Meeting FDA’s Challenge: Robust Integration of External Data for Time-to-Event Endpoints

Recent FDA draft guidances on Bayesian methodology (January 2026) and externally controlled trials (February 2023) have opened the door to integrating external data — from historical trials, registry databases, and real-world sources — into the primary statistical analysis of clinical studies. But opening the door is not the same as providing a roadmap for how to robustly bring in such information. Dynamic borrowing, prior-data conflict, outcome comparability, and the tension between efficiency gains and bias control remain active methodological challenges, particularly for time-to-event endpoints where data scarcity is the rule rather than the exception: small event counts, restricted control arm sizes, and rare diseases routinely undermine the power of standalone trial analyses. This session brings together three talks spanning frequentist and Bayesian paradigms, addressing the integration of external aggregate registry data, nonparametric Bayesian historical borrowing, and survival prediction model transportability across trial populations. Each contribution targets a distinct facet of the external data problem while sharing a common commitment to rigorous bias control and operationally credible solutions. The session concludes with discussion from an accomplished statistician whose career spans academia, FDA, and the pharmaceutical industry, ensuring that the methodological contributions are contextualized against the regulatory expectations and practical constraints that sponsors and reviewers face.