A Bayesian phase I/II trial design to optimizing dose-schedule regimen with competing risk outcomes.

Identifying the optimal dose-schedule regimen in early-phase oncology trials is complicated by competing risks, such as disease progression (DP) and dose-limiting toxicity (DLT). Many existing dose-finding methods fail to adequately address these events or accommodate varying administration schedules. We propose CR-EffTox, a Bayesian adaptive phase I/II trial design that jointly models time-to-event DLT and DP using cause-specific hazard functions. Besides, the model proposed allows for dynamic information borrowing to account for associations among dose-schedule regimes. To guide regimen selection, a novel satisfaction score derived from cause-specific survival curves is introduced to quantify the benefit-risk trade-off. The operating characteristics of the method are evaluated through extensive simulations. The method generally outperforms methods that ignore competing risks or information borrowing, substantially improving correct selection probability and enhancing patient safety by reducing allocation to suboptimal regimens.