A latent outcome variable approach for Mendelian randomization using the stochastic expectation maximization...

Area of Science:

• Genetics
• Epidemiology
• Statistical genetics

Background:

• Mendelian randomization (MR) is crucial for inferring causality between exposures and outcomes.
• Existing MR methods face challenges with invalid instruments, leading to inflated type I errors and biased causal estimates.
• Pleiotropy, where genetic variants affect outcomes through multiple pathways, complicates MR analysis.

Purpose of the Study:

• To develop an enhanced MR method that explicitly disentangles horizontal and vertical pleiotropy.
• To improve the assessment of the exclusion restriction assumption in MR analyses.
• To provide a more precise and robust framework for causal inference using genetic data.

Main Methods:

• Augmenting latent phenotypes of the outcome to separate pleiotropic effects.
• Utilizing the expectation-maximization algorithm for iterative refinement of causal estimates.
• Evaluating performance across diverse simulation scenarios, including various pleiotropy types and Instrument Strength Independent of Direct Effect (InSIDE) assumption violations.

Main Results:

• The proposed method demonstrates superior control of type I error rates and reduced bias compared to established MR approaches.
• It effectively tests for directional horizontal pleiotropy, outperforming MR-Egger.
• The method shows robustness to genetic confounding and accurately identifies violations of the InSIDE assumption, performing well with both individual-level and summary data.

Conclusions:

• The novel MR method offers a more precise and reliable framework for causal inference, particularly in the presence of complex pleiotropy.
• It enhances the validity of MR studies by enabling explicit assessment of key assumptions.
• Application to BMI and metabolic syndrome (MetS) data confirmed its effectiveness, revealing fewer assumption violations than traditional methods, especially for composite MetS scores.