目的:阴性对照在观察性研究中被认为是减轻偏倚的重要工具。本范围审查的目的是总结目前的阴性对照方法(阴性对照暴露[NCE]和阴性对照结果[NCO])。
方法:我们搜索了PubMed,WebofScience,EMBASE和Cochrane图书馆(截至2023年3月9日)关于阴性对照方法的文章。两名评审员独立和一式两份选择了符合条件的研究,并收集了相关数据。我们报告了总数和百分比,并对方法进行了叙述总结。
结果:我们的综述共纳入了37篇相关方法学文章。这些出版物涵盖了NCE(n=11,29.8%),NCO(n=13,35.1%)或两者(n=13,35.1%),最专注于偏差检测(n=14,37.8%),偏倚校正(n=16,43.3%)和P值或置信区间(CI)校正(n=5,13.5%)。其余两篇文章(5.4%),一个讨论了偏差检测和P值orCI校准,另一个讨论了所有三个功能。对于偏差检测,NCE(NCO)和结果(暴露)感兴趣变量之间存在关联,这仅仅表明结果可能存在混杂偏差,选择偏差和/或信息偏差。对于偏差校正,然而,负控制法的算法需要更严格的假设,如秩保持,单调性和线性。
结论:阴性对照可以用于偏倚检测,P值orCI校准,和偏差校正,其中使用阴性对照的偏倚校正是方法学上研究最多的。当前已知的方法需要一些严格的假设来检测或消除偏差。需要更多的方法学研究来优化阴性对照的使用。
OBJECTIVE: Negative controls are considered an important tool to mitigate biases in observational studies. The aim of this scoping review was to summarize current methodologies of negative controls (both negative control exposure [NCE] and negative control outcome [NCO]).
METHODS: We searched PubMed, Web of Science, Embase, and Cochrane Library (up to March 9, 2023) for articles on methodologies of negative controls. Two reviewers selected eligible studies and collected relevant data independently and in duplicate. We reported total numbers and percentages, and summarized methodologies narratively.
RESULTS: A total of 37 relevant methodological articles were included in our review. These publications covered NCE (n = 11, 29.8%), NCO (n = 13, 35.1%), or both (n = 13, 35.1%), with most focused on bias detection (n = 14, 37.8%), bias correction (n = 16, 43.3%), and P value or confidence interval (CI) calibration (n = 5, 13.5%). For the two remaining articles (5.4%), one discussed bias detection and P value or CI calibration and the other covered all the three functions. For bias detection, the existence of an association between the NCE (NCO) and outcome (exposure) variables of interest simply indicates that results may suffer from confounding bias, selection bias and/or information bias. For bias correction, however, the algorithms of negative control methods need more stringent assumptions such as rank preservation, monotonicity, and linearity.
CONCLUSIONS: Negative controls can be leveraged for bias detection, P value or CI calibration, and bias correction, among which bias correction has been the most studied methodologically. The current available methods need some stringent assumptions to detect or remove bias. More methodological research is needed to optimize the use of negative controls.