Baseline Adjustment and Multiplicity Issues of Pre-Clinical Longitudinal Data Analysis

• Activity Number: 34
• Type: Contributed
• Date/Time: Sunday, July 31, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Shubing Wang*+ and Junshui Ma
• Companies: Merck & Co., Inc. and Merck Research Laboratories
• Keywords: pre-clinical ; longitudinal data analysis ; baseline adjustment ; multiplicity ; constrained LDA ; analysis of covariance

Abstract

There are diverse types of statistical models for longitudinal data analysis of randomized clinical trials. The Constrained Maximal Likelihood Longitudinal Data Analysis (cLDA) model assumes that the baselines of all treatment groups are the same. It was claimed that cLDA is optimal (Liu et al., 2009), while some other statisticians gives the edge to Analysis of Covariance (ANCOVA) model, which takes the baseline as a covariate. Unlike randomized clinical trials, pre-clinical studies in general have more than one endpoint. In this situation, completely randomized baselines for all the endpoints are not available. From our simulations, the ANCOVA model is a more powerful and more robust method. Due to its multiple endpoints and that all the between- and within-group tests on a few time points are of interest, we found the method proposed in (Torsten Hothorn et al. 2008) and (Frank Bretz et al. 2010), which derives the simultaneous confidence band directly from the estimated joint distribution of all the test statistics, and provides a natural solution to solve the multiplicity issue of preclinical longitudinal studies.