Objective: To systematically review the application of methods for controlling time-varying confounding in pharmaco-epidemiological studies. Methods: PubMed, Embase, CNKI, and Wanfang were searched for pharmaco-epidemiological studies involving time-varying confounding on June 15th, 2020. The basic characteristics, drug exposure and outcome, time-varying confounders and the application of methods to control these confounders were analyzed. Results: A total of 298 articles were included. An increasing trend was observed in numbers of studies dealing with time-varying confounding in pharmaco-epidemiological studies in recent years. A total of 106 (35.6%) studies involved the safety or effectiveness of medication use in HIV/AIDS patients and 92 of them involved antiretroviral drugs. The most common outcome was mortality, while the most commonly concerned time-dependent confounders were laboratory examination results (179, 60.1%), comorbidities (136, 45.6%), and co-used medications (108, 36.2%). Marginal structure model (MSM) and inverse probability of treatment weighting (IPTW) were the most commonly used methods to control time-varying confounding factors (244, 81.9%). Compared with the results after properly controlling time-varying confounding, traditional methods adjusting only baseline confounders resulted in substantial bias (median 18.2%, interquartile range, 7.4%-40.8%). As for basic assumptions needed for causal methods controlling time-varying confounding, 28.9% and 64.8% of the included studies examined or discussed the assumptions of positivity and no unmeasured confounders, respectively. Conclusions: At present, most of the fields of drug therapy for chronic diseases still pay insufficient attention to time-varying confoundings. Information collected in routine medical practice, such as laboratory tests, comorbidities, and co-used drugs, was the most commonly concerned time-varying confounder. MSM and IPTW were the most commonly applied methods for dealing with time-varying confounding.
目的： 了解药物流行病学研究中时依性混杂控制方法的应用情况。 方法： 系统检索英文数据库PubMed和Embase，中文数据库中国知网和万方数据知识服务平台，纳入发表于2020年6月15日以前且涉及具有中介作用的时依性混杂问题的药物流行病学研究，对其基本特征、药物暴露和结局、时依性混杂及其控制方法等问题进行分析。 结果： 共纳入298篇文献。涉及时依性混杂的药物流行病学研究在近年来明显增长，HIV/AIDS研究领域对该问题关注最多，占全部纳入研究的35.6%（106/298），且其中92个研究涉及抗反转录病毒药物。最常见的研究结局为死亡，而最常关注的时依性混杂因素为实验室检查指标（179，60.1%）、合并症（136，45.6%）和合并用药（108，36.2%）。边际结构模型（MSM）和逆治疗概率加权（IPTW）是最常用的控制时依性混杂的分析方法（244，81.9%）。与合理控制时依性混杂的分析相比，传统方法调整基线混杂引起偏倚的中位数为18.2%（IQR：7.4%~40.8%）。此外，分别有28.9%和64.8%的研究对因果推断的阳性假设和无未观测混杂假设进行了检验或讨论。 结论： 目前大多数慢性病药物治疗领域对时依性混杂关注仍然不足。实验室检查、合并症、合并用药等在常规医疗中容易获得的指标是最普遍考虑的时依性混杂。在控制方法上，MSM和IPTW等相对较简单、结果易理解的方法最常用。.