Frequentist design for two-arm randomized Phase II clinical trials with outcomes from the exponential dispersion family was proposed previously, where the total sample sizes are minimized under multiple constraints on the standard errors of the estimated group means and their difference. This design was generalized from an approach specific for dichotomous outcomes. The two previous approaches measure the central tendency of each group and treatment effect based on mean and difference in means. Other measures such as median or hazard ratio are more appropriate under certain situations. In addition, the frequentist approaches assume that unknown parameters are fixed values. This does not reflect the reality that uncertainty always exists for unknowns. Compared to the frequentist methods, the Bayesian approach offers a flexible way to measure central tendency and treatment effect, and incorporate uncertainty in parameters of interest into considerations. In this article, we generalize a Bayesian design for Phase II clinical trials with endpoints in the exponential family from the two previously developed frequentist approaches. The proposed design minimizes the total sample sizes under pre-specified constraints on the expected length of posterior credible intervals for measures of treatment effect and central tendency in each group. The design is applicable for trials with fixed or optimal randomization allocation ratio and can be applied under adaptive procedure. Examples of method implementations are provided for different types of endpoints from the exponential family in both fixed and adaptive settings.
Keywords: Multiple constraints; natural conjugate prior family; posterior credible interval; sample size.