The Use Of Bayesian Statistics In Multi-regional Clinical Trial

With the globalization of pharmaceutical economics development, multi-regional clinical trial (MRCT) has been used in clinical trials recently. As we see, MRCT may help relive the delay of new drugs on market and cost of clinical trials over spending and other issues. However, the impact of ethnic, cultural and local medical practice and other factors should be taken into consideration when evaluating the results for a specific region. And then, simultaneous global drug development program (SGDDP) was proposed. Nowadays, designs of MRCT and SGDDP focus on sample size calculation, regional consistency and statistical methods, which makes it more challenging.Bayesian methods were well used in MRCT, for the information of non-target ethnic (NTE) population could be regarded as prior of target ethnic (TE) population. And we can contral the impact of evaluation by adjusting prior information.The first part of the study was about Bayesian methods in MRCT. We made a modification of an evaluation rule by combining it with the method 1 in Ministry of Health, Labour and Welfare (MHLW). The three new rules were compared with the old one in assurance probability (AP) and false rate (FR). The simulation test was performed to study AP of rules when assuming efficacy and FR of rules when assuming no efficacy. The results showed that statistical properties of rules differred greatly. So we suggest the new rule to assess regional consistency in global clinical trials.The second part of the study focuses on Bayesian methods in SGDDP. Power prior was utilized in Bayesian approach. Three scenarios were set to consider the power and type I error of the rules with different combinations of low limit of Bayesian credible interval(cutoff), prior parameter(ao) and sample size of local clinical trial(n of LCT). The results showed that, keeping other factors, power increased with the increasing of cutoff or ao or n of LCT. And when higher efficacy was assumed in TE of MRCT, the speed of power to the highest was faster. The value of type I error was the same as that of cutoff when ao was zero, and kept as n of LCT changed. As ao increased, type I error also increased. When ao equalled 1, type I error was also nearly 1. To a conclusion, we can adjust the factors to attain higher power when type I error was well controlled.