临床试验中多个终点变量同时评价的多元logistic模型  

Multivariate Logistic Model of Multivariate Binomial Response in Clinical Trial

作　　者：于浩[1] 丁红[1] 赵杨[1] 苏炳华[2] 丁德云[3] 陈峰[1] 

机构地区：[1]南京医科大学流行病与卫生统计学系,210029 [2]上海交通大学医学院生物数学教研室,200025 [3]浙江大学附属第二医院,310000

出　　处：《中国卫生统计》2007年第3期251-254,共4页Chinese Journal of Health Statistics

基　　金：国家自然科学基金资助项目(39500127)

摘　　要：目的研究临床试验中多个终点变量的同时分析。方法采用多元logistic回归模型,通过对原始资料的格式作适当变换,构造一个虚拟水平,视结果变量为1水平上的观察单位,以患者作为2水平单位,建立2水平logistic模型,对试验组和对照组的疗效,以及患者的年龄,性别,观察指标的基线值,中心效应等协变量进行分析。结果多终点的多元logistic回归模型既可以对单个结果变量进行分析,还可以对多个结果变量进行同时分析,并在扣除组间差异、协变量的影响后,估计两个结果变量之间的相关性。当受试者的多个结果有部分缺失时,该估计方法仍然是有效的。结论多元logistic回归模型可以对多个终点变量进行同时分析。To analyze multiple binomial responses data in clinical trial. Methods A multilevel logistic model was formulated with the responses at pseudo level 1 and subject at level 2. The parameters in the model can be estimated by means of MLwiN software with rearranged data set. Results Using multivariate logistic model, multiple binomial responses can be analyzed simultaneity. The correlation between multiple responses can be estimated in the model. An important feature of the mukilevel models for multivariate responses is that each respondent dose not have to be measured on each response. The estimates will be unbiased and efficient when responses are missing at random. The method was illustrated with a clinical trial data. Conclusion Data with multiple binomial responses can be analyzed with multivariate logistic model.

关 键 词：主要终点变量 多元二分类变量 多元logistic模型 多水平模型 

分 类 号：R195[医药卫生—卫生统计学]