How good is your selected subgroup?

Wednesday, November 7, 2018 - 9:20am - 10:10am
Lind 305
Xuming He (University of Michigan)
Subgroup analysis is frequently used to account for the treatment effect heterogeneity in clinical trials. When a treatment is seen marginally effective for the population of the original study, it is tempting to consider post hoc subgroup identification. When a highly promising subgroup is selected this way, serious questions have to be asked about the potential risks and rewards of the subgroup pursuit. In this talk, we will discuss factors that have direct impacts on the credibility of subgroup pursuit, and then propose a model-free approach to quantify how likely the promise of the selected subgroup is a statistical artifact, and how good the selected subgroup really is. The proposed quantitative analysis of subgroup pursuit can help inform better decisions about any selected subgroup in clinical trials. The talk is based on joint work with Xinzhou Guo.