Adaptive Designs for Clinical Trials with Highly Successful Treatments *

Abstract

We compare the performance of two adaptive designs and equal allocation in a clinical trial with two highly successful treatments and binary outcomes. The measure of interest in the trial is the odds ratio. The goal of the adaptive design is to decrease the total number of failures compared to equal allocation while keeping the power at the same level. One design is based on sequential maximum likelihood estimation, the other on an urn model. We find that the urn model produces a better procedure than the sequential maximum likelihood approach and equal allocation, in that it yields fewer expected treatment failures, maintains the power of the asymptotic test, and is more powerful when the Fisher's exact test is used. We conclude that adaptive designs have attractive properties when both treatments are highly successful.