The "one-size-fits-all'' paradigm is inappropriate for phase II clinical trials evaluating biotherapies, which are often expected to have substantial heterogeneous treatment effects among different subgroups defined by biomarker. For these biotherapies, the objective of phase II clinical trials is often to evaluate subgroup-specific treatment effects. In this article, we propose a simple yet efficient Bayesian adaptive phase II biomarker-guided design, referred to as the Bayesian-order constrained adaptive design, to detect the subgroup-specific treatment effects of biotherapies. The Bayesian order constrained adaptive design combines the features of the enrichment design and sequential design. It starts with a "all-comers" stage, and subsequently switches to an enrichment stage for either the marker-positive subgroup or marker-negative subgroup, depending on the interim analysis results. The go/no go enrichment criteria are determined by two posterior probabilities utilizing the inherent ordering constraint between two subgroups. We also extend the Bayesian-order constrained adaptive design to handle the missing biomarker situation. We conducted comprehensive computer simulation studies to investigate the operating characteristics of the Bayesian order constrained adaptive design, and compared it with other existing and conventional designs. The results shown that the Bayesian order constrained adaptive design yielded the best overall performance in detecting the subgroup-specific treatment effects by jointly considering the efficiency and cost-effectiveness of the trials. The software for simulation and trial implementation are available for free download.
Keywords: Phase II clinical trials; interim analysis; marker-guided design; missing data; sequential design.