In this episode of "In the Interim…", Dr. Scott Berry and Dr. Kert Viele deliver a quick reaction to the FDA’s draft guidance on Bayesian statistics for clinical trials of drugs and biologics. Their assessment addresses the structure, content, and impact of the document, emphasizing evidence-based requirements and guidance scope. The episode breaks down regulatory language, technical expectations, and workflow implications for clinical trial sponsors and statisticians.
Key Highlights
- Clear distinction between trials justified by type 1 error control and trials justified by agreement on Bayesian priors and decision rule.
- Explanation of how informative priors can be created based on external or historical data.
- Technical explanation of dynamic discounting/borrowing, especially in Bayesian hierarchical models for rare populations, pediatric-adult extrapolation, related disease subgroups, and platform and basket trials (e.g., ROAR).
- In-depth look at the necessity of sensitivity and robustness checks for different priors, and the FDA’s design prior and analysis prior terminology.
- FDA’s requirements for accepting external data sources: data provenance, patient-level comparability, recency, and appropriate covariate adjustments.
- Comparison with ICH E20 on adaptive designs, providing context for ongoing regulatory harmonization and possible influence on international regulatory directions.
- Direct warning against attempts to misuse Bayesian methodology as a substitute for scientific rigor; legitimate uses must meet FDA standards and not simply serve to lower evidentiary bars.
