一种基于折扣因子D的贝叶斯方法在MRCT中的应用研究

The Application of a New Bayes Approach Based on Discounting Factor D in Multi-regional Clinical Trials

目的在折扣因子D的基础上,提出一种新的贝叶斯方法用于评价多区域临床试验中目标区药物的疗效,并探讨本方法的可行性。方法以Ⅱ期临床试验中收集到目标药物的种族信息作为样本信息,类似药物或同类药物的临床试验收集到的种族信息作为先验信息,构建折扣因子D的后验分布,并进一步计算加权Z检验统计量Z_W的后验分布,比较先验分布分别为无信息先验、共轭先验和分层先验时D后验分布的特点,并比较不同类型D的后验分布对试验检验效能的影响。结果当历史信息的信息量相对样本信息很小时,则后验均值主要由样本信息决定,后验分布的信息量基本接近样本信息的量,当历史信息的量逐渐增大时,后验均值逐渐向历史信息均值靠拢,后验分布的信息量也逐渐增大。检验效能由D的后验均值决定,与D的变异程度无关。结论本研究提出的贝叶斯方法可以较好地模拟实际情况,具有良好的实际意义和可操作性。

Objective To explore the feasibility of a new Bayes approach based on the discounting factor in multi-regional clinical trials.Methods Taking the racial information collected in phaseⅡclinical trials as sample information and the racial information collected in clinical trials of similar drugs as prior information,the posterior distribution of the discounting factor D,and the posterior distribution of the weighted Z test statistic Z W were constructed.We compared the posterior distributions of D with non-information prior,conjugated prior and hierarchical prior respectively,and the Power with different posterior distributions of D.Results When little historical information was collected relative to the sample information,the posterior mean of D was mainly determined by the sample information and the quantity of posterior information was close to the sample information.The posterior mean of D approached to the mean of historical information and the quantity of information of posterior distribution increased with the increase of the quantity of historical information.Power was determined by the posterior mean of D and was not affected by the variation of D.Conclusion The Bayesian method proposed in this study could well simulate the actual situation and has good practical significance and operability.