Estimation of total mediation effect for a binary trait in a case-control study for high-dimensional omics mediators^†

Area of Science:

• Biostatistics
• Genetics
• Epidemiology

Background:

• Mediation analysis identifies intermediate variables linking exposures to outcomes.
• Traditional methods struggle with high-dimensional mediators and binary outcomes, especially in case-control studies.
• Existing approaches may miss weak effects due to cancellation of opposing component effects.

Purpose of the Study:

• To propose a novel total mediation effect measure for high-dimensional mediators and binary outcomes.
• To provide a causal interpretation applicable to various mediation models.
• To address limitations of existing methods in capturing weak effects and handling binary outcomes.

Main Methods:

• Developed a new total mediation effect measure within the liability framework.
• Utilized a cross-fitted, modified Haseman-Elston regression for estimation.
• Tailored the procedure for case-control studies, adaptable to cohort studies.

Main Results:

• The proposed estimator is consistent, even with non-mediators and weak effect sizes, as shown in simulations.
• Demonstrated the method's efficacy in a real-world study.
• Found that 89% (CI: 73%-91%) of BMI-associated coronary heart disease liability variation is explained by metabolomics in the Women's Health Initiative.

Conclusions:

• The new liability-based mediation measure offers a robust causal interpretation for high-dimensional data.
• The developed estimation procedure is effective for case-control and cohort studies.
• Metabolomics plays a significant role in mediating the relationship between BMI and coronary heart disease liability.