hen performing clinical trials in small populations with a limited number of patients available, it is critical from a scientific and ethical perspective that the maximum amount of information is extracted from the available data. Therefore, traditional statistical approaches may not always be the most appropriate or informative.

In rare diseases and other disease areas where there are significant recruitment challenges due to the patient populations being very small, both for operational and ethical reasons, advances in medicines can become limited when applying traditional statistical approaches, such as a frequentist approach, in clinical trials. One article explains the key differences between the two methodologies as: “the frequentist approach assigns probabilities to data, not to hypotheses, whereas the Bayesian approach assigns probabilities to hypotheses.”

Health authorities are becoming acutely aware of the advances in innovative statistical methods in clinical trials. Developers are strongly encouraging the regulatory community to adopt new guidelines for novel statistical methodologies, and FDA has been pioneering the effort with the release of a draft Bayesian statistics guideline in January 2026.

FDA in the Lead but Can’t Go it Alone

Bayesian statistics may bring benefits to challenging development programs, often through use of prior data that can augment the data within the trial. When applying these methods, one option is to leverage data from another trial in the same program. Pediatric extrapolation is an acceptable use case of Bayesian methods approved by the FDA and the EMA based upon the reliability of the data that has been borrowed. According to the recently published draft guideline, the FDA has further indicated cases of the potential utility of Bayesian approaches, such as:

• Prior information borrowed across two disease subtypes or similar diseases
• Augmenting control arms in clinical trials with external control or nonconcurrent control data
• Prior information borrowed from products in the same class
• Borrowing information between populations in a trial (subgroup analysis)
• Dose-finding trials.

FDA Commissioner Marty Makary said in a recent interview that the use of Bayesian approaches would allow significant efficiencies in clinical development, the transition to continuous trials (from the traditional phased approach), and faster development timelines. FDA’s leadership illuminates the need for other global health authorities to consider similar approaches; convergence is critical when the very nature of rare disease trials requires conducting them globally.

To align with developments at FDA, the EMA’s recently published concept paper (for a subsequent reflection paper) delineates the important questions that European regulators must address better to fully endorse the multiple ICH (most recently, E20 on Adaptive Designs) and other guidelines, such as the EMA guideline on clinical trials for small populations and Q&A for complex clinical trials, that refer to potential use of Bayesian approaches, particularly in confirmatory decision-making. It is important to continue this global dialogue while the European reflection paper is in development during 2026 and 2027.

Overcoming the Inherent Scarcity Problem

A critically important element in Bayesian methods is how prior data is used and identified. The FDA draft guideline emphasizes data quality and reliability; pre-specification; relevance; design of studies from which the data is sought; and availability of patient-level data to sponsor, as important factors in selecting appropriate information to be borrowed, especially where external data is used. Elements such as similarity of estimands, endpoints, and recency are important for data relevance. European regulators have signaled that they are also open to assessing the relevance of Bayesian statistics in rare diseases on a case-by-case basis. However, from their perspective, an inherent dilemma must be addressed—one that generally characterizes drug development for rare populations—to facilitate broader adoption: Where can prior data come from for rare diseases while preserving the nature of confirmatory decision-making? Both emphasize the importance of clearly defined success criteria. In addition, the applicant should identify which questions may be left unanswered and outline why the chosen approaches are appropriate for the research question and statistical plan.

Philosophical Choice on Error Control Impacting Real Patients

Bayesian methods add layers of scientific complexity. They often require extensive simulations to properly assess the operating characteristics of various design and modeling choices. As the concepts of statistical inference when applying Bayesian statistics are based on assigning a probability to a hypothesis, whereas the frequentist method of hypothesis testing either accepts or reject a hypothesis, this reinforces the need for early global alignment with regulatory agencies on concepts such as characterization of the Type I error under various scenarios, success criteria for posterior probabilities, and acceptability of external control data.

European regulators have traditionally put great emphasis on well-understood methods with strict Type I error control and favor simpler models when several options are available. They are very explicit about their role in assessing the Type I error level acceptability. The sponsor is responsible for selecting a sample size that delivers the planned statistical power (i.e., a low Type II error rate) while meeting the pre-specified Type I error standard. In practice, this requires acknowledging an inherent trade-off: Tightening control of Type I error generally reduces power unless the sample size is increased, and boosting power without increasing sample size can inflate the risk of false positives. This approach is challenging for a rare disease with inherently few patients. EMA confirmed that the current practice also opens the door to simulations, but clear justification is always a pre-requisite when they are applied. However, the reflection paper aims to detail better what situations require justification. FDA has a more progressive position and approach: Its draft guideline outlines how to define and justify success criteria, including a description on the use of simulations and a distinction on cases where trials may not need to be calibrated to a fixed Type I error rate.

Innovation Mindset Must Drive

The use of innovative designs is becoming ever more important with increasing availability and access to data and how it may inform regulatory decision-making. As the biology behind many diseases is becoming better understood, the methods to study and demonstrate treatment effects for novel technologies should also evolve to ensure that the right decisions are taken for the benefit of patients. Once the regulatory community is further acquainted with Bayesian methods and this regulatory practice is further established, there may be broader opportunities in other challenging areas (e.g., neurological conditions which have traditionally had a high evidence bar).

One concrete example of applying Bayesian statistics is the recent Alzheimer’s approvals by FDA and EMA (lecanemab) which utilized this approach for dose-finding. The study team used a series of Bayesian interim analyses, identifying doses that had higher response rates, and adapted the trial accordingly. Adapting the prior probability distributions based on patient data that were being collected in the study allowed the trial to learn and to adjust, resulting in a more efficient design. At the same time, the sponsor remained blinded throughout the trial. Leveraging prior data compensated for the absence of good predictive biomarkers or limited evidence of clinical effect, which enabled overall faster development timelines.

Accepting such novel approaches sent an important signal to patients, developers, and investors about regulators’ willingness to expand the toolbox at hand to serve these communities. The emergence of complex endpoints will further promote new statistical considerations and approaches; the sooner regulatory precedents can be established for these new approaches, the better for the patients who are waiting.