Penalized models for analysis of multiple mediators.

Area of Science:

• Statistics
• Genetics
• Psychology

Background:

• Mediation analysis assesses intermediate variables linking exposures to outcomes.
• Traditional methods often analyze one mediator at a time.
• Structural equation models (SEMs) can analyze multiple mediators but may not handle complex correlations efficiently.

Purpose of the Study:

• To develop an advanced statistical method for analyzing multiple mediators simultaneously.
• To extend the utility of SEMs for mediation analysis using penalized models.
• To identify significant mediators in complex biological and psychological pathways.

Main Methods:

• Developed a sparse group lasso penalized model for mediation analysis.
• The model incorporates parameter groupings and encourages sparsity for mediator selection.
• Utilized simulations to validate the method's performance and limitations.

Main Results:

• The novel method effectively evaluates numerous mediators and selects impactful ones.
• Application identified two novel DNA methylation loci mediating childhood trauma and cortisol stress.
• The findings highlight specific biological pathways influenced by early life stress.

Conclusions:

• The proposed sparse group lasso penalized SEM offers a powerful tool for multi-mediator analysis.
• This approach facilitates the discovery of novel mediators in complex associations.
• The developed R package 'regmed' makes these advanced methods accessible to researchers.