Background: Phase IIa dose-screening stroke trials aim to screen candidate dose regimens for initial efficacy and safety before proceeding to definitive evaluation. This often requires simultaneous evaluation across complex endpoints including binary, ordinal, nested, and co-primary outcomes. The Bayesian Optimal Phase II (BOP2) design addresses this through a unified Dirichlet-multinomial model with adaptive posterior probability-based go/no-go stopping rules, but lacks a frequentist analogue that could provide explicit error rate guarantees.
Objective: To evaluate a frequentist confidence-adaptive analogue of BOP2 for Phase IIa dose-screening stroke trials, with explicit Type I error and power control.
Methods: Bayesian posterior probability stopping rules were replaced with confidence distribution tail areas, providing prior-free go/no-go decision criteria with frequentist coverage properties. Futility and safety rules are governed by one-sided confidence bounds calibrated by simulation to control Type I error rates. The design accommodates binary, ordinal, nested and co-primary endpoints under the same framework as BOP2. Performance was evaluated via the ADEMP (Aims, Data-generating mechanisms, Estimands, Methods, Performance measures) simulation framework.
Results: Preliminary simulation results demonstrate that the confidence distribution analogue achieves comparable power and Type I error control to BOP2, without requiring prior distributions. Comparative operating characteristics across endpoint types and scenarios will be presented.
Conclusions: A frequentist confidence distribution analogue of BOP2 offers stroke researchers a prior-free approach to Phase IIa dose-screening trials, with formal error rate control and compatibility with the complex endpoints common in stroke research.