|
Abstract:
|
In subgroup analysis, it is widely recognized that, when a seemingly promising subgroup is selected post hoc, the traditional analysis simply based on the observed effect size and the unadjusted $p$-value of the selected subgroup is overly optimistic. In this paper, we address the issue of bias in subgroup pursuit and propose a bootstrap-based inference procedure for the best selected subgroup effect. We develop a bias-reduced estimate and valid confidence interval on the selected subgroup effect. The proposed procedure is model-free, easy to compute, and asymptotically sharp. We demonstrate the merit of the proposed procedure by working with data from MONET1 study, and show that it can help make a better-informed decision on subgroup pursuit in clinical trials.
|
