Bayesian Adaptive Enrichment Design for Continuous Biomarkers

With the advent of precision medicine and targeted therapies in cancer, new challenges in the statistical design of clinical trials have naturally emerged. Most randomized clinical trial designs incorporating predictive biomarkers (those associated with treatment efficacy) assume biomarkers are dichotomous, or dichotomize naturally continuous biomarkers upfront, or find cut points mid-way through the trial to classify patients as biomarker-positive or biomarker-negative. However, these practices ignore or discard information about continuous and possible nonlinear or non-monotone prognostic or predictive effects. In this article, we propose a novel adaptive enrichment trial design to handle continuous biomarkers with any effect shape, including Bayesian marker-adaptive randomization. We demonstrate that this design can correctly make marker-specific trial decisions with high efficiency, resulting in improved performance and patient-centered decisions compared to adaptive cut-point selection approaches without adaptive randomization that further ignore or oversimplify true underlying marker relationships.