Industry asks FDA to affirm Bayesian methods guidance applies broadly to drugs
Industry groups and drugmakers want the US Food and Drug Administration (FDA) to explicitly clarify that Bayesian statistical methods can be used for products beyond those intended for children and rare diseases. They also asked for additional case examples in the guidance to clarify the agency’s expectations.In January, FDA published a draft guidance that provides recommendations on how sponsors can use Bayesian methods to support the safety and efficacy of their products using information from previous studies. A broad range of stakeholders submitted comments on the proposed guidance, expressing support while also asking the agency to provide specific clarifications and provide new examples for sponsors to use. (RELATED: FDA issues guidance on use of Bayesian methods to support drug development, Regulatory Focus 9 January 2026)
The Pharmaceutical Research and Manufacturers of America (PhRMA) was among the commenters who lauded the draft guidance as a substantial step toward modernizing statistical practices for regulatory decision-making and emphasized that Bayesian statistics have proven particularly useful in pediatric and rare disease trials and in complex trials. The group also expressed that the guidance was an opportunity to increase confidence in clinical trials that use Bayesian approaches and encouraged FDA to work with other regulators through the International Council for Harmonisation (ICH) to develop harmonized guidelines.
Furthermore, PhRMA asked FDA to balance expectations regarding specifications, justification, and documentation with the need to be flexible and reduce burdens on trials that use Bayesian methodology. The group said that some of the expectations for evaluation and documentation outlined in the draft guidance could disincentivize the use of certain Bayesian approaches.
"As the Draft Guidance recognizes, noninformative priors are designed to have the minimum number of assumptions and information, so there is less information to justify and check," said PhRMA. "Since noninformative priors are widely used in Bayesian statistics, the recommendations in this Draft Guidance could disincentivize sponsors from using noninformative priors.
"We recommend that FDA instead distinguish its expectations for noninformative priors, minimally informative priors, and informative, including skeptical priors, and either define these categories within the guidance or provide references to relevant scientific literature," the group added.
PhRMA also asked FDA to expand the scope of the guidance. In particular, the group said that, based on the current language of the guidance, pediatric and rare disease indications are the preferred areas for Bayesian methods, and that other trials will be evaluated on a case-by-case basis. It asked the agency to explicitly state that Bayesian methods could also be used for exploratory and early phase trials.
The pharma lobby group commended FDA’s use of examples in the guidance to describe how sponsors can use Bayesian methods in their regulatory submissions but asked for more visual aids and hosting workshops to help them flush out the agency’s thinking.
“Given the complexity and technical nature of the Draft Guidance, FDA should consider adding visual aids or publishing additional documents to improve the accessibility for readers who are not as familiar with Bayesian methods,” said PhRMA. “PhRMA also recommends that FDA convene workshops with industry and other relevant stakeholders for in-depth discussions on topics discussed in the Draft Guidance.
“Many of the issues discussed in the Draft Guidance are technically complex and could benefit from discussion about how sponsors should successfully employ Bayesian methodologies in clinical trials,” the group added.
The Biotechnology Innovation Organization (BIO) said it was concerned that certain parts of the guidance related to prior specification, borrowing frameworks, operating characteristic calibrations, documentation expectations, and acceptable sensitivity analysis could be seen as too burdensome by sponsors and could lead them to refrain from using Bayesian methods that may be more appropriate. The group requested several changes and clarifications to the guidance to help sponsors adopt Bayesian methods.
BIO also noted that FDA should ensure Bayesian methods are recognized as another general evidential framework rather than a niche tool used in limited circumstances for pediatric and rare disease indications.
"The draft currently emphasizes borrowing in feasibility-constrained settings, even though the scientific value of Bayesian inference extends well beyond feasibility considerations," said BIO. "When relevant and sometimes hard-to-ignore information exists, such as from prior trials, related populations, or ongoing evidence generation, its formal integration may yield a more coherent and efficient assessment of treatment effects than requiring multiple independent, standalone trials that do not leverage accumulated knowledge.
"Recent discussions of evidentiary flexibility in drug development, including the role of single pivotal trials, have also highlighted the potential of Bayesian approaches to strengthen regulatory decision-making frameworks," the group added. "Bayesian methods also support a continuous learning paradigm across development stages."
BIO also asked FDA to ensure that the agency does not impose greater operational or documentation requirements on sponsors who use Bayesian methods than on sponsors who use traditional methods, and to clarify the operating characteristics and calibration.
