A Resampling-based Ensemble Tree Method to Identify Patient Subgroups with Enhanced Treatment Effect
Chakib Battioui
Eli Lilly and Company
Lei Shen
Eli Lilly and Company
Stephen J. Ruberg
Eli Lilly and Company
In this paper we describe an approach to identify patient subgroups with enhanced treatment effect in clinical trials. It utilizes ensemble trees based on resampling and naturally produces two consistency measures for each potential subgroup identified. We compare simple ways to combine these measures into an overall summary of strength. Using stratified permutations and out-of-bag samples, the approach also provides a multiplicity-adjusted p-value and bias-corrected estimate of treatment effect, both of which are important for decision-making in tailored therapeutics applications. A simulation study is performed to evaluate the performance of the proposed method.