Bayesian Dynamic Borrowing to Enhance Evidence for New Therapies.

Abstract

Estimating and interpreting treatment effects (TE) for rare or delayed clinical outcomes is often challenging. To address this, researchers may incorporate additional evidence sources, including historical trial data and concurrent information from intermediate outcomes. In this article, we present Bayesian dynamic borrowing (BDB) as a principled framework for integrating such data while maintaining control of bias and Type I error. Using hypothetical trials of a novel high-efficacy therapy for multiple sclerosis, we provide a step-by-step demonstration of how BDB can be used to combine an imprecise TE estimate for a final outcome with a prediction derived from historical data and information on a concurrent intermediate outcome. Our illustration includes calibration of BDB to meet desired Type I error and power properties, and sensitivity analyses to assess robustness to assumption violations. We also discuss key considerations for applying BDB in regulatory decision making and health technology assessment contexts.