Comb-BOIN12: A Utility-Based Bayesian Optimal Interval Design for Dose Optimization in Cancer Drug-Combination Trials

Abstract

Drug combinations are increasingly utilized in cancer treatment to enhance drug effectiveness through synergistic therapeutic effects. However, determining the optimal biological dose combination (OBDC) in small-scale drug combination trials presents challenges due to the increased complexity of the dose space. To effectively optimize the dose combination of combined drugs, we propose a model-assisted design by extending the single-agent Bayesian optimal interval phase I/II (BOIN12) design. Our approach incorporates a utility function to balance the trade-off between risk and benefit and directly models the utility of each dose by constructing a quasi-beta-binomial model. A key advantage of our design is the simplification of decision-making during interim periods by considering all possible outcomes and pre-including the decision rule in the protocol. Additionally, we present a time-to-event (TITE) version of our design, employing an approximate likelihood approach to mitigate potential late-onset effects. We demonstrate that our proposed design exhibits robust and desirable operating characteristics across various scenarios through extensive simulation studies.

Keywords: Bayesian adaptive design; Dose optimization; Drug combination; Model-assisted design; Risk-benefit trade-off.