Adaptive two-stage sampling in clinical trials

A class of two stage adaptive sampling schemes for comparing the success rates of two treatments in a clinical trial is defined. The goal is to minimize the expected number of patients assigned to the inferior treatment. An analytic approach to the computation of the probability of correct selection and of expected sample sizes is described. Through the use of Monte Carlo simulations, comparisons with alternative adaptive sampling schemes and with the standard sequential approach using vector at a time (VT) sampling are discussed. It is concluded that the adaptive plans discussed herein: (1) have some sample size advantages over other adaptive approaches although the differences are small; (2) are always inferior to VT sampling when success probabilities are less than 0.5; and, (3) can have a substantial advantage over VT sampling in that they require as many as 50% fewer patients on the inferior treatment when success probabilities exceed 0.5.