FDA Guide Outlines Expectations for Bayesian Trial Designs
A new FDA guidance document details how drug developers can appropriately use Bayesian statistical methods in clinical trials intended to support approvals, signaling continued openness to innovative trial designs while emphasizing rigor, transparency, and prespecification. The document, Use of Bayesian Methodology in Clinical Trials of Drugs and Biologics, describes FDA’s current thinking on when and how Bayesian approaches may be used in studies submitted in support of INDs, NDAs, BLAs, and related supplements.
FDA says Bayesian methods can play a role across drug development, including adaptive interim analyses, dose selection, and the design of follow-on studies. The primary focus of the guidance, however, is on the use of Bayesian methods to support primary inference —he main efficacy and safety conclusions relied upon in regulatory decision-making.
Bayesian statistics differ from traditional approaches by formally combining prior information with data generated in a trial to produce a “posterior” estimate of treatment effects. FDA emphasizes that the choice and justification of prior distributions are critical, particularly when external data are incorporated.
The guidance lays out several real-world contexts in which Bayesian methods have already been used in FDA submissions. These include borrowing data from earlier trials of the same product, augmenting randomized control arms with external or nonconcurrent controls, and pediatric extrapolation, where adult data are leveraged to support pediatric indications.
One example cited is the use of Bayesian analysis to incorporate phase 2 data into a phase 3 trial supporting the approval of Rebyota for recurrent Clostridioides difficile infection in 2022. FDA also pointed to oncology platform trials such as GBM AGILE, where Bayesian models have been proposed to account for temporal changes when using nonconcurrent control data.
The agency also highlighted Bayesian approaches in basket trials, subgroup analyses, and early-phase oncology dose-finding studies, where model-based designs may improve efficiency and better identify optimal dosing than traditional maximum tolerated dose strategies.
A substantial portion of the draft guidance addresses how sponsors should define success criteria when Bayesian methods are used. FDA describes multiple acceptable approaches, including calibrating Bayesian success criteria to control Type I error, directly interpreting posterior probabilities when informative priors are justified, and decision-theoretic frameworks that explicitly incorporate benefit–risk considerations. In all cases, the agency stresses the importance of prespecifying thresholds and using simulations to understand trial operating characteristics.