Bayesian causal graphical model for joint Mendelian randomization analysis of multiple exposures and outcomes

Area of Science:

• Epidemiology
• Statistical Genetics
• Causal Inference

Background:

• Current Mendelian randomization (MR) methods often simplify complex real-world relationships between multiple exposures and outcomes.
• Accurate estimation of causal effects requires methods that can model these intricate dependencies.

Purpose of the Study:

• To introduce MrDAG, a Bayesian causal graphical model for summary-level MR analysis.
• To detect and orient dependency relations among multiple exposures and outcomes for improved causal effect estimation.
• To apply the method to understand lifestyle and behavioral exposures impacting mental health.

Main Methods:

• MrDAG utilizes genetic variation as instrumental variables to address unobserved confounders.
• It employs structure learning to determine the directionality of dependencies within and between exposures and outcomes.
• Interventional calculus is used for principled causal effect estimation, assuming known directionality from exposures to outcomes.

Main Results:

• In simulations, MrDAG outperformed existing one-outcome-at-a-time and multi-response Bayesian MR methods.
• The method identified education and smoking as significant intervention points influencing mental health.
• A novel pathway between smoking, genetic liability to schizophrenia, and cognition was discovered.

Conclusions:

• MrDAG offers a robust framework for analyzing complex multi-exposure, multi-outcome causal relationships in MR.
• The findings highlight specific lifestyle factors and their downstream effects on mental health.
• The model facilitates the discovery of complex causal pathways previously unidentifiable with simpler methods.