Inference on Selected Subgroups in Clinical Trials

Speaker: Xuming He, University of Michigan

When existing clinical trial data suggest a promising subgroup, we must address the question of how good the selected subgroup really is. The usual statistical inference applied to the selected subgroup, assuming that the subgroup is chosen independent of the data, will lead to overly optimistic evaluation of the selected subgroup. In this talk, we address the risk of greedy subgroup pursuit and propose quantitative analysis protocols that can help evaluate the potential of selected subgroups.  In particular, we propose a new bootstrap-based inference procedure for the best selected subgroup effect. The proposed inference procedure is model-free, easy to compute, and asymptotically sharp. We demonstrate the merit of our proposed method by re-analyzing the MONET1 trial and show how the subgroup is selected post hoc should play an important role in any statistical analysis.