncompartmental parameters such as: area noncompartm ental analysis 494 Glossary O A under the curve (AUC), Tmax, Cmax, and Lambda Z (the rate constant associated with the terminal elimination phase), among others. Noncompartmental parameters are estimated using the Linear or Linear/Log trapezoidal rule. If Lambda Z can be estimated, then the noncompartmental parameters will be extrapolated to infinity. nonlinear kinetics Kinetics that can be modeled by nonlinear differential equations, for example, V max C dc V ---- = ---------------dt Km + C nonlinear model In nonlinear models, at least one of the parameters appears as other than a coefficient of an independent variable. One such model is the sum of two or more exponentials, such as Y = A1*exp(-B1*X) + A2*exp(-B2*X). A nonlinear model is one for which one or more of the partial derivatives with respect to the model parameters involve model parameters. Note that nonlinear models may arise from compartmental models having either linear or nonlinear kinetics. The process of fitting nonlinear models to data. nonlinear regression O ODBC Open Database Connectivity. A standard for data exchange, in common use on Microsoft Windows platforms. It is an application programming interface for database access that uses Structured Query Language (SQL) as its database access language. In scripting, objects are described by one or more property/ies, and acted upon by methods. In WinNonlin, there are five main objects: data (workbook) objects, text objects, model objects, graph objects, and WinNonlin itself. object P partial tests LinMix tests of whether each model term contributes to the fit after all other model terms have been included. Partial tests are invariant to the model specification order. 495 A WinNonlin User’s Guide percentiles The pth percentile divides the distribution at a point where p percent of the distribution are below this point. Models of drug effect as a function of a drug concentration. pharmacodyn amic (PD) models pharmacokine Models of drug concentration as a function of time. tic (PK) models PK/PD link A PK/PD Link model uses drug concentration data to estimate concentrations at an effect site, which are then used to model the pharmacodynamics. A confidence interval obtained from the tangent planes to the joint confidence ellipsoid of all the parameter estimates. In the Bioequivalence Wizard, the probability of detecting a difference greater than or equal to a specified percentage of the reference mean. A set of observed data to be fit at one time, producing one set of individual model parameters. A profile is identified as data records with a given, unique set of values for all sort variables. Attribute or characteristic of a object in scripting, defining how the object should appear and interact. Properties also set values for an object or engine. planar confidence interval power profile property Q R random effect Also known as covariance parameters, random effects distinguish a linear mixed effects models from standard linear models. In general, random effects are additional unknown random parameters assumed to impact the variability of the data. The variances of the random effects, commonly known as variance components, become the covariance parameters for this structure. Rank of the matrix of partial derivatives. If the matrix is of full rank, Rank= # of parameters. If the Rank is less than the number of parameters, then there is not enough information in the data to estimate all parameters in the model. rank 496 Glossary S regressors A In a LinMix model, regressor variables or covariates are independent variables that employ values from a continuous set (e.g. temperature, body weight), rather than categories. See also classification variables. In PKS, the time that observations or doses really happened, as opposed to relative nominal time. See also relative actual end time. In PKS, the times when infusion doses actually ended, as opposed to relative nominal end time. See also relative actual time. In PKS, scheduled dose and observation times, as set in the study protocol. Compare with relative actual time. See also relative nominal end time. In PKS, the times when infusion doses were scheduled to end, as set in the study protocol. Compare with relative actual end time. See also relative nominal end time. Elapsed time since the first dose or the beginning of the study, period or phase. Observed minus predicted values, where the predicted values come from fitting the specified model. relative actual time relative actual end time relative nominal time relative nominal end time relative time residuals S S Standard error of the weighted residuals: S = ( WRSS ) ⁄ ( DF ) where WRSS is the weighted residual sums of squares, and DF is the number of observations minus the number of zero weights and the rank of the variance/ covariance matrix. The actual times at which the concentration or effects are observed. Schwarz Bayesian Criterion. A measure of goodness of fit based on maximum likelihood. When comparing several models for a given data set, the model with the smallest SBC value is regarded as giving the best fit. Appropriate only for comparing models that use the same weighting scheme. SBC = NOBS × log ( WRSS ) + log ( NOBS ) × NPARM ˆ For LinMix, the SBC calculation is: SBC = – 2 L R + s log ( N – r ) where ˆ ˆ is the restricted log-likelihood function evaluated at final estimates β L R sampling times SBC 497 A WinNonlin User’s Guide ˆ and θ ; r is dim(x); s is the number of parameters; and N is the number of observations. scenario In PKS, a set that includes one structural model, initial parameters, and/or statistical tests to be applied to a data set for fitting or analysis. See also study. Standard deviation. Standard deviation of the logs of the Ys. Standard error. Parameters that are functions of the primary model parameters. For example, AUC and half lives are secondary parameters. Note that, in the case of multiple-dose data, secondary parameters dependent on dose use the first dose. Standard error of Predicted Y. Denotes the “S” shaped nature of certain models. A separate analysis or plot is done for each level of the sort variable(s). If gender and smoking are sort variables, a separate set of analysis results or plots will be created for each of female non-smokers, male non-smokers, female smokers and male smokers. In these models, which can be written as a sum of exponentials, dose does not appear explicitly in the model. Rather, the macro constants are a function of dose. Thus, when fitting macro constant models with user-specified initial estimates, the user must specify the associated dose. For single dose macro constant models, when WinNonlin determines initial parameter estimates by curve stripping, the stripping dose is identical to the administered dose. SD SD of the Logs SE secondary parameters SE - Yhat sigmoid sort variables stripping dose Dose associated with initial parameter estimates for macro constant models. structural model study The set of functions defining the shape of a model, i.e., a structural PK, PD, or PK/PD model. It does not include parameter distributions. In PKS, a study comprises a collection of data that can include observations, dosing, and protocols. A study can also contain one or more scenarios. The variable(s) for which descriptive statistics will be calculated. summary variables 498 Glossary T A T table definition A file (*.TDF) that stores a user-defined table format and settings. See also file “Table definition files” on page 148. Tau (τ) TI Tmax Tmin The (assumed equal) dosing interval for steady-state data. Infusion length. The time of peak concentration. For steady-state data: time of minimum concentration based on samples collected during a dosing interval. A combination of one or more drugs given at specific doses and time(s). treatment U UIR Unit impulse response function, describing the PK profile following an instantaneous dose of one unit of drug into the central compartment. A confidence interval based on standard error. The univariate confidence interval does not take into account correlations with other parameters. Libraries of user-defined ASCII models. See “User Models” on page 217. univariate confidence interval user libraries V V variance Volume of distribution of the central compartment. This is also denoted VC. n The variance, v, in descriptive statistics: ∑ ( Yi – Y ) i=1 2 ⁄ (n – 1) Note: VIF SD = v Variance Inflation Factor, the multiplier of the residual mean square, used in deriving the asymptotic variance of the parameter estimates. Useful in simulations, to find optimal experimental design. Smaller values are better. Theoretical maximum rate of process in Michaelis-Menten kinetics. 499 Vmax A Vss WinNonlin User’s Guide An estimate of the volume of distribution at steady state. Vss = MRTINF*Cl, where MRT = Mean Residence Time and CL= total body clearance. For steady-state date, Vss = MRTINF*Clss. Not computed for model 200, as calculating MRT from oral data requires estimating Mean Input Time, which requires compartmental methods. Volume of distribution based on the terminal phase. For extravascular models, the fraction of dose absorbed cannot be estimated. Therefore, Volume for these models is actually Volume/F, where F is the fraction of dose absorbed. VZ, VZ/F W weight variable The weight variable is a numeric data column used in LinMix fitting to compensate for systematic differences in variance across observations, e.g., variance that increases with observed value. Each observation is multiplied by the square root of its weight. Weight values should generally be proportional to reciprocals of variances. This variable should not be included with classification variables, but can be among regressors or used in the model. Square root of the weight times Y Predicted. Predicted Y times the square root of the weight. Residual Y times the square root of the weight. weighted computed Y weighted predicted Y weighted residual Y weighted - SS Weighted Sum of Squared Residuals. See also WRSS. Westlake confidence interval workspace Confidence intervals that are constrained to be symmetric about 100%. A workspace saves all open windows and settings in WinNonlin, including model and/or analysis settings. Weighted residual sums of squares, an estimate of the variance of the residuals. This is the quantity WNL minimizes during modeling. WRSS 500 Glossary X A X X variable The independent variable to be used in PK modeling or Noncompartmental analysis, usually, time relative to the first dose. Y Y variable The dependent variable to be used PK modeling or Noncompartmental analysis, generally Concentration. 501 A WinNonlin User’s Guide 502 Appendix B WinNonlin Model Libraries Parameterization and equations for WinNonlin compiled and ASCII models WinNonlin is distributed with an extensive library of models, including most of the commonly used one, two and three compartment pharmacokinetic (PK) models, linear, pharmacodynamic (PD), link, indirect responseA and Michaelis-Menten models, as well as seven “models” for noncompartmental analysis of plasma, urine or drug effect data. The following are included in the library of compiled models and detailed in the sections below, following the syntax noted under “Notation and conventions” on page 504. Table B-1. Model libraries Model “Noncompartmental analysis” on page 505 “Pharmacokinetic models” on page 519 “Pharmacodynamic models” on page 533 “PK/PD link models” on page 536 “Indirect response models” on page 539 “Michaelis-Menten models” on page 542 “Simultaneous PK/PD link models” on page 545 “Linear models” on page 546 “Sigmoidal/dissolution models” on page 547 with IVIVC licence only Compiled Yes Yes Yes Yes Yes No No Yes Yes ASCII No Yes Yes No No Yes Yes No No Some of the models are provided in ASCII (text) model libraries; others are compiled into machine language. Each PK and PD model is included in both A. Dayneka NL, Garg V, Jusko W (1993); Jusko W (1990); Nagashima R, O’Reilly RA, Levy G (1969). 503 B WinNonlin User’s Guide ASCII and compiled forms. Compiled models run faster. The ASCII versions provide the user with examples and starting templates for custom models. Note: When using the ASCII models, be sure to examine the ASCII code to ascertain what dosing constants need to be created. Although WinNonlin can be used to estimate the parameters in any system of nonlinear equations, these libraries contain classical PK and PD models. Note that it is possible to create custom model libraries, as described under “User Models” on page 217. Notation and conventions WinNonlin Model Libraries number the central compartment 1, and the exterior to the compartments, zero. First-order rate constants are expressed as Kij, meaning the rate from Compartment i to Compartment j. Thus, K01 is the rate at which drug enters the central compartment from outside the system, and K10 is the rate at which drug leaves the system from the central compartment. Some models compute secondary parameters, such as half-lives and areas, based the estimates of the primary parameters. WinNonlin also estimates standard errors for these secondary parameters. For the two and three-compartment models some pharmacokineticists prefer to estimate the micro rate constants (Kij) as primary parameters; others prefer to estimate the macro rate constants (alpha and beta). The WinNonlin libraries contain both forms. Thus, Model 11, for example, is the two-compartment model with first-order input and alpha and beta as secondary parameters; Model 13 is the same model with alpha and beta as primary parameters and the rate constants K01, K12, and K21 as secondary parameters. Since these models are normally used to fit concentration data as a function of time, t represents the independent variable, and C(t) is the mathematical equation being fit to the observations. All of the models require the dose(s) and time(s) of dosing to be input as a constants. In addition, for those models defined in terms of the macro constants, the dose associated with the initial parameter estimates (this dose is hereafter referred to as the stripping dose) must also be input. The reason for this is as follows. Such models can be written as: C(t) = 504 ∑ Ai × exp ( –αi × t ) WinNonlin Model Libraries Noncompartmental analysis B Note that the dose, D, does not appear directly in the expression for C(t). Instead, the intercepts, {Ai}, are functions of D (as well as the micro constants). To support multiple-dose data, WinNonlin defines the macro constants model as: C(t) = ∑ Dj × ----- × exp ( αi × t ) D j Ai where Dj denotes the jth administered dose. Writing the model in this way enables WinNonlin to fit multiple dose data for the macro constant models. In order to do so, WinNonlin requires the stripping dose (D). With the exception of the Noncompartmental Analysis “models,” WinNonlin models assume that the Time variable has been coded such that the time of the first dose is 0. This applies for both WinNonlin modeling and simulation. The factor that converts dose amount to concentration is denoted as V, volume. In absorption models there is no way to estimate both volume and fraction of dose absorbed, so both of these are included in V (thus V/F is estimated where F is the fraction of the dose that was absorbed). The presentation of the mathematical equations of the models follows FORTRAN notation. For example, exp(-K01 · t) is the constant e with an exponent of the negative product of time, t and the rate constant K01. The models with first-order input are presented with and without a lag time. Since this amounts only to a shift on the time axis, the lag time is not explicitly shown in the expression for C(t), and the estimated Tmax should have the lag time added to it for those models that include a lag time. Note: For discussion of one and two-compartment models with first-order exchange see Gibaldi and Perrier (1982). Noncompartmental analysis The WinNonlin Model Libraries include seven compiled “models” for noncompartmental analysis of plasma, urine or drug effect data. 505 Tau is required for steady-state data only. For additional reading. threshold (optional) Constant infusion Dose. time of last dose a.B WinNonlin User’s Guide Table B-2. Models 210-212 (urine concentrations) assume single-dose data. see Gouyette (1983) and Chan and Gilbaldi (1982). time of last dose Dose time. 506 . dose interval (Tau)a Urine Urine Urine Drug effect Extravascular Bolus IV Any Dose. For steady-state data. dose interval (Tau)a Dose. time of last dose. time of last dose. and that the data are from a “final” dose given at steady-state. baseline effect. dose interval (Tau)a Plasma or blood Extravascular Plasma or blood Bolus IV Plasma or blood Constant infusion Dose. Noncompartmental analysis models Model Type of data 200 201 202 210 211 212 220 Dose input Constants required Dose. Models 200-202 can be used for either single-dose or steady-state data. the computation assumes equal dosing intervals (Tau) for each profile. time of last dose. length of infusion. time of last dose Dose. NCA output parameters and their computation formulas NCA output parameters and their computation are covered under: • • • • “Plasma or serum data” on page 507 “Urine data” on page 512 “Sparse sampling (pre-clinical) data” on page 514 “Drug effect data (model 220)” on page 515 See also “Computational rules for WinNonlin’s noncompartmental analysis engine” on page 181 and “Noncompartmental Analysis” on page 159. When using steady-state data.” on page 507 Table B-4.WinNonlin Model Libraries Noncompartmental analysis B Plasma or serum data Data structure NCA for blood concentration data requires the following input data: • • Time of each sample Plasma or serum concentrations For extravascular and constant infusion models (models 200 and 202). if the input data do not contain an observation at time zero. Dosing_tim Time of last administered dose. • • • Table B-3. Output Models 200-202 estimate the parameters in the following tables. “Parameters available only if λz is estimable. WinNonlin will generate one. it is estimated by back-extrapolating from the first two concentration values according to the rules given under “Insertion of initial time points” on page 182. “Calculations for steady-state data. Tlag is the time prior to the first measurable (non-zero) concentration. “Parameters that do not require λz. Given only for bolus IV models. then that concentration is set equal to the minimum concentration observed in the dosing interval. if the concentration at the dosing time is missing. Assumed to be zero unless otherwise specie fied. Maximum observed concentration divided by dose. It is equal to the first observed concentration value if that value occurs at the dose time. For steady-state data.” on page 508 Table B-5. Parameters that do not require λz C0 Initial concentration. 507 Cmax Cmax_D . Used mainly with steady-state data. Tlag Tmax Extravascular input (model 200) only. the entire curve is considered. For non-steady-state data.” on page 511 Table B-3. Otherwise. or the elapsed time since the time of the first dose. which may code time as the time elapsed since the first dose. Maximum observed concentration. Time of maximum observed concentration. occurring at Tmax. Tmax corresponds to points collected during a dosing interval. For non-infusion models: MRTlast = AUMClast/ AUClast. For infusion models: MRTlast = (AUMClast/AUClast) .B WinNonlin User’s Guide Table B-3. C(0) is set to C1. If the last concentration is non-zero AUClast=AUCall.Lambda_z * Tlast). Otherwise. _upper Note: Parameters that are extrapolated to infinity. zero. The values intercept and lambdaZ are those values found during the regression for Lambda_z. Table B-4. Area under the moment curve from the time of dosing (Dosing_time) to the last measurable concentration. Concentration corresponding to Tlast. the C(0) is estimated by log-linear regression of the first two time points. AUMClast MRTlast No_points_ Number of points used in computing λz. 508 . AUClower (Optional) user-requested area(s) under the curve from time lower to upper. but can be obtained by checking Output intermediate calculations in the Model Options. AUCall will be greater than AUClast as it includes the additional area from the last measurable concentration down to zero. If C2<C1. If λz cannot be estimated. Lambda_z AUC_%Bac Computed for Bolus IV (model 201) only: the percentage of AUCINF that was k_ Ext due to back extrapolation to estimate C(0). The value for intercept is not given in the normal output. which is defined as: Clast_pred = exp(intercept . Parameters that do not require λz (continued) Tlast Clast AUClast AUCall Time of last measurable (positive) concentration. otherwise.TI/2. Mean residence time from the time of dosing (Dosing_time) to the time of the last measurable concentration. in Table B-4 below. Parameters available only if λz is estimable Rsq Rsq_adjusted Goodness of fit statistic for the terminal elimination phase. are calculated two ways: Parameters called “*_obs” use the last observed value for Clast. and parameters called “*_pred” used the predicted value for Clast. Area under the curve from the time of dosing (Dosing_time) to the time of the last observation. Goodness of fit statistic for the terminal elimination phase. adjusted for the number of points used in the estimation of λz. Area under the curve from the time of dosing (Dosing_time) to the last measurable concentration. See also AUCINF_pred.. = Dose/AUCINF_obs. below. log concentration. = Vz_obs.WinNonlin Model Libraries Noncompartmental analysis Table B-4. r Lambda_z_uppe Upper limit on Time for values to be included in the calculation of λz.e. r HL_Lambda_z AUCINF_obs Terminal half-life = ln ( 2 ) ⁄ λ z AUC from Dosing_time extrapolated to infinity. based on the last predicted concentration. Estimated by linear regression of time vs. concentration at the final observation time estimated using the linear regression performed to estimate Lambda Z. AUC_%Extrap _obs Percentage of AUCINF_obs due to extrapolation from Tlast to infinity. based on the last observed concentration (obs). obs = AUClast + -------------------- Clast λz AUCINF_D_obs AUCINF_obs divided by dose.⋅ 100 AUCINF_obs Dose -----------------------------------------λ z ⋅ AUCINF_obs Volume of distribution based on the terminal phase. Lambda_z_lowe Lower limit on Time for values to be included in the calculation of λz. Percentage of AUCINF_pred due to extrapolation from Tlast to infinity. First order rate constant associated with the terminal (log-linear) portion of the curve. = AUCINF_pred – AUClast -------------------------------------------------------------. i.⋅ 100 AUCINF_pred 509 . Parameters available only if λz is estimable (continued) Corr_XY Lambda_z B Correlation between time (X) and log concentration (Y) for the points used in the estimation of λz. Vz_F_obsa AUCINF_obs – AUClast ----------------------------------------------------------. Cl_F_obsa AUCINF_pred Total body clearance for extravascular administration. = Cl_obs. pred = AUClast + ---------------------- Clast λz AUCINF_D_pre d AUC_%Extrap _pred AUCINF_pred divided by dose. AUC from Dosing_time extrapolated to infinity. based on the last observed concentration. i.B WinNonlin User’s Guide Table B-4. For non-infusion models = AUMCINF_pred/AUCINF_pred. MRTINF includes Mean Input Time as well as time in systemic circulation. MRTINF_pred 510 . _obs AUMCINF_obs – AUMClast = ---------------------------------------------------------------------. = Tlast ⋅ Clast obs Clast obs AUMClast + -----------------------------------. based on the last predicted concentration. where TI represents infusion duration. For non-infusion models = AUMCINF_obs/AUCINF_obs.TI/2.+ ---------------------2 λz λz AUMC_%Extrap Percent of AUMCINF_pred that is extrapolated. = Tlast ⋅ Clast pred Clast pred AUMClast + --------------------------------------.TI/2. Vz_F_preda Volume of distribution based on the terminal phase. = Dose/AUCINF_pred Area under the first moment curve (AUMC) extrapolated to infinity. = -------------------------------------------- Dose λ z ⋅ AUCINF_pred Cl_pred. Note that for oral model 200. concentration at the final observation time estimated using the linear regression for Lambda Z. MRTINF includes Mean Input Time as well as time in systemic circulation.+ -------------------2 λz λz AUMC_%Extrap Percent of AUMCINF_obs that is extrapolated. Note that for oral model 200. _pred AUMCINF_pred – AUMClast = -----------------------------------------------------------------------. For infusion models = (AUMCINF_pred/AUCINF_pred) .⋅ 100 AUMCINF_obs AUMCINF_pred Area under the first moment curve (AUMC) extrapolated to infinity.⋅ 100 AUMCINF_pred MRTINF_obs Mean residence time (MRT) extrapolated to infinity. Mean residence time (MRT) extrapolated to infinity..e. For infusion models = (AUMCINF_obs/AUCINF_obs) . where TI represents infusion duration. Parameters available only if λz is estimable (continued) Vz_pred. Cl_F_preda AUMCINF_obs Total body clearance for extravascular administration. Clss_Fa The (assumed equal) dosing interval for steady-state data. the fraction of dose absorbed cannot be estimated.– ---τ 2 AUC 0 τ τ τ τ Infusion: where TI represents infusion duration. For extravascular models (model 200). The following additions and changes are made for steady-state data. 511 . a. AUC from dose time to Tau divided by (dose time + Tau). Time of minimum concentration sampled during a dosing interval. Table B-5.WinNonlin Model Libraries Noncompartmental analysis Table B-4. Calculations for steady-state data Tau Tmin Cmin Cavg Accumulation Index Clss. Mean residence time (MRT) extrapolated to infinity. as MRTINF for oral models includes Mean Input Time as well as time in systemic circulation and therefore is not appropriate to use in calculating Vss. Non-infusion: AUMC 0 + τ ( AUCINF_obs – AUC 0 ) -------------------------------------------------------------------------------------------τ AUC 0 AUMC 0 + τ ( AUCINF_obs – AUC 0 ) TI -------------------------------------------------------------------------------------------. Not computed for model 200. Parameters available only if λz is estimable (continued) VSS_obs VSS_pred B An estimate of the volume of distribution at steady state = MRTINF*Clss. for Cmin and Cmax between dose time and Tau. %_Fluctuation = 100*(Cmax-Cmin)/Cavg. Minimum concentration between dose time and dose time + Tau (at Tmin). therefore Volume and Clearance for these models are actually Volume/F or Clearance/F where F is the fraction of dose absorbed. based on the last observed (obs) or last predicted (pred) concentration. = 1 ---------------------–λz τ 1–e Dose ---------------τ AUC 0 An estimate of the total body clearance = Cmax MRTINF_obs Maximum concentration between dose time and dose time + Tau (at Tmax). based on the last observed (obs) or last predicted (pred) concentration.– ---τ 2 AUC 0 τ τ Infusion: where TI represents infusion duration. For extravascular models (model 200). Vz_Fa = Dose ------------------------τ λ z ⋅ AUC 0 VSS_obs VSS_pred An estimate of the volume of distribution at steady state = MRTINF*Clss. Calculations for steady-state data (continued) MRTINF_pred Mean residence time (MRT) extrapolated to infinity.Starting time) 512 . Urine data Data structure NCA for urine data requires the following input data: • • • Starting and ending time of each urine collection interval Urine concentrations Urine volumes From this data. the fraction of dose absorbed cannot be estimated. Not computed for model 200.B WinNonlin User’s Guide Table B-5. Non-infusion: τ τ AUMC 0 + τ ( AUCINF_pred – AUC 0 ) ---------------------------------------------------------------------------------------------τ AUC 0 AUMC 0 + τ ( AUCINF_pred – AUC 0 ) TI ---------------------------------------------------------------------------------------------. a. models 210-212 compute the following for the analysis: • • Midpoint of each collection interval Excretion rate for each interval (amount eliminated per unit of time) = (Concentration * Volume) / (Ending time . Vz. as MRTINF for oral models includes Mean Input Time as well as time in systemic circulation and therefore is not appropriate to use in calculating Vss. therefore Volume and Clearance for these models are actually Volume/F or Clearance/F where F is the fraction of dose absorbed. Last measurable (positive) rate. No_points_lambda_z Number of points used in the computation of λz. First order rate constant associated with the terminal (log-linear) portion of the curve. Goodness of fit statistic for the terminal elimination phase. Area under the urinary excretion rate curve from time 0 to the last measurable rate. Parameters that do not depend on λz Dosing_time Tlag Tmax_Rate Max_Rate Mid_Pt_last Rate_last AURC_last Amount_ Recovered Vol_UR Percent_ Recovered AURC_all Time of the last administered dose (assumed to be zero unless otherwise specified). Table B-7. Maximum observed excretion rate. adjusted for the number of points used in the estimation of λz. 513 Lambda_z_lower . Midpoint of collection interval associated with the maximum observed excretion rate. Cumulative amount eliminated. Parameters that are estimated only when λz is estimable Rsq Rsq_adjusted Corr_XY Lambda_z Goodness of fit statistic for the terminal elimination phase. Table B-6. log excretion rates. Correlation between midpoints and log excretion rates for the points used in the estimation of λz. = Σ(Concentration * Volume) Sum of Urine Volumes 100*Amount_Recovered/Dose Area under the urinary excretion rate curve from time 0 to the last rate.WinNonlin Model Libraries Noncompartmental analysis B Output Models 210-212 estimate the following parameters. This equals AURC_last if the last rate is measurable. Lambda Z. This is estimated via linear regression of midpoints vs. AUClower_upper (Optional) partial area(s) under the curve from time lower to upper. Lower limit on midpoint for values to be included in λz estimation. Midpoint prior to the first measurable (non-zero) rate. Midpoint of collection interval associated with Rate_last. plus those listed below. SE_AUCall 514 . AURC_INF_pred AURC_%Extrap_pred Percent of AURC_INF_pred that is extrapolated. Parameters that are estimated only when λz is estimable Lambda_z_upper HL_Lambda_z AURC_INF_obs Upper limit on midpoint for values to be included in λz estimation.e. AURC_%Extrap _obs Percent of AURC_INF_obs that is extrapolated. Terminal half-life = ln(2) / λz Area under the urinary excretion rate curve extrapolated to infinity. based on the last observed excretion rate. SE_Cmax SE_AUClast Standard error of data at Tmax (time of maximum mean concentration). i.. The NCA engine computes the mean concentration or rate at each unique time value or interval. where Tlast is the time of last measurable (positive) mean concentration. Standard error of the area under the mean concentration curve from dose time to Tlast. Standard error of the area under the mean concentration curve from dose time to the final observation time.B WinNonlin User’s Guide Table B-7. based on the last predicted rate. See “Sparse sampling computations” on page 188 for additional details of NCA computations with sparse data. it estimates the same parameters normally calculated for plasma or urine data. Plasma or serum concentration parameters Sparse sampling methods for plasma data (models 200-202) compute the following parameters in addition to those listed in “Plasma or serum data” on page 507. Area under the urinary excretion rate curve extrapolated to infinity. Using the mean concentration curve across subjects. Note that AURC_INF is theoretically equal to the amount recovered but will differ due to experimental error. the data are treated as a special case of plasma or urine concentration data. excretion rate at the final midpoint estimated using the linear regression for λz. Sparse sampling (pre-clinical) data When an NCA model is loaded with the Sparse Sampling option (see “NCA model selection” on page 160). such as concentration-effect data. take care to interpret output parameters such as “Tmin” (independent variable value corresponding to minimum dependent variable value) accordingly. generally the time of each observation. along with N (number of nonmissing observations). the mean and standard error for each unique time value (plasma data) or unique midpoint value (urine data). i. SE_AURC_all Individuals by time This sheet includes the individual subject data.WinNonlin Model Libraries Noncompartmental analysis B Urine concentration parameters Sparse sampling methods for urine data (models 210-212) compute the following parameters in addition to those listed under “Urine data” on page 512. entered in the Dosing regimen Baseline response value. and the following constants: • • • Dose time (relative to observation times). SE_Max_Rate SE_AURC_last Standard error of the data at the time of maximum mean rate. entered in the Therapeutic response tab of the Model Properties 515 . For nontime-based data. Standard error of the area under the mean urinary excretion rate curve from dose time through the last interval that has a measurable (positive) mean rate.e. entered in the Therapeutic response tab of the Model Properties or pulled from a Dosing worksheet as described below Threshold response value (optional). Standard error of the area under the mean urinary excretion rate curve from dose time through the final interval. response or effect values (continuous scale) Independent variable. Drug effect data (model 220) Data structure NCA for drug effect data requires the following input data variables: • • Dependent variable.. the user must supply a baseline value (see “Therapeutic response” on page 168). as computed in NCA PK models. if no baseline is provided by the user. baseline). WinNonlin will use the response value at dose time as baseline. In that case. Like other NCA models. Partial areas are set up as described under “Partial areas” on page 184. 