Systemization of Logistic Regression Analysis for Pharmacometric Applications

Conference: ACoP

Abstract

Objectives: Efficacy and safety data are oftentimes collected as binary (yes/no) data in clinical trials during drug development. The implementation and growing use of CDISC standards in data collection and categorization of adverse event data using the MedDRA dictionary has facilitated the standard format of clinical trial data collected across the pharmaceutical industry. A commonly used statistical methodology for analyzing binary data is logistic regression (LR) analysis. The objective of this work was to develop a system to standardize analysis dataset creation, exploratory data review, and LR analysis procedures for exposure-response analyses of binary endpoint data.

Methods: SAS® software was used to develop a code library to transform source clinical trial data into an analysis-ready dataset for use in exposure-response analyses. A library of SAS® code for the creation of standard exploratory graphs and tables was also developed. A systematic approach to statistical analysis using SAS® PROC LOGISTIC and NONMEM was developed, based on standard methods for model building and discrimination, to facilitate the calculation of standard statistics and production of typical diagnostic plots for model building and evaluation.1

Results: The standardized process for dataset creation, exploratory data analysis, and LR was tested on 10 compounds and refined as new variations and additional data checks were identified. This refined process and systematic approach resulted in a greater than 70% decrease in analyst time required for evaluation of exposure-response relationships for binary endpoints. Other positive benefits of system implementation include a reduction in training time for new pharmacometricians and improved quality and consistency of reporting for LR exposure-response analyses.

Conclusions: Standardization of analysis-ready dataset creation, exploratory graphical evaluation, and the LR analysis process for binary endpoints has proven instrumental in generating timely understanding of exposure-response relationships to facilitate model-based decision making under tight timelines and allows for the evaluation of additional endpoints and synthesis of findings across endpoints.

American Conference on Pharmacometrics (ACoP) Annual Meeting: October 4-7, 2015, Crystal City, VA

By Julie Passarell, Caroline Passarell, Darcy Hitchcock, Jill Fiedler-Kelly, Thaddeus H. Grasela