FDA Draft Guidance Sets Expectations for Bayesian Methods in Clinical Trials

The FDA has released a draft guidance for industry on the “Use of Bayesian Methodology in Clinical Trials of Drug and Biological Products,” laying out recommendations for how sponsors can use Bayesian statistics in studies intended to support the effectiveness and safety of drugs and biologics.

The FDA said the document is intended to facilitate the use of Bayesian approaches and help sponsors make better use of available data when designing and evaluating studies.

Put simply, Bayesian statistics analyzes a clinical trial by combining new trial data with existing knowledge. This existing knowledge (such as evidence from prior studies or established scientific understanding) is summarized as a prior distribution, which represents the range of treatment effects considered plausible before the trial began. The trial data form the likelihood. Together, these produce an updated estimate, called the posterior distribution, that reflects all available evidence.

The FDA notes Bayesian methods can be used in several ways across development, including supporting interim decisions in adaptive trials, informing design choices such as dose selection or supporting the main (“primary”) conclusion of a trial. The draft guidance’s main focus is on the last use case: Bayesian methods used to support primary inference for effectiveness and safety in trials that may underpin regulatory submissions.

Bayesian approaches often arise in drug development because it can be difficult to run large, traditional trials in every setting.

The guidance describes scenarios where Bayesian methods have been used or proposed, particularly when sponsors seek to “borrow” information from outside the current trial, such as results from earlier studies of the same product, external or nonconcurrent control data and adult data used to support pediatric studies when extrapolation is scientifically justified.

The FDA also pointed to incorporating information from other sources, including real-world evidence, among examples of how Bayesian approaches may be applied in clinical trials.

Beyond where Bayesian methods may be used, the guidance details what should be decided up front so results are interpretable. One key concept is the trial’s “success criteria,” meaning the rule that determines whether the treatment effect is considered convincing enough.

Related: FDA’s TEMPO Pilot: Boosting Access to Digital Chronic Disease Tools

In traditional trial designs, success is often judged by whether a p-value crosses a predefined statistical threshold. In Bayesian trials, success is more often framed in terms of probability. For example, a sponsor might define success as showing at least a 97.5% posterior probability that the treatment effect exceeds a clinically meaningful margin, such as a minimum improvement over control.

The guidance explains that when Bayesian methods are used for primary inference, especially with borrowed external information, sponsors should justify how such probability thresholds are chosen and, in some cases, show through simulations that the approach controls the risk of false positive conclusions.

The draft also explains how priors are constructed. The FDA recommends that priors be pre-specified, transparent and scientifically justified, particularly when they are informative and intended to borrow from earlier studies or external data sources.

The document discusses several types of informative priors, including power priors and mixture priors, which allow historical or external data to contribute to the analysis while limiting how strongly that information can influence the final result if it conflicts with the new trial data.

The FDA also emphasizes evaluating how the design behaves under different assumptions, often using simulations to assess operating characteristics such as the chance of reaching an incorrect conclusion and the precision of treatment-effect estimates.

When external data are used, the draft highlights the need to assess relevance and data quality, as well as differences in patient populations, endpoints and standards of care over time. It also emphasizes evaluating situations in which new trial results differ from what the prior information would have predicted.

To help reviewers understand how much influence the prior has on the analysis, the FDA recommends quantifying its impact. This can include measures such as effective sample size, which expresses the contribution of the prior in terms of an equivalent number of patients.

The guidance also stresses that Bayesian analyses should be computationally reliable and sufficiently documented so the FDA can understand and, if needed, reproduce key results.

Finally, the draft states that Bayesian methods and decision rules should be specified in advance in the protocol and statistical analysis plan, and that key design choices should be discussed with the FDA early in development.

The FDA noted that the publication of the draft guidance is part of a commitment under the Prescription Drug User Fee Act (PDUFA) VII.

If you want your company to be featured on Xtalks.com, please email [email protected].