516 . If the dosing time is not specified (“Dosing regimen” on page 171). these slopes are reported as their actual value rather than their negative. If the data set does not include a measurement at dose time. Unlike Lambda Z. the user must enter a baseline response value unless working from a PKS study that includes a Dosing worksheet. T = user-supplied threshold effect value. the NCA PD model can calculate the slope of the time-effect curve for specific data ranges in each profile. Instead of Lambda Z calculations. the dosing time is assumed to be zero and WinNonlin will insert the point (0. even for inhibitory effects. See “Lambda Z or slope estimation settings” on page 162 for more information. • • • B = baseline effect value (discussed above).B WinNonlin User’s Guide By default. the PD model can compute partial areas under the curve. “Above” means towards increasing Y values. Estimated parameters Output parameter names use the following conventions. model 220 also computes the following. Minimum observed response value. for PKS studies with no user-supplied baseline value. Table B-9. = AUC_Above_B . = 100 * Time_Below_B/(Tfinal . effect value at dose time. See “Lambda Z or slope estimation settings” on page 162. Area under the response curve that is above the baseline (dark gray areas in the above diagram). Additional parameters for model 220 with threshold Threshold Threshold value used. Upper limit on Time for values to be included in the slope calculation. Rsq_Slope1 (or 2) Goodness of fit statistic for slope 1 or 2. or. (or 2) Slope1_lower or Slope2_lower Slope1_upper or Slope2_upper Rmax Rmin Time_Above_B Time_ Below_B Time_%Below_B Lower limit on Time for values to be included in the slope calculation.WinNonlin Model Libraries Noncompartmental analysis B Table B-8. When a threshold value is provided. Rsq_adj_Slope1 (or 2) Corr_XY_Slope1 (or 2) No_points_Slope1 The number of data points included in calculation of slope 1 or 2. Correlation between time (X) and effect (or log effect. Total time that Response < Baseline.Tdose) where Tfinal is the final observation time and Tdose is dosing time.AUC_Below_B. Slope of the first (or second) segment of the curve. adjusted for the number of points used in the estimation. Total time that Response >= Baseline. for log regression) (Y) for the points used in the slope estimation. Area that is below the baseline and above the response curve (combined blue and pink areas in the above diagram). Maximum observed response value. This is likely to be a negative value for inhibitory effects. Goodness of fit statistic for slope 1 or 2. Estimated parameters for model 220 Baseline AUC_Above_B AUC_Below_B AUC_Net_B Slope1 (or 2) Baseline response (Y) value supplied by the user. 517 . If the log trapezoidal rule is appropriate for the area under a curve.) Use caution with log trapezoidal since areas are calculated both under the curve and above the curve. relative to baseline. the default and recommended method for AUC calculation is the Linear Trapezoidal with Linear Interpolation (set as described under “NCA settings” on page 177.)a Time greater than Tonset at which the curve first crosses back to the baseline side of threshold. Area that is below the threshold and above the response curve (pink area in the above diagram). Time of minimum observed response value (Rmin). as shown in the diagram above. a. (See “NCA settings” on page 177. Use caution in interpreting Tonset and Toffset for noisy data if Baseline and Threshold are close together. Note: For PD data.Tonset Time_Between_BT Total time spent between baseline and threshold (sum of length of green arrows in diagram).Tdose) where Tlast is the final observation time and Tdose is dosing time. the program will use linear trapezoidal for area where the curve is below baseline when computing AUC_Below_B.AUC_Below_T. = AUC_Above_T . = 100 * Time_Below_T/(Tlast . 518 . Total time that Response < Threshold. Tonset = Tdose if the first response value is across the threshold. (Optional) user-requested area(s) under the curve from time lower to time upper. Total time that Response >= Threshold. as shown in the above diagram. The time will be interpolated using the calculation method selected in the model options. Additional parameters for model 220 with threshold (continued) AUC_Above_T AUC_Below_T AUC_Net_T Time_Above_T Time_Below_T Time_%Below_T Tonset Area under the response curve that is above the threshold value (combined light and dark gray areas in the above diagram). and similarly for threshold and AUC_Below_T.B WinNonlin User’s Guide Table B-9.a Toffset Diff_Toffset_Tonset = Toffset . Tmax Tmin AUClower_upper Time of maximum observed response value (Rmax). then it would underestimate the area over the curve. For this reason. Time that the response first crosses the threshold coming from the direction of the baseline value. as follows. Each is available in both compiled and ASCII text forms.WinNonlin Model Libraries Pharmacokinetic models B Pharmacokinetic models The WinNonlin Model Libraries include 19 pharmacokinetic (PK) models. 519 .LIB.LIB . and also in the ASCII library files PK1.PK10. Model Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Model 10 Model 11 Model 12 Model 13 Model 14 Model 15 Model 16 Model 17 Model 18 Model 19 Input IV-bolus IV-infusion 1st order 1st order 1st order 1st order IV-bolus IV-bolus IV-infusion IV-infusion 1st order 1st order 1st order 1st order IV-bolus plus IV-infusion IV-bolus plus IV-infusion IV-bolus plus IV-infusion IV-bolus IV-infusion Number of compartments 1 1 1 1 1 1 2 2 2 2 2 2 2 2 1 2 2 3 3 Parameterization Lag time no no no yes K10 = K01 no K10 = K01 yes micro no macro no micro no macro no micro no micro yes macro no macro yes micro no micro macro macro macro no no no no Elimination rate 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order Note: All models except models 15-17 accept multiple dose data. ASCII models and files WinNonlin PK models are available as compiled models. The models shown in this section give equations for compartment 1 (central). 8 9. 2 3.B WinNonlin User’s Guide Model # 1. 10 11.LIB PK6. 6 7. 16 17.LIB PK9. 18 19 File name PK1. 4 5.LIB PK5.LIB Model 1 One compartment with bolus input and first-order output C (T ) = D/V C (T ) = (D/V) exp(-K10 · T ) Required constants Estimated parameters V = Volume K10 = elimination rate Secondary parameters AUC = D/V/K10 K10 half-life Cmax = D/V CL AUMC MRT Vss Clearance estimated V CL Clearance secondary AUC K10 half-life Cmax K10 AUMC MRT VSS N doses dose N time of dose N (Repeat for each dose) Model 2 One-compartment with constant IV input.LIB PK8. first-order absorption 520 .LIB PK10. 12 13.LIB PK7.LIB PK3.LIB PK2.LIB PK4. 14 15. WinNonlin Model Libraries Pharmacokinetic models B 1 D C ( T ) = ⎛------.[ exp ( – K 10 TSTAR ) – exp ( – K 10 T ) ] ⎝ TI ⎠ VK 10 where TI = infusion length.[ exp ( – K 10 T ) – exp ( – K 01 T ) ] V ( K 01 – K 10 ) Required constants Estimated parameters V_F Secondary parameters AUC = D/V/K10 Clearance Clearance estimated secondary V_F K01 CL_F AUC K01 half-life K10 half-life K10 Tmax N doses dose N (Repeat for each dose) K01=absorption rate K01 half-life CL_F Tmax = time of Cmax time of dose N K10=elimination rate K10 half-life ln ( K01 ⁄ K10 ) = --------------------------------( K01 – K10 ) Cmax = max concentration Cmax 521 . Required constants Estimated parameters V = Volume K10 = elimination rate Secondary parameters K10 half-life Cmax = C(TI) CL AUMC MRT Vss Clearance estimated CL Clearance secondary AUC K10 half-life Cmax K10 AUMC MRT VSS N doses dose N start time N end time N (Repeat for each dose) AUC = D/V/K10 V Model 3 One-compartment with first-order input and output. TSTAR = 0 for T≤TI. no lag time DK 01 C ( T ) = ------------------------------.⎞ ----------. TSTAR = T-TI for T>TI. equal first-order input and output with lag time Identical to Model 5. with an additional estimated parameter: Tlag = lag time. Model 5 One-compartment. 522 .B WinNonlin User’s Guide Model 4 One-compartment with first-order input and output with lag time Identical to Model 3. equal first-order input and output. with an additional estimated parameter: Tlag = lag time. no lag time C (T ) = (D/V) K · T · exp(-K · T ) Required constants Estimated parameters V_F elimination rate Secondary parameters AUC = D/V/K CL_F Tmax = time of Cmax = 1/K Cmax = max concentration Clearance estimated V_F CL_F Clearance secondary AUC K half-life K Tmax Cmax N doses dose N time of dose K=absorption and K half-life N (Repeat for each dose) Model 6 One-compartment. K21)/(ALPHA .BETA) B = (D/V) (BETA .WinNonlin Model Libraries Pharmacokinetic models B Model 7 Two-compartment with bolus input and first-order output.BETA) where -ALPHA and -BETA (ALPHA > BETA) are the roots of the quadratic equation: (r · r + (K12 + K21 + K10) · r + K21 · K10 = 0). 2 to 1 BETA ALPHA half-life BETA half-life A B Cmax = D/V CL AUMC MRT Vss V2 CLD2 523 . micro-constants as primary parameters C (T ) = A exp(-ALPHA · T )+B exp(-BETA · T ) where: A = (D/V) (ALPHA .K21)/(ALPHA . 1 to 2 ALPHA K21 = transfer rate. Required constants Estimated parameters V1 = Volume1 K10=elimination rate Secondary parameters AUC = D/V/K10 K10 half-life Clearance estimated V1 CL V2 CLD2 Clearance secondary AUC K10 half-life ALPHA BETA ALPHA half-life BETA half-life A B Cmax K10 AUMC MRT Vss K12 K21 N doses dose N (Repeat for each dose) time of dose N K12 = transfer rate. TSTAR = 0 for T≤TI. TSTAR = T-TI for T>TI.exp(-BETA · TSTAR)] where TI = infusion length. macro-constants as primary parameters As Model 7 with macroconstants as the primary (estimated) parameters. D ( K21 – ALPHA ) A 1 = ------------.B WinNonlin User’s Guide Model 8 Two-compartment with bolus input and first-order output.exp(-ALPHA · TSTAR)] + B1[exp(-BETA · T ) . microconstants as primary parameters C (T ) = A1[exp(-ALPHA · T ) .-----------------------------------------------------------------TI ( V ) ( ALPHA – BETA )ALPHA 524 . Model 9 Two-compartment with constant IV input and first-order output. Required constants stripping dose Estimated parameters Secondary parameters A B ALPHA BETA AUC = A/ALPHA + B/BETA V1 K10 half-life ALPHA half-life BETA half-life K10 K12 K21 Cmax CL AUMC MRT Vss V2 CLD2 N doses dose N time of dose N (Repeat for each dose) Clearance parameters are not available for Model 8. 525 . 1 to 2 K21 = transfer rate.------------------------------------------------------------TI ( V ) ( ALPHA – BETA )BETA and -ALPHA and -BETA (ALPHA > BETA) are the roots of the quadratic equation: r · r + (K12 + K21 + K10) · r + K21 · K10 = 0. Required constants Estimated parameters V1 = Volume1 K10 = elimination rate K12 = transfer rate. Model 10 Two-compartment with constant IV input and first-order output.WinNonlin Model Libraries Pharmacokinetic models B –D ( K21 – BETA ) B 1 = ------------. macroconstants as primary parameters As Model 9 with macroconstants as the primary (estimated) parameters. 2 to 1 Secondary parameters K10 half-life ALPHA BETA ALPHA half-life BETA half-life A B Cmax = C(TI) CL AUMC MRT Vss V2 CLD2 Clearance estimated CL V2 CLD2 Clearance secondary AUC K10 half-life ALPHA BETA ALPHA half-life BETA half-life A B Cmax K10 AUMC MRT Vss K12 K21 N doses dose N start time N end time N (Repeat for each dose) AUC = D/V/K10 V1 A and B are the zero time intercepts following an IV injection. B WinNonlin User’s Guide Required constants Estimated parameters V1 K21 ALPHA BETA Secondary parameters K10 K12 K10 half-life AUC ALPHA half-life BETA half-life A B Cmax CL AUMC MRT Vss V2 CLD2 N doses dose N start time N end time N (Repeat for each dose) A and B are the zero time intercepts following an IV injection. Clearance parameters are not available in Model 10. first-order output. no lag time and micro-constants as primary parameters C (T ) = A exp(-ALPHA · T ) + B exp(-BETA · T ) + C exp(-K01 · T ) where: ⎛ D⎞ K01 ( K21 – ALPHA ) --⎝ V⎠ A = --------------------------------------------------------------------------------------( ALPHA – BETA ) ( ALPHA – K01 ) ⎛ – D⎞ K01 ( K21 – BETA ) --⎝ V⎠ B = ---------------------------------------------------------------------------------( ALPHA – BETA ) ( BETA – K01 ) 526 . Model 11 Two-compartment with first-order input. first-order output. 2 to 1 BETA ALPHA half-life BETA half-life A B CL_F V2_F CLD2_F Tmaxa Cmaxa a. 527 .WinNonlin Model Libraries Pharmacokinetic models B ⎛ D⎞ K01 ( K21 – K01 ) --⎝ V⎠ C = ----------------------------------------------------------------------------( BETA – K01 ) ( ALPHA – K01 ) and -ALPHA and -BETA (ALPHA > BETA) are the roots of the quadratic equation: r · r + (K12 + K21 + K10) · r + K21 · K10 = 0. Estimated for the compiled model only. with an additional estimated parameter: Tlag = lag time. Required constants Estimated parameters V1_F K01 = absorption rate K10 = elimination rate Secondary parameters K01 half-life K10 half-life Clearance Clearance estimated secondary AUC K01 half-life K10 half-life ALPHA BETA ALPHA half-life BETA half-life A B K10 K12 K21 Tmax Cmax K01 CL_F V2_F CLD2_F N doses dose N time of dose N (Repeat for each dose) AUC = D/V/K10 V1_F K12 = transfer rate. Model 12 Two-compartment with first-order input. 1 to 2 ALPHA K21 = transfer rate. lag time and micro-constants as primary parameters As Model 11. no lag time and macro-constants as primary parameters As Model 11. first-order output. Model 15 One compartment with simultaneous bolus IV and constant IV infusion CB (T ) = (Db/V) exp(-K10 · T ) D IV 1 C IV ( T ) = ⎛ -------. Estimated for the compiled model only. with macroconstants as the primary (estimated) parameters. first-order output.⎞ -----------------. with an additional estimated parameter: TLAG = lag time.B WinNonlin User’s Guide Model 13 Two-compartment with first-order input. BETA half-life V1_F CL_F V2_F CLD2_F Tmaxa Cmaxa N doses dose N time of dose N (Repeat for each dose) Clearance parameters are not available for Model 13.[ exp ( – K10 ⋅ TSTAR ) – exp ( – K10 ⋅ T ) ] ⎝ TI ⎠ V ( K10 ) 528 . Required constants Stripping dose Estimated parameters A B K01 = absorption rate ALPHA BETA Note: C = -(A + B) Secondary parameters K10 K12 K21 AUC = D/V/K10 K01 half-life K10 half-life ALPHA half-life a. lag time and macro-constants as primary parameters As Model 13. Model 14 Two-compartment with first-order input. WinNonlin Model Libraries Pharmacokinetic models B C (T ) = CB (T ) + CIV (T ) where TI = infusion length. micro constants as primary parameters CB (T ) = A1 exp(-ALPHA · T ) + B1 exp(-BETA · T ) where: D B ( ALPHA – K21 ) A 1 = -----.------------------------------------------------------------TI ( V ) ( ALPHA – BETA )BETA 529 . TSTAR = T-TI for T>TI.exp(-BETA · TSTAR)] where: D IV ( K21 – ALPHA ) A 2 = ------------.exp(-ALPHA · TSTAR)] + B2[exp(-BETA · T ) .--------------------------------------------V ( ALPHA – BETA ) and: CIV (T ) = A2[exp(-ALPHA · T ) .-----------------------------------------------------------------TI ( V ) ( ALPHA – BETA )ALPHA – D IV ( K21 – BETA ) B 2 = ------------.--------------------------------------------V ( ALPHA – BETA ) – D B ( BETA – K21 ) B 1 = --------. TSTAR = 0 for T≤TI Required constants bolus dose IV dose length of infusion = TI Estimated Secondary parameters parameters V = Volume K10 CL K10 half-life Clearance Clearance estimated secondary V CL K10 K10 half-life Model 16 Two compartment with simultaneous bolus IV and constant infusion input. Required constants bolus dose IV dose length of infusion (= TI) Estimated parameters V1 K10 K12 K21 Secondary parameters K10 half-life ALPHA BETA ALPHA half-life BETA half-life A B CL V2 CLD2 Clearance estimated V1 CL V2 CLD2 Clearance secondary K10 half-life ALPHA BETA ALPHA half-life BETA half-life A B K10 K12 K21 Model 17 Two compartment with simultaneous bolus IV and constant infusion input and macro constants as primary parameters As Model 16 with macroconstants as the primary (estimated) parameters. 530 . TSTAR = T-TI for T>TI.B WinNonlin User’s Guide where TI = infusion length. Required constants stripping dose bolus dose IV dose length of infusion (= TI) Estimated parameters A B ALPHA BETA K10 K12 K21 K10 half-life ALPHA half-life Secondary parameters BETA half-life V1 CL V2 CLD2 Clearance parameters are not available in Model 17. TSTAR = 0 for T≤TI C (T ) = CB (T ) + CIV (T ) and -ALPHA and -BETA (ALPHA > BETA) are the roots of the quadratic equation: r · r + (K12 + K21 + K10) · r + K21 · K10 = 0. WinNonlin Model Libraries Pharmacokinetic models B Model 18 Three-compartment with bolus input. 531 . first-order output and macro constants as primary parameters C (T ) = A exp(-ALPHA · T ) + B exp(-BETA · T ) + C exp(-GAMMA · T ) Required constants stripping dose Estimated parameters A B C ALPHA BETA GAMMA Cmax V1 K21 K31 K10 K12 K13 K10 half-life ALPHA half-life BETA half-life Secondary parameters GAMMA half-life AUC CL AUMC MRT Vss V2 CLD2 V3 CLD3 N doses dose N time of dose N (Repeat for each dose) Clearance parameters are not available in Model 18. TSTAR = T-TI for T>TI.exp(-GAMMA · T )] where: D ⎛ ------.exp(-ALPHA · T )] + B1 [exp(-BETA · TSTAR ) . TSTAR = 0 for T≤TI Required constants Estimated parameters V1 K21 K31 Secondary parameters Cmax K10 K12 K10 half-life ALPHA half-life BETA half-life MRT Vss V2 N doses dose N start time of dose N 532 .⎞ ( K21 – ALPHA ) ( K31 – ALPHA ) ⎝ TI ⎠ A 1 = --------------------------------------------------------------------------------------------------------------------------------V ( ALPHA ) ( GAMMA – ALPHA ) ( BETA – ALPHA ) D ⎛ -------⎞ ( K21 – BETA ) ( K31 – BETA ) ⎝ TI ⎠ B 1 = ----------------------------------------------------------------------------------------------------------------------V ( BETA ) ( GAMMA – BETA ) ( ALPHA – BETA ) D ⎛ -------⎞ ( K21 – GAMMA ) ( K31 – GAMMA ) ⎝ TI ⎠ C 1 = --------------------------------------------------------------------------------------------------------------------------------------V ( GAMMA ) ( ALPHA – GAMMA ) ( BETA – GAMMA ) where TI = infusion length.exp(-BETA · T )] + C1 [exp(-GAMMA · TSTAR ) .B WinNonlin User’s Guide Model 19 Three compartment model with constant IV infusion. macro constants as primary parameters C (T ) = A1 [exp(-ALPHA · TSTAR ) . Pharmacodynamic models Load and set up pharmacodynamic (PD) models as described under “Loading the model” on page 193 and “Model properties” on page 196. B and C are the zero time intercepts following an IV injection.WinNonlin Model Libraries Pharmacodynamic models Required constants end time of dose N (Repeat for each dose) Estimated parameters ALPHA BETA GAMMA Secondary parameters K13 A B C GAMMA half-life AUC CL AUMC CLD2 V3 CLD3 B Clearance parameters are not available in Model 19. 102 103. Available PD models in the WinNonlin Model Libraries include the following. 104 105. Model # 101. Model Model 101 Model 102 Model 103 Model 104 Model 105 Model 106 Model 107 Model 108 Description Simple Emax Model Simple Emax Model Inhibitory Effect Emax Model Inhibitory Effect Emax Model Sigmoid Emax Model Sigmoid Emax Model Effect at C=0 0 E0 Emax Emax 0 E0 Effect at C=infinity Emax Emax 0 E0 Emax Emax 0 E0 Inhibitory Effect Sigmoid Emax Model Emax Inhibitory Effect Sigmoid Emax Model Emax ASCII library These models are available as compiled models and also as ASCII library files.LIB PK53. The ASCII library files are stored in the WINNONLN\LIB subdirectory.LIB 533 . 106 File name PK51.LIB PK52. A. B WinNonlin User’s Guide Model # 107. EC50 Secondary parameter: Emax .LIB Table B-10. 108 File name PK54. EC50 Model 102 Simple Emax model with a baseline effect parameter E = E0 + (Emax .E0 ) [C / (C + EC50)] Estimated parameters: Emax.E0) Drug-induced maximal effect Shape parameter Model 101 Simple Emax model E = (Emax · C) / (C + EC50) Estimated parameters: Emax.E0 Gamma Definition Maximum effect Baseline effect Drug concentration required to produce 50% of the maximal effect (Emax. Model notation Term Emax E0 EC50 Emax. E0.E0 534 . EC50 .WinNonlin Model Libraries Pharmacodynamic models B Model 103 Inhibitory effect model E = Emax [1 . EC50 Model 104 Inhibitory effect model with a baseline effect parameter E = Emax . EC50. Gamma 535 . E0 Secondary parameter: Emax .(C / (C + EC50))] Estimated parameters: Emax.E0 ) [C / (C + EC50)] Estimated parameters: Emax.E0 Model 105 Sigmoid Emax model E = (Emax · Cγ ) / (Cγ + ECγ 50)] Estimated parameters: Emax.(Emax . EC50. Gamma Model 108 Sigmoid inhibitory effect model with a baseline effect parameter E = Emax . E0. E0.E0) [ Cγ / (Cγ + ECγ 50)] Estimated parameters: Emax.E0) [ Cγ/ (Cγ +ECγ 50)] Estimated parameters: Emax.E0 PK/PD link models When pharmacological effects are seen immediately and are directly related to the drug concentration. Gamma Secondary parameter: Emax .Cγ / (Cγ + ECγ 50)] Estimated parameters: Emax. EC50. a pharmacodynamic model is applied to characterize the relationship between drug concentrations and effect.(Emax . Gamma Secondary parameter: Emax .B WinNonlin User’s Guide Model 106 Sigmoid Emax model with a baseline effect parameter E = E0 + (Emax . When the pharmacologic response takes time to develop and the observed response is not 536 . EC50.E0 Model 107 Sigmoid inhibitory effect model E = Emax [1 . To link any PK and PD models: 1. 537 . Model parameter information will be required for the PK model in order to simulate the concentration data. and generate concentrations at the effect site to be used by the PD model. see “Simultaneous PK/PD link models” on page 545. Thus.WinNonlin Model Libraries PK/PD link models B directly related to plasma concentrations of the drug a link model is usually applied to relate the pharmacokinetics of the drug to its pharmacodynamics. the PK data are not modeled. Effect at time 0 Rate of drug loss from the effect compartment. 2. Click the PK/PD/NCA Analysis Wizard button on the tool bar or choose Tools>PK/PD/NCA Analysis Wizard from the menus. Notation Term Emax ECe50 E0 Ke0 Emax–E0 Gamma Maximum effect Concentration required to produce one-half of the maximal drug effect. The PK model is used to predict concentrations. and these concentrations are then used as input to the PD model. Drug induced maximal effect Shape parameter Definition Linking PK and PD models WinNonlin can generate concentrations at effect sites using any PK model in the compiled library. The Ke0 T1/2 is the half life of the amount of time required to collapse the hysteresis curve. the PK/PD Link models treat the pharmacokinetic parameters as fixed. Note: To fit PK and PD data simultaneously. The PK/PD Link models can use any combination of WinNonlin compiled Pharmacokinetic models and Pharmacodynamic models. and using these concentrations in any PD model. Select PK/PD Link in the Model Types dialog. Output parameters The output parameters for the PD models differ from the parameters for these models when they are not linked. The Selection Summary dialog appears. Select the PD Model tab. one for the PK Model and one for the PD Model. Enter initial estimates and bounds. 538 . Notice that the Model Parameters dialog now has a two tabs. 5. 7. Click Finish. Click the Model Parameters button or choose Model>Model Parameters from the menus. Click OK. Note: The units for PK concentration can be entered in the PK Concentration Unit field or set using the Units Builder button. and model options as for other compartmental models. Model Parameters. Enter the Data Variables. To assign parameter values for the PK model: 1. 4. 5. 2. The initial parameter values must be entered. 4. 3. Click Next. Dosing Regimen/ Constants. Click Apply or OK. The model parameters are set and the dialog closes. as described under “Model properties” on page 196. To assign initial estimates and bounds for the PD model: 1. 6. Select a PD model from those detailed under “Pharmacodynamic models” on page 533. Enter the parameter values.B WinNonlin User’s Guide 3. Select a PK model from those detailed under “Pharmacokinetic models” on page 519. 3. Click Next. Table B-11 lists the PD model parameters. Select the PK Model tab. 2. 539 . E0. Gamma. When the pharmacologic response takes time to develop and the observed response is not directly related to plasma concentrations of the drug a link model is usually applied to relate the pharmacokinetics of the drug to its pharmacodynamics.WinNonlin Model Libraries Indirect response models B Table B-11. and Jusko (1993). ECe50. KEO. ECe50. Link model output parameters PD model 101 Simple Emax 102 Simple Emax with baseline effect 103 Inhibitory effect 104 Inhibitory effect with a baseline effect 105 Sigmoid Emax 106 Sigmoid Emax with a baseline effect 107 Sigmoid inhibitory effect 108 Sigmoid inhibitory effect model with a baseline effect Estimated parameters Emax. ECe50. Garg. Four basic models have been developed for characterizing indirect pharmacodynamic responses after drug administration. ECe50. KE0 Emax. E0. Gamma. ECe50. KE0 Emax. ECe50. KE0 Emax. KE0 Emax-E0 Emax-E0 Emax-E0 Emax-E0 Secondary parameter Indirect response models When pharmacological effects are seen immediately and are directly related to the drug concentration. E0. KE0 Emax. and then use these concentrations as input to the indirect response model. One kind of link model available in WinNonlin is the family of indirect pharmacodynamic response (IPR) models proposed by Jusko and co-workers Jusko (1990) and Dayneka. These models are based on drug effects (inhibition or stimulation) on the factors controlling either the input or the dissipation of drug response. KE0 Emax. ECe50 Emax. KE0 Emax. a pharmacodynamic model is applied to characterize the relationship between drug concentrations and effect. E0. Gamma. ECe50. Gamma. The indirect response models differ from other models in the WinNonlin Model Libraries in that they use a PK model to predict concentrations. = k in – k out ⎛ 1 – ---------------------.⎞ R ⎝ dt C p + IC 50⎠ Estimated parameters Kin Kout IC50 540 .B WinNonlin User’s Guide Note: These models are available only as compiled models Models Notation Term R kin kout Cp Ce IC50 Emax EC50 Measured response to a drug The zero order constant for the production of response The first order rate constant for loss of response Plasma concentration of a drug Drug concentration at the effect site The drug concentration that produces 50% of maximum inhibition Maximum effect The drug concentration that produces 50% of maximum stimulation Definition Model 51 Inhibition of input Cp dR = k ⎛ 1 – ---------------------.⎞ – k R ----in ⎝ out dt C p + IC 50⎠ Estimated parameters Kin Kout IC50 Model 52 Inhibition of output Cp dR ----. 5. displaying the Model Types dialog. Select Indirect Response from the selections in the Compartmental Models group box. 6. If everything described in this summary box is correct. The Selection Summary box appears. (Otherwise.= k in ⎛ 1 + -----------------------. Select a PK model. Click Next. 7. Click Next. 541 .⎞ R ----in out ⎝ dt C p + EC 50⎠ Estimated parameters Kin Kout EC50 Emax Running an indirect response model To select an Indirect Response model: 1.) The modeling window appears. 3. The PK/PD Analysis Wizard appears. The list of Indirect Response Models appears. 4. The WinNonlin Compiled Models dialog appears. Select an Indirect Response model. 2.WinNonlin Model Libraries Indirect response models B Model 53 Stimulation of input Emax ⋅ C p dR ----. click Back and return to correct the mistake. click Finish. Click the PK/PD/NCA Analysis Wizard button on the tool bar or select the PK/PD/NCA Analysis Wizard from the Tools menu.⎞ – k out R ⎝ dt C p + EC 50⎠ Estimated parameters Kin Kout EC50 Emax Model 54 Stimulation of output Emax ⋅ C p dR = k – k ⎛ 1 + -----------------------. 3. and Model Options precisely as for other models. 