药物流行病学研究中的时依性变量处理方法简介及比较  

Introduction and Comparison of Time-dependent Variable Processing Methods in Pharmacoepidemiology Studies

作　　者：梁际洲 郭晓晶 许金芳 陈晨鑫 韦连慧 陈枭 郑轶 迟立杰 叶小飞 贺佳 Liang Jizhou;Guo Xiaojing;Xu Jinfang;Chen Chenxin;Wei Lianhui;Chen Xiao;Zheng Yi;Chi Lijie;Ye Xiaofei;He Jia(Department of Health Statistics,Faculty of Health Service,Naval Medical University of PLA,Shanghai 200433,China)

机构地区：[1]中国人民解放军海军军医大学卫勤系卫生统计学教研室,上海200433

出　　处：《药物流行病学杂志》2022年第3期190-197,共8页Chinese Journal of Pharmacoepidemiology

基　　金：国家自然科学基金项目(编号:82073671);上海市自然科学基金项目(编号:18ZR1449500);上海市第四轮公共卫生三年行动计划重点学科建设项目循证公共卫生与卫生经济学(编号:15GWZK0901);上海市卫计委优秀青年医学人才培养计划(编号:2018YQ47);上海市公共卫生学科带头人(编号:GWV-10.2-XD22);上海市公共卫生优青计划(编号:GWV-10.2-YQ33)。

摘　　要：基于纵向数据的因果推断,进而评价药物安全性与有效性是药物流行病学的重要工作之一。但在现实研究中由于存在不同程度的混杂,无法直接计算药物效应值。混杂中时依性混杂最为常见,却难以通过常规方式消除影响。除了混杂因素,在一些试验中暴露因素同样具有时依性。本文基于领域相关研究,试图对时依性变量的种类进行辨析,简要介绍含时依系数的Cox回归、边缘结构模型、结构嵌套的加速失效时间模型以及序列条件平均模型基本思想与计算方法,总结4种方法的优缺点与应用方向,以期为科研人员在分析中对时依性变量控制有所借鉴与启示。The evaluation of drug safety and efficacy based on causal inference from longitudinal data is one of the most important work in pharmacoepidemiology study.However,it is difficult to calculate the drug effect value directly because of the different degree of confounding in real research.Time-dependent variable is the most common in variable,but it is difficult to eliminate the influence by conventional means.In addition to confounding factors,exposure factors are also time dependent in some trials.In this paper,based on the relevant research in the field,we try to differentiate and analyze the kinds of time dependent variables,this paper briefly introduces the basic ideas and calculation methods of traditional time-dependent Cox regression,marginal structural model,structure nested accelerated failure time model and sequential conditional mean model,and summarizes the advantages,disadvantages and application direction of these four methods.It is hoped that it can be used for reference and enlightenment for the control of time dependent variables in studies.

关 键 词：时依性变量 COX回归 边缘结构模型 加速失效模型 序列条件平均模型 

分 类 号：R181.35[医药卫生—流行病学]