Bayesian mendelian randomization with study heterogeneity and data partitioning for large studies.

Area of Science:

• Epidemiology
• Statistical Genetics
• Biostatistics

Background:

• Mendelian randomization (MR) is crucial for causal inference in observational studies when randomized controlled trials are not feasible.
• Study heterogeneity is often overlooked in MR, and Bayesian MR methods can be computationally prohibitive for large datasets.

Purpose of the Study:

• To address study heterogeneity in Mendelian randomization by proposing a random effect Bayesian MR model.
• To overcome computational challenges in large studies using a subset posterior aggregation method.

Main Methods:

• Developed a random effect Bayesian MR model accommodating multiple exposures and outcomes.
• Implemented a subset posterior aggregation technique by dividing data into subsets and combining estimated causal effects.
• Evaluated the method through simulations with partly missing exposure data.

Main Results:

• Random effect Bayesian MR demonstrated superior performance compared to conventional inverse-variance weighted estimation.
• Data partitioning for large studies minimally impacted estimate variation but affected unbiasedness with weak instruments and high missing data rates.
• The 'divide and combine' approach enhanced computational efficiency with acceptable bias for large subsets.

Conclusions:

• The developed Bayesian MR method explicitly accounts for study heterogeneity and eases computational burden for large studies.
• The approach offers modeling flexibility for integrating heterogeneous studies and improving computational practicality in MR.
• This work highlights the utility of Bayesian MR with subset posterior aggregation for robust causal inference.