6. Note: Initial parameter values for the PK model must be specified. It is possible to choose to have WinNonlin generate the initial parameter values or to specify user-supplied initial values. Select the PK Model tab located under the parameters table. To assign initial estimates and bounds for the IPR model: 1. Dosing Regimen/ Constants. 5. 2. Select the IPR Model tab in the Model Properties dialog. 4. Michaelis-Menten models The WinNonlin library contains four Michaelis-Menten models: 542 . Enter the parameter values for the PK model. Click Apply or OK to specify the settings and close the dialog. (Optional) Enter the PK units in the PK Concentration Unit field. 2. Enter initial estimates and bounds. Dosing Regimen/Constants.B WinNonlin User’s Guide Note: Enter the Data Variables. as described under “Model properties” on page 196. Initial estimates and bounds can be specified for the IPR model. and model options as for other compartmental models. Notice that this dialog now has tabs for both the PK Model and the indirect response model. Click Apply. Model Parameters. 3. To assign parameter values for the PK model: 1. Click the Model Parameters button or choose Model>Model Parameters from the menus. Enter the Data Variables. times in the data set must correspond to the dosing times. be sure to examine the ASCII code to determine which dosing constants need to be created. WinNonlin assumes that the time of the first dose is zero. even if these times contain no observations. Required constants Estimated parameters V = Volume VM = max elimination rate KM = Michaelis constant Secondary parameters AUC = (D/V)(D/V/2 + KM)/ VMAX N doses Dose N Time of dose N 543 . with initial condition C(0) = D/V. Model 301 One-compartment with bolus input and Michaelis-Menten output C (T ) is the solution to the differential equation: dC/dt = (-VMAX · C)/( KM + C ). For these models. As with all WinNonlin ASCII models. include a column for Weight on the data set. In this case.WinNonlin Model Libraries Michaelis-Menten models B Model Description Model 301 One compartment with bolus input and Michaelis-Menten output Model 302 One compartment with constant IV input and Michaelis-Menten output Model 303 One compartment with first order input. These models parameterize Vmax in terms of concentration per unit time. For discussion of difficulties inherent in fitting the Michaelis-Menten models see Tong and Metzler (1980) (1980) and Metzler and Tong (1981) (1981).LIB. Michaelis-Menten output and a lag time These models are stored as ASCII library files in \LIB\MM. and weight these observations as 0. Michaelis-Menten output and no time lag Model 304 One compartment with first order input. with initial condition C(0) = 0.B WinNonlin User’s Guide Model 302 One-compartment with constant IV input and Michaelis-Menten output C (T ) is the solution to the differential equations: dC/dt = (D/TI/V) .VM · C /(KM+C ).VMAX · C /(KM+C ) for T <=TI. with initial condition C(0) = 0. dC/dt = . Required constants Estimated parameters V = Volume VM = max elimination rate KM = Michaelis constant Secondary parameters (none) N doses Dose N Start time for dose N End time for dose N Model 303 One-compartment with first-order input. Required constants Estimated parameters V_F K01 = absorption rate VM = max elimination rate KM = Michaelis constant Secondary parameters K01 half-life N doses Dose N Time of dose N 544 .VMAX · C /(KM+C ) for T > TI. Michaelis-Menten output and no lag time C (T ) is the solution to the differential equation: dC/dt = K01 · (D/V) exp(-K01 · T ) . . Time 0 0. Data The data for these models must contain the concentration and effect data in the same column. Simultaneous PK/PD link models WinNonlin has a library of link models that simultaneously fit PK and PD data... 1 148 145 123 .... but easy to rewrite them for a different PD model. 1 2 2 2 . An example is shown below. Function 2 should include the time and effect data. with a function variable in a separate column to distinguish the two. As it is difficult to rewrite these models for changes in the PK model. 2 ASCII library These models are stored as ASCII library files in the \LIB subdirectory.. be sure to examine the ASCII code to see what dosing constants need to be created. this library includes each PK models with PD Model 105. 36 0 0. As with all of the WinNonlin ASCII models. 36 Response 0 5 15 .WinNonlin Model Libraries Simultaneous PK/PD link models B Model 304 One-compartment with first-order input..5 1 ..5 1 .. Michaelis-Menten output and lag time As Model 303. Function 1 should include the time and concentration data. 545 . with one additional estimated parameter: Tlag = lag time... These models are described below. 121 Function 1 1 1 . bolus input + constant infusion 2 compartment . equal input and output with a lag time 2 compartment . defined in macroconstants 2 compartment .bolus input. includes a lag time 2 compartment .bolus input. defined in macroconstants 2 compartment .extravascular input. defined in macroconstants. defined in microconstants 2 compartment .LIB KEO2.1st order input. defined in macroconstants 3 compartment .bolus input + constant infusion.LIB KEO7.1st order input.LIB Model 19 Linear models WinNonlin includes a selection of models that are linear in the parameters.LIB KEO2.LIB PK model Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Model 10 Model 11 Model 12 Model 13 Model 14 Model 15 Model 16 Model 17 PK model description 1 compartment .LIB KEO5.1st order input. defined in macroconstants.LIB KEO3.extravascular input 1 compartment .extravascular input.bolus input 1 compartment .LIB KEO4. defined in microconstants 2 compartment . defined in microconstants 2 compartment .LIB KEO1. defined in macroconstants 2 compartment .LIB KEO6. equal input and output rates 1 compartment .LIB KEO6.LIB KEO5. defined in macroconstants KEO10. defined in macroconstants 3 compartment . 546 .constant IV input 1 compartment .constant infusion.bolus input + constant infusion.LIB KEO4.LIB KEO8.constant IV input. with a lag time 1 compartment .extravascular input. includes a lag time 1 compartment .bolus input.LIB KEO9.B WinNonlin User’s Guide Model 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 File KEO1.LIB KEO3.1st order input.LIB KEO8.LIB KEO7.LIB Model 18 KEO10.constant IV input. defined in microconstants 2 compartment . See also “Linear Mixed Effects Modeling” on page 323 for more sophisticated linear models. Note: The default computation method. Gauss-Newton (Hartley) (Model Options>PK Settings). Sigmoidal/dissolution models If you have purchased and installed an IVIVC Toolkit for WinNonlin license. 547 .WinNonlin Model Libraries Sigmoidal/dissolution models B Table B-12. The linear models require no dosing information (constants). See “Units” on page 208 to set preferred units for the output parameters. Linear models Plot Model # Name 501 Constant Model Estimated parameters (units) y(t) = constant CONSTANT (y-unit) 502 Linear y(t) = INT + SLOPE * t INT (y-unit) SLOPE (y-unit / t-unit) 503 Quadratic y(t) = A0 + A1*t + A2*t2 A0 (y-unit) A1 (y-unit / t-unit) A2 (y-unit / t-unit2) 504 Cubic y(t) = A0 + A1*t + A2*t2 + A3*t3 A0 (y-unit) A1 (y-unit / t-unit) A2 (y-unit / t-unit2) A3 (y-unit / t-unit3) Where y-unit is the preferred unit of the y (concentration) variable and t-unit is the preferred unit of the t (time) variable. Double Weibull and Makoid-Banakar. the WinNonlin model library includes the following models for smoothing dissolution data as part of analysis of in vivo-in vitro correlations (IVIVC): Hill. Weibull. and the model option “Do Not Use Bounds” (Model Options>Model Parameters) are recommended for all linear models. i. you may need to fix a parameter while estimating the others. especially the slope factor b. using the preferred units for y.tlag)b) where U(x) = x if x >= 0.tlag)b) / (MDTb + U (t .int)*tb) / (MDTb + t b) where the estimated parameters are: Finf = amount released at time infinity.e. using the preferred units for y MDT = mean dissolution time. It may not be possible to estimate all parameters simultaneously. Model 601 Hill y(t) = int + ((Finf . The IVIVC models require no dosing (constants). zero (default) or estimated intercept.: y(t) = int + ((Finf . This model includes an optional lag time in the form of a step function. See “Units” on page 208 to set preferred units for the output. in the preferred units for x (time) b = slope factor (no units) int = y-intercept.B WinNonlin User’s Guide For these models. 548 . the parameter Tlag may be strongly correlated with other parameters.int)*U(t . Model parameters can be estimated or set to a fixed value. See “Model parameters” on page 197 for details. U(x) = 0 if x < 0. WinNonlin Model Libraries Sigmoidal/dissolution models B Model 602 Weibull y(t) = int + (Finf .: y(t) = int + (Finf . This model includes an optional lag time in the form of a step function.int) * (1 . Model 603 Double Weibull y(t) = int + f1 * (Finf .exp[-(t / MDT)b]) where the estimated parameters are: Finf = amount released at time infinity. (x units) 549 .e.tlag) / MDT)b]) where U(x) = x if x >= 0. zero (default) or estimated intercept.int) * (1 . setting the fraction due to each Weibull (no units) Finf = amount released at time infinity. MDT2 = mean dissolution time for each Weibull. i.int) (1-exp[(-t / MDT1)b1]) + (1-f1) * (Finf .int) (1-exp[(-t / MDT2)b2]) where the estimated parameters are: f1 = weighting factor. using the preferred units for y MDT1.exp[-(U(t . U(x) = 0 if x < 0. in the preferred units of x (time) b = slope factor (no units) int = y-intercept. using the preferred units for y. using the preferred units for y MDT = mean dissolution time. exp[-(U(t .tlag)/Tmax ) ] where U(x) = x if x >= 0.int) * (t/Tmax)b * exp[b * (1 .exp[-(U(t . U(x) = 0 if x < 0. This model includes an optional lag time in the form of a step function.int) (1 . This model includes an optional lag time in the form of a step function. i.int) (1 . using the preferred units for y Tmax = time of maximum y value.t/Tmax)]. i. using the preferred units for y. using the preferred units for x (time) b = slope factor (no units) int = y-intercept.tlag) / MDT2)b2]) where U(x) = x if x >= 0.e.U(t . zero (default) or estimated intercept. zero (default) or estimated intercept.: y(t) = int + f1 * (Finf .B WinNonlin User’s Guide b1.int) * [ U(t . Model 604 Makoid-Banakar y(t) = int + (Fmax . for t <= Tmax y(t) = Fmax. using the preferred units for y. 550 .: y(t) = int + (Fmax . b2 = slope factors (no units) int = y-intercept. for t > Tmax where the estimated parameters are: Fmax = maximum y value.e.tlag)/Tmax]b * exp[ b * (1 .tlag) / MDT1)b1]) + (1-f1) * (Finf . U(x) = 0 if x < 0. Number list is a comma-separated list of up to 30 numbers or cell references. arguments.Text represents a text string. .Precision is the number of decimal places allowed. wherein: . . ref_type[. Creates a cell address as text.Expression list is a comma-separated list of up to 30 expressions. .a1] [. In DBCS systems this function returns a copy of text in which the double-byte characters are converted to single-byte characters.[ ] denotes optional arguments. column.Appendix C Worksheet Functions Syntax.Value list is a comma-separated list of up to 30 values or cell references. . B5. False if at least one argument is false. e. and examples for functions in WinNonlin worksheets WinNonlin worksheets support the functions detailed in Table C-1. Worksheet functions Function(arguments) ABS(number) ACOS(number) ACOSH(number) ADDRESS(row. . Returns True if all arguments are true. if possible.sheet] ) AND(logical_list) ASC(text) Description Absolute value of a number. . Table C-1. Returns the arcsine of a number. Returns the arc cosine of a number. 551 ASIN(number) . .. . Characters that cannot be converted are left unchanged.g.Serial number refers to the stored value for a date or time entered in a cell.Number is a number or reference to a cell that contains a number.Significance is the multiple to which to round. Returns the inverse hyperbolic cosine of a number. month. life. Returns the number of non-blank values in the supplied list. Worksheet functions (continued) Function(arguments) ASINH(number) ATAN(number) ATAN2(y. Joins several text items into one item. factor] ) DOLLAR(number [. Returns the specified number as text. Returns the day of the month that corresponds to the given date. Returns a character that corresponds to the supplied ASCII code. Returns a numeric code representing the first character of the supplied text. Returns the number of values in the supplied list. Rounds the specified number up to the nearest even integer. Returns the serial number of the supplied date. Returns the arithmetic mean of the supplied numbers. salvage. period [.…) COS(number) COSH(number) COUNT(value list) COUNTA(expression list) COUNTIF(range. Removes all nonprintable characters from the supplied text. salvage. item_list) CLEAN(text) CODE(text) COLUMN(reference) COLUMNS(range) CONCATENATE(text1. Returns the real depreciation of an asset for a specific period of time using the fixed-declining balance method. DAY(6-21-94) returns 21. 552 . precision] ) EVEN(number) Description Returns the inverse hyperbolic sine of a number. Returns the number of cells within a range that meet the given criteria. Returns the number of columns in a range reference. Rounds a number up to the nearest multiple of a specified significance. Example DATEVALUE (“3/6/94”) returns 34399. Example: DAY(34399) returns 6. Returns the hyperbolic cosine of a number. Returns the column number of the supplied reference. significance) CHAR(number) CHOOSE(index. Returns the arctangent of a number. period [. Returns the serial number of a date supplied as a text string.months] ) DDB(cost. life. Returns the inverse hyperbolic tangent of a number. criteria) DATE(year.C WinNonlin User’s Guide Table C-1. x) ATANH(number) AVERAGE(number list) CEILING(number. text2. Returns the depreciation of an asset for a specific period of time using the double-declining balance method or a declining balance factor. Returns the cosine of an angle. Returns the arctangent of the specified coordinates. using the local currency format and the supplied precision. day) DATEVALUE(text) DAY(text) DB(cost. Returns a value from a list of numbers based on the index number supplied. significance) Description Compares two expressions for identical. Determines if the specified expression is a number.Worksheet Functions C Table C-1. payment [. start_position] ) FIXED(number [. true_value. a1] ) INT(number) IPMT(interest. Returns the hour component of the specified time in 24-hour format.5.precision][. guess] ) ISBLANK(reference) ISLOGICAL(expression) ISNONTEXT(expression) ISNUMBER(expression) ISREF(expression) ISTEXT(expression) Searches the top row of a table for a value and returns the contents of a cell in that table that corresponds to the location of the search value. Returns the future value of an annuity based on regular payments and a fixed type] ) interest rate. Returns internal rate of return for a series of periodic cash flows. range_number] ) INDIRECT(ref_text [. True is returned if the expressions are identical. text [. expression2) EXP(number) FACT(number) FIND(search_text. Searches for a string of text within another text string and returns the character position at which the search string first occurs. column] [. pv] [. [fv]. pv. Returns the factorial of the specified number. formats the number in decimal format.5. Rounds a number down to the nearest multiple of specified significance. Returns the contents of a cell from a specified range. Determines if the specified expression is not text. -1. nper. FIXED(2009. Determines if the specified expression is text. False is returned if they are not.500. Searches for a string of text within another text string and returns the byte position at which the search string first occurs. Rounds the supplied number down to the nearest integer. Worksheet functions (continued) Function(arguments) EXACT(expression1. case-sensitive matches. Returns the interest payment of an annuity for a given period. Determines if the specified expression returns a logical value. FV(interest. row] [. [type] ) IRR(cash_flow [. per. HLOOKUP(search_item. and returns the result as text. text [. search_range. Returns the contents of the cell referenced by the specified cell. 553 . Determines if the specified expression is a range reference. false_value) INDEX(reference [. row_index ) HOUR(serial_number) IF(condition. start_position] ) FINDB(search_text. Rounds a number to the supplied precision. 1) returns 2010. based on regular payments and a fixed periodic interest rate.no_commas] ) FLOOR(number.000. Returns e raised to the specified power. Determines if the specified cell is blank. 3) returns 2. Example: FIXED(2000. Tests the condition and returns the specified value. nper. Returns the month that corresponds to the supplied date. num_chars) MIDB(text. type] ) NPV(discount_rate. Returns the base 10 logarithm of a number. Returns True if at least one of a series of logical arguments is true. LEN(“1-3”) returns 3. pf [.base]) LOG10(number) LOWER(text) MAX(number list) MID(text. Returns the modified internal rate of return for a series of periodic cash flows. Returns the left-most byte from the specified text string. Returns the natural logarithm of a number.num_bytes] ) LEN(text) LENB(text) LN(number) LOG(number. Returns the number of characters in the supplied text string. finance_rate. Examples: LEN(“3rd Quarter”) returns 11. [. omitting the base assumes a base of 10. (Empty parentheses are required). beginning with the specified starting position. Returns the minute that corresponds to the supplied date. start_position. num_chars]) LEFTB(text [. pmt. Returns the specified number of bytes from a text string. 4. Tests the supplied value and returns the value if it is a number. start_position. 2) returns 04. Worksheet functions (continued) Function(arguments) LEFT(text [.C WinNonlin User’s Guide Table C-1. value_list) ODD(number) OR(logical_list) PI() Description Returns the left-most characters from the specified text string. Returns the remainder after dividing a number by a specified divisor. Returns the net present value of an investment based on a series of periodic payments and a discount rate. Example: MID(“07/04/96”. Returns the current date and time as a serial number. 554 . Returns the number of periods of an investment based on regular periodic payments and a fixed interest rate. Examples: LEFT(“2nd Quarter”) returns 2. beginning with the specified starting position. reinvest_rate) MOD(number. LEFT(“2nd Quarter”. Rounds the specified number up to the nearest odd integer. num_bytes) MIN(number list) MINUTE(number list) MIRR(cash_flows. Returns the value of p. Returns the largest value in the specified list of numbers. Returns the logarithm of a number to the specified base. fv] [. Changes alphabetic characters in the specified string to lower case. 3) returns 2nd. Returns the smallest value in the specified list of numbers. divisor) MONTH(serial_number) N(value) NOW() NPER(interest. Returns the number of bytes in the supplied text string. Returns the specified number of characters from a text string. RAND() Returns a number selected randomly from a uniform distribution greater than or equal to 0 and less than 1. fv] [. PV(interest. num_chars] ) ROUND(number. Returns the interest rate per period of an annuity. nper. Example: REPLACE(“For the year: 1996". OfDigits) ROUNDUP(number. text [. pmt. fv] [. pv [. type] [. num_chars] ) RIGHTB(text [. based on regular payments and a fixed periodic interest rate. repl_text) REPT(text. Returns the sign of the specified number. (Empty parentheses are required). SEARCHB(search_text. Repeats a text string the specified number of times. numberOfDigits) ROWS(range) Rounds the given number up to the supplied number of decimal places. number. pmt [. text [. number) RIGHT(text [. Rounds the given number to the supplied number of decimal places. “07/04/96”) returns 3. ROUNDDOWN(number. [fv]. REPLACE(orig_text. Returns the right-most bytes from the given text string. pv. pv [. type] Returns the present value of an annuity. Multiplies a list of numbers and returns the result. type] ) PPMT(interest. num_chars. RATE(nper. per. fv] [.Rounds a number down. start_position. Returns the second that corresponds to the supplied date. SEARCH(search_text. start_position] ) SECOND(serial_number) SIGN(number) SIN(number) Locates the position of the first byte of a specified string within another string.Worksheet Functions C Table C-1. 1. num_bytes. Returns the right-most characters from the given text string. start Locates the position of the first character of a specified text string within position]) another text string. Returns the principle paid on an annuity for a given period. Worksheet functions (continued) Function(arguments) PMT(interest. nper. Returns the sine of the supplied angle. considering a series of constant pay) ments made over a regular payment period. repl_text) REPLACEB(orig_text. precision) Replaces part of a text string with another text string. “7”) returns “For the year: 1997". Replaces part of a text string with another text string. Returns the number of rows in a range reference. start_position. nper. given a series of constant guess] ) cash payments made over a regular payment period. Example: SEARCH(“/”. 18. 555 . [type] ) PRODUCT(number list) Description Returns the periodic payment of an annuity. VDB(cost. Returns the variance of a population based on a sample of up to 30 values. Returns the depreciation of an asset for a specified period using a variable start_period. sum_range) SUMPRODUCT(range_list) SUMSQ(number list) SYD(cost. salvage. Returns the depreciation of an asset for a specific period of time using the straight-line balance method. Returns the current date as a serial number. Squares each of the supplied numbers and returns the sum of the squares. column_index) WEEKDAY(serial_number) YEAR(serial_number) 556 Searches the first column of a table for a value and returns the contents of a cell in that table that corresponds to the location of the search value. Replaces a specified part of a text string with another text string.C WinNonlin User’s Guide Table C-1. Returns the day of the week that corresponds to the supplied date. Returns the depreciation of an asset for a specified period using the sum-ofyears method.precision]) TYPE(expression) UPPER(text) VAR(number list) Description Returns the hyperbolic sine of the specified number. minute. search_range.format) TIME(hour. Changes the characters in the specified string to uppercase characters. Returns the hyperbolic tangent of a number. Worksheet functions (continued) Function(arguments) SINH(number) SLN(cost. criteria. instance]) SUM(number list) SUMIF(range. factor] method of depreciation. Returns the given number as text. method] ) VLOOKUP(search_item. [. Returns the sum of the specified numbers. Truncates the given number to an integer. Returns the standard deviation of a list of as many as 30 numbers. old_text. Returns the tangent of the specified angle. . second) TIMEVALUE(text) TODAY() TRIM(text) TRUNC(number [. “-”. using the specified formatting. Example: SUBSTITUTE(“07-04-97”. Returns a serial number for the supplied text representation of time. end_period [. Returns the sum of the specified cells based on the given criteria. Returns the year that corresponds to the supplied date. life. “/”) returns 07/04/97. Removes all spaces from text except single spaces between words. life) SQRT(number) STDEV(number list) SUBSTITUTE(text. Returns the argument type of the given expression. Multiplies the cells in the given ranges and returns the sum of those products. period) TAN(number) TANH(number) TEXT(number. new_text [. life. Returns a serial number for the supplied time. salvage. Returns the square root of the specified number. salvage. BASE 0 BEGIN COMMAND Usage Description Example BEGIN Indicates that all commands have been specified and processing should start. Arrays and Functions Commands and syntax for ASCII text user models Notation Items enclosed in brackets. See “Drug effect data (model 220)” on page 515 for details. i.e. upper_val.. #.” E BASELINE COMMAND Usage Description Example BASE N Sets the (optional) baseline value for use analyzing effect data with NCA model 220. 557 . BEGIN BTIME COMMAND Usage BTIME lower_val.Appendix D Commands. The symbol “|” means “or. etc. [ ]. are optional. #. numexcl. D WinNonlin User’s Guide Description Selects points within the Lambda Z time range to be excluded from computations of Lambda Z. In the following example, Lambda Z is to be computed using times in the range of 12 to 24, excluding the concentrations at 16 and 20 hours. These “excluded” points will, however, still be used in the computation of the pharmacokinetic parameters. BTIME 12, 24, 2, 16, 20 If there are no times to be excluded, set numexcl to zero or leave it blank. Do not mark a point for exclusion by setting the case weight to 0 as that will also exclude this data point from the calculation of the PK parameters. Example Notes F CARRYALON G COMMAND Usage Description Examples CARRY #, etc. Specifies variables from the input data grid to be copied into each of the output worksheets. These variables are not required for the analysis. CARRY 3 CARR 1 CARRYALONG 2 See the NCARRY command. How Carry Along Data is Migrated to Output Multigrids The data for carry along columns is moved to all output worksheets from PK (PK/PD) or NCA analyses, with the exception of the last worksheet, named Settings. For all output worksheets other than the Summary Table, the carry along data will appear in the right-most column(s) of the worksheet. The first cell of data for any given profile within a carry along column will be copied over the entire profile range. There is one exception: generation of the “Summary Table” worksheets. The carry along columns will be the first columns, preceded only by any sort key columns. Also, as there are almost the same number of observations in the raw data set as in the Summary Table, all the cells within a profile of a carry along column are copied over to the Summary Table—not just the first value. There is one special case here as well: an NCA without a first data point of t=0, c=0. When such a data point does not exist on the raw data set, depending on which NCA model is selected, the core program may generate a pseudo-point of t=0, c=0 to support all necessary calculations. In this event, the first cell within a carry along column is copied 'up' to the core-generated data value of 0,0. Notes 558 Commands, Arrays and Functions F COMMANDS D MODEL BLOCK Usage COMMANDS Statements END The COMMANDS block allows WinNonlin commands to be permanently inserted into a model. Following the COMMANDS statement, any WinNonlin command such as NPARAMETERS, NCONSTANTS, PNAME, etc., may be given. To end the COMMAND section use the END statement. COMMANDS NCON 2 NPARM 4 PNAME 'A', 'B', 'Alpha', 'Beta' END Description Example CON ARRAY Usage Description Example Note CON(N) Contains the values of the constants, such as dosing information, used to define the model. The index N indicates the Nth constant. CON(3) The values specified by the CONSTANT command, described below, correspond to CON(1), CON(2), etc. respectively. CONSTANTS COMMAND Usage Description Examples Note CONSTANTS [are | =] C1, C2, …, CN Specifies the values of the constants (such as dosing information) required by the model. The arguments C1, C2, …, CN are the values. CONSTANTS 20, 2.06, -1 CONS are 5, 2 The NCONSTANTS command must appear before CONSTANTS. CONVERGEN COMMAND CE Usage CONVERGENCE [is | =] C Description Examples Specifies the convergence criterion, C. 0 < C < 0.1. The default value is 10-4. CONVERGENCE = 1.e-6 CONV .001 559 D WinNonlin User’s Guide Note The test for convergence is SSnew – SSold ------------------------------------- < C , where SS is the residual sum of squares. SSold When CONVERGENCE = 0, convergence checks and halvings are turned off (useful for maximum likelihood estimates). G DATA COMMAND Usage Description DATA [[are in,] 'path'] Begins data input and optionally specifies an external data file. If arguments are omitted, the data are expected to follow the DATA statement. If the arguments are used, path is the full path of the data set, up to 64 characters. DATA DATA 'C:\MYLIB\MYFILE.DAT' Examples DATE COMMAND Usage Description Example DATE Places the current date in the upper right corner of each page in ASCII output. DATE DHEADERS COMMAND Usage Description Example Note DHEADERS Statement indicating that data included within the model file contain variable names as the first row. DHEADERS If a DNAMES command is also included, it will override the names in the data. DIFFERENTI AL MODEL BLOCK Usage DIFFERENTIAL Statements END Statements in the DIFFERENTIAL - END block define the system of differential equations to fit. The values of the equations are assigned to the DZ array. Description 560 Commands, Arrays and Functions G Example DIFFERENTIAL DZ(1) = -(K1+KE)*Z(1) + K2*Z(2) DZ(2) = K1*Z(1) - K2*Z(2) END D Note NDERIVATIVES must be used to set the number of differential equations to fit. NFUNCTIONS specifies the number of differential equations and/or other functions with associated data. DINCREMEN T COMMAND Usage Description Examples DINCREMENT [is | =] D Specifies the increment to use in the estimation of the partial derivatives. The argument D is the increment. 