BEACON : An Adaptive Bayesian Early‐Phase Oncology Trial Design Integrating Dose Optimization and Proof‐of‐Concept Across Survival and Binary Endpoints

Conventional dose selection in oncology, which focuses on the maximum tolerated dose (MTD), has certain limitations for targeted agents and immunotherapies, where the efficacy-toxicity relationship may not be monotonic. Regulatory initiatives such as FDA's Project Optimus highlight the need for improved dose optimization methods early in drug development. Although recent Bayesian dose-finding designs enhance flexibility by integrating efficacy data and adopting adaptive monitoring, most depend on binary response outcomes and lack sufficient proof-of-concept (PoC) assessment before advancing doses to confirmatory trials. To address these gaps, we propose BEACON, an extended Bayesian optimization framework for randomized phase II oncology trials. BEACON incorporates time-to-event endpoints into both dose selection and PoC assessment. Extending from the DODII method (Bayesian dose optimization for randomized phase II trials), BEACON combines Bayesian safety and futility monitoring with a pick-the-winner strategy for both survival and binary outcomes. The design dynamically borrows information across dose levels during PoC to enhance decision-making. By allowing both survival and binary outcomes within a unified dose optimization with PoC strategy, BEACON enhances design flexibility and delivers robust confirmation of clinical benefit, aligned with regulatory expectations for dose justification in early-phase oncology studies. Simulation studies demonstrate that the BEACON design exhibits favorable operating characteristics, effectively controlling false go and selection error rates for the recommended doses and reducing the total sample sizes of the trial. With adaptive borrowing across dose levels, the design robustly maintains the desired proof-of-concept power.