CoxMDS: multiple data splitting for high-dimensional mediation analysis with survival outcomes in epigenome-wide...

Area of Science:

• Genetics and Bioinformatics
• Statistical Genomics
• Epidemiology

Background:

• Causal mediation analysis identifies intermediate variables (mediators) in exposure-outcome relationships.
• Existing high-dimensional mediation methods struggle with FDR control in finite samples, particularly with correlated or non-Gaussian data, common in DNA methylation studies.
• Reliable identification of causal mediators is crucial for understanding disease mechanisms and developing targeted interventions.

Purpose of the Study:

• To introduce CoxMDS, a novel multiple data splitting method for causal mediation analysis in high-dimensional settings.
• To ensure finite-sample false discovery rate (FDR) control, especially for correlated or non-Gaussian mediators in survival outcome analyses.
• To improve statistical power for identifying causal mediators compared to existing methods.

Main Methods:

• Developed CoxMDS, a multiple data splitting approach utilizing Cox proportional hazards models.
• Applied CoxMDS to simulated datasets to evaluate its performance in maintaining FDR control and power.
• Utilized CoxMDS for causal mediation analysis on DNA methylation data from The Cancer Genome Atlas (TCGA) and the Alzheimer's Disease Neuroimaging Initiative (ADNI).

Main Results:

• Simulations demonstrated that CoxMDS effectively controls FDR in finite samples and outperforms existing methods in statistical power.
• CoxMDS identified eight CpG sites in TCGA data suggesting DNA methylation mediates smoking's effect on lung cancer survival.
• Two CpG sites were identified in ADNI data, indicating DNA methylation may mediate smoking's effect on time to Alzheimer's disease conversion.

Conclusions:

• CoxMDS offers a robust solution for causal mediation analysis with high-dimensional, complex data, ensuring reliable FDR control.
• The method successfully identified potential DNA methylation mediators for smoking's impact on lung cancer and Alzheimer's disease survival.
• CoxMDS advances the field of genetic epidemiology by providing a powerful tool for dissecting complex biological pathways.