0 < D < 0.1. The default value is 0.0001 DINCREMENT is 0.001 DINC 0.005 DNAMES COMMAND Usage Description DNAMES [are | =] 'name1' , 'name2' , . . . , 'nameN' Assigns names to the columns on the data set. The arguments 'name1', 'name2', . . . ,'nameN' are comma- or space-separated variable names. Each may use up to 8 letters, numbers or underscores, and must begin with a letter. DNAMES 'Subject', 'Time', 'Conc' The names appear in the output and may also be used in the modeling code. If both DNAMES and DHEADERS are included, DNAMES takes precedence. Example Note DTA ARRAY Usage Description Example DTA(N) Returns the value of the Nth column for the current observation. T = DTA(1) W = DTA(3) DTIME COMMAND Usage Description DTIME # Specifies the time of the last administered dose. This value will appear on the Final Parameters table as Dosing_time. Computation of the final parameters will be adjusted for this dosing time. 561 D WinNonlin User’s Guide Example DTIME 48 DZ ARRAY Usage Description Example Note DZ(N) Contains the current values of the differential equations. The index N specifies the equation, in the order listed in the DIFFERENTIAL block. DZ(3) = -P(3)*Z(3) The maximum value of N is specified by the NDERIVATIVES command. H F VARIABLE Usage Description Example F Returns the value of the function for the current observation. F = A * exp(-B*X) FINISH COMMAND Usage Description Example FINISH Included for compatibility with PCNonlin Version 4. FINISH should immediately follow the BEGIN command. BEGIN FINISH FNUMBER COMMAND Usage Description FNUMBER [is | =] N Indicates the column in the data set that identifies data for the FUNCTION block when NFUNCTIONS > 1. The function variable values must correspond to the FUNCTION numbers, i.e., 1 for FUNC 1, 2 for FUNC 2, etc. FNUMBER is 3 FNUM 3 FNUMBER should be specified prior to DATA. Examples Note 562 Commands, Arrays and Functions I FUNCTION D MODEL BLOCK Usage FUNCTION N Statements END The FUNCTION - END block contains statements to define the function to be fit to the data. The argument N indicates the function number. FUNC 1 F=A*EXP(B) + C*EXP(D) END FUNC 1 F=A*EXP(B) + C*EXP(D) END FUNC 2 F=Z(1) END A function block must be used for every data set to be fit. NFUNCTIONS specifies the number of function blocks. The variable F must be assigned the function value. WT can be assigned a value to use as the weight for the current observation. Description Examples Note I INITIAL COMMAND Usage Description Examples INITIAL [ARE | =] I1, I2, I3, . . . , IN Specifies initial values for the estimated parameters. The arguments I1, I2, I3, . . . , IN are the initial values, separated by commas or spaces. INITIAL = 0.01, 5.e2, -2.0 INIT are 0.02, 10, 1.3e-5 INIT 0.1 .05 100 INITIAL must be preceded by NPARAMETERS. Note ITERATIONS COMMAND Usage Description Examples ITERATIONS [are | =] N Specifies the maximum number of iterations to be executed, N. ITERATIONS are 30 ITER 20 563 D WinNonlin User’s Guide Note If ITERATIONS = 0, the usual output is produced using the initial estimates as the final values of the parameters. J LOWER COMMAND Usage Description Example Note LOWER [are |=] L1, L2, L3, . . . , LN Specifies lower bounds for the estimated parameters. The arguments L1, L2, L3, . . . , LN are the lower bounds, separated by commas or spaces. LOWER 0 0 10 If LOWER is used, a lower bound must be provided for each parameter, and UPPER must also be used. LOWER must be preceded by NPARAMETERS. K MEANSQUAR COMMAND E Usage MEANSQUARE [is | =] M Description Allows a specified value other than the residual mean square to be used in the estimates of the variances and standard deviations. The argument M replaces the residual sum of squares in the estimates of variances and standard deviations. If given, MEANSQUARE must be greater than 0. MEANSQUARE = 100 MEAN=1.0 Examples METHOD COMMAND Usage METHOD [is ⏐ =] N 564 Commands, Arrays and Functions K Description D In modeling, the METHOD command specifies the type of fitting algorithm WinNonlin is to use. The argument N takes the following values: 1. Nelder-Mead Simplex 2. Levenberg and Hartley modification of Gauss-Newton (default) 3. Hartley modification of Gauss-Newton In noncompartmental analysis, METHOD specifies the method to compute area under the curve. The argument N takes on the following values: 1. Lin/Log - Linear trapezoidal rule up to Tmax, and log trapezoidal rule after 2. Linear trapezoidal rule (Linear interpolation) (default) 3. Linear up/Log down rule – Uses the linear trapezoidal rule any time the concentration data is increasing and the logarithmic trapezoidal rule any time that the concentration data is decreasing. 4. Linear Trapezoidal (Linear/log Interpolation) METHOD is 2 METH 3 Example MODEL COMMAND Usage Description MODEL [N[[,] 'path']] In a command file, MODEL specifies the compiled or source model to use. If “MODEL N” is specified, model N is used from the built-in library. If “MODEL N, PATH” is used, model N from the source file path is used. MODEL 5 MODEL 7, 'PK' MODEL 3 'D:\MY.LIB' Examples MODEL LINK COMMAND Usage Description MODEL LINK M1 [,] M2 MODEL with the option LINK specifies that compiled models are to be used to perform PK/PD or Indirect Response modeling. The arguments M1 and M2 are model numbers where: M1 is the PK model number, and M2 is the model number of the PD model or Indirect Response model. When this option is used a PKVAL command must also be used. MODEL LINK 4 101 PKVAL 10 1.5 1.0 1.0 MODE LINK 1, 53 PKVA 10, 2.5 See also PKVALUE. Examples Note 565 D WinNonlin User’s Guide L NCARRY COMMAND Usage Description Examples Note NCARRY [are | =] N Specifies the number of carry-over variables specified. The argument N is the number of carry-over variables. NCAR are 3 NCAR 1 NCARRY must be specified before CARRYALONG. NCATRANS COMMAND Usage Description NCATRANS When using NCA, NCATRANS specifies that the Final Parameters output will present each parameter value in a separate column, rather than row. See “Transpose final parameters” in the WinNonlin User’s Guide. NCATRANS NCAT Examples NCONSTANT S COMMAND Usage Description Examples NCONSTANTS [are | =] N The NCONSTANTS command specifies the number of constants, N, required by the model. If used, NCONSTANTS must precede CONSTANTS. NCONSTANTS are 3 NCON 1 NDERIVATIV ES COMMAND Usage Description Examples NDERIVATIVES [are | =] N Indicates the number, N, of differential equations in the system being fit. NDERIVATIVES are 2 NDER 5 NFUNCTION S COMMAND Usage NFUNCTIONS [are | =] N 566 Commands, Arrays and Functions L Description Examples Note D Indicates the number, N, of functions to be fit (default = 1). There must be data available for each function. If used, NFUNCTIONS must precede DATA. NFUNCTIONS are 2 NFUN 3 The functions can contain the same parameters so that different data sets can contribute to the parameter estimates. When fitting multiple functions one must delineate which observations belong to which functions. To do this, include a function variable in the data set or use NOBSERVATIONS. See FNUMBER or NOBSERVATIONS. NOBOUNDS COMMAND Usage Description Examples Note NOBOUNDS Tells WinNonlin not to compute bounds during parameter estimation. NOBOUNDS NOBO Unless a NOBOUNDS command is supplied WinNonlin will either use bounds that the user has supplied, or compute bounds. NOBSERVATI COMMAND ONS Usage NOBSERVATIONS [are | =] N1, N2, N3, . . . , NF Description Provides backward compatibility with PCNonlin command files. When more than one function is fit simultaneously, NOBS can indicate which observations apply to each function. The arguments N1, N2, . . . , NF are the number of observations to use with each FUNCTION block. The data must appear in the order used: FUNC 1 data first, then FUNC 2, etc. Note that this statement requires counting the number of observations for each function, and updating NOBS if the number of observations changes. NOBS must precede DATA. NOBS 10, 15, 12 The recommended approach is to include a function variable whose values indicate the FUNCTION block to which the data belong. (See FNUMBER.) Example Note NOPAGE COMMAND Usage Description Examples NOPAGE Tells WinNonlin to omit page breaks in the ASCII output file. NOPAGE NOPAGE BREAKS ON OUTPUT 567 D WinNonlin User’s Guide NOPLOTS COMMAND Usage Description Example Note NOPLOTS Specifies that no plots are to be produced. NOPLOTS This command turns off both the high resolution plots and ASCII output. NOSCALE COMMAND Usage Description NOSCALE Turns off the automatic scaling of weights when WEIGHT is used. By default, the weights are scaled such that the sum of the weights equals the number of observations with non-zero weights. NOSCALE This command will have no effect unless WEIGHT is used. Example Note NOSTRIP COMMAND Usage Description Example NOSTRIP Turns off curve stripping for all profiles. NOSTRIP NPAUC COMMAND Usage Description Examples Note NPAUC # Specifies the number of partial area under the curve computations requested. NPAUC 3 See PAUC. NPARAMETE RS COMMAND Usage Description Examples Note NPARAMETERS [are | =] N Sets the number of parameters, N, to be estimated. NPARAMETERS must precede INITIAL, LOWER or UPPER. (LOWER and UPPER are optional.) NPARAMETERS 5 NPAR 3 See PNAMES. 568 Commands, Arrays and Functions M NPOINTS D COMMAND Usage Description Example Notes NPOINTS [is | =] N Determines the number of points in the Predicted Data file. Use this option to create a data set to plot a smooth curve. 10 <= N <= 20,000. NPOINTS 200 For compiled models without differential equations the default value is 1000. If NDERIVATIVES> 0 (i.e., differential equations are used) this command is ignored and the number of points is equal to the number in the data set. The default for user models is the number of data points. This value may be increased only if the model has only one independent variable. NSECONDAR COMMAND Y Usage NSECONDARY [are | =] N Description Examples Note Specifies the number, N, of secondary parameters to be estimated. NSECONDARY 2 NSEC = 3 See SNAMES. NVARIABLES COMMAND Usage Description Examples NVARIABLES [are | =] N Specifies the number of columns in the data grid. N is the number of columns (default = 2). If used, NVARIABLES must precede DATA. NVARIABLES 5 NVAR 3 M P ARRAY Usage Description Examples P(N) Contains the values of the estimated parameters for the current iteration. The index N specifies the Nth estimated parameter. P(1) T=P(3) - A 569 D WinNonlin User’s Guide Note The values of the INITIAL command correspond to P(1), P(2), etc., respectively. NPARAMETERS specifies the value of N. PNAMES can be used to assign aliases for P(1), P(2), etc. PAUC COMMAND Usage Description PAUC #, # etc. Lower, upper values for the partial AUC computations. If a specified time interval does not match times on the data set, then WinNonlin will generate concentrations corresponding to those times. PAUC 0, 5, 17, 24 See NPAUC. Examples Note PKVALUE COMMAND Usage Description Examples PKVALUE N1, N2,. . . ,NPK Used to assign parameter values to the PK model when using MODEL LINK. MODEL LINK 4 101 PKVAL 10 1.5 1.0 1.0 MODE LINK 1, 53 PKVA 10, 2.5 PNAMES COMMAND Usage Description PNAMES [are | =] 'name1', 'name2', . . . , 'nameN' Specifies names associated with the estimated parameters. The arguments 'name1', 'name2', . . . ,'nameN' are names of the parameters. Each may have up to eight letters, numbers, or underscores, and must begin with a letter. These names appear in the output and may be used in the modeling code. PNAMES are 'ALPHA', 'BETA', 'GAMMA' PNAM 'K01', 'K10', 'VOLUME' The number of names specified must equal that for NPARAMETERS. Examples Note PUNITS COMMAND Usage Description PUNITS [are | =] 'unit1', 'unit2', . . . , 'unitN' Specifies the units associated with the estimated parameters. The arguments 'unit1', 'unit2', . . . ,'unitN' are units and may be up to eight letters, numbers, operators or underscores in length. These units are used in the output. 570 Commands, Arrays and Functions N Examples Note PUNITS are 'hr', 'ng/mL', 'L/hr' PUNI 'hr' The number of units specified must equal that for NPARAMETERS. D N REMARK COMMAND Usage Description Examples Note REMARK [comment] Allows comments to be placed in a command file. REMARK Y transformed to log(y) REMA GAUSS-NEWTON If a command takes no arguments or a fixed number of arguments, comments may be placed on the same line as a command after any required arguments. REWEIGHT COMMAND Usage Description REWEIGHT W In the PK modeling module, REWEIGHT specifies that iteratively reweighted least squares estimates of the parameters are to be computed. The weights have the form WT = FW where W is the argument of REWEIGHT and F is the current value of the estimated functions. If W < 0 and F < 0 then the weight is set to 0. REWEIGHT 2 REWE -.5 REWEIGHT is useful for computing maximum likelihood estimates of the estimated parameters. The following statement in a model is equivalent to a REWEIGHT command: WT = exp(W log(F)), or WT = F**W This command is not used with noncompartmental analysis. Examples Note O S ARRAY Usage S(N) 571 only the estimated parameters and predicted values from the initial estimates are computed.END block. Aliases may be assigned for S(1). The index N specifies the Nth secondary parameter. if one is included. etc. When it is encountered. using SNAMES. Model size sets default memory requirements and may range from 1 to 6. Assignments are made to the array S. Description Examples Note SIMULATE COMMAND Usage Description Example Note SIMULATE Causes the model to be estimated using only the time values in the data set and the initial values. SECONDARY MODEL BLOCK Usage SECONDARY Statements END The secondary parameters are defined in the SECONDARY .693/K10 END NSECONDARY must be used to set the number of secondary parameters. S(2). S(1) = -P(1)/P(2) The range of N is set by NSECONDARY. It must precede the DATA statement. no data fitting is performed.693/K10 END Or. SIZE COMMAND Usage Description SIZE N N is the model size. The secondary parameters are defined in the SECONDARY section of the model. SIMULATE SIMULATE may be placed in any set of WinNonlin commands. if the command SNAMES 'K10_HL' and a PNAMES command with the alias K10 have been included: SECONDARY K10_HL=0. The default value is 2.D WinNonlin User’s Guide Description Example Note Contains the estimated secondary parameters. All other output such as plots and secondary parameters are also produced. 572 . SECONDARY S(1)=0. 'name2'. 1 SORT 1 3 2 SORT 1. . etc. 'nameN' are names of the secondary parameters. . . SORT 1. 'name2'. 'nameN' Specifies the names associated with the secondary parameters.e. numbers or underscores in length. Arguments s1 and the options s2. Causes WinNonlin to segment a large data file into smaller subsets. The 1 that follows the command SORT has no function.. .] s1 [. 2 SORT must be on one line. They appear in the output and may also be used in the programming statements. It is included to provide compatibility with PCNonlin version 4. 'Tmax' SNAM 'K10_HL' 'K01_HL' Examples SORT COMMAND Usage Description SORT 1 [. etc. See “Sparse sampling computations” on page 188. 1. it can not use the line continuation symbol '&. . aliases for S(1). Each subset will be run through the modeling separately.Commands. . . STARTING MODEL BLOCK Usage STARTING Statements END 573 . are the column numbers of the sort keys. i. Arrays and Functions O Example Note SNAMES SIZE 3 Applicable only when using ASCII models. . SNAMES are 'AUC'. s3. S(2). and must begin with a letter. . Each may be up to eight letters. The arguments 'name1'.’ Examples Note SPARSE COMMAND Usage Description SPARSE Causes WinNonlin to use sparse data methods for noncompartmental analysis. The default is to not use sparse methods. 'Cmax'.] s2 . D COMMAND Usage Description SNAMES [are | =] 'name1'. Values are placed in the Z array. . 'unitN' Specifies the units associated with the secondary parameters. START Z(1) = CON(1) Z(2) = 0 END Example STEADY STATE COMMAND Usage Description Examples STEADY STATE # Informs WinNonlin that data were obtained at steady state. The arguments 'unit1'. Description 574 . 'mL/hr' SUNI '1/hr' The number of units specified must equal NSECONDARY. STEADY STATE 12 STEA 12 SUNITS COMMAND Usage Description SUNITS [are | =] 'unit1'. 'unit2'.END block specify the starting values of the differential equations. so that different PK parameters are computed. . . These units are used in the output. .D WinNonlin User’s Guide Description Statements in the STARTING . numbers.END block define global temporary variables that can be used in the MODEL statements. SUNITS are 'ml'. the (assumed equal) dosing interval. . Parameters can be used as starting values. . . . 'unit2'. operators or underscores in length. # is Tau.'unitN' are units and may be up to eight letters. Examples Note P TEMPORARY MODEL BLOCK Usage TEMPORARY Statements END Statements placed on the TEMPORARY . THRE 0 TIME COMMAND Usage Description Example TIME Places the current time in the upper right hand corner of each output page.Commands. 575 Description . 1 <= N <= 5. Arrays and Functions P Example TEMP T=X DOSE1 = CON(1) TI=CON(2) K21=(A*BETA + B*ALPHA)/(A+B) K10=ALPHA*BETA/K21 K12=ALPHA+BETA-K21-K10 END D THRESHOLD COMMAND Usage Description Example THRE N Sets the (optional) threshold value for use analyzing effect data with NCA model 220.END block are executed for each observation before the estimation begins. To specify any title other than the first. TITLE Experiment EXP-4115 TITLE 1 Experiment EXP-4115 TITLE 2 Subject #1 Examples TRANSFORM MODEL BLOCK Usage TRANSFORM Statements END Statements placed in the TRANSFORM . a value N must be included with TITLE. See “Drug effect data (model 220)” on page 515 for details. Up to five titles are allowed per run. TIME TITLE COMMAND Usage Description TITLE [N] Indicates that the following line contains a title to be printed on each page of the output. D WinNonlin User’s Guide Example TRANSFORM Y = exp(Y) END Q UPPER COMMAND Usage Description Example Note UPPER [are |=] U1. U3. UPPER 1. and the concentration in the urine. and only the first four letters of each keyword is required. 200 If UPPER is used.C4 give their column numbers. the volume of urine collected. . . . URINE LOWER=C1 UPPER=C2 VOLUME=C3 CONC=C4 URINE CONC C1 VOLU C2 LOWE C3 UPPE C4 The keywords may appear in any order. UN Sets upper bounds for the estimated parameters. The arguments U1. U2. UPPER must be preceded by NPARAMETERS. URINE COMMAND Usage Description URINE LOWER [is | =] C1 UPPER [is | =] C2 VOLUME [is | =] C3 [is | =] CONCENTRATION [is | =] C4 Used to specify variables when doing noncompartmental analysis for urine data (Models 210-212). the ending times of the urine collection interval. and LOWER must also be specified. The four keywords LOWER UPPER VOLUME. Variables must be specified for all four keywords. an upper bound must be provided for each parameter. Examples Note R WEIGHT COMMAND Usage WEIGHT [W] 576 . and CONCENTRATION identify the required data columns. . … . The user must supply: the beginning time of the urine collection interval. U3. UN are the bounds. 1. U2. C1 . separated by commas or spaces. Examples Note WT VARIABLE Usage Description Example Note WT In PK modeling. REWEIGHT and NOSCALE.5 Uses as the weight 1/Sqrt(Y) WEIGHT WTNUM 7 Uses the values in Column 7 as the weights The weights are scaled so that their sum = the number of observations with non-zero weights. F = exp(-X) By default. a column in the data set provides weights— by default. WTNUMBER 4 WTNU is 7 If WEIGHT is used and WTNUMBER is not. the values of X are taken from the first column of the data set unless the XNUMBER command has been used. WT = 1/(F*F) If the variable WT is included in a model. See also WTNUMBER. Another column may be set using WTNUMBER. If W is not set. Not used for NCA. the third column. The argument W sets the weight value: YW. when used in a command file. See also “WEIGHT” on page 576. 577 . WEIGHT Uses the values in Column 3 as the weights WEIGHT -. S X VARIABLE Usage Description Example Note X Contains the current value of the independent variable. WTNUMBER COMMAND Usage Description Examples Note WTNUMBER [is | =] N Specifies which column of the data set contains weights. Arrays and Functions S Description D Activates weighted least squares computations. WT contains the weight for the current observation.Commands. then any weighting indicated via the Variables and Weighting dialog will be ignored. WT. the default column for weights is 3. The index N specifies the Nth differential equation. XNUMBER is 3 XNUM 1 XNUMBER should precede DATA. YNUMBER COMMAND Usage Description Examples Note YNUMBER [is | =] N The YNUMBER command specifies the column number. YNUMBER is 4 YNUM 3 The YNUMBER command should precede the DATA command. U Z ARRAY Usage Description Example Note Z(N) Contains the current values of the solutions of the differential equations. Y is assumed to be in the 2nd column unless YNUMBER is used. XNUMBER indicates the column number that the differential equations are integrated with respect to. Examples Note T Y VARIABLE Usage Description Example Note Y Current value of the dependent variable. the default value is 2.D WinNonlin User’s Guide XNUMBER COMMAND Usage Description XNUMBER [is | =] N Sets the column number. The default value is 1. When running Version 4 command files. If differential equations are being fit. N. 578 . N. of the dependent variable. B = A*Z(2) The maximum value of N is set by NDERIVATIVES. of the independent variable. Y = log(Y) By default. syntax and usage For instructions on writing and executing WinNonlin scripts. refer to “Scripting” on page 417.AlphaTerms(1)=.229 Toolbox.84 579 .AggregateVar = “Conc” AlphaTerms() Property Applies To Syntax Remarks See also Example Toolbox Toolbox. Only one Aggregate Variable may be used. Table. The Aggregate Variable is divided into separate columns based on values of the cross variable(s). Table.AlphaTerms(n)=number (n) corresponds to the number set by NumTerms NumTerms Toolbox.Appendix E Scripting Commands WinNonlin scripting language commands. See the Examples Guide for example scripts. sets the variable from Data Set 1 that composes the body of the table.AlphaTerms(2)=3.AggregateVar = VariableName Similar to a Variable in tables 4-8. A AggregateVar Property Applies To Description Syntax Remarks Example Table For tables 9 and 10. When using Append: • If the two data sets use different missing value codes.Delimiter.Append See also Example AppendFilename Property Applies To Description Syntax See also Data Specifies a file to be appended to an existing data object.E WinNonlin User’s Guide Append Method Applies To Description Syntax Remarks Data Add content of a file to another file. AppendFilename. objectname. AppendFilename 580 . Use Find/Replace to standardize the missing value codes.Headers. The ReadOnly status of the original file will be maintained.AppendFilename = string Append. if the file type is ASCII. Objectname.AppendFileType = EXCEL MYDATA. • Formulas from Excel files will be lost. objectname.AppendFilename = “C:\MYDIR\newdata.xls” MYDATA. . AppendFileType AppendFileType Property Applies To Description Syntax Remarks See also Data Specifies the file format of the file to be appended to an existing data object. Append. Only the values are appended. When using the Data.AppendFileType = {ASCII | EXCEL} . etc. but it is not possible to append a . Delimiter. • All load options are in effect. just as the Data.PWO). • The ReadOnly status is not recalled for the appended file.PWO to another file. AppendFileType.Append method.Load method would. WinNonlin will not update these automatically.Append Data can be appended to a Pharsight workbook object (*. Headers. properties.PWO file types are not supported with this property. Missing MYDATA. it will use the . 3 Toolbox. Used only with the Append method.Aterms(n)=number The argument (n) corresponds to the number set by NumTerms.AppendFileType = EXCEL MYDATA.Aterms(2)=8.Scripting Commands A AppendFileType (continued) Property Example MYDATA.Append E AppendSheet Property Applies To Description Syntax Remarks Example Data Selects the sheet of a multisheet workbook to be used for an append operation (similar to the Multigrid Property). MYDATA1. Toolbox.Aterms(1)=12.AppendFilename=”Test2” MYDATA1.AppendFilename = “C:\MYDIR\newdata.Filename=”Test1” MYDATA1.AppendSheet=”Final Parameters” Used with the Append method of the Data property.Append Arrange Method Applies To Description Syntax WinNonlin Arranges the icons representing any minimized windows in WinNonlin. MYDATA.Multigrid=”Sheet 3” MYDATA1.xls” MYDATA.1 581 .AppendSheet=”Sheet 1” <-selected sheet to append MYDATA1.Load MYDATA1. WinNonlin Arrange Aterms() Property Applies To Syntax Remarks Example Toolbox Toolbox. Bolus={True|False} Toolbox. The default is True (yes).AutoSort = {True | False} Plot.Bolus=True C CarryData Property Applies To Description Syntax Merge Sets whether or not “Carry along data for like sort levels” is used.CarryData = {True | False} Cascade Property Applies To 582 WinNonlin . Toolbox.AutoFileOptions = True/False AutoSort Property Applies To Description Syntax Example Plot If AutoSort = “True” (default). Plot. the X variable column will be sorted before the plot is made. DataObject. Merge.AutoSort = False B Bolus Property Applies To Description Syntax Example Toolbox Sets whether WNL will constrain the estimated input rate to zero at the initial time (lag time).E WinNonlin User’s Guide AutoFileOptions Property Applies To Description Syntax Data Turns on automatic file options when loading a data file. CaseSensitive = {True | False} Use with Include. RenameCol and Replace.CloseAll Any information that had not been saved will be lost. MYDATA.Operation=”=” MYDATA.Tolerance=”0” MYDATA.Scripting Commands C Cascade (continued) Property Description Syntax See also Example Causes the open objects to be cascaded in the WinNonlin window.Find=”3” MYDATA. WinNonlin. WinNonlin.CaseSensitive=False MYDATA. Objectname.SearchArea=”CONC” MYDATA. Exclude. Save and Unload Column Property Applies To Data 583 . Remove.Cascade E CaseSensitive Property Applies To Description Syntax Remarks Example Data Specifies whether a search is case sensitive (True) or not (False).Cascade TileHorizontal and TileVertical WinNonlin.Replace CloseAll Method Applies To Description Syntax Remarks See also WinNonlin Closes all window objects within WinNonlin.With=”Missing” MYDATA. RemoveCol. Confidence = {True | False} The default is False.Column=”a” MYDATA. Note that WinNonlin recognizes only columns that contain data. 584 . by column name (or letter). Objectname. Percentiles Desc.” Load treats consecutive delimiters as one.Column = string Use with RemoveCol and RenameCol. In order to create and rename a new column.ConcCol=”Concentration” Confidence Property Applies To Description Syntax Remarks See also Example Desc Sets whether confidence intervals will be calculated with descriptive statistics (True) or not.g. using a Paste or Append operation) then rename it. Interval.With=”Subject” MYDATA. Desc. put data in the column first (e.ConcCol=column name Toolbox. MYDATA..Confidence = True ConseqDelim Property Applies To Description Syntax Remarks Data When ConseqDelim = “True.E WinNonlin User’s Guide Column (continued) Property Description Specifies a column to be renamed or removed. the system will use the value specified on the most-recent Load method.ConseqDelim = {True | False} Used only when loading ASCII data. If a value is not set.RenameCol Syntax Remarks Example ConcCol Property Applies To Description Syntax Example Toolbox Sets the input column for Concentration data. objectname. Toolbox. CrossVar(2) = “Form” CrossVars Property Applies To Table 585 . Missing.CrossVar(1) = “Subject” Table.Copy = string Optionally. For table 9. 7.Filename = <file name> MYDATA. EndCol. Objectname.EndCol=b MYDATA.Copy MYDATA. the cross variable must be from Data Set 1.Scripting Commands C ConseqDelim (continued)Property See also Delimiter.CrossVar(Variable no.StartCol=a MYDATA. Headers E Copy Method Applies To Description Syntax Remarks Data Copies a specified range of data and stores it for use with Paste. 5. which define the columns when breaking a variable into separate columns within a table.StartCol=e MYDATA. RegVar( ) Table. specify a selection of cells for copying using StartCol. MYDATA.) = Variable Name Applies to tables 4. To specify an entire row (or entire rows) use only StartRow and EndRow. CrossVars sets the number of cross variables. 9. StartRow. To specify an entire column (or entire columns) use only StartCol and EndCol.StartRow=1 MYDATA.Paste Example CrossVar( ) Property Applies To Description Syntax Remarks See also Example Table Defines cross variable(s).StartRow=1 MYDATA. 6. 8. and EndRow. IDVar( ).EndRow=20 MYDATA. Table. CrossVars. GroupVar( ). MYDATA. StartRow.StartCol=A MYDATA. GroupVars.EndCol=J MYDATA. EndCol.DefinitionFile = string 586 . except the data file(s) to use.Filename = <file name> MYDATA.E WinNonlin User’s Guide CrossVars (continued)Property Description Syntax Remarks See also Example Defines the number of cross variables for the Table engine.EndRow=10 MYDATA.StartRow=5 MYDATA. and EndRow. StartRow and EndRow.StartRow=1 MYDATA.CrossVars = 2 Cut Method Applies To Description Syntax Remarks Example Data Cuts a specified range of data from a data set and stores it for use with the Paste function. Table. 8.Paste D DefinitionFile Property Applies To Description Syntax Table Loads a definition file containing all specifications for a table. use StartCol. IDVars. 6. To select an entire column (or columns) use only StartCol and EndCol. 5. Table.Cut = <string> To select cells to cut. 7.StartCol=A MYDATA.CrossVars = integer Applies to tables 4. RegVars Table. Objectname. for rows.Cut MYDATA. 9. Table.DeleteFiles Directory names and *.” MYDATA.Delimiter = “ ” Delta Property Applies To Description Toolbox User supplied smoothing parameter. ConseqDelim. It must be created and saved by the table Generator.Scripting Commands D DefinitionFile (continued)Property Remarks E Any input data file that includes the exact variable names used in the table definition may be used with a definition file. FileMask WinNonlin.tdf” Example DeleteFiles Method Applies To Description Syntax Remarks See also Example WinNonlin Deletes a file selected using the FileMask property.Delimiter = character Used only while loading ASCII data files. WinNonlin. Headers MYDATA.DeleteFiles Delimiter Property Applies To Description Syntax Remarks See also Example Data Specifies the field delimiter to use with the Load method. Missing.Delimiter = “. The definition file is machine readable only. It is necessary to turn a table definition file “off” if a script will use a table definition file.DefintionFile=“ ” to nullify the definition file before using the commands. 587 . then create another table using Scripting Language commands. If this value is not specified.xls” WinNonlin. Use the command: TABLE.DefinitionFile=“c:\winnonln\mytable.* are not allowed in file specifications. the system will use the value specified for the most-recent Load method.FileMask = “data. objectname. Trans.Delta=2 DestArea Property Applies To Description Syntax Remarks See also Example Trans If DestCol is defined as a new column name. is required even if there is only one data source for the plot.DestCol = string Spaces are not allowed in the new column name.DestArea = {Adjacent | Append} If DestCol uses an existing column. DestArea Trans.DnErr(Data Set no.DnErr(1) = “Minimum” 588 .DestArea = Append DestCol Property Applies To Description Syntax Remarks See also Example Trans This property defines the destination column for the transform function. then DestArea is ignored. then DestArea determines its location.E WinNonlin User’s Guide Delta (continued)Property Syntax Example Toolbox. The column can be either inserted adjacent to the SourceCol or appended to the end of the data set.Delta=number Toolbox. DestCol Trans.DestCol = “LnConc” DnErr( ) Property Applies To Description Syntax Remarks See also Example Plot Defines the column containing down error data for a plot with error bars. Trans. ErrType Plot. UpErr( ). Plot.DnErr(1) = “SE” Plot.) = string Data Set no. DoseFile = “c:\dosedata. See “Special file formats” on page 430 for details on this file. Toolbox.DoseUnit={unit} Toolbox. Toolbox. Its value is a string (column name) or number (position). 