Hierarchical Bayesian Approaches to the Subgroup Analysis Problem in Infectious Disease Clinical Trials

• Activity Number: 307
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Carl Dicasoli*+ and Cynthia DeSouza and Lan Lan
• Companies: Vertex Pharmaceuticals and Vertex Pharmaceuticals and Vertex Pharmaceuticals
• Keywords: hierarchical Bayes ; interaction ; Type I/II error ; power ; subgroup analysis ; shrinkage

Abstract

In clinical trials, a particularly relevant problem includes the examination of subgroup analyses within the context of a larger, general effect. In most frequentist setups, comparing treatments separately within subgroups defined by a partition of subjects into mutually exclusive subgroups, increases both Type 1/Type 2 errors and is likely to produce misleading results. Hence, there is a need not to implement hypothesis testing within subgroups but rather to employ shrinkage methods within a Bayesian framework that specify that qualitative treatment-by-subgroup interactions are a priori unlikely. Because of the exchangeability of interaction terms, shrinkage occurs in the Bayesian posterior point and interval estimates. Furthermore, if the prior distribution for every parameter of interest is well-conceived, the corresponding posterior probabilities do not need adjustment according to the number of subgroups tested. In our work, we will conduct simulation studies to compare the performance of a Bayesian hierarchical model under different shrinkage models to frequentist approaches by analysing the required sample size, power, and width of credible/confidence intervals.