摘要
目的探讨观察性研究中用于混杂偏倚控制的倾向性评分匹配、马氏距离匹配和遗传匹配三种方法的性能。方法针对连续型结局变量,设定混杂变量与处理分组变量之间具有不同复杂度的回归模型结构,采用Monte-Carlo模拟方法比较三种匹配方法在处理组间效应估计和匹配前后自变量均衡的区别,进而对三种方法性能进行评估。结果在给定的模拟情形下,相比于倾向性评分匹配和马氏距离匹配,遗传匹配法得出的效应估计偏差最小,匹配后两组自变量均衡性最好。结论遗传匹配在三种匹配方法中表现出较好的统计性能,可考虑作为观察性研究中控制混杂偏倚优先推荐的匹配方法。
Objective To evaluate the performance of controlling confounding bias in observational study using propensity score matching(PSM),Mahalanobis distance matching(MDM)and genetic matching(GM).Methods In our Monte-Carlo simulation,we focus on continuous outcome and conduct series of scenarios that differ in the degree of complexity in the true propensity score model.We compare the difference between the three methods in treatment effect estimation and covariate balance before and after matching.Results Our simulation results show that genetic matching has smallest bias to estimate treatment effect and highest balance of covariate compared with PSM and MDM.Conclusion Genetic matching shows better statistical performance among the three matching methods,which can be considered as a preferred recommended matching method to control confounding bias in non-randomized observational studies.
作者
陈文松
刘曼
刘玉秀
许敏怡
熊殷
巩浩雯
李维勤
Chen Wensong;Liu Man;Liu Yuxiu(Department of Biostatistics,School of Public Health,Nanjing Medical University(211166),Nanjing)
出处
《中国卫生统计》
CSCD
北大核心
2022年第3期322-328,共7页
Chinese Journal of Health Statistics
基金
国家自然科学基金面上项目(81473066)。
关键词
遗传匹配
倾向性评分
马氏距离
观察性研究
Genetic matching
Propensity score
Mahalanobis distance
Observational study