589 .Ecol=”Effect” EndCol Property Applies To Description Data Sets the last column in a range. Toolbox.dat” DoseUnit Property Applies To Description Syntax Example Toolbox Specifies unit for dosing. Model.Ecol=column name Not used in calculations. LamzFile.Scripting Commands E E DoseFile Method Applies To Description Syntax Remarks See also Example Model Loads an ASCII data file containing model dosing information. but the column number (position from the left) must be used to designate a new column. LinkFile.DoseUnit=”ng/ml” E Ecol Property Applies To Description Syntax Remarks Example Toolbox Specifies the input column for effect data. PartFile Model. ParmFile. Note that the column name can be used to reference an existing column.DoseFile=string See “Special file formats” on page 430. StartRow MyData.Copy MyData. The following example removes all rows after row 4. Set StartCol <= EndCol.EndCol = MyData. StartRow <= EndRow.EntireRow = {True | False} Use with Include and Exclude.RemoveRow Example EntireRow Property Applies To Description Syntax Remarks Data Specifies that the entire row. Objectname.StartCol MyData. StartRow.EndRow=-1 MYDATA. as opposed to individual cell(s). StartCol.EndRow = MyData. is to be included or excluded.StartCol MyData.StartRow MyData.StartRow=5 MYDATA.EndCol = MyData. Copy or Font and FontSize. If StartRow is set and is not –1. setting EndRow = –1 selects all rows below and including StartRow.EndCol = string | number Use with RemoveRow.EndRow = number Use with the RemoveRow.EndRow = MyData.Copy = 1 = “Time” 12 “Time” = 1 = 1 12 3 EndRow Property Applies To Description Syntax Remarks Data Sets the last row in a range. Objectname.E WinNonlin User’s Guide EndCol (continued) Property Syntax Remarks See also Examples Objectname. EndRow MyData. Cut and Copy methods. 590 . MYDATA. Cut. Exclude E ErrType Property Applies To Description Syntax Remarks See also Example Plot Determines handling of error data by the Plot engine.Scripting Commands E EntireRow (continued)Property Example MYDATA.SearchArea=”entire data set” MYDATA. EntireRow.Find=”0” MYDATA.ErrType = Absolute Exclude Method Applies To Description Syntax See also Example Data Excludes the selected data. using the same options as PrintAll.Exclude ExportAll Method Applies To Description WinNonlin Exports all open objects to Microsoft Word. Use the “Print” properties to set the object type(s) to export. DnErr( ) Plot.Operation=”<” MYDATA. Plot.SearchArea=”entire data set” MYDATA. 591 .CaseSensitive=False MYDATA. Operation. CaseSensitive MYDATA.ErrType = {Absolute | Relative} The default value is Relative.Tolerance=”0” MYDATA.Find=”censor” MYDATA. objectname. Tolerance. See “Data exclusion and inclusion” on page 60. UpErr( ).Exclude Find.EntireRow=True MYDATA. SearchArea.EntireRow=True MYDATA. EndRow. The example below exports all charts. StartRow. See “Error bars” on page 120. Each transformation requires different information.Extra = {string | number} Function used TRANS_LN TRANS_LOG10 TRANS_SQRT TRANS_ABS TRANS_CHANGEBASE TRANS_PCTCHANGE TRANS_RATIOBASE TRANS_MULTCOL TRANS_METABOLITE TRANS_ADDCOL TRANS_SUBTRACTCOL TRANS_E TRANS_INV TRANS_POWER TRANS_MULT TRANS_DIV TRANS_ADD TRANS_SUBTRACT Use Not used Not used Not used Not used Variable Name .defines the denominator.defines Column Y String .Extra = “Hour” TRANS_POWER: Trans.E WinNonlin User’s Guide ExportAll (continued)Method Syntax See also Example WinNonlin. DestCol.Extra = 2 592 . Trans.PrintData=False WinNonlin.PrintGraph=True WinNonlin. as detailed under “Remarks” below.ExportAll PrintAll WinNonlin.specifies Time column String .PrintModel=False WinNonlin. DestArea TRANS_CHANGEBASE: Trans.specifies Time column Variable Name . SourceCol.PrintText=False WinNonlin.specifies Time column Variable Name . Extra.defines Column Y String .ExportAll Extra Property Applies To Description Syntax Remarks Trans Contains additional information required for some transformations. Column Y String .defines Column Y Not used Not used number number number number number See also Example InData. WSP). WinNonlin.DeleteFiles Filename Property Applies To Description Syntax Remarks Example WinNonlin. Filename.dat” WinNonlin. Text Sets a file name for Load and Save.FileType = “PWO” MyData. the FileType must be set to PCO. Graph. For the WinNonlin object. Data..load or MyData.\data\profiles. Text Sets the file format for Load and Save.FileMask=”bguide2. WTO. Save 593 See also .Filename = “. WinNonlin can load WDO.Filename = <absolute or relative path to data file> Use with FileType except when using with the WinNonlin object (assumes file type . WinNonlin. Model. Graph.wsp” WinNonlin. objectname.filename = <absolute or relative path to workspace file> or objectname.FileMask =string Use with DeleteFiles. specifies a workspace to load. Data. WMO.filename = “c:\winnonln\examples\profiles. Model. WinNonlin. WGO.Load FileType Property Applies To Description Syntax Remarks ANOVA.pwo” MyData.Scripting Commands F E F FileMask Property Applies To Description Syntax Remarks Example WinNonlin Specifies a file in the current working folder. WinNonlin.FileType = {ASCII | ANV | EXCEL | EXCEL95 | EXCEL97 | PWO | BMP | WMF | PCO | PMO | RTF | PTO | LML | SAS | ODBC | PKS}a When loading graph objects. Load. Save is equivalent to export. Do not use directory names or *. When the FileType is set to anything other than PCO.* to select files. and EXCEL file types but cannot save back to them. Font=”Arial” FontSize Property Applies To Description Syntax 594 Data Sets the font size for the current selection. StartCol.StartCol=1 MyData.Replace Font Property Applies To Description Syntax See also Example Data Sets the font for the current selection.Font = string FontSize.SearchArea=”CONC” MYDATA. EndCol MyData.With=”0” MYDATA.FileType = EXCEL a.001” MYDATA.EndRow=1 MyData. Remove and Replace.E WinNonlin User’s Guide FileType (continued)Property Example MyData. Use Excel97 to load Excel2000 files. Include.StartRow=1 MyData.Tolerance=”0.Find=”0” MYDATA. objectname.Find = string Use with Exclude.FontSize = number . ObjectName. MYDATA.EndCol=10 MyData. ObjectName. EndRow.Operation=”=” MYDATA. StartRow. Find Property Applies To Description Syntax Remarks Example Data Specifies data to search for. StartRow=1 MyData.FootnoteFontSize = 12 Function Property Applies To Description Transform Defines the function to be used by the transform engine. EndCol MyData. ObjectName.FootnoteFont = “Arial” FootnoteFontSize Property Applies To Description Syntax See also Example Graph Sets the font size for the chart footnote.StartCol=1 MyData.EndRow=1 MyData.EndCol=-1 MyData. StartCol.FootnoteFont=string FootnoteFontSize MyGraph.Scripting Commands F FontSize (continued)Property See also Example Font.FontSize=12 E FootnoteFont Property Applies To Description Syntax See also Example Graph Sets the font for the chart footnote. EndRow. 595 . StartRow.FootnoteFontSize=number FootnoteFont MyGraph. ObjectName. SourceCol. Extra. DestArea Trans. Table. which sets the number of group variables. DestCol. InData. ) when using the Plot engine. IDVar( ).GroupVar(1) = “Subject” 596 . )). Table. GroupVar applies to all tables except 9 and 10 (see JoinGroupVar( . GroupVarBreak( ).Function = TRANS_LN G GroupVar( ) Property Applies To Description Syntax Remarks See also Example Table Defines the group variables for the Table engine. GroupVars.Function = {TRANS_LN | TRANS_LOG10 | TRANS_SQRT | TRANS_ABS | TRANS_CHANGEBASE | TRANS_PCTCHANGE TRANS_RATIOBASE | TRANS_MULTCOL | TRANS_METABOLITE | TRANS_ADDCOL| TRANS_SUBTRACTCOL| TRANS_E | TRANS_INV | TRANS_POWER | TRANS_MULT | TRANS_DIV | TRANS_ADD | TRANS_SUBTRACT} LN(x) Log10(x) Square Root(x) Absolute Value(x) Change from Baseline %Change from Baseline Ratio from Baseline x*y x/y x+y x-y e^x 1/x x^n x*n x/n x+n x-n Remarks See also Example The string must match exactly (not case-sensitive) one of the strings above. RegVar( ) See GroupVar( .GroupVar(Variable no.) = string This property must be specified after GroupVars. CrossVar( ).E WinNonlin User’s Guide Function (continued)Property Syntax Trans. GroupVars(1) = 2 GroupVarBreak( ) Property Applies To Description Syntax Table Specifies whether text output has a page break when the group variable values change. RegVar( ) Plot.) = string This property must be specified after GroupVars( ).1) = “Subject” Plot. IDVars. When using table 9 or 10. Plot. The example given below sets the first GroupVar for Data Set 1 to Subject.) = integer Plot. CrossVars.GroupVar(1.GroupVars = 1 GroupVars( ) Property Applies To Description Syntax Example Plot Sets the number of group variables from a given data set.GroupVars(Data set no. for the Plot engine. GroupVar( ).Scripting Commands G E GroupVar( .GroupVar(2. which sets the number of group variables.GroupVar(Data set no. IDVar( ).GroupVars = integer Applies to all tables except tables 9 and 10. Variable no. GroupVar( ).) = {True|False} 597 . CrossVar( ). RegVars Table.1) = “Patno” GroupVars Property Applies To Description Syntax Remarks See also Example Table Sets the number of group variables for the Table engine. Table.. Plot. Table. ) Property Applies To Description Syntax Remarks See also Example Plot Defines the group variables for each data set for the Plot engine.GroupVarBreak (Variable no. see JoinGroupVars. ). Missing. see JoinIDVar( . 5.Halt=”See the output” MsgBox Headers Property Applies To Description Syntax Remarks See also Data When Headers = “True. CrossVar( ). WinNonlin. 2. 8. 10.E WinNonlin User’s Guide GroupVarBreak( ) (continued) Property See also Example GroupVar( ). Selecting No stops the run.” the first row of a data file contains column names rather than data.IDVar(Variable no. RegVar( ) . Similar to MsgBox. When using table 9. 3.Headers = {True | False} Not used when loading PWO files. 7.GroupVarBreak(1) = True H Halt Method Applies To Description Syntax Remarks Example See also WinNonlin During runtime. Table.Halt= string Useful in debugging scripts. leaving WNL and all windows open.) = string Applies to table no. GroupVar( ). objectname. 6. The number of ID variables is set by IDVars. ConseqDelim I IDVar( ) Property Applies To Description Syntax Remarks See also 598 Table Defines the ID variables for a table. Delimiter. WinNonlin. Halt stops the script and presents a dialog asking “Do You Want to Continue?” with Yes/No buttons. The default value is that used in the last Load operation. GroupVars Table. Variable no. EntireRow. IDVar( ) defines each variable. 2.IncludeVar(2. Table.Scripting Commands I IDVar( ) (continued) Property Example Table.1) Merge.2) Merge.IDVar(2) = “Time” E IDVars Property Applies To Description Syntax Remarks See also Example Table Sets the number of ID variables for the Table engine. CaseSensitive MYDATA. Tolerance. RegVars Table.) = string IncludeVars( ) Merge.IncludeVar(1.Include IncludeVar( .IncludeVar(2.IDVars = integer Applies to table no. Operation. Find..IncludeVar(Data Set no. When using table 9.SearchArea="conc" MYDATA. SearchArea. Merge. 8.Find="4.2) = = = = “AUC_PO” “Cmax_PO” “AUC_IV” “Cmax_IV” 599 . 5. objectname. 3.1) Merge.IncludeVar(1.Include Exclude.IDVar(1) = “Subject” Table. 6. Set IncludeVars( ) first. CrossVars. ) Property Applies To Description Syntax See also Example Merge Tells the Merge engine which columns to include from each data set. see JoinIDVars. 7.IDVars = 2 Include Method Applies To Description Syntax See also Example Data Includes a range of data. GroupVars.3" MYDATA. 10. Must be set after NumData. ) Merge.IncludeVars(Data Set no.E WinNonlin User’s Guide IncludeVars( ) Property Applies To Description Syntax See also Example Merge Defines the number of variables to be included from each data set in a merge operation. NumData. Plot Defines the data source for the above engines.InData(1) = MyOral Merge.InData = MYDATA InData( ) Property Applies To Description Syntax Remarks See also Example Table. Merge.IncludeVars(1) = 3 Merge. engine.IncludeVars(2) = 2 InData Property Applies To Description Syntax Remarks See also Example Desc.InData(2) = MyIV where MyOral and MyIV are previously defined data objects InitialDose Property Applies To 600 Toolbox . NumData.InData( ) = string The value supplied must be a valid Data object in the script file. engine. InData Merge. Trans Defines the data source for the above engines.InData = string The value supplied must be a valid Data object in the script file.) = integer IncludeVar( . InData( ) Trans. Merge. InputLag={True|False} Toolbox.) = Variable Name Both arguments are required even if there is only one data source and/or join cross variable. CrossVar( ). Toolbox.JoinCrossVar(Data Set no. ) 601 . ). JoinIDVar( . JoinGroupVar( . JoinCrossVars sets the number of join cross variables. Variable no. JoinCrossVars.Scripting Commands J InitialDose (continued) Property Description Syntax Example Sets a value for initial dose. Table.InputLag=False Interval Property Applies To Description Syntax See also Example Desc Specifies the value of the confidence interval for descriptive statistics.Interval = number Confidence. ) Property Applies To Description Syntax Remarks See also Table Defines a common cross variable when merging two data sets using table 9. Percentiles Desc. The default is 95.. Toolbox.Interval = 95 J JoinCrossVar( .InitialDose=number Toolbox. Desc.InitialDose=11 E InputLag Property Applies To Description Syntax Example Toolbox Selects whether WinNonlin will constrain the derivative of the estimated input rate to zero at the initial time (lag time). ) (continued)Property Example Table.JoinCrossVar(1. 602 .1) = “Form” Table.1) = = = = “Form” “Subject” “Form” “Subject” JoinCrossVars Property Applies To Description Syntax Remarks See also Example Table Defines the number of join cross variables for table 9.JoinCrossVar(2. ) Property Applies To Description Syntax Remarks See also Example Table Defines a common group variable when merging two data sets with table 9.JoinGroupVar(1. GroupVar( ).1) Table.JoinGroupVar(1. JoinGroupVars sets the number of join group variables.E WinNonlin User’s Guide JoinCrossVar( .JoinCrossVar(1. JoinIDVar( . Table. JoinCrossVar( .) = Variable Name Both arguments are required even if there is only one data source and/or join group variable. Variable no.JoinCrossVars = integer The two data sets must have the same number of cross variables. JoinIDVars.1) Table. CrossVars. ). ). Table.1) Table. JoinCrossVar( . JoinGroupVars Table.JoinCrossVars = 2 JoinGroupVar( . Table.JoinGroupVar(Data Set no..JoinGroupVars = integer The two data sets must have the same number of group variables. ) Table. JoinGroupVars.1) = “Trt” JoinGroupVars Property Applies To Description Syntax Remarks Table Defines the number of join group variables when merging two data sets with table 9.JoinCrossVar(2. Keo=number Toolbox. JoinIDVars Table. JoinIDVar( .JoinIDVar(1.2) = “PatNo” JoinIDVars Property Applies To Description Syntax Remarks See also Example Table Defines the number of join ID variables from each data set for table 9. Must be from 0 to 1. JoinCrossVars. IDVars. JoinIDVars sets the number of join ID variables. Table. JoinCrossVars Table.Keo=. IDVar( ). ). JoinGroupVar( .JoinIDVars = integer The two data sets must have the same number of ID variables.JoinIDVar(Data Set no.Scripting Commands K JoinGroupVars (continued) Property See also Example GroupVars. Variable no.JoinIDVar(1. ) Table. JoinCrossVar( . JoinIDVars. JoinGroupVars. Toolbox.JoinGroupVars = 1 E JoinIDVar( .) = Variable Name Both arguments are required even if there is only one data source and/or one join ID variable. ). ).1) = “Subject” Table.JoinIDVars = 1 K Keo Property Applies To Description Syntax Example Toolbox Sets the value for Keo. ) Property Applies To Description Syntax Remarks See also Example Table Defines a common ID variable when merging two data sets with table 9.5 603 . JoinGroupVar( .. Table. LamzFile = “c:\lamzdata. . Plot.Legend(1) = “” LegendFont Property Applies To Description Syntax See also Example Graph Sets the font for the chart legend. Plot. Model.LegendFont=string LegendFontSize MyGraph.LamzFile = string See “Special file formats” on page 430 for details on this file. ObjectName.Legend(Data Set no.E WinNonlin User’s Guide L LamzFile Method Applies To Description Syntax Remarks See also Example Model Loads an ASCII data file containing Lambda_z information for NCA models. DoseUnit.) = string Applies only when creating overlaid plots.Legend(1) = “Plasma” Plot.dat” Legend( ) Property Applies To Description Syntax Remarks Example Plot Defines a string to appear in the legend of an overlay plot (instead of the data set name).LegendFont = “Arial” LegendFontSize Property Applies To Description 604 Graph Sets the font size for the chart legend. PartFile Model.Legend(2) = “Urine” Plot. LegendFontSize=number LegendFont MyGraph.LinkFile = string When using compiled. WinNonlin Loads a file into the system using the current file name. or User-ASCII models.Scripting Commands L LegendFontSize (continued) Property Syntax See also Example ObjectName. Text. Save MyGraph. ASCII PK. and possibly lower/upper bounds.LegendFontSize = 12 E LinkFile Method Applies To Description Syntax Remarks Model Load an ASCII data file containing model parameters information for PK/PD and IPR models. ParmFile Model. Filename.Load When using graph objects.Load See also Example LoadTemplate Method Applies To Graph 605 . this method loads workspaces. the Load method works only for file type PCO.LinkFile = “c:\linkdata. loading the Data Object as file type PKS with the file name pointing to a PKS file will load the study with or without a scenario. Graph.dat” See also Example Load Method Applies To Description Syntax Remarks Data. objectname. FileType. ParmFile reads in the model parameters and possibly lower/upper bounds. LinkFile reads in the PK model parameters and ParmFile reads the PD or IPR parameters. Model. Model. DoseUnit. When using the pre-compiled PK/PD or PK/IPR models. This method should not be used for NCA models. When used with the Data Object associated with a PKS study. See “Special file formats” on page 430 for details of these ASCII files. When used with the WinNonlin object. The scenario name is captured from the PKS file pointed to by the Data Object Filename property. Missing = string ConseqDelim. Toolbox.MoveFirst See also Example 606 .MoveLast SortLevel MyGraph. MoveFirst MovePrevious MoveNext MoveLast Methods Applies To Description Syntax Graph In a group of graphs created with a sort variable.MaintenanceDose=number Toolbox.MaintenanceDose=11 Missing Property Applies To Description Syntax See also Example Data Sets the alphanumeric string that represents missing data.MoveFirst objectname. objectname.MoveNext objectname.LoadTemplate “c:\winnonln\temps\xy. Headers MYDATA.LoadTemplate = string SaveTemplate MyGraph.gtp” M MaintenanceDose Property Applies To Description Syntax Example Toolbox Sets a value for maintenance dose. selects which graph is active or displayed. objectname. Delimiter. objectname.MovePrevious objectname.E WinNonlin User’s Guide LoadTemplate (continued)Method Description Syntax See also Example Loads a template on the current graph.Missing = . Weighted Predicted Y. Observed Y Predicted Y MyGraph. Observed Y vs. Weighted Predicted Y vs. and Sort commands. If Multigrid is used on a data set that is currently showing the history.MultiGraph = Graph Name where available Graph Names are: • Modeling: X vs. objectname.MultiGraph = “X vs.. Weighted Residual Y. Weighted Residual Y. use this command to select the appropriate worksheet before specifying the SortVars. SortVar( ).MsgBox = string The script will pause until the user clicks OK to clear the message box from the screen. N Npoints Property Applies To Toolbox 607 . objectname.Scripting Commands N E MsgBox Method Applies To Description Syntax Remarks Example WinNonlin Presents a message box during script runtime. Partial Derivatives • NCA: X vs. X vs. e. Similar to Halt. Observed Y and Predicted Y” Example MultiGrid Method Applies To Description Syntax Remarks Data Selects which sub-grid (worksheet) in a multigrid (workbook) is active. the data will be toggled into view for that object. Final Parameters When sorting data. Observed Y and Predicted Y.g. WinNonlin. or displayed.MultiGrid = Grid Name where Grid Name is the worksheet name. WinNonlin.MsgBox = “Script complete” MultiGraph Method Applies To Description Syntax Graph Changes the visible graph type in multigraphs. NumData = integer When used with Merge. and Replace. NumData must be 1 or 2.NumTerms=4 O Operation Property Applies To Description Syntax Remarks Data Specifies search operator. Value must be 2 or greater.E WinNonlin User’s Guide Npoints (continued) Property Description Syntax Example Number of output data points. 608 .Npoints=100 NumData Property Applies To Description Syntax Remarks See also Example Table.NumData = 1 NumTerms Property Applies To Description Syntax Remarks Example Toolbox Number of exponential terms in the unit impulse response. NumData must be 2. Objectname. Toolbox. with table or Plot.Npoints=number Toolbox. engine. Plot. InData( ) Table. Toolbox.NumTerms=number Range is 1 to 9 Toolbox. Include.Operation = {<> | < | > | <= | >= | = } Use with Exclude. Remove. Merge Indicates the number of data sets to be used with an engine. Desc.With="Missing" MYDATA.OutData = “” specifies a Null alias. created when Run is invoked.CaseSensitive=False MYDATA.OutGraph = “MyGraph” OutText Property Applies To Description Syntax Remarks Example ANOVA.Scripting Commands O Operation (continued)Property Example MYDATA. if any. engine. PK. LinMix. Bioeq. PK. PK.Find="3" MYDATA. Merge Sets the name of the output data object generated by the Run method of any engine. PK Creates a graph object during the Run of an engine.Operation="=" MYDATA.Tolerance="0" MYDATA. no output will be generated. engine. table.SearchArea="CONC" MYDATA. Desc Stats (Box and Whiskers plot) Sets the name of the text object. no output will be generated. no output will be generated.OutText = string engine.OutGraph = string engine.OutGraph = “” specifies a Null alias.OutText = “MYTEXT” 609 . Plot.OutData = “MYDATA” OutGraph Property Applies To Description Syntax Remarks Example Plot. engine.Replace E OutData Property Applies To Description Syntax Remarks Example ANOVA. PK.OutText= “” specifies a Null alias.OutData = string engine. ParmFile reads in the initial values for model parameters. However. and. Accepts a quote delimited string up to 255 characters. DefinitionFile Table. DefinitionFile Table. Model. PageFooter.PageHeader=“This is my Title” ParmFile Method Applies To Description Syntax Remarks Model Loads an ASCII data file containing the model parameter and boundaries information. Table. Accepts a quote delimited string up to 255 characters.PageFooter=“This is my Footnote” PageHeader Property Applies To Description Syntax Remarks See also Example Table Sets a title for a scripted table. when using SCI PK/PD or PK/IPR models. if both are used in a single run. Table. and optional lower/upper bounds.dat” See also Example 610 .ParmFile = “c:\parmdata. LinkFile reads in the PK model parameters and ParmFile reads the PD or IPR parameters. DoseUnit. if both are used in a single run.ParmFile = string When using compiled or ASCII PK models or User-ASCII models.PageFooter = string A footer specified using this command will override any footers specified in a table definition file. lower/upper bounds. See “Special file formats” on page 430 for details on these files. optionally. This method should not be used for NCA models. PageHeader. LinkFile Model.PageHeader = string A title specified using this command will override any titles specified in a table definition file.E WinNonlin User’s Guide P PageFooter Property Applies To Description Syntax Remarks See also Example Table Sets a footer for a scripted table. StartRow.PartFile = string See “Special file formats” on page 430 for the required format of these ASCII files.EndCol=J MYDATA. MYDATA. Interval 611 . Objectname. LamzFile Model. DoseUnit. Desc.PartFile = “c:\partdata.Filename = <file name> MYDATA. Confidence.StartCol=A MYDATA.Paste Use with the Cut or Copy method.StartCol=A MYDATA.dat” Paste Method Applies To Description Syntax Remarks Example Data Pastes a range of data selected with the Cut or Copy property to a data set. EndCol.StartRow=5 MYDATA.Cut MYDATA.Scripting Commands P E PartFile Method Applies To Description Syntax Remarks See also Example Model Loads an ASCII file containing time ranges for the computation of partial areas in NCA. Model.Paste StartCol.StartRow=1 MYDATA.Percentiles = {True | False} The default is False. EndRow Applicable Properties Percentiles Property Applies To Description Syntax Remarks See also Desc When Percentiles = “True.EndRow=10 MYDATA.” percentile calculations will be included with descriptive statistics. PrintModel. PrintGraph. PrintText WinNonlin.E WinNonlin User’s Guide Percentiles (continued) Property Example Desc. PrintData.PrintAll Print. objectname. PrintText WinNonlin. WinNonlin. If the history of a data object is visible. only the current data grid will be printed.PrintData = {True | False} PrintAll. PrintGraph. PrintModel.Percentiles = True Print Method Applies To Description Syntax Remarks See also Example Data. Model.Print PrintAll Method Applies To Description Syntax See also Example WinNonlin Prints all open windows or data grids. the history will be printed rather than the data. WinNonlin. PrintAll MyData.PrintAll PrintData Property Applies To Description Syntax See also Example WinNonlin PrintData = False excludes Data windows from printing with PrintAll. including all worksheets of open workbooks.PrintData=False PrintGraph Property Applies To 612 WinNonlin .Print If used on a multigrid data object. Text Prints the specified object. The default is True. Graph. PrintGraph = {True | False} PrintAll. PrintMultiDown PrintMultiDown Property Applies To Chart 613 . PrintGraph WinNonlin.PrintModel=False PrintMulti Property Applies To Description Syntax See also Chart If True. The default is True. multiple charts are printed per page. Chartobject. PrintData. WinNonlin. PrintMultiDown. WinNonlin. PrintAll PrintMultiAcrossProperty Applies To Description Syntax See also Chart Sets the number of charts printed across the page. Chartobject. PrintText.PrintMultiAcross = number PrintMulti.PrintMulti = {True | False} PrintMultiAcross. PrintData WinNonlin.Scripting Commands P PrintGraph (continued)Property Description Syntax See also Example E PrintGraph = False excludes Graph windows from printing with PrintAll. PrintModel.PrintModel = {True | False} PrintAll.PrintGraph=False PrintModel Property Applies To Description Syntax See also Example WinNonlin PrintModel = False excludes the Model window from PrintAll. PrintText. The default is True. For table 8 only one regular variable may be specified. The default is True. WinNonlin.RegVar(1) = “Tmax” Table. IDVar( ). PrintData.RegVar(2) = “Cmax” Table. RegVars sets the number of regular variables. GroupVar( ) Table. the regular variables are the variables from Data Set 2.PrintText=False Q R RegVar( ) Property Applies To Description Syntax Remarks See also Example Table Defines a ”regular” variable that composes the body of a table. For tables 9 and 10.E WinNonlin User’s Guide PrintMultiDown (continued)Property Description Syntax See also Sets the number of charts printed down the page.RegVar(3) = “AUClast” RegVars Property Applies To 614 Table . Chartobject. CrossVar( ). Table.RegVar(Variable no.PrintMultiDown = number PrintMulti. PrintModel. PrintMultiAcross PrintText Property Applies To Description Syntax See also Example WinNonlin PrintText = False excludes Text windows from printing with PrintAll. PrintGraph WinNonlin.) = string Applies to all table templates.PrintText = {True | False} PrintAll. Operation.Scripting Commands R RegVars (continued)Property Description Syntax Remarks See also Example Defines the number of regular variables for a table.Remove Find.RegVars = 3 E Remove Method Applies To Description Syntax See also Example Data Removes rows fitting search criteria from a workbook.Column="a" MYDATA. Tolerance. Objectname.RegVars = integer Applies to all table templates. RegVar( ). CaseSensitive MYDATA. objectname. For table 8 only one regular variable may be specified.RemoveCol RemoveRow Method Applies To Description Data Removes a range of rows from a workbook. Identify the first and last row in the range by row number.SearchArea="TIME" MYDATA. CaseSensitive MYDATA. For tables 9 and 10. IDVars.Operation="=" MYDATA. SearchArea.RemoveCol Column. 