摘要
Abstract
In causal mediation analysis,the selection of covariates determines the accuracy of mediation effect estimation.However,prior studies addressing effect estimation in causal mediation analysis involving correlated covariates remained limited.To tackle this issue,this paper introduced the Generalized Outcome Adaptive Lasso(GOAL)method into the covariate screening process within causal mediation analysis.The GOAL method integrated the advantages of both adaptive weighting penalties and the elastic net,and utilized the minimization of the weighted absolute mean difference(wAMD)for parameter tuning.This approach more effectively handled the selection of correlated covariates,reduced bias,and enhanced estimation efficiency.Through multiple simulated scenarios and comparisons with the Outcome Adaptive Lasso(OAL)method,the superior performance of our proposed approach was verified,demonstrating its applicability in causal mediation analysis.The method was applied to the China Health and Retirement Longitudinal Study(CHARLS)dataset to evaluate the mediating role of depressive symptoms(DS)in the relationship between adverse childhood experiences(ACEs)and chronic lung diseases(CLDs).关键词
变量选择/GOAL方法/因果中介分析/最小化加权绝对均数差/不良童年经历/CHARLSKey words
variable selection/GOAL method/causal mediation analysis/weighted absolute mean difference/adverse childhood experiences/CHARLS分类
信息技术与安全科学
