Bayesian Phase I Clinical Trial — Dose-Finding Design
Bayesian Phase I Clinical Trial (Dose-Finding Design) · Also known as: Bayesian dose-finding trial, CRM trial, continual reassessment method trial, Bayesian dose-escalation study
A Bayesian Phase I clinical trial uses prior probability models and sequential Bayes updating to find the maximum tolerated dose (MTD) of a new agent. Unlike the traditional 3+3 rule-based escalation, the Bayesian approach revises a dose-toxicity curve continuously as each patient's outcome is observed, allowing faster convergence to the true MTD while minimising exposure of patients to unsafe or subtherapeutic doses.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use a Bayesian Phase I design when the primary objective is to identify the MTD or recommended Phase II dose of a new drug, biologic, or combination regimen in oncology or other disease areas where toxicity is the dose-limiting factor. It is particularly valuable when the anticipated sample size is small (15–40 patients) and efficient use of each patient's outcome is critical. Prefer a Bayesian design over the traditional 3+3 when the goal is to estimate the MTD with quantified uncertainty rather than merely to satisfy a rule-based algorithm. Do not use when the dose-toxicity relationship is highly non-monotonic (e.g., immunotherapy with U-shaped toxicity curves) without model adaptation; the standard one-parameter CRM assumes monotone toxicity, and violations can lead to dangerous recommendations. Also avoid when regulatory requirements or institutional culture demand a rule-based escalation scheme, as Bayesian CRM requires more statistical infrastructure and real-time data access.
Strengths & limitations
- Targets the MTD directly via a probabilistic model rather than a rigid algorithm, improving estimation accuracy.
- Adaptively allocates patients to doses near the target toxicity rate, reducing unnecessary exposures at clearly safe or unsafe doses.
- Provides a quantified posterior distribution over the MTD, enabling transparent uncertainty communication to regulators and ethics boards.
- Sample size is used more efficiently than 3+3; simulations consistently show better MTD accuracy for the same or smaller N.
- Readily extended to handle multiple toxicity grades, delayed outcomes (TITE-CRM), or combinations of agents.
- Requires pre-specification of a prior model and target toxicity probability; model misspecification or a poorly calibrated prior can bias early escalation.
- Demands real-time data monitoring and statistical support throughout the trial, increasing operational complexity and cost.
- Regulatory familiarity is lower than for rule-based designs in some regions, requiring additional justification in submissions.
- The standard CRM assumes monotone dose-toxicity; it is not appropriate for agents with non-monotone or plateau toxicity profiles without model modification.
Frequently asked
How is a Bayesian Phase I trial different from the traditional 3+3 design?
The 3+3 design applies a fixed rule: if 0 of 3 patients experience a DLT, escalate; if 1 of 3, add 3 more; if 2 or more, de-escalate and declare the previous dose as MTD. No model is fitted and no uncertainty is quantified. The Bayesian CRM continuously fits a dose-toxicity curve and assigns each new cohort to the dose estimated to be closest to the target DLT rate. Simulation studies consistently show the CRM identifies the true MTD more accurately and treats more patients near it.
What is a 'prior skeleton' and how do I choose one?
The prior skeleton is the initial vector of expected DLT probabilities assigned to each dose level before any patients are treated — for example, (0.05, 0.10, 0.20, 0.35, 0.50). These values encode your prior belief about the dose-toxicity shape. Choosing the skeleton involves consulting pre-clinical data, pharmacokinetic modelling, and expert opinion. Running simulations under multiple skeleton scenarios and selecting the one that gives the best operating characteristics across plausible true curves is standard practice.
How large a sample is needed?
A typical Bayesian Phase I trial enrolls 20–40 patients, though sample size is determined by simulation rather than formula. Investigators run thousands of simulated trials under a range of assumed true dose-toxicity curves and assess the proportion of trials that correctly identify the MTD, the proportion that recommend an overly toxic dose, and average patient allocation at each level. The sample size is the minimum N at which these operating characteristics meet pre-specified thresholds.
Can Bayesian Phase I methods be used outside oncology?
Yes, though oncology is the dominant application because toxicity is the primary endpoint and the sample size constraint is acute. Bayesian dose-finding has been applied in vaccine development, anesthesiology, and early-phase psychiatric drug trials. The key requirement is a binary or ordinal endpoint that increases monotonically with dose; the statistical machinery transfers directly.
Are regulatory agencies (FDA, EMA) comfortable with Bayesian Phase I designs?
Both the FDA and EMA have issued guidance acknowledging adaptive and Bayesian designs in early-phase trials. FDA's 2019 guidance on adaptive designs and the EMA's reflection paper on adaptive designs both permit Bayesian dose-finding when the design is pre-specified, the operating characteristics are demonstrated via simulation, and the decision rules are transparent. Pre-IND or scientific advice meetings are recommended to align on the design before initiation.
Sources
- O'Quigley, J., Pepe, M., & Fisher, L. (1990). Continual reassessment method: a practical design for phase 1 clinical trials in cancer. Biometrics, 46(1), 33–48. DOI: 10.2307/2531628 ↗
- Chevret, S. (Ed.). (2006). Statistical Methods for Dose-Finding Experiments. Wiley. ISBN: 978-0470861769
How to cite this page
ScholarGate. (2026, June 3). Bayesian Phase I Clinical Trial (Dose-Finding Design). ScholarGate. https://scholargate.app/en/epidemiology/bayesian-phase-i-clinical-trial
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Adaptive Phase I Clinical TrialEpidemiology↔ compare
- Adaptive Randomized Clinical TrialEpidemiology↔ compare
- Bayesian Randomized Clinical TrialEpidemiology↔ compare
- Dose-Response AnalysisEpidemiology↔ compare
- Phase I Clinical TrialEpidemiology↔ compare