观察性研究中混杂因素控制的敏感性分析方法

Methods of Sensitive Analysis Controlling Confounding Factors in Observational Studies

随机对照试验是因果推断的金标准。然而由于药品上市前研究的局限性,需要在上市后进一步开展药品不良反应监测。上市后药品不良反应监测主要为基于真实世界的观察性研究,无法完全替代随机对照临床试验的最主要原因就是其控制混杂因素的效果不如随机对照试验。现在有很多方法可有效控制已知混杂因素,但未知混杂因素的控制仍是一大难点。目前比较流行的控制未知混杂因素的方法(如工具变量法、双重差分法等)适用条件过于苛刻,而敏感性分析法应用条件较为灵活,但使用相对复杂,并未得到广泛使用。本文介绍三类实用性较强的敏感性分析法,阐明了其原理及使用方法,以便研究者在观察性研究中适当应用。

Randomized controlled trials are golden standard of causal inference.However,they still have limitations,resulting in long-term post-marketing surveillance of adverse drug reactions.Instead of randomized controlled trials,observational studies based on real-world data are mostly used to conduct post-marketing surveillance.The major defect of observational study is the difficulty to control confounding factors,causing the irreplaceability of randomized controlled trials.Methods controlling known confounding factors are developed and proved to be effective but these controlling unknown confounding factors are difficult to be as effective as much.Prevalent sensitive analysis methods controlling unknown confounding factors(like instrument variable and difference-in-differences analysis)are rigorous in assumptions.In general,sensitive analysis is flexible in application but complicated in practice,preventing generalization of this method.In this article,we introduce three kinds of strong practicality sensitive analysis methods,illustrating their mechanism and usage,hoping to help researchers understand robustness of results in observational studies.