A Bayesian approach to Mendelian randomization using summary statistics in the univariable and multivariable settings...

Area of Science:

• Genetic Epidemiology
• Statistical Genetics
• Causal Inference

Background:

• Mendelian randomization (MR) utilizes genetic variants as instrumental variables for causal inference.
• Genome-wide association studies (GWAS) provide numerous genetic variants for polygenic MR, increasing statistical power but also susceptibility to bias from invalid instruments.
• Correlated pleiotropy, violating the "instrument strength independent of direct effect" assumption, is a key challenge in MR.

Purpose of the Study:

• To propose a flexible Bayesian framework for Mendelian randomization that enables valid causal inference in general settings.
• To introduce MR-Horse and MVMR-Horse methods that can handle correlated and uncorrelated pleiotropy.
• To develop methods usable with summary statistics from genome-wide association studies (GWAS) without requiring individual-level data.

Main Methods:

• A novel Bayesian framework for Mendelian randomization (MR) was developed.
• The MR-Horse and MVMR-Horse methods were proposed, utilizing summary statistics from GWAS.
• The framework accommodates both correlated and uncorrelated pleiotropy.

Main Results:

• Simulation studies demonstrated that the proposed approach maintains Type I error rates below nominal levels, even in scenarios with high pleiotropy.
• Applied examples in univariable and multivariable settings, including those with very weak instruments, showcased the practical utility of the methods.
• The methods provide valid causal inference without access to individual-level data.

Conclusions:

• The proposed Bayesian framework and associated MR-Horse/MVMR-Horse methods offer a robust approach to Mendelian randomization.
• These methods effectively address bias from correlated pleiotropy and can be applied using readily available GWAS summary statistics.
• The framework facilitates more reliable causal inference in genetic epidemiology, even under challenging conditions like weak instruments and high pleiotropy.