|
Abstract:
|
There are major challenges in the rigorous identification of patient subgroups with enhanced treatment effect based on data from clinical trials. A typical late phase drug development program consists of multiple studies, which present complexities-since there is more data and more analysis options-as well as opportunities for subgroup identification. We propose an approach to this problem with three key aspects: (1) the multiple analyses should be considered jointly as a learn-and-confirm paradigm; for example, the choice of analysis method for the first trial should be determined with the subsequent analysis in mind; (2) a number of recently developed subgroup identification methods should be utilized, as they provide increased rigor and statistical power over traditional post hoc subgroup analyses; (3) thoughtful simulations should be performed to optimize the learn-and-confirm approach to subgroup identification for a specific application prior to unblinding, with proper considerations given to prior knowledge and meaningful effect sizes. I will present both general considerations and a case study.
|
