A Robust Quantitative Risk Screening for Subgroup Pursuit in Clinical Trials

In clinical studies, when to recommend or decide further pursuit of the most promising subgroup that has been observed from an existing trial is a very important question. It is well recognized that the working models in assessing subgroup effects might be misspecified and the observed treatment effect size of the best selected subgroup tends to be too optimistic. Therefore, a careful and robust statistical quantification of risk is useful before any decision of subgroup pursuit is made. Via the newly established bootstrap consistency for the misspecified proportional hazard model, the issue of selection bias and model misspecification in subgroup pursuit is addressed, and a robust risk quantitative measure directly based on the observed treatment effect of the selected subgroup that might be used in the decision-making of subgroup pursuit is provided. Two earlier studies are reviewed to demonstrate what can be learned from the proposed risk index.