BOOST: An Adaptive Biomarker-Guided Simon Two-Stage Design with Explicit Control of Pooled and Subgroup Inference

Biomarker-stratified phase II trials need to efficiently identify treatment activity in heterogeneous subpopulations while maintaining rigorous multiplicity control. Conventional approaches, such as parallel Simon two-stage designs, are often inefficient and provide limited guidance on balancing pooled and subgroup-specific efficacy claims under regulatory constraints. We propose a Biomarker-guided Optimized Outcome Simon Two-stage (BOOST) design, which extends the classical Simon framework to adaptive, biomarker-stratified settings. BOOST incorporates explicit stage one futility rules, adaptive enrichment toward promising subgroups, and a unified final-stage decision logic accommodating pooled and subgroup-specific efficacy claims. The design is formulated within a constrained optimization framework that minimizes expected sample size subject to prespecified power and family-wise error rate (FWER) constraints under either weak or strong control. Importantly, the definitions of Type I error rate and power are explicitly aligned with the operational decision rules of the design. Simulation studies demonstrate that BOOST achieves desired Type I error rate control and target power while reducing expected sample size compared with parallel Simon two-stage designs and existing adaptive approaches. Efficiency gains are most pronounced in settings with heterogeneous biomarker-defined treatment effects, with reductions in expected sample size and minimal additional cost when strengthening from weak to strong FWER control.