倾向指数匹配法的模拟研究

Simulation study of propensity score matching method

目的 采用SAS程序模拟来研究倾向指数匹配法在处理非随机化试验数据中的效果.方法 利用蒙特-卡罗（Monte Carlo）模拟法产生有3个协变量（连续性变量X1,二分类变量X2,X3）的2组随机样本,以分组变量为因变量,以协变量为自变量建立logistic回归模型,并计算研究对象的倾向指数,然后按照倾向指数做组间无放回的卡钳匹配,得到一个各协变量均衡的处理组与对照组样本.用假设检验法和标准差异法分别评价匹配前后2组之间协变量的均衡性,并估计匹配前后2组间的处理效应.结果 假设检验法评价组间均衡性的结果为：匹配之前,协变量X1,X2,X3在2组间均有统计学差异,表明协变量X1,X2,X3在2组间不均衡;匹配之后,协变量X1,X2,X3在2组间均无统计学差异,表明协变量在2组间均衡.标准差异法评价组间均衡性的结果为：匹配之前,X1,X2,X3标准差异的均值分别为1 967.03%,117.29%,63.74%,均远远大于10%,表明协变量X1,X2,X3在处理组和对照组间都不均衡;匹配之后,X1,X2,X3标准差异的均值分别为19.46%,7.37%,6.85%,表明协变量X1,X2,X3在匹配后都基本变的均衡.可见使用基于倾向指数的卡钳匹配法对非随机化数据进行处理,协变量间不均衡的2个处理组在匹配以后达到了均衡.对处理效应的估计结果为：匹配之前,2组间的处理效应有统计学差异,但在匹配之后,2组间的处理效应变得没有统计学差异,表明匹配之前2组间的统计学差异是由协变量的不平衡引起的.结论 倾向指数法是一种有效的处理非随机化试验数据的方法,具有重要的应用价值.

Objective To evaluate the effect of propensity score matching method in analyzing non-randomized data with SAS program simulation. Methods To generate two groups of random samples of three covariates （continuous covariate, binary covariateand binary covariate） with Monte Carlo method. To estimate the propensity score with a logistic regression model in which the dependent variable is group variable and the independent variables are covariates and get a new sample with covariates balanced between treatment and control groups by using caliper matching without replacement according to the propensity score. With hypothesis test and standardized difference to evaluate the balancing of the covariates between two groups before and after matching respectively and estimate the treatment effect before and after matching. Results The result of hypothesis test evaluating the balancing of the covariates between two groups is ： there are significant differences between two groups before matching which indicates they are unbalanced between the two groups, arid there are no significant differences between two groups after matching which indicates they are balanced between the two groups. The result of standardized difference evaluating the balancing of the covariates between two groups is： before matching, the means of the X1 ,X2 ,X3 are 1 967.03%, 117.29% ,63.74% respectively, which are all greater than 10% and indicate that they are unbalanced between the two groups, and after matching, the means of the X1 ,X2 ,X3 are 19.46% ,7.37% ,7.37% respectively, which indicate they get balanced basically between the two groups. It is clear that analyzing the non-randomized data with caliper matching method based on propensity score can adjust the imbalance of covariates between two groups. The result of the treatment effect between two groups is ： there are significant differences before matching, but there are no significant differences after matching, which indicate the significant differences before matching is caused by the imbalance of the covariates. Conclusion Propensity score is an effective means to analyze the observational data and has important application value.