获得现实世界证据的因果推断统计学思考

医学研究中，时常观察到相关关系（association），但因果推断（causal inference）才是临床研究的最终目标。因果关系的判定标准包含关联的时序性、强度、可重复性、特异性、一致性、剂量反应关系、生物合理性以及实验证据8个方面。为了获得因果关系，临床研究设计与分析中蕴含了众多因果推断元素。本文解析混杂因素的存在对因果关系的影响，并针对随机分组、分析数据集及亚组分析3个重要问题，探讨其中的因果推断元素。医学工作者应当充分认识到临床研究中的因果要素，从而正确认识研究所能提供的证据等级，并在实际工作中产生高等级的医学证据。

Statistical consideration about causal inference to obtain real-world evidence

Association is often observed in medical research, but causal inference is the ultimate goal of clinical study. The criteria for determining causality include association temporality, strength, consistency, specificity, coherence, dose-response relationship, biologic plausibility and experimental evidence. In order to obtain causality, there are many causal inference elements in clinical study design and analysis. This study analyzes the influence of confounders on causality, and discusses the causal inference elements in three important topics:randomization, analysis of data sets and subgroup analysis. Medical researches should be fully aware of the causal elements in clinical study, so as to currently understand the level of evidence that can be provided by study, and try to produce high-level medical evidence in practice work.