Group-sequential response-adaptive designs for multi-armed trials

Abstract Several experimental treatments are often compared with a common control in a clinical trial nowadays. A group-sequential design incorporating response-adaptive randomization can help to increase the probability of receiving a more promising treatment for patients in the trial and to detect a treatment effect early so as to benefit the whole population of interest. With such ethical advantages, the trial design has invoked investigation using the Bayesian approach. In the frequentist approach, the type I error rate of a multi-armed trial may involve two error elements, the inflated error rates caused by multiple treatment comparisons and sequential testing. In this study, a group-sequential global test was considered. By monitoring the response-adaptive design at a continuous information time, calculation of the information time and two optimal response-adaptive sampling rules for multi-armed trials were described. Operating characteristics of the designs were investigated via simulation for censored exponential survival outcomes and using patient data sampled from a four-armed binary trial to demonstrate their practical applicability. Our results showed that, in general, the adaptive designs preserved ethical advantages in terms of reducing the average numbers of patients and failures compared with a group-sequential non-adaptive randomized design, while not adversely affecting the power.