High-Dimensional Mediation Analysis With Confounders in Survival Models.

Area of Science:

• Biostatistics
• Environmental Health
• Genomics

Background:

• Mediation analysis is crucial for understanding environmental exposure mechanisms on health.
• Existing high-dimensional mediation models often overlook crucial confounding variables in observational studies.

Purpose of the Study:

• To develop a robust high-dimensional mediation analysis procedure that incorporates propensity score adjustment for confounders.
• To evaluate the performance of the proposed method in mediator selection and effect estimation.

Main Methods:

• Proposed a concise and efficient high-dimensional mediation analysis procedure utilizing propensity scores for confounder adjustment.
• Conducted simulation studies to compare the proposed method against approaches ignoring confounders.
• Applied the procedure to The Cancer Genome Atlas (TCGA) lung cancer dataset.

Main Results:

• The proposed procedure demonstrated superior performance in mediator selection and effect estimation compared to methods that disregarded confounders.
• As sample size increased, the method's performance in variable selection and mediation effect estimation matched that of models including all confounders.
• Identified specific methylation markers (cg21926276, cg20707991) mediating the increased risk of death in lung cancer patients with a history of heavy smoking.

Conclusions:

• The propensity score-adjusted high-dimensional mediation analysis offers an efficient and accurate approach for analyzing complex exposure-mediator-outcome relationships.
• This method provides a valuable tool for identifying key biological mediators in environmental health research, as exemplified by its application in lung cancer.