A Bayesian Optimal Interval Design Considering Efficacy and Toxicity in Early Phase Basket Trials.

Abstract

Oncology drug development has increasingly shifted toward determining optimal biological doses rather than maximum tolerated doses (MTDs), particularly for targeted therapies and immunotherapies that exhibit complex dose-efficacy relationships. Concurrently, basket trials have emerged as an efficient approach for evaluating investigational treatments across multiple cancer types sharing common molecular targets. We propose the BOIN-ETB design, a model-assisted dose-finding design that addresses optimal dose (OD) identification in phase I/II basket trials by incorporating both toxicity and efficacy endpoints. The proposed approach employs common toxicity boundaries across cancer types while implementing cancer-specific efficacy boundaries to account for differential efficacy responses between baskets. OD selection utilizes utility functions that quantify efficacy-toxicity trade-offs. Through comprehensive simulation studies across Fourteen realistic scenarios, the BOIN-ETB design demonstrates robust performance in identifying true ODs while maintaining acceptable safety profiles across diverse cancer populations. The design provides superior consistency compared to alternative approaches, particularly in scenarios with heterogeneous dose-efficacy relationships between cancer types, making it well-suited for contemporary oncology dose-finding basket trials.