Sign in Subscribe

Bayesian Methods in Regulatory Science: A Complete Guide

Last updated: February 2026

Over the past few months, I've written a growing collection of in-depth posts exploring how Bayesian thinking applies to FDA decisions, clinical trial design, and regulatory science. This directory organizes the complete collection, from formal Bayesian methods to critical examinations of where the framework breaks down.

Start Here: Understanding the Landscape

The FDA's Bayesian Guidance: Learning in Theory, Pre-Specification in Practice - The FDA endorsed Bayesian methods but demands pre-specification. What the new guidance actually means for trial design and why the philosophical promise of learning meets the operational reality of regulatory constraints.

Calibrated Bayes: The Framework You're Already Using - If you've ever justified a Bayesian design to a regulator, you've been practicing calibrated Bayes—designing for frequentist operating characteristics while interpreting with posterior probabilities. The middle path between philosophical camps that most working biostatisticians have quietly converged on.

What I Submitted to FDA on the Bayesian Guidance - Three public comments on the draft guidance: name the hybrid framework, operationalize pre-specification for prior-data conflict, and evaluate composed designs end-to-end. The formal version of what this series has been building toward.

Formal Bayesian Analysis: How It Actually Works

REBYOTA: Inside the First FDA-Approved Bayesian Analysis - Deep dive into the first FDA approval using Bayesian methods as the primary analysis. Shows what formal Bayesian regulatory submissions look like: dynamic borrowing, prior specification, and sensitivity analysis in practice.

Stop the Zombie Trial: The "Kill Switch" for Failed Experiments - How Bayesian Predictive Power (BPP) stops futile trials before they waste resources. Includes the hydroxychloroquine case study and why BPP beats Conditional Power for interim decision-making. Features a free calculator for running your own scenarios.

The COVID-19 Vaccine Trial That Put Bayesian Sequential Design on the Map - The BNT162b2 trial's primary analysis wasn't frequentist group sequential. It was a Bayesian posterior probability monitoring framework with thresholds calibrated to control Type I error. How it worked, why it mattered, and what Zhou and Ji's review of Bayesian sequential methods means for trial design going forward.

FDA Decisions Under Uncertainty

Qalsody: The Probability of Harm - How to interpret FDA's accelerated approval despite p = 0.97. When should we approve drugs under uncertainty in fatal diseases? A Bayesian lens on harm trade-offs, biomarker sufficiency, and decision-making when functional endpoints fail.

What FDA's Recent Rare Disease Approvals Teach Us About Single-Arm Trial Design - Between late 2024 and late 2025, FDA approved six rare-disease therapies supported primarily by single-arm trials. What separated success from rejection wasn't luck or regulatory leniency. It was understanding what evidence compensates for the absence of a control arm.

Error Asymmetry: How FDA Decides Which Mistakes Matter More - Approving an ineffective drug and rejecting an effective one are both errors, but they don't carry the same weight. Aduhelm shows how error-asymmetry judgments fail. Oncology accelerated approvals show how they succeed. The difference is whether the conditions that justify asymmetric weighting are actually present.

When Tumor Shrinkage Doesn't Mean Living Longer - Thirty years of accelerated approvals show a 61% conversion rate and 19% withdrawal rate. What separates surrogates that predict survival from surrogates that don't, and what trial designers can see before the first patient is enrolled.

When Things Go Wrong: Critical Examinations

The BATTLE Trial: When Adaptive Designs Fail - Critical examination of adaptive randomization in practice. Why methodological sophistication doesn't translate to better trials when assumptions about homogeneity and platform stability break down.

The Post-Hoc Problem: Why Bayesian Pre-Specification Matters More Than the Philosophy - When sponsors pivot from failed functional endpoints to successful biomarkers after unblinding, how do we know if the pivot is legitimate? Pre-specified Bayesian frameworks (priors on surrogacy, loss functions, decision thresholds) create accountability that post-hoc narrative construction never can.

The Square Peg Problem: Why FDA's Bayesian-Frequentist Truce Still Hurts - When FDA requires Bayesian methods to be calibrated to frequentist Type I error, philosophical tension becomes operational friction. The bronchial thermoplasty case study shows what happens when calibration backfires: a trial with 96% posterior probability fails because of an interim analysis that never occurred.

These posts complement the Bayesian series with practical frameworks for trial design:

Reading Paths

New to Bayesian regulatory science?
Start: FDA's Bayesian Guidance → Calibrated Bayes → REBYOTA → Qalsody → My FDA Comments

Designing a Bayesian trial?
Start: Calibrated Bayes → REBYOTA → Zombie Trial → Post-Hoc Problem → Square Peg Problem → My FDA Comments

Interested in FDA decision-making?
Start: Qalsody → Rare Disease Approvals → Error Asymmetry → Post-Hoc Problem → FDA's Bayesian Guidance

Want critical perspectives on when methods fail?
Start: BATTLE → Square Peg Problem → Post-Hoc Problem → 50:50 Randomization

Managing resource allocation in trials?
Start: Zombie Trial → BATTLE → 50:50 Randomization

Understanding the philosophical tensions?
Start: Square Peg Problem → Calibrated Bayes → Post-Hoc Problem

Interested in trial design fundamentals?
Start: Randomization Scheme [Coming soon] → What Randomization Can't Fix [Coming soon] → 50:50 Randomization → Group Sequential Design

Have questions or topics you'd like to see covered? Contact me or comment on any post.