"FDA appropriately recognizes that Bayesian decision criteria do not always need to be calibrated to traditional Type I error thresholds," said the group. "However, the guidance would benefit from clearer delineation of when calibration is expected, when it is optional, and when alternative approaches, such as simulation-based evaluation to avoid misleading conclusions or decision-theoretical criteria, are appropriate.
"Specifically, the guidance should distinguish among confirmatory trials, adaptive designs, borrowing nuisance parameters versus treatment effects, and small-population settings," it added.
Additionally, BIO asked FDA to ensure the guidance allows for flexibility when incorporating relevant evidence using Bayesian methods and takes into account that rare diseases and small patient populations require tailored considerations. The group also asked for clearer expectations regarding covariate adjustment in the borrowing framework, multiplicity across endpoints, sequential development programs, simulation expectations, software validation, and reporting in regulatory submissions.
Furthermore, Bio asked FDA to clarify how manufacturers should present their Bayesian findings in product labels and how predictive distributions can be used for decision-making during interim analysis.
The American Academy of Pediatrics (AAP) also expressed support for the draft guidance but stated that, due to the unique characteristics of the pediatric patient population, it wanted clarifications in several areas to ensure that the agency's proposed extrapolations and data borrowing were scientifically justified and appropriately protected children. The group emphasized that pediatric conditions tend to have limited data to extrapolate robust conclusions from, and that using data from adult studies needs to be justified on a case-by-case basis.
"To further clarify the guidance, AAP recommends language specifying that pediatric extrapolation and borrowing strategies should also account for developmental stage and physiological maturity, and not treat pediatric populations as a single homogenous group," said AAP. "In particular, adult data may be more relevant for adolescents (e.g., >_ 13 years of age) or pediatric patients with body weights comparable to adults (e.g., >40kg), whereas adult data may be less relevant for younger or smaller children.
"Such distinctions would help sponsors and reviewers apply Bayesian methods in a manner that reflects pediatric biological differences," the group added. "AAP also suggests that the guidance clarify that a prior distribution reflects the incorporation of information from another distribution or data source into the proposed study to improve interpretability."
AAP asked FDA to consider the use of Bayesian dose-finding approaches beyond oncology trials. It also asked the agency to clarify that there is no rigid cut-off time for the recency of data used for extrapolations, and that such data should be considered based on whether older data remain applicable and comparable to current clinical practice.
The Duke Margolis Institute for Health Policy also asked FDA to provide more examples in the draft guidance of when Bayesian approaches may be used. The institute was especially grateful for the guidance, as China's National Medical Products Administration (NMPA) and the European Medicines Agency (EMA) are the only other regulatory agencies to have published guidance or reflection papers on the use of Bayesian methods.
Duke Margolis noted that the guidance asks sponsors to provide details on the software used to analyze data using Bayesian methods, most likely so that FDA reviewers can analyze the raw data themselves. However, the institute said sponsors may not always be able to provide raw data, complete datasets, or statistical algorithms for several reasons. With that in mind, it asked the agency to provide specific clarifications, such as expectations for sponsor documentation and instances when the agency may need to inspect software tools and related activities.
Duke Margolis also noted that the guidance states that real-world data (RWD) can be used as a source of external information and that shrinkage estimation via a one-way Bayesian hierarchical model can be used to estimate the treatment effect in one subgroup as a weighted average of the overall estimated treatment effect. The institute asked for clarification on how both such methods can be used by sponsors.
"We believe that addressing these limitations in the final guidance, versus on a case-by-case basis at the time of or after regulatory review, aligns with the spirit of “early and often” engagement between sponsors and the FDA as they contemplate appropriate uses/applications of RWD/E in regulatory submissions," said Duke Margolis.
Several pharmaceutical companies also commented on the guidance, including EMD Serono. It asked FDA for clarification on its expectations for trials that use interim decision-making where success criteria other than a Type I Error Rate are used. More specifically, it asked for additions to the guidance, such as asking sponsors to prespecify the anticipated frequency of interim analyses for trials not calibrated to a Type I error rate.
"Clarifying FDA’s expectations for interim Bayesian decision-making is critical for both operational feasibility and regulatory alignment," said EMD Serono. "Such clarifications would help sponsors design interim strategies that uphold trial integrity while leveraging the flexibility of Bayesian methods."
EMD Serono also asked FDA to affirm that Bayesian approaches may be used for safety analyses in later-phase development, where they can help with interpretability and efficiency, and clarify how to use different Bayesian strategies, such as hierarchical models, joint modeling, and composite probability-based criteria. The company also asked the agency to recommend using a tiered documentation structure for when submitting information on prior descriptions.
