Meta-analytic Phase I Clinical Trial

A standard CRM prior is typically elicited from clinical judgment alone and is non-informative or weakly informative. The MAP prior is empirically derived by fitting a hierarchical meta-analytic model to actual dose-toxicity data from historical Phase I trials. It encodes not just a belief about the dose-toxicity curve shape but also the uncertainty arising from between-trial variability, making it both more data-driven and more transparent about where the prior information comes from.

There is no hard minimum, but the heterogeneity parameter τ is estimated with poor precision when fewer than three or four historical trials are available. With only one or two prior studies, the uncertainty in τ is so large that the MAP prior adds limited value over a standard weakly informative prior, and a robust mixture that down-weights historical data heavily may be more appropriate. Simulation studies are recommended to assess the operating characteristics under varying numbers of historical trials.

A robust mixture prior is a weighted combination of the MAP prior (derived from historical data) and a non-informative vague prior component. The mixture protects against prior-data conflict: if the new trial yields DLT rates that are inconsistent with historical experience, the posterior automatically down-weights the historical component and relies more on the new data. Without this safeguard, an overconfident MAP prior can anchor the model too strongly to historical rates even when current patients behave very differently.

Regulatory acceptance is growing but not uniform. The FDA's guidance on Bayesian adaptive designs and the ICH E20 guideline acknowledge meta-analytic prior borrowing as a valid strategy provided that the prior is pre-specified, sensitivity analyses are performed, and heterogeneity assumptions are transparent. Early engagement with regulatory agencies and pre-specification of the prior in the protocol or statistical analysis plan are essential.

Yes. Although the methodology was largely developed and popularized in oncology, the meta-analytic Phase I framework applies wherever dose-toxicity data exist from earlier studies — including rare diseases, vaccines, and biologics. The key requirement is that historical and current trials share a comparable dose-response mechanism and that heterogeneity in population and protocol can be characterized and modeled.