The Bayesian Edge: how the new FDA guidance is enabling nimble, efficient clinical trials

In recent years, the US Food & Drug Administration (FDA) has been on a mission to modernise its programmes, adding more flexibility to clinical trial designs and accommodating the complexities of indications such as cancer and rare diseases.

As part of the push, new draft guidance from the FDA¹ was recently released on the use of Bayesian statistical methods in trials. As the guidance states, “Bayesian methods provide various applications in clinical trials. For example, Bayesian calculations can be used to govern the timing and adaptation rules for an interim analysis in an adaptive design, to inform design elements (eg, dose selection) for subsequent clinical trials, or to support primary inference in a trial.”¹

The Bayesian method is hardly new. So why now? And what does this mean for clinical trial designs and sponsors looking to leverage the new opportunities?

Enabling a smart and efficient route

Bayesian modelling² is a statistical approach that enables one to continuously update the probability of an outcome of interest as new information becomes available. Bayes’ theorem was posthumously published in 1763, but its application is more ubiquitous than ever in modern life. Google Maps navigation is a prime example: Bayesian inference powers the ‘smart’ real-time parts of Google Maps navigation, while graph algorithms handle the ‘fastest route’ part. Bayesian inference facilitates the dynamic process that ensures the best and latest information is used.

This approach of leveraging real-time information also has application in clinical trials. Researchers can incorporate prior data, such as results from earlier preclinical- or clinical-stage studies and even initial data from emerging trials, to refine estimates of a treatment’s effectiveness or safety in flight. Traditionally, these aspects would be fixed and blinded until the end of a study, when results are analysed using predetermined frameworks. A Bayesian approach framework allows sponsors to use accumulating evidence to drive more adaptive and potentially more efficient study designs while a trial is underway. While frequentist methods also support many adaptive features, Bayesian frameworks integrate priors and real-time probabilistic updating more naturally.

However, as with traditional approaches, the methods and plans for use of data or adaptations should be made in a structured manner.

This mindset also echoes a broader shift in leadership thinking towards strategic agility. Rather than locking into one course of action, many biopharmaceutical executives are increasingly building systems or clinical trial frameworks that allow decisions to evolve alongside the evidence rather than rigidly sticking to pre-determined plans or protocols. The ability to calculate predictive probabilities from the posterior distribution helps assess the likelihood of future outcomes based on current, observed information.

A long-complicated relationship with clinical trial designs

While the idea of adaptive, flexible and efficient clinical trials is broadly popular, much of what Bayesian methodology engenders goes against historical precedent for how clinical trials should be run. The FDA (and other regulatory agencies) historically was often reluctant to accept Bayesian methods in clinical trials, particularly in pivotal trials where the application employed relies on informative prior assumptions that can influence the final results, particularly if they are poorly chosen or biased. Without the proper amount of transparency around borrowing plans and consideration for the appropriate weight that evidence should have relative to the analysis of a contemporary trial, an adaptive analytical or statistical method for assessing trial results could bias conclusions drawn for study of a therapeutic candidate.

More broadly, Bayesian inference often requires complex computing without closed-form solutions, so discussions that were typical in all drug programmes around concepts like type I error were not as straightforward. Historically, before Bayesian methods became more commonplace, the statistical complexity of Bayesian approaches could be intimidating for many parties involved in clinical research. People often believed that posterior probabilities would be harder to interpret compared to traditional trial analysis techniques such as p-value–based hypothesis testing and confidence intervals. A 2023 survey of clinical researchers³ found that insufficient knowledge of Bayesian approaches was perceived as the main barrier to implementing Bayesian methods. The survey respondents also reported broad interest in learning more about and maximising the potential use of this statistical tool to transform trial design.

Gradual acceptance by the FDA

Fortunately, advances in computing and methodology are now making Bayesian approaches more easily implementable. In January 2026, the FDA recognised this with the release of new draft guidance on how Bayesian statistical methods can be used in clinical trials for investigational therapies. While this ‘guidance’ document is not yet legally binding, it indicates the agency’s changing attitude towards the use of Bayesian methods, including what is acceptable and how these statistical methods might be used in future clinical trials. The guidance’s primary focus is on supporting primary inference, encouraging sponsors to consider where they might incorporate prior evidence, such as historical trial data or real-world evidence, provided they can justify their assumptions and also demonstrate the results are both strong and reproducible. For therapeutic areas with small patient populations, such as certain types of cancer or rare diseases, the agency’s draft guidance may support meaningful gains in trial speed, flexibility and efficiency with more external information to draw from. In some instances, this could be the difference between a trial being feasible or not.

Oncology is particularly well suited for Bayesian statistical approaches because cancer research is inherently complex and uncertain. Patient populations are often small and highly heterogeneous, with different genomic profiles, disease stages, exposure to prior treatments and other varying factors. The oncology field also moves rapidly, as new trial results and insights are published by both academia and industry regularly. In this environment, traditional study designs that set fixed protocols from trial initiation through to completion can be inefficient or position sponsors to be slow to respond to new insights. Bayesian frameworks offer a flexible alternative, allowing researchers to continuously incorporate new information and adapt trial designs and protocols as evidence accumulates and as appropriate. The ability to refine decisions in real time may help move towards more optimal clinical trial designs and advance promising therapies faster while minimising further investment in ineffective treatments.

