You were among the first to submit and defend a Bayesian trial design to the FDA around 20 years ago. How were Bayesian methods viewed by regulators at that time?
At that time, the statistical community was still quite split between frequentists and Bayesians, so it seemed additional educational work was required. In the early 2000’s, Bayesian approaches were viewed mostly as exploratory tools, essentially for early phases, in indications where standard trial design and recruitment were impossible. For example, the FDA and Health Canada were open to consider a phase 1-2 Bayesian adaptive design for a treatment following spinal cord injury. [1] Interactions with FDA were very ad hoc, with early face-to-face interactions.
What stands out most to you about how the FDA’s attitude toward Bayesian methods has evolved leading up to the 2026 draft guidance?
The FDA statisticians were overall very open and supportive from the beginning. Yet at that time the lack of experience, precedents, validated tools and programming environments, regulatory guidance was limiting adoption. On the methodological side, the primary issue perceived by practitioners and regulators was the use and the robust definition of a prior distribution. With time, confidence and experience has been built.
Bayesian borrowing and extrapolation can be recognized as powerful but risky. What does the guidance make clear that sponsors must get right for regulators to be comfortable?
The FDA guidance clarifies how to properly define and evaluate prior distributions, and the required sensitivity analyses and evaluations of operating characteristics. [2]
From your experience, what are the biggest mistakes sponsors still make when proposing Bayesian designs to regulators today?
Some of the key mistakes include:
- Being reacting instead of proactive, early engagement and planning is a key success driver
- The lack of thorough simulation-based evaluations before submitting a Bayesian approach
- The misspecification of “trial success” definition
For biotechs in particular, how should early regulatory conversations evolve when considering Bayesian approaches across development and decision-making?
Biotech sponsors are often even more legitimate users of Bayesian trials and more broadly for Bayesian approaches for decision making. As a matter of fact, biotechs need even more than big pharmas to quantify their current development risk and expected drug value over drug development time, based on the evidence available beyond just trials. While the same regulatory processes apply to all sponsors, the benefit of quantifying and monitoring such risks and uncertainty is enhanced for innovative biotechs and smaller companies.
Looking ahead, do you see this guidance as a turning point, or a formalization of expectations that were already there?
It is both, as this formalization is unlocking many Bayesian design opportunities and clarifying implementation for sponsors.
Have questions about implementing Bayesian approaches in your trials?
References
- Amzal, B., Racine, A., & Mercier, F. (2006). Accelerated drug development strategy via Bayesian approach: Application to a spinal cord injury compound. In Proceedings of the 8th World Meeting on Bayesian Statistics (ISBA), Benidorm, Alicante, Spain, June 1–6, 2006.
2. https://phastar.com/knowledge-centre/blogs/expert-opinion-on-fda-recommendations-on-bayesian-methods-for-drug-development/
