In January 2026, FDA issued the draft guidance "Use of Bayesian Methodology in Clinical Trials of Drug and Biological Products," marking the agency's clearest endorsement to date of modern Bayesian approaches for primary inference in drug and biologic approvals. While Bayesian methods have long been accepted for certain products, their role in pivotal drug trials has historically been limited. This guidance signals broader regulatory acceptance, provided sponsors clearly justify how prior data are incorporated into the analysis and demonstrate that doing so will not bias results.

A maturing regulatory framework. The guidance makes clear that FDA views Bayesian methods as suitable not only for adaptive designs and dose-finding studies, but also for confirmatory trials, including those relying on external data borrowing (e.g., pediatric extrapolation, rare diseases, platform trials, subgroup analyses). The guidance outlines three approaches for pre-specifying success criteria:

• Calibrate Bayesian success criteria to Type I error (e.g., controlling one-sided α = 0.025).
• Rely on direct posterior probabilities, where regulatory decisions hinge on the probability that treatment benefit exceeds a clinically meaningful threshold.
• Base success criteria on benefit-risk or decision-theoretic approaches, balancing benefit, safety, disease seriousness, and regulatory error consequences.

Prior construction is the centerpiece. The guidance devotes substantial detail to how prior distributions must be constructed, justified, and pre-specified in study documentation, including:

• Systematic identification of all relevant external data (including unfavorable studies).
• Explicit modeling of relevance, exchangeability, and bias.
• Use of dynamic discounting methods (e.g., hierarchical models, mixture priors, commensurate priors) to protect against prior-data conflict.
• Quantification of prior influence using effective sample size (ESS) rather than Type I error inflation.

In practical terms, FDA is making clear that borrowing must be disciplined, transparent, and proportionate to the strength and relevance of the external data.

CDRH vs. CDER/CBER. While CDRH has historically been more accepting of Bayesian designs in device trials, some recent experience indicates that the device Center is more comfortable with Bayesian designs that can be justified in frequentist terms – such as control of Type I error – even when posterior probability was the natural inferential metric. Although this guidance has not been adopted by CDRH, it affirms that CBER and CDER may base regulatory decisions on posterior probability, where external data are credibly incorporated into a prior distribution.

Overall, the 2026 guidance encourages transparent, well-justified, and pre-specified use of Bayesian methods, especially when borrowing external information, to support regulatory decisions. Protocols utilizing Bayesian approaches should be discussed with FDA prior to execution.