In a world where we can order a pizza with a thumbprint and get a self-driving car to drop us at the airport, the world of clinical trial statistics has not kept pace.
For nearly a century, the Frequentist approach has reigned supreme, originally for good reason. It produces robust, reproducible science using standardised methods, shaped when pen, paper, and statistical tables still anchored trial analysis. But this approach treats every trial as an island (at the level of statistical inference) which means ignoring much of what we already know about a drug the moment enrolment begins. The computational barriers to Bayesian methods disappeared decades ago and earlier this month, the FDA signalled that the regulatory barriers are coming down too.
Reverend Thomas Bayes published his theorem in 1763. Now, 263 years later, the FDA is the first major regulatory body to say “Let’s talk about Bayes properly and comprehensively”.
On 9 January, the FDA published comprehensive draft guidance on Use of Bayesian Methodology in Clinical Trials of Drug and Biological Products. While not replacing frequentist methods, the guidance recognises Bayesian approaches that ask probability based questions like “how likely is this drug to help?” can be as useful and meaningful as traditional frequentist hypothesis-testing frameworks that might ask “is the p-value <0.05?”. This is a bold, welcome step toward modernising how everyone thinks about evidence in clinical trials.
For us, it's a landmark acknowledgment of what we've long believed: Bayesian methods aren't fringe math. Implemented properly, they offer a more efficient and more ethical path to getting effective treatments to patients. At Presentient, we’ve consistently said that the traditional frequentist trials (typically run as fixed-design studies) often waste time, money, and most importantly, patient hope.
Although the FDA has permitted Bayesian methods for medical devices since 2010, for drugs and biologics there was no clear guidance so sponsors defaulted to frequentist methods to avoid regulatory uncertainty. That uncertainty just got a whole lot smaller.
The FDA now explicitly outlines how sponsors can use Bayesian methods for:
- Stopping Trials Early: If the math shows a 99% probability that the drug works (or a 99% probability it’s a dud), Bayes lets you stop the trial and get the result. No more maintaining patients on placebo for three more years just to satisfy a rigid, pre-set sample size.
- Borrowing Strength: Why pretend we know nothing about a drug just because we’re testing it in a new population like paediatrics? Bayes allows us to use "priors" (data from previous studies) to inform the current one. It’s how humans naturally learn and a powerful source of signal.
- Adaptive Designs: The guidance finally provides a clear framework for using Bayesian methods to change a trial while it's running (e.g. adjusting doses, reallocating patients, stopping arms) without fear of regulatory pushback.
The FDA’s new stance looks like a green light for the industry to stop being afraid of the "B-word." At Presentient, we’ve spent the last two years building the tools to make ubiquitous "Early Stopping" a reality and we are thrilled to see the FDA issue this guidance, which is a win for sponsors, a win for innovation, and a massive win for patients who shouldn't have to wait for a 500-person trial to finish when the answer was clear at person 150 (about the average fraction at which our algorithm stops trials).
It’s no coincidence that the first application the FDA highlights is “determining futility or success earlier in adaptive trials”. This is the single highest impact problem that all blinded clinical trials face, costing the industry tens of billions and it’s precisely what BRAKES solves.