Precision oncology seeks to match patients with treatments that maximize clinical benefit. Identifying clinically interpretable subgroups with differential treatment effects from randomized clinical trials, however, remains challenging. We propose a subgroup-discovery framework based on dense random survival forests, with a splitting criterion that combines prognostic discrimination and treatment-effect heterogeneity. The framework aggregates patient proximity matrices across multiple forest configurations, identifies subgroups using the resulting treatment-informed similarity representation, and summarizes them through concise covariate rules. An empirical calibration procedure assesses evidence for treatment-effect heterogeneity. Extensive simulations demonstrate the ability of the framework to distinguish heterogeneous from homogeneous treatment-effect settings and recover treatment-responsive subgroups, with empirical Type I error control under the homogeneous scenarios considered. An application to a Phase III oncology trial identifies a treatment-responsive patient profile defined by baseline characteristics, illustrating the potential of the framework to inform subgroup-focused clinical research and future trial design.
翻译:暂无翻译