A Bayesian phase I/II platform design with data augmentation accounting for delayed outcomes

Delayed outcomes, such as late-onset toxicity and efficacy, present substantial challenges in dose optimization during Bayesian phase I/II platform trials for drug therapies across multiple indications. To address these challenges and ensure timely, reliable dose selection, we propose an enhanced Bayesian phase I/II platform design with data augmentation method accounting for delayed outcomes (BPDD). The BPDD design uses an iterative augmentation process to predict unobserved outcomes, combining observed and imputed data to refine dose-toxicity and dose-efficacy estimates, which not only supports optimal decision-making but also significantly reduces the time required for clinical trials. By borrowing information across indications through hierarchical modeling, our approach improves the accuracy of parameter estimates and facilitates efficient risk-benefit trade-off analyses. At each interim analysis, these updated estimates guide decisions on dose escalation, de-escalation, and optimal biological dose identification for each indication. Simulation studies demonstrate the robustness and flexibility of our approach in managing delayed outcomes, highlighting its accuracy and time efficiency in dose optimization. This framework has the potential to enhance clinical trial efficiency, accelerate drug development, and improve decision-making in multi-indication platform trials.