TL39: Analysis of Change from Baseline Data in the Presence of Covariate-by-treatment Interaction
*Jihao Zhou, Allegan Pharmaceuticals Inc. 


Change from baseline analysis continues to be of great interest in the field of clinical trials (Senn (2006), Stat Med.; Liu, et al (2009). Stat Med). The commonly used statistical approach is the analysis of covariance (ANCOVA) model. Under the ANCOVA model assumption of homogeneity of slopes, comparative treatment effect can be estimated from the between-group least squares mean difference. However, in the case of heterogeneous slopes, the interpretation of the main treatment effects is controversial. (ICH E9 (1998), CPMP (2003), Points to consider on adjustment for baseline covariates). If interaction of covariate-by-treatment is identified, then many questions emerge. What about the baseline value? Should it be included in the model or not and what is the impact? How should the treatment effect be reported in the product label? To circumvent the issue, alternative methods such as nonparametric approaches and data transformation are used; but doing so leads to a different causal inference. In this roundtable, participants shall discuss appropriate approaches to estimating the treatment effect and to exchange experiences or ideas from both the FDA and industry perspectives.