Abstract

The aim of this study is to evaluate the use of Simple Bayesian Adaptive Randomization Design (SBARD) and predictive probability in clinical trial. The trial was planned to randomize 418 women with abnormal uterine bleeding to have pathologic evaluation by 2 devices. Total of 5,000 simulations were generated to evaluate the performance of SBARD and predictive probability under three different scenarios. The results from the SBARD were compared with the actual results. We found that the results of SBARD under the first scenario required 92 subjects in control arm and 61 subjects in treatment arm with total sample size of 253 (60% of actual total number) were similar to the actual results (135 subjects in control arm and 140 in treatment arm with total sample size of 275) . However, subjects are equally assigned in the real study, while SBARD assign higher subjects in control arm (92 subjects), indicating treatment might inferior than control. Moreover, if the trial is continued until the end of study, we can not conclude that treatment is better than control under all scenarios due to the predictive probabilities and frequentist p-value (p<0.001, p = 0.162) do not reach pre specific cutoff points too.