非实验性药物流行病学研究数据缺失的预防、检查和处理

Prevention,Examination and Treatment of Missing Data in Nonexperimental Pharmacoepidemiologic Studies

与精心开展的随机化临床试验相比,非实验性研究中的缺失数据会对有效性形成更大的威胁。然而,非实验性真实世界研究可以更加准确地描述医疗干预措施在实际情境中对不同的患者群体的作用,从而补偿这些限制条件。如果研究者认识到缺失一定数量的数据(无论关于暴露、结局还是混杂因素)不可避免,就应该在研究开始时针对缺失数据制定计划,尽可能防止缺失数据,同时为处理重要变量的缺失数据制定规划。如不能获得所有患者的全部数据元素,就必须认真检查和处理缺失数据。可采用多重填补等统计技术填补空缺。这些方法都需要尽可能了解导致数据缺失的因素的相关假设及其与研究结果的关联。此文描述了预防缺失数据的策略和处理非实验性真实世界研究中缺失数据的分析方法,并加入了例证说明。

In nonexperimental research,missing data pose an increased threat to validity compared with well-conducted randomized clinical trials. However,nonexperimental real-world studies can compensate for these limitations by providing a more accurate picture of how well medical interventions work in real-life settings and for diverse patient populations. Recognizing that some amount of missing data is inevitable,whether for the exposure,outcomes or confounding factors,it is best to plan for missing data at the start of a study by preventing missing data where possible and planning for handling missing data for important variables. In situations where all data elements are not available for all patients,careful examination and handling of missing data is required. Statistical techniques such as multiple imputation may be used to fill in gaps. These methods require understanding,to the extent possible,of assumptions concerning the factors leading to the missing values and how they relate to the study outcomes. Here we describe strategies to prevent missing data and analytic methods to handle missing data in nonexperimental real-world studies,and include an illustration.