Keywords: adaptive randomization, Bayesian design, clinical trial, estimation bias, group sequential design
In recent years, many outcome adaptive randomization (OAR) methods have been proposed and used to conduct comparative clinical trials, motivated by the ethical desire to give better treatments to more patients in the trial. OAR remains controversial, and some of its properties are not well understood by the clinical trials community. This talk presents results of a computer simulation study to evaluate properties of a 200-patient trial conducted using one of four Bayesian OAR methods, and compare them to a fairly randomized group sequential design. The simulations show that OAR has several very undesirable properties. These include a surprisingly high probability of a sample size imbalance in the wrong direction, with many more patients assigned to the inferior treatment arm, the opposite of the intended effect. Compared to a fairly randomized design, OAR also produces less reliable final inferences, including a greatly overestimated actual treatment effect difference. This bias increases if the prognosis of the accrued patients either improves or worsens systematically (drifts) during the trial. These problems decrease potential benefit to future patients, and to patients enrolled in the trial. When considering whether to use OAR, these problems should be weighed against its putative ethical benefit.