High-dimensional mediation analysis for continuous outcome with confounders using overlap weighting method in...

Area of Science:

• Genomics and Bioinformatics
• Statistical Genetics
• Causal Inference

Background:

• High-dimensional mediation analysis (HDMA) is crucial for understanding complex biological pathways.
• Traditional methods struggle with confounding and extreme propensity score distributions.
• Existing HDMA methods often use propensity score (PS) adjustments, which can lead to biased estimates.

Purpose of the Study:

• To develop a robust HDMA procedure that effectively handles confounding.
• To improve the accuracy of mediator selection and effect estimation in high-dimensional data.
• To offer a more reliable alternative to PS-based methods in HDMA.

Main Methods:

• Integrated overlapping weighting (OW) for confounding adjustment within the HDMA framework.
• Employed sure independence screening (SIS) and de-biased Lasso penalization for variable selection.
• Utilized joint-significance testing with a mixture null distribution for robust analysis.

Main Results:

• The proposed OW-based HDMA procedure demonstrated superior performance in simulations.
• Achieved higher true positive rates and lower mean square error compared to PS-based methods.
• Identified specific methylation markers (cg13917614, cg16893868) mediating smoking's effect on NK cells in empirical data.

Conclusions:

• The novel HDMA method offers enhanced power, accurate mediation effect estimation, and effective false discovery rate control.
• The procedure is practical for use even when confounders are present.
• This OW-integrated HDMA approach is a valuable tool for genetic and epidemiological research.