临床试验中率差及其置信区间的估计方法  

Methods of estimating rate difference and associated confidence interval in clinical trials

作　　者：韩景静 曾新[1] 王骏[1] 

机构地区：[1]国家食品药品监督管理总局药品审评中心,北京100038

出　　处：《中国新药与临床杂志》2016年第4期255-259,共5页Chinese Journal of New Drugs and Clinical Remedies

摘　　要：目的探讨临床试验二分类数据率差及其置信区间的估计方法。方法分别介绍不存在协变量、只存在一个协变量、存在多个协变量三种情况时,率差及置信区间的估计方法,然后利用实例进行说明。结果当不存在协变量时,可用两组反应率估计值进行计算;当只存在一个分层因素时,针对不同情况可选择CMH、逆方差、最小风险加权方法进行估计;当存在多个协变量时,可结合logistic回归模型和delta方法、bootstrap方法计算。结论临床试验二分类数据率差及其置信区间的准确估计,需要结合试验设计,考虑是否需要校正协变量以及协变量的数量进而选择不同的统计方法。AIM To explore methods of estimating rate difference（ RD） and associated confidence interval（ CI） for binary data in clinical trials. METHODS The methods to estimate RD and its CI in three scenarios were introduced, with no covariate, one covariate, and more than one covariate in the statistical model, respectively. An example was provided for further illustration. RESULTS When there was no covariate,the estimated response rates of the two groups were used to calculate RD and its CI. When there was only one covariate, CMH, inverse of varianc e or minimum risk weighting methods was applied. The delta or bootstrap method combined with the logistic regression was used in case of more than one covariate. CONCLUSION It is necessary to choose appropriate method to calculate RD and its CI for binary data considering whether to adjust for covariates and the possible number of covariates according to the trial design in clinical trials.

关 键 词：临床试验 统计学 置信区间 

分 类 号：R195[医药卫生—卫生统计学]