Bayesian Hierarchical Borrowing: Adaptive Evidence Sharing for Treatment Effect Estimation in Platform Trials (Motivated by the I-SPY 2 Adaptive Randomization Study of Neratinib in Early Breast Cancer by Park et al.)

Bayesian hierarchical borrowing has become an essential methodological approach for modern platform trials characterized by biological heterogeneity, small subgroup sample sizes, and adaptive decision-making. By enabling partial pooling of information across related subgroups, hierarchical models improve estimation precision while preserving clinically meaningful differences in treatment effects. This report examines the rationale, implementation, and implications of Bayesian hierarchical borrowing, motivated by the I-SPY 2 adaptive randomization study of neratinib in high-risk early breast cancer. Using this trial as a case study, we describe how covariate-adjusted hierarchical logistic models supported response-adaptive randomization, early stopping for futility, and therapy graduation based on predictive probabilities of success in future confirmatory trials. We outline key dataset requirements, model assumptions, and reporting standards, including posterior estimates, borrowing diagnostics, and decision-focused visualizations. Strengths and limitations of hierarchical borrowing are discussed, with emphasis on exchangeability assumptions, prior sensitivity, and interpretability. Finally, we highlight future directions for methodological refinement, integration with adaptive infrastructures, and alignment with precision medicine goals. Together, this report demonstrates how Bayesian hierarchical borrowing transforms fragmented subgroup data into coherent, decision-relevant evidence for efficient and transparent clinical trial evaluation.