基于孟德尔随机化方法的纵向中介分析

Longitudinal mediation analysis based on Mendelian randomization

工具变量法广泛应用于中介分析,能够有效避免传统因果推断方法面临的难题,即由于未观测到的混淆因素和逆向因果造成的对因果效应的估计偏差.现有的工具变量方法大多服务于横断面研究,但纵向数据相较于截面数据能更好地反映因果路径。现有的文献中没有针对纵向中介分析的工具变量方法.为此本文开发了一种新的工具变量法用来估计纵向中介效应,同时建立了新方法的大样本性质,包括相合性和渐近正态性.另外,一系列模拟研究的结果展示了新方法的有限样本性质.

Instrumental variables(IVs)are widely used in mediation analysis to effectively reduce causal effect bias due to unobserved confounding factors and reverse causal direction that cannot be handled with conventional causal inference methods.Most IV methods in the literature are designed for cross-sectional studies.Longitudinal data can better reflect causal paths than cross-sectional data,which provides observations of individual patterns of changes and measurements of event duration.To our knowledge,there is no IV method specifically tailored for longitudinal mediation analysis in the literature.A new IV method is proposed to estimate longitudinal mediation effects.Large sample properties,including consistency and asymptotic normality,are established for the new IV method.Simulation studies are provided to demonstrate the desired finite sample properties of the new method.