Docs/Bayesian Toolkit

Bayesian Toolkit Overview

FDA-grade Bayesian trial design in five integrated steps. The Bayesian Toolkit provides a complete workflow for designing clinical trials using Bayesian methodology, aligned with the FDA's January 2026 Bayesian guidance.

1. When to Use Bayesian Methods

Bayesian methods are particularly valuable in these scenarios:

1

You have relevant historical data

Phase II results, published trials, or registry data can be formally incorporated via informative priors—potentially reducing sample size by 20–40% while maintaining statistical rigor.

2

Sample sizes are constrained

Rare diseases, pediatric populations, and oncology basket trials often can't achieve traditional frequentist power. Bayesian borrowing provides a principled way to augment limited data.

3

Interim decisions matter

Predictive probability of success (PPoS) answers the question stakeholders actually care about: “Given what we've seen so far, how likely is this trial to succeed?”

4

Regulators expect it

The FDA's January 2026 guidance explicitly endorses Bayesian methods for pivotal drug and biologic trials—not just devices. The guidance cites REBYOTA as a successful example.

Bayesian vs. Frequentist: A Practical View

AspectFrequentistBayesian
Prior informationIgnored (or informal)Formally incorporated
Interim interpretationConditional power at assumed effectPPoS across posterior uncertainty
Result statement“p < 0.05” or “95% CI excludes null”“92% probability treatment is effective”
Regulatory statusStandardAccepted with documentation (FDA 2026)

Hybrid Designs

The toolkit supports hybrid designs—Bayesian priors and interim monitoring with frequentist final analysis—which is the FDA's recommended approach for most pivotal trials.

2. The 5-Step Workflow

The Bayesian Toolkit follows a sequential workflow. Each step produces outputs that feed into subsequent calculators.

1
PRIOR
2
BORROWING
(optional)
3/4
SAMPLE SIZE
5
POWER
(PPoS)

Step 1: Prior Elicitation

What it does: Translates clinical knowledge into Beta distribution parameters (α, β) for binary endpoints.

Three methods: Quantile matching, ESS-based, Historical data

Output: Beta(α, β) prior with documented ESS and justification

Prior Elicitation Documentation

Step 2: Bayesian Borrowing (Optional)

What it does: Formally incorporates external control data with appropriate discounting.

Three methods: Power prior (static δ), Commensurate prior, MAP prior

Output: Effective prior, prior-data conflict diagnostics, sample size comparison

Bayesian Borrowing Documentation

Step 3: Single-Arm Sample Size

What it does: Determines sample size for single-arm Bayesian trials with operating characteristics.

Key outputs: Recommended N, Type I error rate, power curve, sensitivity analysis

Bayesian Sample Size Documentation

Step 4: Two-Arm Design

What it does: Sizes randomized two-arm Bayesian trials (superiority or non-inferiority).

Design options: Superiority, Non-inferiority, Allocation ratios (1:1, 2:1, 3:1)

Two-Arm Bayesian Design Documentation

Step 5: Predictive Power (Interim PPoS)

What it does: Calculates probability of trial success given interim data.

Key outputs: PPoS with decision gauge, sensitivity analysis, posterior visualization

Bayesian Predictive Power Documentation

PPoS Decision Framework

PPoSRecommendation
≥ 90%Predicted Success — Verify current posterior meets significance
20–90%Continue — Insufficient evidence for early stopping
< 20%Stop for Futility — Very low probability of success

3. Which Calculators Do I Need?

Single-Arm Trial

Do you have historical data to incorporate?

YES → 1. Prior Elicitation → 2. Bayesian Borrowing → 3. Sample Size → 5. PPoS

NO → 1. Prior Elicitation → 3. Sample Size → 5. PPoS

Two-Arm Randomized Trial

Do you have historical control data?

YES → 1. Prior Elicitation → 2. Bayesian Borrowing → 4. Two-Arm Design → 5. PPoS

NO → 1. Prior Elicitation → 4. Two-Arm Design → 5. PPoS

Interim Monitoring Only

Trial already designed, need interim decision support?

1. Prior Elicitation → 5. PPoS (with interim data)

4. Common Design Patterns

Pattern A: Phase II Oncology (Single-Arm with Historical Borrowing)

Single-arm Phase II evaluating ORR against historical control rate.

StepCalculatorPurpose
1Prior ElicitationConvert Phase I/II data to Beta prior with δ = 0.5 discount
2Bayesian BorrowingQuantify sample size reduction from borrowing
3Sample SizeSize trial for 80% power, verify Type I ≤ 0.05
5PPoSMonitor at 50% enrollment for futility

Example: Historical ORR = 12%, target ORR = 25%, Phase II data: 24/200 responders

  • • Prior: Beta(13, 89) with ESS = 102 after 50% discount
  • • Sample size: ~85 patients (vs. ~110 without borrowing)
  • • Interim PPoS threshold: < 20% → stop for futility

Pattern B: Confirmatory RCT (Two-Arm Superiority)

Phase III RCT with expert-elicited prior, no historical borrowing.

StepCalculatorPurpose
1Prior ElicitationQuantile matching from expert opinion (skeptical)
4Two-Arm DesignSize for superiority with 1:1 allocation
5PPoSInterim futility monitoring at 50% information

FDA hybrid approach: Use Bayesian PPoS for interim decisions, frequentist test for final analysis.

Pattern C: Rare Disease (Maximal Borrowing)

Small population, strong historical data from natural history study.

StepCalculatorPurpose
1Prior ElicitationHistorical data method with minimal discount (δ = 0.8)
2Bayesian BorrowingMAP prior from multiple natural history cohorts
3Sample SizeAggressive sample size reduction justified by ESS
5PPoSContinuous monitoring given small N

Regulatory context: FDA Section IV.B.2 addresses non-calibrated designs for rare diseases where traditional operating characteristics aren't feasible.

5. Regulatory Alignment

All five calculators are designed to satisfy FDA January 2026 guidance requirements:

RequirementHow the Toolkit Addresses It
Prior justification (Section V.D)Prior Elicitation documents source, method, and ESS
Discounting rationale (Section V.D.4)Bayesian Borrowing provides power prior δ with justification
Operating characteristics (Section IV.A)Sample Size calculators report Type I error and power via simulation
Sensitivity analysisAll calculators support prior sensitivity across optimistic/skeptical scenarios
Decision thresholds (Section IV.A)PPoS calculator implements FDA's three approaches to success criteria

SAP Documentation Checklist

Each calculator generates outputs for your Statistical Analysis Plan:

  • Prior specification with α, β (or μ, σ²), source, and ESS
  • Discounting method and justification (if borrowing)
  • Decision rule: threshold γ, comparison metric, success criterion
  • Sample size with operating characteristics
  • Sensitivity analysis plan across prior specifications
  • Interim monitoring rules with PPoS thresholds

Ready to Start?

Begin with Prior Elicitation to define your prior, then follow the workflow.