615 . the regular variables are the variables from Data Set 2.Remove RemoveCol Method Applies To Description Syntax See also Example Data Removes a specified column from a workbook. CrossVars. Table.Find="3" MYDATA. GroupVars Table. Identify the column by column name or letter. CaseSensitive MYDATA. objectname.StartRow=1 MYDATA.Find="3" MYDATA.CaseSensitive=False MYDATA.With="Missing" MYDATA. With. EndRow Applicable Properties RenameCol Method Applies To Description Syntax See also Example Data Renames a column. Identify the column to be renamed by column name or letter.SearchArea="CONC" MYDATA. With.RemoveRow MYDATA.With=”Subject” MYDATA.Replace 616 . Tolerance.EndRow=18 MYDATA.RenameCol Column. Operation. SearchArea.E WinNonlin User’s Guide RemoveRow (continued)Method Syntax Example objectname. CaseSensitive MYDATA.RemoveRow StartRow. objectname.RenameCol Replace Method Applies To Description Syntax See also Example Data Replaces data specified by the Find property with data specified by the With property.Tolerance="0" MYDATA.Operation="=" MYDATA.Replace Find.Column=”a” MYDATA. use WordExportFilename to have Word automatically save the output file. Trans.pmo” MyModel. Run.Load PK.Response=column name Toolbox. table. creates a new document.RunWordExport 617 . engine.Response=”Response” Run Method Applies To Description Syntax Remarks Example ANOVA. Plot. RunWordExport requires Word to be installed on the same machine as WinNonlin.WordExportFilename = “bg1. and populates it with the PK output. WordExportFilename Set MyModel = Model MyModel.FileType = “PMO” MyModel. PK.RunWordExport Optionally.Filename = “bg1. LinMix. Bioeq. Toolbox. Desc. PK.Run Be sure to name all possible output objects prior to using the Run method.Run RunWordExport Method Applies To Description Syntax Remarks See also Example PK Runs the PK analysis and launches Word (if not already running).Scripting Commands R E Response Property Applies To Description Syntax Example Toolbox Specifies column for response data. Desc.doc” PK. Merge Invokes the specified engine with its currently defined properties. The Model. DnErr( ) Plot.) = {True | False} The argument is required even if there is only one data source. File names use SaveAllPrefix plus a number. If True. The PKS file contains all information necessary to save the workspace.E WinNonlin User’s Guide S SameErrCol( ) Property Applies To Description Syntax Remarks See also Example Plot Determines whether the same column is to be used for both up and down error bars. Filename. objectname. Model.SaveAll MYOUTGRAPH. if one exists.Save When working with graph objects. FileType. must be saved as a PMO. Graph.SameErrCol(Data Set no. Plot. The default is False.SaveAll 618 . WinNonlin Saves a file to disk using the current file name. if the FileType is anything other than PCO. then the column defined by UpErr( ) is used for the error data. Saving Workspaces: Saving a workspace requires that all WinNonlin objects be saved first.SaveAllPrefix="graf" MYoutgraph. Save is equivalent to export.Save See also Example SaveAll Method Applies To Description Syntax Example Graph Saves each graph in a multipart graph to a file. objectname. Load MyGraph. Text. Saving PKS: Saving a PKS workspace requires that the save operation be performed on the study workbook data object as a file of type PKS. All multigrids must be saved as PWO's. UpErr( ).SameErrCol(1) = True Save Method Applies To Description Syntax Remarks Data.filetype="BMP" MYOUTGRAPH. SaveAll SaveTemplate Method Applies To Description Syntax See also Example Graph Saves the current graph settings to a file. Include.CaseSensitive=False MYDATA. as a graph template. rema Save all plots as separate . objectname.Find="3" MYDATA.Operation="=" MYDATA. objectname.Scripting Commands S E SaveAllPrefix Property Applies To Description Syntax Remarks Example Graph Specifies file name prefix for saving multipart graphs. MYDATA.SearchArea = {column name | “entire data set”} It is not possible to search a “Current Selection” using the Scripting Language.WMF files MYOUTGRAPH.filetype="wmf" MYOUTGRAPH.gtp” SearchArea Property Applies To Description Syntax Remarks Example Data Specifies where the Exclude. Use with SaveAll.SaveTemplate = string LoadTemplate MyGraph.Replace 619 . Remove or Replace should focus. objectname.SaveAllPrefix="plot" MYOUTGRAPH.SearchArea="CONC" MYDATA.SaveAllPrefix = string This prefix must be between 1 to 4 characters in length.SaveTemplate “c:\winnonln\temps\xy.With="Missing" MYDATA.Tolerance="0" MYDATA. objectname.Sort SortVar( ).Smoothing={Automatic|None|User} Toolbox.ShowUnits={True|False} XUnits. SortVars SortLevel Method Applies To 620 Graph . YUnits Plot. Toolbox.ShowUnits=False Smoothing Property Applies To Syntax Example Toolbox Toolbox.Sequence=”Sequence” ShowUnits Property Applies To Description Syntax See also Example Plot When ShowUnits = True.Sequence = column name Toolbox. The default value is True.Smoothing=”Automatic” Sort Method Applies To Description Syntax See also Data Sorts a data set using the specified sortkeys.E WinNonlin User’s Guide Sequence Property Applies To Description Syntax Example Toolbox Specifies input column for sequence data. Plot. any units defined by XUnits or YUnits will appear after the respective axis title. both arguments are required even if there is only one data source and/or sort key.) = string For the Plot engine. Desc. separated by a single space. GroupVar( ) Data. Must be set afterSortVars or SortVars( ). no. Data. there must be a value even if there is only one sort key.Scripting Commands S SortLevel (continued)Method Description Syntax Remarks See also Example E Allows navigation among plots containing sort keys by going directly to a specified sort level. SortVars( ). objectname. Merge Defines the number of sort keys to be used by the specified object or engine. ).) = string SortVar( . ) Property Applies To Description Syntax Remarks See also Plot. Must be set after SortVars. Engine.. SortLevel must contain a value for each key in order.SortLevel = string If there is more than one sort key.SortVar(Data Set no. engine.SortVar(1) = “Form” Data. SortVars( ). Desc Defines the sort keys to be used by the specified engine.SortVars = integer or objectname. Sort Var.SortVars = integer 621 . MovePrevious. MoveFirst. GroupVar( ) SortVars Property Applies To Description Syntax Data.SortVar(Variable no.SortLevel = “SubjA Cap” SortVar( ) Property Applies To Description Syntax See also Example Data. Merge Defines the sort keys to be used. For the Desc engine. MoveLast MyGraph. MoveNext.SortVar(2) = “Time” SortVar( . stacked={True|False} True=stacked. Plot. SortVars. even though there are two data sources. and not SortVars( ).stacked=True 622 .) = integer There must be a value even if there is only one data source.SortVars = 1 SortVars( ) Property Applies To Description Syntax Remarks See also Example Plot Defines the number of sort keys to be used by the Plot engine.SortVars(Data Set no.SourceCol = “Conc” Stacked Property Applies To Description Syntax Remarks Example Toolbox Sets whether treatment data are in separate columns or stacked in one column in the input. GroupVars( ) Desc. SortVar( ). Trans. the number of SortVars must be the same for each. Toolbox. SortVar( ). SortVars( ). GroupVars( ) Plot.SortVars(1) = 2 SourceCol Property Applies To Description Syntax See also Example Trans Defines the source column for transformations.SourceCol = string DestCol Trans. False=separate Toolbox. Therefore SortVars is used.E WinNonlin User’s Guide SortVars (continued)Property Remarks See also Examples When working with the Merge engine. StartCol can be a string identifying the column name (for columns already containing data) or a number identifying the column position. Set StartCol <= EndCol. Both are required. Cut and Copy.StartCol MyData. The latter is required to designate a new column.EndRow = MyData.Scripting Commands S E StartCol Property Applies To Description Syntax Remarks Data Specifies the first column in a range. objectname. all rows will be selected.EndRow=18 MYDATA. If StartRow is –1 and EndRow is not set.StartCol = {string | number} Use with RemoveRow.StartRow = number Set StartRow <= EndRow. EndRow.StartCol MyData. EndCol MyData. Cut and Copy or Font and FontSize.EndCol = MyData.StartRow MyData. StartRow.StartRow MyData.RemoveRow Stats Property Applies To Table 623 .EndRow = MyData.Copy See also Examples = 1 = “Time” 15 “Time” = 1 = 1 15 3 StartRow Property Applies To Description Syntax Remarks Example Data Specifies the first row in a range for RemoveRow. unless StartRow = –1.EndCol = MyData. Objectname.Copy MyData. MYDATA.StartRow=1 MYDATA. SummaryVars = 1 624 .Stats = {True|False} Applies to table no. 7. Causes all of the available statistics to be printed.) = string SummaryVars Desc.Subject = column name Toolbox. 10. 3. Desc. Table. 1. Desc. Must follow SummaryVars. 4. 6.E WinNonlin User’s Guide Stats (continued)Property Description Syntax Remarks Example Sets whether to generate descriptive statistics in the Table engine.SummaryVars = integer SummaryVar( ) Desc.SummaryVar(Data Set no. 8.Stats = True Subject Property Applies To Description Syntax Example Toolbox Specifies input data set column for subject identifiers. Toolbox.SummaryVar(1) = “Conc” SummaryVars Property Applies To Description Syntax See also Example Desc Sets the number of summary variables for descriptive statistics.Subject=”Subject” SummaryVar( ) Property Applies To Description Syntax See also Example Desc Specifies a variable to for which to calculate descriptive statistics. 9. Table. Tab = {Data | History} If MultiGrid is used on a data set that is currently showing the history. TableNum Property Applies To Description Syntax Remarks See also Example Table Identifies a table template number. or using a table definition file.TerminalLower = number Toolbox. DefinitionFile Table. Table.TerminalLower=1 625 . objectname.Tau=1 TerminalLower Property Applies To Description Syntax Example Toolbox Lower range for terminal phase. Toolbox.Tau = number Toolbox.TableNum = integer Tables may be defined in a script file either by using scripting language commands.TAB value will automatically be set to Data–that is.Scripting Commands T E T Tab Property Applies To Description Syntax Remarks Data Activates the Data or History tab of a data object. Toolbox. including TableNum. the data will be toggled into view for that object.TableNum = 2 Tau Property Applies To Description Syntax Example Toolbox Specifies value for Tau. the . 0=False.tilehorizontal TileVertical Property Applies To Description Syntax See also Example WinNonlin Arranges open objects to fill the WinNonlin window without overlapping. If True. optimized for height. WinNonlin. Toolbox.TileHorizontal TileVertical and Cascade WinNonlin. WinNonlin.TerminalPhase=1 TerminalUpper Property Applies To Description Syntax Example Toolbox Upper range for terminal phase. use with TerminalLower and TerminalUpper.TerminalUpper = number Toolbox.TerminalPhase = { 1 | 0 } 1=True. Toolbox.TileVertical TileHorizontal and Cascade WinNonlin. Toolbox.E WinNonlin User’s Guide TerminalPhase Property Applies To Description Syntax Remarks Example Toolbox Sets whether to include a user-defined terminal phase with Nonparametric Superposition.TerminalUpper=5 TileHorizontal Property Applies To Description Syntax See also Example WinNonlin Arranges open objects to fill the WinNonlin window without overlapping.tilevertical 626 . optimized for width. Toolbox.TitleFont = “Arial” 627 . Plot.Scripting Commands T E TimeCol Property Applies To Description Syntax Example Toolbox Specifies input data column for Time.Title = “My Plot Title” TitleFont Property Applies To Description Syntax See also Example Graph Sets the font for the chart title. ObjectName.TimeCol = column name Toolbox.TimeRepeat = number Toolbox. Toolbox.TitleFont = string TitleFontSize MyGraph.TimeRepeat=5 Title ‘Property Applies To Description Syntax See also Example Plot Defines the string to be displayed in the title of the plot. YTitle.TimeCol=”Time” TimeRepeat Property Applies To Description Syntax Example Toolbox Sets repeat dosing interval to every number time units.Title = string XTitle. Legend( ) Plot. Remove and Replace. MYDATA.Tolerance="0" MYDATA.Find="3" MYDATA. Toolbox.Tolerance = number Use with Exclude.TreatmentA = ”Treatment1” TreatmentB Property Applies To Description Syntax 628 Toolbox Specifies input data set column to be used for second treatment. Include.TitleFontSize=12 Tolerance Property Applies To Description Syntax Remarks Example Data Specifies tolerance range for numeric data searches.With="Missing" MYDATA.E WinNonlin User’s Guide TitleFontSize Property Applies To Description Syntax See also Example Graph Sets the font size for the chart title.TreatmentB = column name .TreatmentA = column name Toolbox.SearchArea="CONC" MYDATA.CaseSensitive=False MYDATA.TitleFontSize = number TitleFont MyGraph. objectname.Operation="=" MYDATA.Replace TreatmentA Property Applies To Description Syntax Example Toolbox Specifies input data set column to be used for first treatment. Toolbox. ObjectName. Objectname.Filename="MODEL.DoseFile="DOSES.OutData=OUTDATA1 Pk.Run Example 629 .LIB" MYMODEL.Column = ”Concentration” MyData.txt" Pk.FileType="ASCII" MYMODEL.Uniform = {True | False} MYOUTGRAPH.Units="units. The units file can be created in a text editor or using the Save button on the Units tab of the Model Options dialog.Scripting Commands U TreatmentB (continued)Property Example Toolbox.UNIFORM = True Unit Property Applies To Description Example Data Object Assigns a unit to a column. MyData.DAT" Pk. See also “Special file formats” on page 430.OutText=OUTTEXT1 Pk. Set MYMODEL=Model MYMODEL.LamzFile="LAMBDAZ.Load MYMODEL.TreatmentB=”Treatment2” E U Uniform Property Applies To Description Syntax Example Graph Specifies that plots created with a sort variable should have uniform axes across sort levels.DAT" MYMODEL. Used in conjunction with Column.Unit = ”ng/ml” Units Property Applies To Description PK engine Assigns a units file to provide preferred parameter units for non-compartmental analysis. UpErr(1) = “Maximum” Plot. Objectname. Load MyModel. Desc. Desc.Weight = string If a weight variable is not specified.UpErr(1) = “SE” V W Weight Property Applies To Description Syntax Remarks Example Desc Sets the weight column for weighted descriptive statistics.UpErr(Data Set no. SameErrCol( ) Plot.Weight = “Weight” 630 . uniform weighting will be used.E WinNonlin User’s Guide Unload Method Applies To Description Syntax Remarks See also Example Plot Specifies that plots created with a sort variable should have uniform axes across levels.Unload UpErr( ) Property Applies To Description Syntax Remarks See also Examples Plot Defines the column containing up error data for a plot with error bars. DnErr( ).) = string The argument is required even if there is only one data source.” then UpErr( ) is used for both up and down error. Using Unload on the WinNonlin object causes the application to close. If SameErrCol( ) = “True. Plot.Uniform = {True | False} Unload should be applied to any object that will not be referenced any longer in a script. With = string MYDATA.Column=”a” MYDATA. RunWordExport will create. Model.WindowState = Minimized With Property Applies To Description Syntax Example Data Sets a replacement string for use with RenameCol and Replace. Using this property on the WinNonlin object allows a script to run with the application minimized. but not save.Filename = “bg1. Word must be installed on the same machine as WNL.RunWordExport WorkingDir Property Applies To WinNonlin 631 . objectname.pmo” MyModel. Text.RenameCol WordExportFilename Property Applies To Description Syntax Remarks Example PK Sets the file name for the Word export method.WordExportFilename = string Use only with RunWordExport. WinNonlin Sets the window state of the specified object.With=”Subject” MYDATA. Objectname.WindowState = {Minimized | Maximized | Normal} WinNonlin.doc” PK. Set MyModel = Model MyModel. If WordExportFilename is not set. the output document. Graph.Scripting Commands W E WindowState Method Applies To Description Syntax Example Data.Load PK.FileType = “PMO” MyModel.WordExportFilename = “bg1. PK. ” as it removes the need to specify absolute paths for data files in the script—i.. Plot. WorkingDir = string This is particularly useful for making script files “portable. file name alone will suffice.XLabelsFont = string XLabelsFontSize MyGraph.e. WorkingDir = “C:\Newdir” Example X XLabelsFont Property Applies To Description Syntax See also Example Graph Sets the font for the X axis labels. Legend( ) 632 .E WinNonlin User’s Guide WorkingDir (continued)Property Description Syntax Remarks Sets the directory in which WinNonlin will load and save files. YTitle. ObjectName.XLabelsFont = “Arial” XLabelsFontSize Property Applies To Description Syntax See also Example Graph Sets the font size for the X axis labels.XLabelsFontSize = 12 XTitle Property Applies To Description Syntax See also Plot Defines the string to be displayed in the x-axis title.XTitle = string Title. ObjectName.XLabelsFontSize = number XLabelsFont MyGraph. if the files are located in the working directory. ObjectName.XTitleFont = string XTitleFontSize MyGraph.XUnits = ‘hr” XUnits Property Applies To Description Plot Defines the units to be displayed in the x-axis title (in parentheses after the title).XTitle = “Time” E XTitleFont Property Applies To Description Syntax See also Example Graph Sets the font for the X axis title.XTitleFont = “Arial” XTitleFontSize Property Applies To Description Syntax See also Example Graph Sets the font size for the X axis title. ObjectName.XTitleFontSize = 12 XUnits Property Applies To Description Syntax See also Example Graph Defines the units to be displayed in the x-axis title (in parentheses after the title).Xunits = string YUnits MyGraph. ObjectName. 633 .XTitleFontSize = number XTitleFont MyGraph.Scripting Commands X XTitle (continued) Property Example Plot. Plot.) = string The argument is required even if there is only one data source.E WinNonlin User’s Guide XUnits (continued) Property Syntax See also Example Plot.YLabelsFont = “Arial” YLabelsFontSize Property Applies To Description Syntax See also Example Graph Sets the font size for the Y axis labels.YLabelsFont = string YLabelsFontSize MyGraph. ObjectName.YLabelsFontSize = number YLabelsFont MyGraph. ObjectName. Plot.XUnits = “hr” XVar( ) Property Applies To Description Syntax Remarks Example Plot Defines the X variable column for a plot.XVar(1) = “Time” Y YLabelsFont Property Applies To Description Syntax See also Example Graph Sets the font for the Y axis labels.XVar(Data Set no. YUnits Plot.XUnits=string ShowUnits.YLabelsFontSize = 12 634 . YUnits = “mg/ml” 635 .YTitle = string Title.YTitleFontSize = number YTitleFont MyGraph. Plot. ObjectName.YUnits = string XUnits MyGraph.YTitleFont = string YTitleFontSize MyGraph.Scripting Commands Y E YTitle Property Applies To Description Syntax See also Example Plot Defines the string to be displayed in the y-axis title.YTitle = “Concentration” YTitleFont Property Applies To Description Syntax See also Example Graph Sets the font for the Y axis title.YTitleFont = “Arial” YTitleFontSize Property Applies To Description Syntax See also Example Graph Sets the font size for the Y axis title.YTitleFontSize = 12 YUnits Property Applies To Description Syntax See also Example Graph Defines the units to be displayed in the y-axis title of the plot (in parenthesis after the title). Legend( ) Plot. ObjectName. ObjectName. XTitle. YVar(Data Set no.YUnits = string ShowUnits. Plot.YVar(1) = “Concentration” 636 .E WinNonlin User’s Guide YUnits Property Applies To Description Syntax See also Example Plot Defines the units to be displayed in the y-axis title of the plot (in parenthesis after the title).YUnits = “mg/ml” YVar Property Applies To Description Syntax Remarks Example Plot Defines the Y variable column for a plot. Plot. XUnits Plot. Plot.) = string The argument is required even if there is only one data source. Program terminated.g. The preceding statement contains a syntax error (e. See Microsoft’s documentation for information on error messages from Visual Basic and the Windows OCX's. 10001 Invalid statement. 10002 Unmatched parenthesis. as follows: Source Microsoft Visual Basic & OCX'sa Compartmental modeling LinMix and Bioequivalence wizards Table Wizard Descriptive statistics Noncompartmental analysis IVIVC Scripting language Message numbers 1-9999 10000-10999 11000-11999 12000-12999 13000-13999 14000-14999 15000-15999 19000-19999 a. Program terminated. C = A+ *B). Each has a unique set of error message numbers.Appendix F Warnings and Error Messages Messages by category There are six sources of error messages in WinNonlin. The preceding statement contains an extra left parenthesis. Compartmental modeling See also “Troubleshooting modeling problems” on page 209. 637 . 10004 Missing END statement. 0-9. &) 10010 Model too complex. Workspace is not large enough to accommodate the model. Program terminated. Program terminated. Function numbers must be integers. assign expressions to temporary variables (in the TEMP section) before using them to define the model. Rerun with larger model size (SIZE command). Program terminated. -. An invalid FUNCTION number was assigned. To help alleviate this problem. please bring it to the attention of Pharsight customer support. SECONDARY. This is an unusual error. 10005 Invalid number.. greater or equal to 1. 10011 Unmatched parenthesis. (). 10012 Invalid function number. DIFFERENTIAL. DZ(1. (i. An “END” statement was missing for one or more of the following labeled groups of commands: TRANSFORM. FNUMBER. +.1) or ABS (A. The model was too complex for the default memory settings. STARTING. 10007 Invalid variable name or character.F WinNonlin User’s Guide 10003 Improper statement found when evaluating model. not A-Z. There was an unmatched or missing right parenthesis. The previous equation contains an invalid character. 10006 Model too complex. or too many arguments were encountered in a subscribed variable or function (e. Program terminated. or the Model size option on the PK Settings tab. **. 638 . TEMPORARY. *. in the Model Options dialog box). Program terminated. If it occurs. Try the model again after increasing the model size (via the SIZE command. Program terminated. The previous equation contains an illegal number (constant). and less than or equal to the number of functions in the NFUNCTIONS command.g. COMMANDS.e. /. Program terminated.B)). Program terminated. The maximum number of variables and operators has been exceeded (2000). Rerun with larger model size (SIZE command). or is missing the ending colon “:”. For example. Program terminated. Program terminated.g. An invalid label was assigned to a group of commands (e. Rerun with larger model size (SIZE command). Program terminated.B)) 10015 Model section does not exist. Split it up into two or more parts using temporary variables. with a GOTO) which was never defined.Warnings and Error Messages Compartmental modeling 10013 F Invalid group identifier. STRT in place of START). 10014 Missing argument in function or array. for an ELSE statement. 10023 No matching IF. Either the label is missing entirely. Program terminated. 10019 Missing ENDIF statement. Program terminated. The maximum number of temporary variables and labels (100) has been exceeded.g. 10017 Equation too complicated. An ENDIF statement is missing. Program terminated. 10020 Undefined (GOTO) label. At least one of the equations was too complicated to evaluate. THEN.. Program terminated. A labeled section does not exist. MAX(A) instead of MAX(A. 639 . Change the label to a valid name. Check to see if NSEC > 0 but no secondary parameters are defined. but the secondary parameters were never defined. An ENDIF statement was encountered without a matching IF statement. A label was referenced (e. 10018 No IF corresponding to ENDIF..g.. Program terminated. Program terminated. An argument was missing for a function (e. or if the FUNCTION block(s) is missing. Program terminated. an NSEC command was specified. 10021 Invalid (GOTO) label name. The label name is invalid. 10022 Model too complex. 10029 Invalid statement encountered when evaluating model. Program terminated. The argument for loge(x) was invalid. 10027 Invalid array subscript. Perhaps X was undefined. or 0 or a negative number. Change the label to a new name. Program terminated. 10028 Invalid statement encountered when evaluating model. Program terminated. Program terminated. The argument for log10(x) was invalid. Program terminated. 10024 IF statements nested too deeply. An invalid branching occurred during execution. Program terminated. The model has gone outside allowable limits when accessing some variable which is an array (e. IF statements can only be nested as deep as 10 levels. 10026 Duplicate label name. 640 . Check to see if subscript exceeded array bound. Program terminated.F WinNonlin User’s Guide An ELSE statement was encountered without corresponding IF . or 0 or a negative number. Perhaps X was undefined. DTA (11)). Check to see that the model is specified correctly. This is not supported. Program terminated. 10025 Cannot branch out of program section. A duplicate label name was encountered. Program terminated. An attempt was made to branch from one block of statements (TEMP. Attempt to divide by zero.THEN statements. 10034 Invalid argument for LOG10(X). FUNC. etc.g. An invalid comparison occurred during execution.) to a labeled section in another block of statements using a GOTO. Invalid argument for LOGE(X). Program terminated. Check to see that the model is specified correctly. 10030 10031 10033 Operation attempted using A missing VALUE. An invalid array access was attempted. NEXT statement missing corresponding DO. 10036 Invalid argument for ATAN2(X. 10051 Too many nested DO statements. 10050 DO statement missing a NEXT statement. X must be the column number of a defined variable. Program terminated. No input record may exceed 256 characters in length. Perhaps X was undefined or a negative number. Notify Pharsight technical support for assistance at [email protected]. Program terminated. A NEXT statement ends a DO loop. Program terminated. 10054 10055 10056 Maximum length of DO statement is 788 characters. Check all DO statements for corresponding NEXT statements.Y). DO statements may only be nested 10 deep. or perhaps only one argument was given. Program terminated. 10037 Problem exceeds available resources. A NEXT statement must end each DO loop. Program terminated. 10052 Invalid DO statement. Program terminated. Program terminated. Please check the model specification for problems. Check all NEXT statements for corresponding DO statements. 10057 String length exceeds 256. Program terminated. Program terminated. The argument for lag(x) was invalid. Perhaps either X or Y was undefined or a negative number. Program terminated. The argument for SQRT(x) was invalid. 10058 Invalid lag function argument. Maximum number of DO statements is 999. Program terminated. 10053 Maximum model size is 64. The argument for ATAN2(x. Check the arguments and syntax of any DO statements.Warnings and Error Messages Compartmental modeling 10035 F Invalid argument for SQRT(X). Program terminated. 641 .y) was invalid. The program was unable to read NPARM numbers on the INITIAL command. 10109 Error encountered when reading initial parameter values. LOWER and UPPER commands. Turning off of convergence checks is only supported for integrated equations.g. were not given. The program was unable to read NFUN numbers on the NOBS command. No NPARAMETERS command was encountered. Using default units. INIT. 10106 The number of parameter values. No INITIAL command was encountered.2 . 10102 A DATA command was not encountered. 642 . The number of observations to be input (NOBS). must be specified prior to INIT. 10110 Error encountered when reading the number of NOBS for each function. was not given. must precede DATA. OR UPPER NPARAMETERS must precede INITIAL. must be specified prior to the DATA command. LOWER.F WinNonlin User’s Guide 10101 The number of data observations. The preferred units were inconsistent with the default units or were not acceptable units. or one of the numbers was not carefully defined (e. 10103 The number of parameters to be estimated. The number of data observations to be input (NOBS) was not defined. was not defined. 10111 The number of functions must be greater than 0. NPARM.E-3). 10104 The initial estimates of the parameters. 10108 The number of data observations. 10107 Error converting units for next value. Missing DATA command. NOBS. NPARM. NOBS. Either fewer than NPARM initial estimates were given. 10105 CONV = 0 is only supported for integrated functions (NDER=0). 1. The specified number of functions must be between 1 and 10. 643 .Warnings and Error Messages Compartmental modeling 10112 F The number of parameters to be estimated must be greater than 0. Method 3: Hartley's Modification of the Gauss-Newton Algorithm. Program defaulting to method 2. 10118 The number of rows in the plots must be between 10 and 50. An invalid value was assigned to the CONVERGENCE command. An invalid value was entered. The specified METHOD must be 1. A maximum of 10 variables (columns) can be read in from a data file. 10113 The number of derivatives must be greater than 0. Execution continuing. Program defaulting to 1. 2 or 3. execution continuing. YNUMBER) must be between 1 and 10. The column numbers associated with X and Y (XNUMBER. Program terminated. An invalid value was entered. 10116 The number of variables in the input file must be greater than 0. PRogram defaulting to 50. execution continuing. 10115 The maximum number of iterations must be greater than or equal to zero. 10119 The number of columns in the plots must be between 10 AND 100. 10114 The specified method for estimating the residual sum of squares must be 1. Program defaulting to 100. 2 or 3. execution continuing. 10117 The column numbers corresponding to X and Y must be greater than 0 and must be less than or equal to NVAR. Method 2: Hartley's and Levenberg's modification of the Gauss-Newton Algorithm (default). execution continuing. Program terminated. Method 1: Nelder-Mead. 10120 The specified convergence criterion must be greater than or equal to 0.E-4. Program defaulting to 50. If the system includes differential equations. The specified number of parameters must be between 1 and 10. An invalid value was assigned to maximum number of ITERATIONS. the number of differential equations in the system must be between 1 and 10. Execution continuing without limits on the parameter space. Fewer than NOBS observations were on the data file. Check to see that the model was correctly specified. Either the program was expecting a number to be specified and none was given. If LOWER limits are specified.E-3).g. b. 10127 If lower (or upper) limits are specified then upper (or lower) limits must also be specified. An error was encountered when trying to numerically integrate a system of differential equations. The increment used for estimating the partial derivatives DINC must be between 0.F WinNonlin User’s Guide 10121 The increment for estimating partial derivatives must be between 0 AND 0. This can usually be attributed to an improperly defined model. Either fewer than NPARAMETERS numbers were on the card or one or more invalid numbers were specified (e. 10123 The variable number corresponding to weight must be greater than 0. 10128 An improper call was made to the numerical integration routine. An error was encountered when reading the lower (or upper) parameter limits. 10125 Error on data file An error was encountered when reading the data file. they must have an assigned column number greater than one.2 . There are several ways this can happen: a. An error was encountered when trying to read the (first) number from the preceding command. Fewer than NVAR numbers were on one or more rows on the data file.E-3).2 . or the number was not correctly defined (e. Program defaulting to 0. Check your input and model. If weights are to be read in from the data file.g. 1. Execution continuing without limits on the parameter space.0 and 0.001. then UPPER limits must also be specified (and vice-versa).g. X and Y cannot be assigned to the same column in the input data file. 10122 The variable names for X and Y cannot be equal. One or more illegal numbers (e. execution continuing.1. 10124 Error encountered when trying to read a number on the preceding command. 644 . Program terminated.2 .E-3). 1. c.1. 1. 