An exact method for link parameter estimation in error benchmarking: an application to Phase II two-stage single arm oncology trials

Bayesian trial designs rely on the specification of a vague data prior to the determination of trial outcome. Phase II two-stage single arm oncology trials have a generally small sample size which may not provide the necessary amount of data to outweigh a misspecified or overly optimistic prior. To account for this issue, clinicians may wish to benchmark Bayesian errors with frequentist measures of trial validity, Type I and Type II errors. A confidence prior, reflecting the clinician's confidence in the efficacy of the treatment before data accrual, is used to equate the errors across design strategies. An exact method, modelling both the inherent small sample size and multi-stage decision process of Phase II single arm trials, is presented for constructing the confidence prior.