Abstract

Multiregional trials are increasingly used at the dose-finding stage, where an optimal biological dose (OBD) must be selected for each region despite limited regional sample sizes and possible between-region heterogeneity. Existing methods address this in isolation: dose-finding methods assume a pooled population, consistency methods assess regional agreement but stop short of dose selection, and borrowing methods improve regional estimation yet are rarely tied to dose optimization. We propose a unified framework for region-specific dose optimization. The framework simultaneously monitors binary toxicity and binary efficacy outcomes and selects the OBD for each region by weighing the two through a utility function. To inform this selection, longitudinal tumor size measurements are used to describe treatment response profiles across regions, providing a basis for assessing how similar the regions are. Between-region similarity is quantified through a pairwise likelihood-ratio distance, which in turn is used to construct power priors for the efficacy outcome, so that information is borrowed adaptively—more strongly between regions with similar response profiles and less between those that diverge. We evaluate the framework in a simulation study spanning scenarios. Our proposed method achieves strong information borrowing in homogeneous scenarios, improving the accuracy of OBD estimation and the detection of cross-regional OBD consistency. In heterogeneous scenarios, it automatically switches to weak borrowing. The framework provides a coherent basis for dose optimization in multiregional dose-finding trials.