Some Statistical Issues in Estimating Slope Effect for Repeated Measures

• Activity Number: 170
• Type: Topic Contributed
• Date/Time: Monday, August 5, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Tao Song*+
• Companies: Biogen Idec
• Keywords: slope effect ; linear mixed model ; type I error ; power

Abstract

MMRM is now often used in analyzing confirmatory trials with repeated measures. The commonly used MMRM is a full multivariate model. It assumes an unstructured time and the primary interest is to evaluate treatment effect at last visit. This model is generally considered appropriate at the study design stage when the underlying function of time is little known before the trial data is collected. However, in presence of underlying true linearity for the repeated measures, slope effect during the entire course of the study can serve better clinical interpretation. A linear mixed model can be statistically efficient and gain power. We have conducted a simulation study to address a few statistical issues in using a linear mixed model: 1) to examine the type I error control when the linearity assumption is violated; 2) to examine the extent of power improvement over MMRM and the implication to study design parameters (e.g., number of repeated measures); 3) to compare with MMRM under MNAR scenarios.