Robust multivariable Mendelian randomization based on constrained maximum likelihood.

Area of Science:

• Epidemiology
• Genetics
• Biostatistics

Background:

• Mendelian randomization (MR) uses genetic variants to infer causality from observational data.
• Multivariable MR (MVMR) enhances causal inference by accounting for pleiotropy and mediating exposures, surpassing univariable MR (UVMR).
• Existing MVMR methods lack robust theoretical properties and efficiency assessments.

Purpose of the Study:

• To develop an efficient and robust MVMR method with strong theoretical foundations.
• To evaluate the performance of the proposed method against existing MVMR techniques.
• To apply the method to investigate causal relationships between cardiometabolic risk factors and coronary artery disease (CAD).

Main Methods:

• Proposed a novel MVMR method based on constrained maximum likelihood (cML), termed MVMR-cML.
• Conducted extensive simulations to assess method performance and robustness.
• Applied MVMR-cML to large-scale genome-wide association study (GWAS) summary data for cardiometabolic risk factors and CAD.

Main Results:

• MVMR-cML demonstrated superior performance compared to existing MVMR methods in simulations.
• The method effectively addressed genetic pleiotropy and mediating effects.
• Identified direct causal effects of triglycerides, LDL cholesterol, and systolic blood pressure on CAD.
• Observed diminished effects for HDL cholesterol, diastolic blood pressure, and body height after accounting for other factors.

Conclusions:

• MVMR-cML provides a theoretically supported, efficient, and robust approach for MVMR analysis.
• The method enhances understanding of complex causal pathways in cardiometabolic disease.
• Confirms direct causal roles for specific lipid fractions and blood pressure in CAD etiology.