10126 Error encountered when reading in the upper or lower parameter limits. One or more of the initial parameter estimates is outside the specified upper and lower limits. 10132 The number of constants must be specified before assigning them values. Check to see that the model was correctly specified.Execution continuing. Relative error and-or absolute error tolerances were reset. The NCON command must appear prior to the CONSTANTS command. If CONSTANTS are used. Execution continuing A very large number of function evaluations are having to be made to obtain the desired precision. Execution continuing. 645 . The program is having difficulty obtaining the necessary precision when integrating a system of differential equations. Either fewer than NCON values were specified. 10131 The number of constants must be greater than 0. 10134 10135 Parameter estimate for <parameterName> is at or near boundary. The referenced library contains a model statement without a model number. 10133 An error was encountered when reading the values assigned to the constants.g. Check your model for possible singularities. 1. 10138 A convergence failure occurred during a singular value decomposition.2 .g. MODEL 7). Program terminated. 10130 The program is having to work very hard to obtain the desired accuracy. or one of the values was not correctly defined (e. Parameter limits ignored . Check the model for possible singularities. Check the model for possible singularities. Each MODEL statement in a WinNonlin library must also have a corresponding model number (e. the number of user input constants must be between 1 and 500. Make sure your model is correct. Program terminated. An initial parameter estimate is outside of the specified upper and lower limits.E-3).Warnings and Error Messages Compartmental modeling 10129 F Too much relative accuracy was requested for this problem. An error was encountered when trying to read the NCON values assigned to constants. Program terminated. 10136 10137 An error was encountered when reading constant names. If secondary parameters are specified. the number of secondary parameters must be between 1 and 50. 10143 An error was encountered when reading parameter. data or unit names. The specified TITLE number must be 1. 4. An error was encountered reading the parameter or secondary parameter names. 2. 10144 10145 The specified model number does not exist in the specified library. 10142 NSEC must be greater than 0. 3. 10147 Scale values could not be computed for the next plot. 10146 The specified number of plot points must be between 10 and 20000. 10139 NCON was specified but the constants were not input. This could happen.F WinNonlin User’s Guide Check to make sure that the model has been correctly defined. The specified number of points to be included in the output plot file must be between 10 and 20000. Axis scale values could not be computed for a plot. Make sure that the number of input names matches NPARAMETERS (or NSECONDARY) and that each name is enclosed in single quotes. 3. The NSECONDARY command must precede the SNAMES command. Program continuing with default names. 4. OR 5. Plot deleted. TITLE 3). 10148 The number of constants (NCON) is inconsistent with the number of administered doses. for example. The specified title number must be 1. 2. 10140 NSEC must be specified prior to the SNAME command. An NCONSTANTS command was encountered but the constants were never defined (CONSTANTS). NPOINTS reset to default.g. 646 . or 5 (e. if all values to be plotted were identical. secondary. The number of constants (NCONSTANTS) is not consistent with the number of doses given. Program terminated. Try using different initial estimates. An error occurred when reading the parameter values for the PK model. Check to make sure NVAR is correct.Warnings and Error Messages Compartmental modeling 10149 F The WEIGHT command cannot be used when performing simulations. 10157 The value of FNUM is invalid. Make sure that the missing value code is eight characters or less in length. Check model definition statements to make sure WT is defined correctly. The Gauss-Newton algorithm was unable to converge. Weight set to missing. Check for proper specification of the model file. 10151 The algorithm is unable to converge. 10161 Inadequate system resources available for this problem. Convergence assumed. or try rerunning the problem using the Nelder-Mead algorithm. The variance-covariance matrix for the model parameters is singular. These are the values for selecting the Lambda Z time interval. REWEIGHT will be used. The WEIGHT command was changed to REWEIGHT. Program continuing with default estimation of time range. 10150 The parameter variance-covariance matrix is singular. FNUMBER must be the number of a column on the input data set.com. Program terminated. 10162 Error encountered when reading link model PK constant values. Notify Pharsight technical support for assistance at support@pharsight. it is inappropriate to use the WEIGHT command. 10152 10153 The residual sum of squares is zero. 647 . Program continuing with default missing value code. and surrounded by quotes. Program terminated. 10154 The specified library file does not exist. 10156 An error was encountered when reading the missing value code. Program terminated. since no observed data exist. 10155 An error was encountered when reading beta time range. When performing simulations. One or more weights became negative. Infusion model should not be used if end time is less than or equal to start time. Please modify them and rerun your model. IV dose <= zero. DNAME. Internal parsing file error. 10171 The lower and upper bounds were equal for one or more model parameters.F WinNonlin User’s Guide 10163 NOBS command ignored since FNUM specified. 10166 10167 Command encountered when reading model statements. 10165 The PNAME. 10164 The NOBS values are inconsistent with the number of records in the data file. try adding bounds so that the program can grid search for initial estimates. Check to see if a compiled model was specified by user. Use bolus model if start time equals end time. SNAME. PUNIT and SUNIT values cannot contain embedded blanks. Model statements were contained in the command file. The sum of the NOBSERVATIONS values is not equal to the number of observations on the data set. CNAME. See “User Models” on page 217. Models 15. See also “Model structure” on page 220. or infusion length <= zero. Initial estimates cannot be determined for this model. See “FNUMBER” on page 562. Check whether the model is missing an EOM statement or any commands in the model are not inside a block. The number of observations must be greater than the number of parameters.com. 10170 Improper use of END statement. Please report to technical support. Error degrees of freedom less than or equal to zero. and 17 do not support bolus dose <= zero. Error variance and standard errors or parameters cannot be estimated. Use END statements only in user models. If no bounds were used. Initial estimates cannot be found for this model. It is recommended to use a function variable for data with more than one function. Contact Pharsight Technical Support for assistance: support@pharsight. The function variable specified by FNUMBER will be used to determine which observations belong to which functions. 10172 10173 10174 10175 10201 648 . 16. try adding bounds so that the program can grid search for initial estimates. 10205 Due to failure of the curve stripping method. Automatic estimation of initial estimates is available for single dose data only. less compartments. Please try fitting a simpler model. Execution continuing without limits on parameter space. Contact Pharsight Technical Support. 11051 11060 Internal error during calculations. Closing other WinNonlin windows would also help. If no bounds were used. Try fitting a less complex model. A grid search will be used to obtain initial estimates. a grid search will be performed. 11001 Memory allocation error. The initial parameter estimates are outside of the LOWER and UPPER limits. 10206 The initial estimates are outside of the lower or upper parameter limits. If other applications are running. The files should be extracted first. close them and try again. Initial estimates for the number of exponential terms in the specified model could not be determined. 10203 An error occurred during curve stripping. LinMix and Bioequivalence wizards When errors occur using the LinMix or Bioequivalence wizards a message box will appear with the error number and a description of the error.Warnings and Error Messages LinMix and Bioequivalence wizards 10202 F Curve stripping is only done for single dose data.com. 10204 The requested number of exponentials could not be fit. save your work in these applications. Insufficient memory. 649 . Contact Pharsight technical support at support@pharsight. Initial estimates could not be found for the model. Contact Pharsight Technical Support. Initial estimates could not be determined for this model via curve stripping. 10998 Execution error. Corrupted data set. Internal parsing error. Note that this may occur when extracting PWO’s directly from zipped files. 11002-11032 11050. Error in Satterthwaite DF. Error opening file. If this occurs when using the default bioequivalence model with Subj(Seq) as a random effect. i. Using method of moments. There are no degrees of freedom for residual. Try random instead of repeated. The residual variance is not modeled. Dependent variable must be numeric. There is no residual variance. Model may be over-specified.e. No statistical tests are possible. Newton's algorithm converged with modified Hessian. it most likely indicates that the within-subject variance (residual) is greater than the between-subject variance. Character variables in models must be class variables. Failed to converge in allocated number of iterations. Try a different model.F WinNonlin User’s Guide 11061 11062 11063 11064 11065 11066 11067 11068 11069 11070 11071 11075 11089 11090 Input data not rectangular. A more appropriate model would be to move Subj(Seq) out of the random model and into the fixed model. Information matrix is deemed singular. Consider FA0 structure instead of UN.. Consider omitting this VC structure. Too many levels of a classification variable. Output is suspect. . Try the Satterthwaite option. 11091 11092 11093 11094 11095 650 Negative final variance parameter. Starting estimates supplied by user are invalid. Try using Residual DF option. Error in Variance Matrix. Seq+Subj(Seq)+Formulation+Period. Asymptotic covariance matrix not computed. There are no residual degrees of freedom. Bioequivalence statistics could not be computed for some test formulations. Output is suspect. Negative final variance component. Weight variable must be numeric. More than 10 sequences not allowed for population/individual bioequivalence. Negative final variance parameter for AR structure. Negative final variance parameter for TOEP structure. Do you want to continue with shortened values? Alpha variables are limited to a length of 128 characters. More than 2 treatment formulations not allowed for population/individual bioequivalence. This error occurs when there is an error on one of the tabs. More than 10 periods not allowed for population/individual bioequivalence. Program termininating. 11097 11098 11099 11101 11102 11103 11104 11105 11106 11107 11108 11109 11121 11122 11201 11206 There is an error on the following tab(s): tab name(s). More than 200 subjects not allowed for population/individual bioequivalence.Warnings and Error Messages LinMix and Bioequivalence wizards 11096 F Negative final block term for CS. Fewer than 2 sequences had complete design. Consider omitting random model for this structure. Subject <name> had incomplete design and was discarded. Program terminating. Fewer than two periods. The column ColName contains alphanumeric value(s) that are greater than maxsize in length. 651 . Sequence < > had less than 2 complete subjects and was discarded. Program termininating. Possibly the model is missing. or the dependent variable is missing. Ln-transform requested for data that is zero or negative causing an incomplete design. Fewer than two treatment formulations. Negative final standard deviation for ARH/CSH structure. We suggest creating a new variable with re-coded values that are less than or equal to this maximum length. Missing data for dependent variable causing an incomplete design. variable will be temporarily truncated for use by LinMix. 11211 File data filename not found This error occurs when the data file referenced by the data statement in the command file is not found. 11210 Unable to find file command filename This error occurs when the specified command file does not exist. 12999 Memory allocation error Insufficient memory. 11215 Table Wizard When errors occur with the Table Wizard a message box will appear with the error number and a description of the error. The LinMix module has a limit of using no more than 256 variables.F WinNonlin User’s Guide 11207 11208 Failed to create model as specified in the model file. 13000 13999 652 Floating Point error Memory allocation error . close them and try again. Variable name too long for LinMix. If other applications are running. Descriptive statistics When errors occur in the Descriptive Statistics module a message box will appear with the error number and a description of the error. There are greater than 256 variables in your data set. save your work in these applications. This error occurs when the user loads a model command file from the tool bar and this file does not contain any valid DATA statement. Unable to find any DATA command in command filename. 11212 11214 Second delimiter in filename for DATA statement not found. Closing other WinNonlin windows would also help. Please decrease the number of variables on the data set then rerun the LinMix module. Maximum variable name length is 256 characters. or select Sparse Sampling in NCA. Insufficient random access memory. Error opening file. Incompatible units for summary table Amount values. using default units. Only 1 non-zero concentration value found. and it was at dosing time. Noncompartmental analysis Error messages 14001 Memory allocation error. Warnings 14002 14021 14022 14024 14050 14061 14062 14064 14066 14501 14502 14503 Incompatible units for partial AUCs and summary AUCs. Incompatible units for summary table AUMC values. program terminating. or negative volume or concentration. close them and try again. If other applications are running. Invalid urine data: lower time >= upper time. save work. Contact Pharsight Technical Support. All volume values are zero. 653 . Contact Pharsight Technical Support. using default units. Internal parsing error. save your work in these applications. Duplicate time values on data file. If other applications are running. using default units. Use Descriptive Statistics to compute mean data and perform NCA on the means. close the applications and try again.Warnings and Error Messages Noncompartmental analysis F Insufficient memory. Closing other WinNonlin windows would also help. Closing other WinNonlin windows may also help. Data error: same subject cannot have multiple samples at same time. No data in steady state dosing interval. All concentration values are zero. No grid output for this subject. Therapeutic window calculations cannot be done for negative data. All volume values are zero. No parameters could be extrapolated to infinity. save work. Lambda_z could not be estimated. Adjusting for infusion has caused negative MRTlast value. All concentration values are zero.F WinNonlin User’s Guide 14504 Incompatible units for final parameter <default parameter name>. No response at dosing time. No grid output for this subject. close the applications and try again. Insufficient random access memory. All final parameters are set to missing. All final parameters are set to missing. threshold will not be used. The value at dosing time could not be determined by log-linear regression of the first two measurements. Only 1 non-zero concentration value found. 14511 14512 14513 14514 14515 14520 14521 14522 14524 14523 14530 14531 14532 14533 14534 IVIVC 15001 Memory allocation error. MRT parameters are adjusted for length of infusion. 654 . Adjusting for infusion has caused negative MRTinf(obs) value. and was set to the first observed measurement. Baseline is set to 0. Adjusting for infusion has caused negative MRTinf(pred) value. using default units. You may also try using a constant infusion model. and it was at dosing time. Threshold cannot equal baseline. AUC_TAU could not be estimated. Steady state PK parameters could not be estimated. If other applications are running. Only one non-missing observation. Less than two non-missing observations. Closing other WinNonlin windows may also help. Contact Pharsight Technical Support. Cannot produce Wagner-Nelson table. If no errors exist. 15010 15011 15101 15102 15103 15104 15105 Scripting language Errors that are generated when using the Scripting Language are written to the data log file. If other applications are running. Internal convolution error. Error opening one or more input files. Try specifying AUCINF value. Insufficient random access memory. 655 . If other applications are running. General 20000 and above Internal error. Cannot have equal time values in input dataset. Use Descriptive Statistics to compute mean data to perform analysis on the means. Please consult this log for guidance in resolving problems with script files. LambdaZ could not be estimated by WNL. close them and try again. (Loo-Riegelman only. Cannot continue with Loo-Riegelman. close the applications and try again. Contact Pharsight Technical Support. Contact Pharsight Technical Support. Number of non-missing observations for a function is zero. Closing other WinNonlin windows may also help.) Memory allocation error. (only for Wagner-Nelson) LambdaZ could not be estimated by WNL. Contact Pharsight technical support at support@pharsight. save work. 19999 Memory allocation error Insufficient memory. save your work in these applications. Duplicate time values on data file.Warnings and Error Messages Scripting language 15002 15003 15004 F Error opening file. this log file will not be created.com. Closing other WinNonlin windows would also help. Contact Pharsight Technical Support. Internal parsing error. F WinNonlin User’s Guide 656 . 12:2663-92. Marcel Dekker. Int J Clin Pharm Ther 32(7): 376-378.Appendix G References Further reading on modeling and related analyses See the following headings for an alphabetized listing of related references. A new procedure for testing equivalence in comparative bioavailability and other clinical trials. The bioavailability of erythromycin sterate versus enteric-coated erythromycin base taken immediately before and after food. 657 . • • • • • • “Bioequivalence” on page 657 “In vitro-in vivo correlation and formulation science” on page 658 “Pharmacokinetics” on page 659 “Regression and modeling” on page 660 “Sparse data analysis methods” on page 662 “Statistics” on page 663 Bioequivalence Anderson and Hauck (1983). Luus. 2nd ed. Inc. Hsuan and Holder (2000). Chow and Liu (2000). Statist Medicine 19:2885-97. Commun Stat Theory Methods. Hyslop. Hauschke. Int Med Res 9:470-7. Presentation of the intrasubject coefficient of variation for sample size planning in bioequivalence studies. Clayton and Leslie (1981). A small sample on confidence interval approach to assess individual bioequivalence. Design and Analysis of Bioavailablilty and Bioequivalence Studies. Steinijans. Elze and Blume (1994). Schall. Diletti. Orthonormal bases of compactly supported wavelets. AAPS Press. J Pharmacokinet Biopharm 24:282. Haskel and Hanson (1981). Population and Individual Approaches to Establishing Bioequivalence. Programs 21. Cutler (1978). Loo and Riegelman (1968). VA. Madden. Chappell MJ. Comparison of six deconvolution techniques. Inc. In vitro-in vivo correlation and formulation science Charter and Gull (1987). J Pharmaceut Sci 57:918. Bioavailability and Bioequivalence Studies for Orally Administered Drug Products—General Considerations. Stockholm. Iman and Conover (1979). Gabrielsson and Weiner (2001). Godfrey. Gillespie (1997).499-509. Pharmacokinetics. J Pharmacokinet Biopharm 6(3):243-63. Scientific Foundation for Regulating Drug Product Quality. Average. J Pharmacokinet Biopharm 15. Alexandria. 98-118. J Pharmacokinet Biopharm 6(3):227-41. Numerical deconvolution by least squares: use of polynomials to represent the input function. In Amidon GL. Daubechies (1988). Gibaldi and Perrier (1975). 658 . Statistical Approaches to Establishing Bioequivalence. Numerical deconvolution by least squares: use of prescribed input functions. Modeling Strategies for In Vivo-In Vitro Correlation. Robinson JR and Williams RL. New method for calculating the intrinsic absorption rate of drugs. Cutler (1978). Communications on Pure and Applied Mathematics 41(XLI):909-96. A comparison of the two one-sided tests procedure and the power approach for assessing the equivalence of average bioavailability. Pharmacokinetic and Pharmacodynamic Data Analysis: Concepts and Applications. Math. Hovroka R and Bates RA (1996). US FDA Guidance for Industry (August 1999). J Pharcokinet Biopharm 15:657-680. Swedish Pharmaceutical Press. US FDA Guidance for Industry (October 2000).. 645-55. 3rd ed. US FDA Guidance for Industry (January 2001). Marcel Dekker.G WinNonlin User’s Guide Schuirmann (1987). eds. 21. New York. Technometrics. Comments on two recent deconvolution methods. Estimation of statistical moments and steady-state volume of distribution for a drug given by intravenous infusion. Dayneka. In Amidon GL. Cheng and Jusko (1991). Alexandria.. Pharmacol Ther 16:143. Holford and Sheiner (1982). New York. Pharmacokinetics. Robinson JR and Williams RL. AAPS Press. Direct linear plotting method for estimation of pharmacokinetic parameters. Corticosteroid pharmacodynamics: models for a broad array of receptor-mediated pharmacologic effects. Pharmacokinetics: Statistical moment calculations. 2nd ed. J Pharmacokinet Biopharm 18(5):483-99. Polli (1997). In Amidon GL. Scientific Foundation for Regulating Drug Product Quality. Robinson JR and Williams RL. Scientific Foundation for Regulating Drug Product Quality. J Pharm Sci 70:1093-4. IVIVC Examples. (1981). 2nd ed. Gouyette (1983). 659 . Jusko (1990). Gibaldi and Perrier (1982). ESTRIP. Garg and Jusko (1993). AAPS Press. Linear inverse--theory and deconvolution. Kinetic Data Analysis: Design and Analysis of Enzyme and Pharmacokinetic Experiments. Analysis of In Vitro-In Vivo Data. J Pharm Sci 80:202. Treitel and Lines (1982). Arzneim-Forsch/Drug Res 33 (1):173-6. Geophysics 47(8):11539.References Pharmacokinetics G Meyer (1997). New York. Swedish Pharmaceutical Press. ed. J Clin Pharmacol 30:303. A BASIC computer program for obtaining initial polyexponential parameter estimates. Kinetics of pharmacological response. eds. Endrenyi. Noncompartmental determination of mean residence time and steady-state volume of distribution during multiple dosing. VA. J Pharm Sci 67:1687-91. J Pharmacokin Biopharm 21:457. VA. Gabrielsson and Weiner (1997). Chan and Gilbaldi (1982). J Pharm Bioph 10(5):551-8. Verotta (1990). Alexandria. Marcel Dekker. Plenum Press. Koup (1981).. Comparison of four basic models of indirect pharmacodynamic responses. Pharmacokinetic and Pharmacodynamic Data Analysis: Concepts and Applications. Pharmacokinetics Brown and Manno (1978). Stockholm. eds. Lifetime Learning Publications. Munson and Rodbard (1978). A modified form of Levenberg's correction. radioligand assay and physiological dose-response curves. Kuh and Welsch (1980). John Wiley & Sons. Kinetics of pharmacologic effects in man: The anticoagulant action of warfarin. A semicompartmental modeling approach for pharmacodynamic data assessment. Bard (1974). Biometrics 18:10413. Technometrics. O’Reilly and Levy (1969). Wagner (1975). Corbeil and Searle (1976). Draper and Smith (1981). 18:31-8. Metzler and Tong (1981). Computational problems of compartmental models with MichaelisMenten-type elimination. Annals of the Institute of Mathematical Statistics 30:A9-14. Regression Diagnostics. Nonlinear Parameter Estimation. John Wiley & Sons. Fundamentals of Clinical Pharmacokinetics. Posterior probabilities for choosing a regression model. Belmont. Applied Regression Analysis. NY. J Pharmaceutical Sciences 70:733-7. Analyzing Experimental Data By Regression. Chapter 12 in Numerical Methods for Non-linear Optimization. Clin Pharmacol Ther 10:22. Belsley. New York. Davies and Whitting (1972). John Wiley & Sons. Allen and Cady (1982). Academic Press. Bates and Watts (1988). CA. New York. Beck and Arnold (1977). DeLean. Restricted maximum likelihood (REML) estimation of variance components in the mixed models. Illinois. A method for fitting linear combinations of exponentials. John Wiley & Sons. New York. Parameter Estimation in Engineering and Science. New York.G WinNonlin User’s Guide Kowalski and Karim (1995). Cornell (1962). Nagashima. New York. J Pharmacokinet Biopharm 23(3):307-22. Academic Press. Am J Physiol 235(2):E97-E102. Drug Intelligence. Regression and modeling Akaike (1978). Simultaneous analysis of families of sigmoidal curves: Application to bioassay. Nonlinear Regression Analyses and Its Applications. 660 . 2nd ed. J Pharmacokinet Biopharm 7:537-41. New York. Comput Biol Med 9:39-48. John Wiley & Sons. J Pharmacokinet Biopharm 23:307-22. Scientific Consulting Inc. Computing Journal 7:308-13. J Stat Comp Sim 554:363-78. A semicompartmental modeling approach for pharmacodynamic data assessment. The use of nonparametric methods in the statistical analysis of the two-period change-over design. Vol. A method of exponential curve fitting by numerical integration. Hartley (1961). Kennedy and Gentle (1980). Leferink and Maes (1978). Biometrics 26:815-21. Parsons (1968). The modified Gauss-Newton method for the fitting of nonlinear regression functions by least squares. Arzneim Forsch 29:1894-8. An interactive curve fit programme for pharmacokinetic analyses. 661 . Longley (1967). AUTOMOD: A Polyalgorithm for an integrated analysis of linear pharmacokinetic models. Foss (1970). Murray and Wright (1981). PCNonlin. 1: Unconstrained Optimization. Gomeni and Gomeni (1979). Biometrics 577-83. Practical Optimization. Peck and Barrett (1979)..References Regression and modeling G Fai and Cornelius (1996). USA. Kowalski and Karim (1995). A simplex method for function minimization. Jennrich and Moore (1975). Nelder and Mead (1965). Amer Stat Assoc Proceedings Statistical Computing Section 57-65. 69:819-41. Gill. Technometrics 3:269-80. North Carolina. Fletcher (1980). Nonlinear least-squares regression programs for microcomputers. STRIPACT. Academic Press. Biological problems involving sums of exponential functions of time: A mathematical analysis that reduces experimental time. Giesbrecht and Burns (1985). Math Biosci 2:123-8. Statistical Computing. Approximate f-tests of multiple degree of freedom hypotheses in generalized least squares analysis of unbalanced split-plot experiments. Journal of the American Statistical Association. Biometrics 41:477-86. Two-stage analysis based on a mixed model: Large sample asymptotic theory and small-sample simulation results. New York. Koch (1972). Maximum likelihood estimation by means of nonlinear least squares. Practical methods of optimization. Marcel Dekker. G WinNonlin User’s Guide Ratkowsky (1983). Nuclear Instrum Meth 205:479-83. Nonlinear Regression Modeling. Holder (2001). New York. Transaction of the American Mathematical Society 54. Miller and Ham (1979). J Pharm Sci 87(3):372-8. Biometrics Bulletin 2:110-4. Pharm Res 12:124-8. Improved resolution in the analysis of the multicomponent exponential signals. J Pharm Sci 65:1001-10. J Biopharm Stat 8(2):317-28. Sheiner. An approximate distribution of estimates of variance components. Simultaneous modeling of pharmacokinetics and pharmacodynamics: application to d-tubocurarine. Marcel Dekker. Tang-Liu and Burke (1988). Tong and Metzler (1980). Mager and Goller (1998). Tests of statistical hypotheses concerning several parameters when the number of observations is large. Applying Bailer's method for AUC confidence intervals to sparse sampling. An extension of Satterthwaite's approximation applied to pharmacokinetics. J Biopharm Stat 11(1-2):75-9. Pharm Res 5:238-41. J Pharmacokinet Biopharm 16:303-9. Resampling methods in sparse sampling situations in preclinical pharmacokinetic studies. SIAM Review 18:376-411. Satterthwaite (1946). Wald (1943). Watts and Davenport (1976). Nedelman. Testing for the equality of area under the curves when using destructive measurement techniques. The effect of azone on ocular levobunolol absorption: Calculating the area under the curve and its standard error using tissue sampling compartments. Stanski. Sparse data analysis methods Bailer (1988). Mathematical properties of compartment models with MichaelisMenten-type elimination. Solving nonstiff ordinary differential equations . Comments on Nedelman and Jia's extension of Satterthwaite's approximation applied to pharmacokinetics. Estimating the dimension of a model. Shampine. Annals of Statistics 6:461-4. 662 . Gibiansky and Lau (1995).the state of the art. Clin Pharm Ther 25:358-71. Mathematical Biosciences 48:293-306. Nedelman and Jia (1998). CSTRIP. Schwarz (1978). Vozeh. Smith and Nichols (1983). A FORTRAN IV Computer Program for Obtaining Polyexponential Parameter Estimates. Sedman and Wagner (1976). New York. The American Statistician. Conover (1980). 2nd ed. Statistics Allen (2001). Principles and Procedures of Statistics. Winer (1971). Linear Models. ASA Proceedings of the Biopharmaceutical Section 74-81. 663 . A Biometrical Approach. Dixon and Massey (1969). et al.References Statistics G Yeh (1990). McGraw-Hill Book Company. McGraw-Hill Book Company. John Wiley & Sons. New York. Biometrics 577-84. Hypothesis testing in linear models (Eisenhart Model 1). Searle (1971). New York. New York. Estimation and Significant Tests of Area Under the Curve Derived from Incomplete Blood Sampling. 28(3):98-100. New York. John Wiley & Sons. Steel and Torrie (1980). Statistical Principles in Experimental Design. Koch (1972). Private correspondence. Practical Nonparametric Statistics 2nd ed. Introduction to Statistical Analysis. 3rd ed. 2nd ed. McGraw-Hill Book Company. Kutner (1974). The use of non-parametric methods in the statistical analysis of the two-period change-over design. G WinNonlin User’s Guide 664 . wmf. 483 Actual clock time defined.map.exp. 15. 14 *. 15 *. 350.cmd. 15 *.wsc.dep. 487 *. 37 *. 483 Actual time defined.pto. 148. 15.bmp. 15 *. 103.pks. 105 creating.imp. 48 fill. 16 *. 15. 106 *. 579 Akaike Information Criterion AIC defined.wgo. 350. 