Two recent trials provide a window into how this might look in practice. The Biomarker-integrated Approaches of Targeted Therapy for Lung Cancer Elimination (BATTLE) trial⁴ applied a Bayesian adaptive randomisation strategy in 334 patients with advanced non–small cell lung cancer (NSCLC). As the trial progressed, patient biomarker profiles and treatment responses were used to update the probability that each drug or drug combination being assessed would work for specific subtypes of NSCLC. New patients were then increasingly assigned to the treatments most likely to benefit them, enabling the study to learn in real time which therapeutic regimens were most effective for specific biomarker-defined groups.

Another Phase II trial, the I-SPY 2 trial in breast cancer,⁵ uses a Bayesian adaptive design to test multiple investigational therapies simultaneously. The study is in the neoadjuvant setting, evaluating treatments given early – even before surgical intervention. As patient outcome data accumulate, Bayesian models continuously update the probability that each drug will succeed in a future Phase III trial for specific biomarker-defined patient subgroups. Treatments showing strong signals of benefit can ‘graduate’ early for further development, while ineffective therapies are sidelined. This approach allows researchers to identify promising drug–biomarker combinations more quickly and efficiently from an expanded pool of options.

Implications for clinical development

Examples such as the I-SPY 2 and BATTLE trials illustrate how using Bayesian methods can make clinical research more flexible and agile, allowing investigators to respond to emerging evidence in real time. Another important component is the ability to borrow data from external sources, including prior studies, natural history data, real-world evidence or parallel treatment arms, which can all be incorporated during a trial and when analysing results.

By integrating real-time data, insights or other sources into trial design, sponsors may also be able to reduce the size of control groups through the use of historical controls and perhaps introducing more ‘synthetic control arms’, without compromising the statistical power of a study. Smaller control groups hold a lot of appeal, both in reducing the number of patients who need to be recruited (and hence timelines) and also in attracting participants in the first place if, by design in the protocol, a majority are randomised to receive an investigational therapy versus placebo. Trials of rare disease drugs could particularly benefit, where enrolling large numbers of participants is not often feasible.

The ability to refine decisions in real time may help move towards more optimal clinical trial designs and advance promising therapies faster while minimising further investment in ineffective treatments”

At IMUNON, we explored Bayesian borrowing while designing our Phase III trial of IMNN-001 in newly diagnosed advanced ovarian cancer. We assessed the viability of using historical control data to reduce the control arm size in Phase III and incorporating our own Phase II data as an informative prior in the approximately 500-patient Phase III study. Although the historical data from the literature with similar inclusion criteria proved too dated due to advances in maintenance and later-line therapies, we concluded that formal use of our Phase II results would have been both productive and acceptable by regulators. We ultimately opted not to pursue it but the exercise nonetheless validated the strong potential of Bayesian methods for future programmes where prior data are more contemporaneous or patient populations are smaller.

Embracing the philosophical shift

Bayesian statistics are more than just number-crunching applications. These methods require a mindset shift that could reshape clinical trial design over the coming years. They reflect a broader philosophy of decision making that values learning and adaptability over rigid adherence to an initial plan based on pre-conceived notions of how clinical trials must be conducted. Drug development is inherently variable and uncertain, thus taking new information into account simply makes sense. The new FDA guidance acknowledges the ever-changing reality of drug R&D and provides a structured way to incorporate continuous learning throughout a trial while maintaining the scientific rigour needed for clinical advancement and regulatory approvals. For the industry, this creates a clearer path towards broader adoption of Bayesian approaches and the potential benefits that come with them.

References

1. Use of Bayesian Methodology in Clinical Trials of Drug and Biological Products. [Internet] US Food and Drug Administration. 2026. Available from: https://www.fda.gov/regulatory-information/search-fda-guidance-documents/use-bayesian-methodology-clinical-trials-drug-and-biological-products

2. ‌van de Schoot R, Depaoli S, King R, et al. Bayesian Statistics And Modelling. Nature Reviews Methods Primers. [Internet]. 2021;1(1):1–26. Available from: https://www.nature.com/articles/s43586-020-00001-2

3. ‌Muehlemann N, Zhou T, Mukherjee R, et al. A Tutorial on Modern Bayesian Methods in Clinical Trials. [Internet] Therapeutic Innovation & Regulatory Science. 2023;57(3):402–16. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10117244/

4. ‌[Internet] Clinicaltrials.gov. 2026. Available from: https://clinicaltrials.gov/study/NCT01248247

5. ‌CTG Labs – NCBI. [Internet] clinicaltrials.gov. Available from: https://clinicaltrials.gov/study/NCT01042379