有向无环图在混杂因素识别与控制中的应用及实例分析  

Application of directed acyclic graphs in identifying and controlling confounding bias

作　　者：刘慧鑫[1] 汪海波[2] 汪宁[3] Liu Huixin;Wang Haibo;Wang Ning(Peking University People’s Hospital,Beijing 100044,China;Peking University Clinical Research Institute,Beijing 100191,China;National Center for AIDS/STD Control and Prevention,Chinese Center for Disease Control and Preventionz,Beijing 100026,China)

机构地区：[1]北京大学人民医院,100044 [2]北京大学临床研究所,100191 [3]中国疾病预防控制中心性病艾滋病预防控制中心,北京100026

出　　处：《中华流行病学杂志》2020年第4期585-588,共4页Chinese Journal of Epidemiology

基　　金：国家自然科学基金(81602939)。

摘　　要：观察性研究是流行病学病因研究常用的研究设计,但应用观察性研究进行因果推断时,常由于未经识别、校正的混杂因素的存在,歪曲暴露因素与研究结局之间的真实因果关系。传统混杂因素判断标准在实际应用中不够直观,且有一定局限性,有时甚至出现混杂因素的误判。有向无环图(DAGs)可以直观识别观察性研究中存在的混杂因素,将复杂的因果关系可视化,判断研究中需要校正的最小校正子集,并可避免传统混杂因素判断标准的局限性,结合DAGs还可以指导混杂因素校正方法的选择,在观察性研究中因果推断具有重要指导价值,DAGs在未来的流行病学研究中将有更多的应用。Observational study has been viewed as the most convenient method in designing etiological studies.However,the presence of confounders always challenge the researchers in study design,since unadjusted confounders may lead to biased results.The traditional definition of a confounder is not intuitional in application and sometimes leading to inappropriate adjustment of nonexistent"confounders"which might induce new bias to merge.The use of directed acyclic graphs(DAGs)may identify confounders easier and more intuitional,as well as avoiding superfluous adjustment.It can also contribute to the identification of adjustment methods,and be useful in causal inference of observational studies.

关 键 词：因果推断 混杂因素 有向无环图 

分 类 号：O15[理学—数学]