48 center across cells.dll. 16 placeable header information. 193.lml. 418 *. 483 in LinMix.pco.anv. 451 Alignment. 15. 15 *. 79 *. 15.xpt. 193 *. 145 Alpha. 484 *. 48 table cells.psc. 102. 215 Alias PKS. 193 *. 483 AggregateVar. 487 *. 103. 213 in PK workbook output. 14 *. 373 in PK text output. 120 Accumulation index.jpeg. 493 saving. 15 *.lib. 396.wdo. 15.wto. 483 Absolute error bars. 15.pwo.wsp. 218 *. 78 A A defined. 19 *. 87 *.txt. 484 665 . 16 *. 499 *. 395. 16 *. 102 saving. 418 *.Index Symbols *.wmo. 363 in LinMix vs SAS PROC MIXED. 455 *.tdf. 360 *. 87 creating.pmo. 170 Beta B. 213 defined. 222 comparison operators. 117 Autoregressive variance. 397 errors. 485 defined. 396 Bioequivalence Wizard. 483 half-life. 232 statements. 233 structure. 581 AUC. 485 AUClast. 354 Banded unstructured variance. 485 average. 397 dependent variables. 485 666 AUMClast. 485 Banded no-diagonal factor analytic. 484 AlphaTerms(). 379 fixed effects. 354 Banded Toeplitz variance. 580 AppendFileType. 485 Bibliography. 220 variables. 485 Balanced design. 485 calculation formulas. 354 AutoSort. 235 operators. 183 calculation methods. 221 comments. 580 AppendSheet. 579 AND. 580 AppendFilename. 177 compartmental modeling. 470 AUMC AUMCINF. defined. 354 Bar charts. 582 Automatic sorting. 231 weights. 396 . 649 fixed effects (average). 235 creating. 226 functions. 396 Axes log. 582 Average bioequivalence. 484 ANOVA. 581 ASCII data files. 213 AUCall. 236 Bioequivalence. 657 Bioassay functions. 484 partial area calculations. 243 Aterms(). 126 log base. 218 examples. 122 uniform. 117 Baseline command. 484 AUCINF. 396 recommended models. 470 reason. 485 AutoFileOptions. 379 setting up. 375. 233 reserved variable names. 375. 485 defined. 127 options. 484 statistic. 184 Audit trail. 386 pval.Pharsight WinNonlin User’s Guide A. 395 classification variables. 233 programming language. 372 APPEND. 16 ASCII models block content. 581 ARRANGE. 235 Anderson-Hauck. 123 B B defined. 485 half-life. 557 NCA model 220. 115 error bars. 398 weight variables. 16 scatter plots. 486 Cell references. 118 intervals. 381 linear model. 486 C++.Index C fixed effects (pop. 152. 43 merge. 486 Box and whiskers. 397 Bitmaps. 117 page setup. 197 defined. 400. 117 box and whiskers. 46 formulas. 124 copy and paste. 122 log. 399 sort variables. 48 cell references. 117 axis options. 217 Carry along data for like sort levels. 16 saving to graphics formats. 44 center across cells./individ. 403 population/individual. 49 filling adjacent. 582 CASCADE. 582 CaseSensitive. 402 general options. 126 multiple X-Y variables. 59 Carry along variables. 48 Center across cells. 122 sort variables. 486 C C defined. 582 Bounds. 486 changing chart type. 115 Charts. 33 save as *. 16 Blinding type. 119 histograms. 496 random effects. 127 group variables. 398 regressor variables.wmf. 378 options. 122 667 . 401 power. 46 formatting. 123 legend. 130 creating. 122 bar charts. 485 variable name limitations. 49 deleting values. 120 grid lines. 401 least squares means. 48 Chart Wizard. 119 templates. 129 title. 44 including or excluding. 60 inserting. 395 types. 48 deleting. 378 missing data. 115 automatic sorting. 25 Pharsight Chart Objects. 43 deleting formats. 117 sort key title. 397 sums of squares. 486 BQL defined. 123. 454 BOLUS. 402 output. 583 Cavg. 378 variance structure. 397 repeated variance. 486 CarryData.). 381 type. 44 Cells alignment. 122 axis titles. 15 pink. 117 overlay. 369 Convolution. 83 Convergence criterion. 500 Confidence limits Univariate. 53 COMMAND. 583 Columns cell references. 223 Command file. 116 CL. 44 deleting. 122 X-Y variables. 454 defined. 221 Command block. 207. 487 LinMix. 295 COPY. 14 worksheet cell fill-down. 212 descriptive statistics. 487 Constrained optimization. 397 defined. 152 planar. 266. 487 CLR. 123 units display default. 56 including or excluding. 130 workbooks. 31 use group headers. 496 univariate. 352. 46 Correlation matrix. 83 Coefficients. 583 Cmax. 213 Covariates. 536 COND. 487 CMT. 487 Clss. 297 units. 346. 44 freezing. 222 Comparison operators. 223 Comments comment character. 451 Connection string. 488 LinMix. 584 Confidence intervals. 487 Clss/F. 60 inserting. 262. 373 defined.Pharsight WinNonlin User’s Guide uniform axes. 499 Westlake. 82 in ASCII models. 352 Clock time. 43 names. 90 ODBC export. 349 LinMix. 584 Concentration-effect models. 87 CONSEQDELIM. 383 Classification variables bioequivalence. 487 Cmin. 488 degrees of freedom. 354 668 CON array. 232 ConcCol. 397 . 585 Copying charts. 293 settings. 487 CloseAll. 584 Constants defined. 487 Condition number. 356 nonestimability. 487 COMMANDS. 235 Compiling Using C++. 365 COLUMN. 487 CONFIDENCE. 238 Compound. 53 units. 212. 487 Classical and Westlake Confidence Intervals. 348 Control stream. 213 Configuration PKS. 7 Contrasts defined. 102 ODBC import. 487 Compound symmetry variance. 299 variables. 260 references. 35 saving. 79 tables. 215 Cubic model. 188. 587 Dependencies. 254 separate columns. 586 Degrees of freedom LinMix contrasts. 49 cells. 62. 42 DELIMITER. 165 defined. 514 transformations. 29 NCA calculation method. 279 numeric. 37 origins. 587 Deleting cell formats. 547 Cumul_Amt_Abs_V. 489 Custom query builder. 10 licensing. 28 modeling output. 488 Crossed factors. 44 rows. 278 Curve stripping.Index D defined. 11 user feedback. 261. 33 file type limitations. 13 inclusions and exclusions. 35 Dependent variables bioequivalence. 57 reset selections. 36 refreshing output. 269 Wagner-Nelson. 83 Description 669 . 60 merging. 31 workbooks. 97 Customer support. 28 Defect reports. 64 Data variables compartmental modeling. 352 NONMEM export. 349 LinMix versus PROC MIXED. 567 Data dependencies. 259 CROSSVAR( ). 514 toxicokinetic. 44 worksheets. 587 DELTA. 43 columns. 585 CROSSVARS. 488 Crossover design. 488 exponential. 453 DCP description. 585 CSS. 14 file types. 397 LinMix. 278 DefaultNCAParameterNames. 262 Loo-Reigelman. 372. 196 noncompartmental analysis.map. 269 units and UIR. 11 DefinitionFile. 586 D DATA. 146 Deconvolution. 258 stacked in one column. 373 Satterthwaite’s approximation. 180 Defaults missing value indicator. 33 *. 31 units. 49 cell values. 161 Dates. 29 page breaks in tables. 658 UIR. 62 sorting. 57 sparse sampling. 461 Decimal places SAS Transport export. 11 CUT.dep. 370 DeleteFiles. 454 Descriptive statistics. 213. 224 DnErr( ). 234 Emax defined. 588 Detach. 230 DTA array. 120 Error messages. 649 noncompartmental analysis. 202 compiled PK models. 589 EndRow. 333 DF. 225 Differential equations. 202 Down variable. 472 simulation. 215 DIFFERENTIAL. 165 DNAMES. 349. 590 EQ operator. 225 Directories options. 358 Examples Examples Guide. 590 Engines. 231 DVID. 171. 589 Eigenvalue defined. 157 DestArea. 430. 413 Tau. 234 Document manager. 423 defined. 588 Do. 652 Errors. 201 data variables. 151 KS_pval. 28 Disable curve stripping. 11 EntireRow. 120 relative. 364 Else. 232 E E. 120 DTA. 83 DZ array. 200 scenario-specific. 652 IVIVC. 364 LinMix. 589 DoseUnit. 588 DestCol. 221. 84 PKS. 488 ECOL. 654 LinMix. 489 scripting language. 653 scripting. 637 bioequivalence. 489 EC50 defined. 448. 171 worksheet data. 36 Determining estimability. 221 EndCol. 2 Excel 670 . 418 Enhancement requests. 202 dose normalization. 210 ErrType. 637 descriptive statistics. 4 user models. 201 file structure. 591 Estimates fixed effects. 432 noncompartmental analysis. 589 Dosing ASCII models. 203 units. 649 compartmental modeling. 235 Error bars. 489 Eigenvalues. 467 DoseFile.Pharsight WinNonlin User’s Guide PKS studies. 171 NONMEM export. 655 tables. 472 regimen. 488 E0. 120 absolute. 489 END. 215 Schwarz Bayesian Criterion. 225. 235 G Gamma defined. 336 average bioequivalence. 490 half-life. 16 saving workbooks to Excel. 364 pop. 223.0 files. 595 FootnoteFontSize. 352 parameter estimates. 402 Fixed parameters. 595 Formatting tables. 591 Exporting NONMEM. 217 user models. 198 FNUMBER. 489 Export definition. bioequivalence. 16 EXCLUDE. 221. 46 Final parameters confidence intervals. 231 FONT. 212 FIND. 46 Formulas. 489 Fdiss defined. 592 F F. 237 FUNCTION. 76 Word export options. 489 Functions. 78 EXTRA. 396 defined. 78 FileType. 50 find next. 441 SAS Transport. 594 FontSize. 14 data files. 174 Exclusions. 336. 231 Fabs defined. 15 files. 215 Fixed effects. 13 PKS files. 286 reset selections. 78 SigmaPlot. 103 ExportAll. 51 replace. 591 Exclude profiles with insufficient data. 77. 60 based on criteria. 45 Formulation defined. 62 Exponential curve stripping. 80 ODBC.Index F Excel 4. 102 SAS Transport files. 17 FileMask. 594 FootnoteFont. 85 S-PLUS. 166. 146 worksheet cells. 489 LinMix. 51 Fit Akaike Information Criterion. 86 Word export. 60 from Lambda Z calculations. 594 Find. 489 File types WNL and WNM compatibility. 227. 60 by selection. 223 Function variable./individ. 595 multiple FUNC blocks. 489 FORTRAN. 490 671 . 44 operators. 16 data file limitations. 593 Files ASCII. 593 Fill-down. 102. 491 parameter estimates and bounds. 596 GroupVar( . 597 GroupVarBreak( ). 18 Hill model. 111 672 ID variables. 118 intervals. 541 defined. 60 Inclusions. 490 Hartley modification. 539. 598 IDVars. 207. 600 Independent variable. 42 Interaction. 542 Individual bioequivalence. 6. 16 IncludeVar( . 354 compound symmetry. 376 INITIAL. 598 Hessian eigenvalues. 6. 37 I IConsumer interface. 60 based on criteria. 354 Hidden windows. 463 Group variables. 38 worksheets. 599 Include placeable header information. 490 charts. 233 Importing ODBC. 597 GT operator. ). 43 rows. 490 Go to. 568 Initial parameter estimates. 60 by selection. 62 Increment for partial derivatives. 9 Levenberg-type modification. 454 Indirect response models. 38 items logged. 601 Inserting cells. 233 Green scenario icon. 491 IDVar( ). 600 Including based on criteria. 177. 491 Intercept. 60 selected cells. 491 Indication. 6 GE operator. 197 InitialDose. 598 Harmonic mean. 597 GroupVars( ). 123 History "cannot be processed". ). 542 parameters. 597 GroupVars. 9 Levenberg Hartley modification. 235 H HALT. 292 linear trapezoidal with linear/log. 177. 599 If Then Else. 600 InData( ). 52 GOTO.Pharsight WinNonlin User’s Guide Gauss-Newton algorithm. 119 GroupVar( ). 600 InputLag. 43 worksheets. 548 Histograms. 490 HEADERS. 364 Heterogeneous autoregressive variance. 491 Interpolation linear trapezoidal. 292 . 599 IncludeVars( ). 87 INCLUDE. 235 Geometric mean defined. 43 columns. 491 InData. 60 reset selections. 178. 123 IPR. ). 492 not estimable. 492 Lambda z. 430. 381 defined. 604 Legends. 601 . 4. 492 merge data sets. 279 minimizing the wizard. 601 Intervals. 242. 305 Loo-Reigelman. 313 project status. 304 saving projects. 2 unit impulse response fit. 57 tables. 492 half-life. 143 JoinCrossVar( . 492 least squares fitting. 287 predicting PK. 491 IVIVC. ). 292 linear/log trapezoidal. 277 toolkit introduced. 305 toolkit. 603 KM. 306 dissolution models. 310 dissolution data and models. 492 Kolmogorov-Smirnov normality test. 602 JoinIDVar( . 122 Levy plots. 307 Levy plots. 304 model validation. 604 LE operator. 5 Iterative reweighting. 302 JoinCrossVars. 157 L LABEL. 278 wizard. 476 673 J Join variables. 603 Joint Contrasts. 310 options. 16 K K half-life. 492 K01. 492 LinMix. 164 range file structure. 233 Lag time defined. 492 K10. 307 use raw data. 347 JPEG files. 300 Libraries. 604 LegendFontSize. 157 KS_pval. 277 convolution. 491 IQueryBuilder interface. 603 JoinIDVars. 492 K0. 602 JoinGroupVar( . 307. 547 error messages. 110 Iterations. 162 defined. 235 Least squares means. 293 correlation models. 365 procedure. 654 fixed parameters. 602 JoinGroupVars. 300 loading projects. 492 KE0. 359 Legend( ). 315 Wagner-Nelson. 431 LamzFile. 344. 208 history. ). 292 INTERVAL. 198 fraction absorbed. 604 LegendFont. 315 output options.Index J linear up/log down. 178. 62 Loo-Reigelman.Pharsight WinNonlin User’s Guide add or update. 177. 11 expiration warning. 360 Hessian eigenvalues. 353 weight variables. 605 Loading ODBC import query. 363 classification variables. 292 Linear up/log down. 344. 352 coefficients. 369 crossed factors. 177. 479 Logarithmic axes. 356 residuals. 372 compared to PROC MIXED. 349. 333 estimates. 362 singularity tolerance. 364 hypothesis tests. 365 command file. 328 LLOQ. 356 convergence criterion. 363 sequential tests. 493 use when less than LLOQ. 96 LoadTemplate. 184 trapezoidal rule. 334 output. 336. 477 loading from PKS. 545 LinkFile. 493 Linear models. 206 Linear trapezoidal rule. 536 simultaneous. 74 LOAD. 488 dependent variables. 355 regressor variables. 364 Satterthwaite’s approximation. 649 estimability. 430. 459 defined. 279 . 361 X matrix. 31 Linear interpolation rule. 370 Schwarz’s criterion. 362 initial variance parameters. 366 contrasts. 292 with linear/log interpolation. 372 computations. 3 Linear regression. 341. 352 errors. 368 non-estimable functions. 493 minimization algorithm. 326 LinMix Wizard. 352 text output. 366 repeated effects. 183. 346. 352 REML. 478 Licensing customer support. 605 Locked. 337 sort variables. 292 Linear/log trapezoidal rule. 336 parameters. 350 Newton’s algorithm. 3. 360 iteration history. 361 transformation of the response. 351 variance structure. 123 interpolation rule. 350 linear mixed effects model. 546 defined. 292 Link models. 358 fixed effects. 361 output parameterization. 493 compared to ANOVA. 352 fixed effects estimates. 364 random effects. 359 674 limits and constraints. 477 status. 178. 184 Linear kinetics. 338. 365 partial tests. 337 variable name limitations. 605 LinMix. 184 Logarithms transformations. 364 general options. 365 least squares means. 323 Akaike’s criterion. 178. 362 predicated values. 339. 352 workbook output. 233 Maximum likelihood estimates. 606 Makoid-Banakar model. 207 exclude profiles with insufficient data. 288 workbook output. 549 mdllib_mdl604. 193 675 . 288 LOWER. 498 MaintenanceDose. 542 defined. 200 Model size. 28 in NCA parameter estimates. 568 LT operator. 198 Model properties compartmental models. 207 iterations. 208 Mean square variance. 543 MODEL. 203 convergence criterion. 494 Microsoft Visual C++. 493 Merge cells. 173. 206 units. 287 weighting. 221 Model 220. 206 model size. 4 Mm. 206 MISSING. 199 ParmFile. 208 minimization. 161 Modeling.lib. 543 Vmax. 493 stripping dose. 196 dosing. 4 mdllib_mdl602. 197 initial estimates. 167 slopes data ranges. 549 mdllib_mdl603. 175 Model parameters. 162 threshold. 418 Methods defined. 167 partial areas.lib. 206 title for NCA text output. 174 increment for partial derivatives. 208 weighting for NCA. 59 Method scripting language. 76 Minimization algorithm. 550 Marking valid/invalid data. 198 title. 197 bounds. 212. 207 output. 167 Model options. 208 number of predicted values. 550 Mean defined. 493 equation. 493 Michaelis-Menten. 35 Mathematical operators. 48 Merging. 239 Microsoft Word export. 196 noncompartmental analysis. 57 carry along for like sort levels. 208 Model variables compartmental modeling. 235 M Macro rate constants. 606 Missing values default indicator.Index M output options. 208 mean square. 182 Mm. 176 transpose final parameter table. 515 baseline. 206 propagate final estimates. 499 Micro rate constants. 174 PK settings. 178. 493 Mean square. 204 output intermediate calculations. 546 link. 637 errors. 504 number of predicted values. 53 Native format. 86 start. 223 NDERIVATIVES. 15 sigmoidal/dissolution. 205 Models concentration-effect. 205 S-PLUS export. 549 Hill. 223 Nmiss. 494 Multiple linear regression. 550 676 Michaelis-Menten. 505 pharmacokinetic. 208 linear models. 494 . 234 NFUNCTIONS. 223 Nelder-Mead simplex algorithm. 206 refreshing output. 80 notation and conventions. 217 Weibull. 542 noncompartmental. 494 Nested factors. 539 linear. 33 output options. 3 mean square. 204 PK settings. 536 creating ASCII models. 210 output dependencies. 221 convergence criterion. 428 N N modeling output. 208 NONMEM export. 491. 170 model 220 threshold. 494 No Intercept. 494 Name objects. 6. 373 MRT. 218 cubic. 160 noncompartmental analysis. 519 Pharsight Model Objects. 548 indirect response. 170 NCONSTANTS. 200 error messages. 29 output. 494 Nested variables. 494 MRTlast. 368 Next. 545 types. 208 weights. 203 model settings. 547 simultaneous link. 196 default output.Pharsight WinNonlin User’s Guide ASCII model number. 193 Makoid-Banakar. 549 Modified likelihood. 207 data variables. 206 model options. 210 initial parameter estimates. 494 MRTINF. 208 minimization algorithm. 209 stop. 536 loading. 209 units. 35 scaling of weights. 197 iterations. 209 troubleshooting. 472 Naming columns. 328 No Scaling of Weights. 471 NCA model 220 baseline. 206 Nobs. 29 dosing. 352 Newton’s algorithm. 3 Multiple profiles in script files. 547 double Weibull. 207 options. 196 model size. 8 defined. 193 user models. 83 dependent variables. 182 data pre-treatment. 171 drug effect data. 174 computational rules. 181 default calculation method. 83 occasions. 102 677 . 160 model variables. 178 weighting. 495 ODBC export. 29. 16 Nonparametric superposition. 186 units. 506 parameter names. 160 therapeutic response window. 3. 102 building an export definition. 168 therapeutic window calculations. 159 AUC formulas. 495 Nonlinear regression. 181 data exclusions. 182 model selection. 167 partial areas file structure. 161 models. 568 NSECONDARY. 103 export definition. 30 default parameter names file. 47 Number of predicted values. 174 parameter calculations. 183 chart output. 83 problem. 80 CMT. 494 Noncompartmental analysis. 653 exclude profiles with insufficient data. 180 Lambda z.Index O NOBSERVATIONS. 162 Lambda z file structure. 85 other data items. 174 Nonlinear kinetics defined. 421 Occasions. 5 methods. 82 control stream. 431 missing parameter values. 180 defined. 354 Nominal time defined. 174 include parameter in output. 245 Normality test. 102 definition. 175 workbook output. 84 DVID. 174 output options. 85 ODBC defined. 515 errors. 174 page breaks in text output. 223 Number format. 173 output text. 162 sparse sampling data. 223. 104 connection string. 207 O Object defined. 83 dosing. 157 NPARAMETERS. 179 slope data ranges. 439 precision of output. 494 disable curve stripping. 102 creating an export. 567 No-diagonal factor analytic. 83 comment character. 173 output intermediate calculations. 505 output. 184 partial areas. 495 Nonlinear models. 8 NONMEM export. 495 Objects Scripting Language. 165 dosing. 90. 3. 82 Non-numeric data troubleshooting. 418. 179 partial area computations. 31 Page setup. 9 scientific and consulting services. 495 Password PKS. 401. 491 Partial tests. 198. 174 table defaults. 400. 362. 117 P P array. 87 loading a saved connection. 87 building a query. 33 PK settings. 7 increment. 28 OR. 173 NCA output.Pharsight WinNonlin User’s Guide loading a saved connection. 92 save password. 83 Output compartmental models. 519 Pharsight. 235 Origins. 467 Pharmacodynamic models. 89 Percentiles descriptive statistics. 402 directories. 360 models. 107 Watson DMLIMS. 365 names for NCA. 198 LinMix output. 87. 35 unlocking. 430 Partial areas. 436 fixed. 15 Pharsight Workbook Objects. 232 678 Page breaks NCA output. 180 ParmFile. 10 technical support. 90 custom query builder. 5. 31 WinNonlin options. 8. 77. 36 Other data items NONMEM export. 10 Pharsight Chart Objects. 10 training courses. 152 Performance PKS. 4. 11 products. 28 license expiration warning. 430 PartFile. 33 refreshing. 92 saving export settings. 184 Partial derivatives. 235 Options. 89 software developer’s kit. 105 ODBC import. 28 Word export. 29 NCA model. 28 bioequivalence. 97 Operators. 93 connection string. 180 preferred names. 452 save ODBC password. 15 Pharsight Model Objects. 207. 179 ParmFile. 36 Output intermediate calculations. 10 customer feedback. 430 preferred default names. 174 Overlay chart. 78 workbooks. 167 computation of. 434 file structure for link models. 210 pink. 97 custom query builder example. 15 Pink output windows. 21 Parameters file structure. 111 import settings file. 9 contact information. 89 saving. 533 Pharmacokinetic models. 31 LinMix. 173 tables. 206 . 15 Pharsight Text Objects. 1 accessing. 450. 446 study data requirements. 212 Planar confidence interval. 454 Power. 441 green icon. 386 Precision NCA output. 545 PKS. 478 load. 496 simultaneous. 463 loading libraries. 450 studies. 448. 450 password. 21. 451 create study. 449 save as. 450. 79 Pre-clinical data. 479 libraries. 451 sessions. 461 dosing. 496 PUNIT. 472 server. 496 Programming language. 450. 477 logging on. 229 Population bioequivalence. 449 creating scenarios. 82 PROC MIXED. 514 Preferred units. 445 PLANAR. 452 performance. 449 add columns. 447 user name. 451 non-WinNonlin documents. 467 red icon. 452 study metadata. 224. 466 other documents. 401 fixed effects. 211 PK/PD link models. 465 DCP description. 156 Queries building ODBC queries. 477 alias. 25 PrintMulti. 401 data requirements. 476 library and share status. 25 page setup. 178 Printing. 188. 27 print preview. 198 Properties defined. 475 status. 450. 463 save. 450 add or update library. 536 defined. 464 scenarios. 376 Population/individual bioequivalence. 224 Q Quantiles computation. 447 scenario-specific dosing. 613 Problem NONMEM. 233 Project. 453 time. 402 Portfolio PKS studies. 463 information. 472 file format. 454 Propagate final estimates. 373 Profile. 450 share. 449 load study or scenario. 451 append/update study. 179 SAS Transport export. 613 PrintMultiDown. 447. 93 679 . 496 PNAMES. 449 scenario search. 450 configuration.Index Q PK text output. 613 PrintMultiAcross. 21 print all. 452 using with WinNonlin. 397 LinMix. 615 RemoveRow. 213 defined. 339. 356 REPLACE. 614 Relative actual end time. 657 Refresh. 618 Sample number. 373 LinMix. 615 RenameCol. 16 Rows cell references. 60 inserting. 497 Relative actual time. 44 deleting. 616 Repeated bioequivalence. 79 export. 497 estimate of residual standard deviation. 213. 470 Red scenario icon. 341. 68 Ratio tables.Pharsight WinNonlin User’s Guide ODBC import queries. 368 Rich text files. 463 References. 406 Read only. 617 S S. 497 Relative error bars. 398 LinMix. 497 REMARK. 497 Relative time. 35 Regression methods. 215 S array. 615 RemoveCol. 51. 211 Rank.dll. 78 Satterthwaite’s approximation. 92 QueryBuilderInterfaces. 496 Random effects bioequivalence. 617 RunWordExport. 399 Repeated effects LinMix. 614 RegVars. 44 freezing. 8 Gauss-newton algorithm. 366 Newton’s algorithm. 616 Replace. 370 . 352 Regressors defined. 497 RegVar( ). 355 RANK. 458 SAS share. 232 SameErrCol( ). 366 Newton’s algorithm. 120 Relative nominal end time. 475 SAS Transport files data set name. 497 Relative nominal time. 109 R Random effect defined. 8 Regressor variables bioequivalence. 368 680 REMOVE. 43 Rule sets. 71 RUN. 496 Rank transformations. 617 Restricted maximum likelihood LinMix. 479 Reason. 56 including or excluding. 222 REML. 66 Residuals. 9 Nelder-Mead simplex. 497 RESPONSE. 466 adding non-WinNonlin documents. 28 ShowUnits. 413 model parameters. 212 Simulation. 79 tables. 618 SaveAllPrefix. defined. 146 Simplex algorithm. defined. 412 secondary parameter estimates. 426 SD. 363 in LinMix vs SAS PROC MIXED. 226 Select Data for Modeling dialog. 208 Share.Index S SAVE. 415 Simultaneous PK/PD link models. 621 SortVars. 465 defined. 214 simulation. 620 SortVar( ). 498 bioequivalence. 496 script file (. 618 SaveAll. 6 Singularity tolerance. 478 Show formulas. 6. 415 dosing. 498 SearchArea. 447 adding columns. 224 SORT. 498 SDK. 619 SaveTemplate. 205. 620 Sort key title. ). 409 comparing dose schemes. 466 creating. 337. defined. 226. 493 object. 6 Singular value decomposition. 418 script file components. 57 SortLevel. 374 Slopes disable curve stripping. 620 SNAMES. 214 user models. 498 search. 497 Scientific notation. 352 Sorting worksheet data. defined. 179 SAS Transport export. 221. 545 Singular or near-singular equations. 655 file paths (Filename property). 362 Server PKS. 57. 593 method. 620 Sequential tests. 621 681 . 107 SE. 250 SEQUENCE. 621 SortVar( .PSC). 464 Schwarz Bayesian Criterion in LinMix. 361. 415 standard error computation. 48 Scripting commands. 440 property. 619 SECONDARY. 145. 620 SigmaPlot export to. 215 SBC defined. 619 Scaling of weights. 122 Sort variables. 489 error messages. 117 Scenarios. 579 engines. 213 in PK workbook output. 451 Setting preferred units. 227 Secondary parameters. 397 charts. 242 Scatter plots. 373 in PK text output. 209 Semicompartmental modeling. 85 Significant digits NCA output. 475 status. 85 version. 495 PKS. 165 SMOOTHING. 119 LinMix. 624 SummaryVars. 625 Tau. 144 Summary variables. 145. 454 SUBJECT.Pharsight WinNonlin User’s Guide SortVars( ). 623 STARTING. 146 significant digits. 133 font. 224 . 498 description. 134 units. 147 units display default. 623 STATS. 31 page break defaults. 624 Sums of squares. 144 templates. 499 Technical support. 622 Source changed. 224 Temporary variables. 35 SourceCol. 134 TEMPORARY. 479 name. 221. 622 Standard error. 514 references. 453 PKS. 31 TAU. 225 StartRow. 215 STACKED. 11 Templates charts. 224 T TAB. 145 summary statistics. 662 use sparse sampling. 203. 221 StartCol. 498 SummaryVar( ). 454 design. 381 682 SUNIT. 446 start and end dates. 146 error messages. 188 models. 498 Studies balanced design. 133 options. 485 defined (PKS). 146 join variables. 146 data only. 129 tables. 10 licensing. 493 SSR. 160 S-PLUS export to. 147 significant digits. 131 cell alignment. 146 decimal places. 625 Table Wizard. 212 START. 147 formatting. 71 Stripping dose defined. 454 locked. 145 column alignment. 148 templates. 144 table definition files. 74 rule sets. 623 Status code transformations. 146 summary statistics. 31 page numbers. 652 excluded data. 624 Summary statistics. 453 type. 86 version. 86 SS/df (mean square). 143 missing data. 145 formatting. 131 cell alignment. 151 tables. 69 LLOQ. 134 TableNum. 622 Sparse data computations. 625 Tables. 66 LinMix response. 497 relative nominal end time. 499 TileHorizontal. defined. 299 dose. 362 Text page setup. 627 TimeRepeat. 497 relative. 487 nominal time. 15 pink. 629 Unit impulse response See also UIR. 123 UNIT. 221. 7 Types of models. 178 Trapezoidal rule. 54 noncompartmental analysis. 497 TimeCol. 170 TI. 483 actual time. 309 ULOQ. 459 Undo. 627 TitleFontSize. 354 TOLERANCE. 68 replace. 33 Therapeutic response window. 208 converting. 16 Trust region. 629 Uniform axes. 206 non-compartmental modeling. 29 Treatment PKS mapping. 206 TitleFont. 54 summary statistics. 628 Toeplitz variance. 626 Time actual clock time. 171 file. 497 relative nominal time. 186 Threshold command. 66 status codes. 497 relative actual time. 626 TerminalUpper. 213 default method for NCA. 626 TileVertical. 483 clock time. 178 options. 430 mass. 628 Tolerance. 244 Transformations. 50 UNIFORM. 626 Tests of hypotheses. 193 U UIR. 625 TerminalPhase. 64. 29 compartmental modeling. 337 multiple. 10 TRANSFORM. 66 equations. 24 Text windows Pharsight Text Objects. 462 Troubleshooting. 627 TITLE. 269 automatic generation. 309 Units assigning. 53 compartmental modeling. 55 convolution. 31 prefixes.. 64 rank.Index U TerminalLower. 51 Training courses. 494 relative actual end time. 222. 153 supported unit types. 209 loading tabular data. 176. 627 Title. 575 NCA model 220. 69 Transpose final parameter table. 62 action. 54 683 . 67 filter. 168 calculations. 630 UPPER. 212. 398 bioequivalence. 224 transform block. 354 mean square. 354 compound symmetry. 225 libraries. 232 sort. 415 VIF defined. 217 ASCII examples. 499 Valid data. 224 weight variable. 54 volume. 217 secondary parameters block. 354 heterogeneous compound symmetry. 452 V V. 231 User name PKS. 116 classification. 354 unstructured. 354 banded unstructured. 222 WinNonlin variables. 221 commands block. 493 no-diagonal factor analytic. 212 Univariate confidence interval. 491 join. 225 structure. 490 ID. 119. 354 banded no-diagonal. 378 LinMix. 486 charts. 499 Variance structure. 498 temporary. 499 UNLOAD. 378. 499 heterogeneous autoregressive. 568 Use group headers. 492 model variables. 54 Univariate command. 354 components. 487 function variable. 351 684 Variables carry along. 232 regressors. 35 Validation IVIVC.Pharsight WinNonlin User’s Guide time. 300 Variable names Bioequivalence. 338. 160 User models. 498 summary variables. 36 Unstructured variance. 489 group. 315 Use sparse sampling. 237 function block. 223 creating an ASCII model. 354 Toeplitz. 353 types. 120 Update Reason dialog. 226 block content. 353. 491 independent. 307. 354 Up variable. 122 Use raw data. 225 FORTRAN. 161 names. 226 starting values block. 354 Visual Basic custom ODBC query example. 354 defined. 471 UpErr( ). 398 LinMix. 354 Banded Toeplitz. 231 Variance autoregressive. 354 Variance Inflation Factor. 196 model variables for NCA. 111 . 500 WinNonlin variables. 119. 220 temporary variables. 497 reserved names. 630 Unlocking derived data. 218 differential equation block. 631 Word export. 175 Loo-Reigelman. 591 WordExportFilename. 243 iterative reweighting. 17 options. 631 WinNonlin. 549 double. 87. 205 uniform. 44 freezing rows and columns. 76 export options. 231 WinNonlin object files. 241 Weighted summary statistics. 288 noncompartmental analysis. 500 VZ. 45 options. 241 weights from data file. 28 new. 157 Weighting. 352 variables. 500 W Wagner-Nelson. 242. 631 Worksheets cell references. 50 formulas. 35 saving to Excel. 16 show formulas. 242 Westlake confidence interval. 15 WinNonlin Script File. 16 unlocking. 64 dependencies. 397 variables (LinMix). 215 Weibull model. 16 WindowState. 77. 637 Watson DMLIMS. 42 fill-down. 1 Enterprise. 631 Workbooks copying. 1 variables in user models. 21 Pharsight Workbook Objects. 630 Weight variables (bioequivalence). 97 WCSS. 28 status codes. 242 scaling of weights. 499 Vss. 17 WITH. 33 refreshing. 549 WEIGHT. 175 scaling. 46 find. 33 missing values. 36 WorkingDir. 288 Warnings and errors.Index W Vmax. 500 Weighted least squares. 175 Wagner-Nelson. defined. 28 Professional. 78 ExportAll command. 44 deleting. 14 copying with transformations. 500 Windows hiding. 500 VZ/F. 69 troubleshooting. 56 685 . 491 least squares. 278 output options. 241 compartmental modeling. 287 weighting. 288 weighted least squares. 205 in ASCII models. 15 pink. 14 operators. 18 Windows metafiles. 418 WinNonMix file compatibility with WNL. 1 file compatibility with WNM. 35 page setup. 288 workbook output. 28 out of date. 232 X X. 37 inserting. 51 scientific notation. 52 history. 57 naming. 48 sorting. 215 WT. 19 defined. 42 merge cells. 500 WSSR. 633 XUnits. 232 X matrix. 328 XLabelsFont. 48 merging. 20 new. 632 XTitleFont. 48 WRSS. 635 YUnits.Pharsight WinNonlin User’s Guide go to. 48 Workspaces. 632 XLabelsFontSize. 57 transformations. 62. 632 XTitle. 53 wrap text. 35 Wrap text. 19 saving dependencies. 634 YTitle. 633 XTitleFontSize. 20 saving. 633 XVar( ). 635 YTitleFont. 634 YLabelsFontSize. 634 Y YLabelsFont. 635 686 . 64 units. 636 Z Z array. 232 YTitleFontSize. 500 loading. 635 YVar. 42 replace.