Recovery of a Treatment Effect from Pre-Specified Subgroups

• Activity Number: 563
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Jitendra Ganju*+ and Kefei Zhou and Yunzhi Lin
• Companies: Gilead and Amgen and AbbVie
• Keywords: randomized clinical trial ; subgroups ; multiplicity

Abstract

In clinical trials it is often known in advance that some subgroups are more likely to demonstrate larger treatment effects than other subgroups. When this is the case, we show that fitting models to subsets of data defined by subgroups can give higher power than the conventional method of fitting one model to the entire data. The proposed method involves combining p-values from subgroups in a way to control the type I error rate at its designated value. When such subgroups can be identified in advance the gain in power is large enough to recommend use of this approach over the conventional method. Even when the subgroups are incorrectly identified, the loss in power is not much.