论文部分内容阅读
目的利用观察性数据进行因果效应的推断一直是研究者关注的焦点。在药物流行病学领域,探索药物疗效或不良反应也需要因果推断。由van deer laan提出的目标最大似然估计(targeted maximum likelihood estimation,TMLE)被证明具有良好的特性,本文对该方法的原理及应用现状做一介绍。方法因果推断常用的方法包括逆概率加权法,G-算法(G-formulation)及一些双稳健的估计方法。通过文献检索,对TMLE法的原理、模型构建、效应估计、统计推断、性质及应用等方面进行回顾与综述,同时比较TMLE法与其他几种因果推断方法的异同。结果与其他因果推断方法相比,TMLE法具有一定的优势,然而其实现过程更复杂。结论 TMLE法是一种双稳健的因果推断方法,能产生目标参数的有效无偏估计,可为TMLE法在药品不良反应主动监测因果推断中的应用提供参考,以期进一步加强药品风险管理。
Purpose The use of observational data for causal inference has been the focus of researchers. In the field of pharmaceutical epidemiology, to explore the efficacy of drugs or adverse reactions also require causal inference. The target maximum likelihood estimation (TMLE) proposed by van deer laan has been proved to have good characteristics. This paper introduces the principle and application status of this method. Methods Methods commonly used in causal inference include inverse probability weighting, G-formulation and some bi-stable estimation methods. Through the literature search, this paper reviews and summarizes the principles, model construction, effect estimation, statistical inference, nature and application of TMLE method, and compares the similarities and differences between TMLE and other causal inference methods. Results Compared with other causal inference methods, TMLE has some advantages, however, its implementation process is more complicated. Conclusion TMLE is a bi-stable causal inference method, which can produce an effective unbiased estimation of target parameters. It can provide a reference for the application of TMLE in the active monitoring of causal inference of adverse drug reactions, with a view to further strengthening drug risk management.