Bayesian methods for meta-analysis of causal relationships estimated using genetic instrumental variables

Area of Science:

• Genetics
• Biostatistics
• Epidemiology

Background:

• Mendelian randomization (MR) uses genetic variants as instrumental variables to infer causal relationships.
• Existing MR methods often analyze single markers or studies, limiting scope and power.
• Integrating multiple markers and studies is crucial for robust causal inference.

Purpose of the Study:

• To extend existing Mendelian randomization methods for multiple genetic markers and multiple studies.
• To develop a flexible Bayesian hierarchical model for meta-analysis of individual participant data in MR.
• To provide a framework for estimating causal relationships and assessing heterogeneity across studies.

Main Methods:

• Reformulated single-marker MR as regression with heterogeneous error, using a Bayesian approach.
• Extended Bayesian methods to incorporate multiple genetic markers.
• Developed a hierarchical model for meta-analysis of MR studies with potentially different marker sets.

Main Results:

• Demonstrated a flexible Bayesian framework for multi-marker, multi-study MR.
• Successfully estimated the causal relationship between C-reactive protein and fibrinogen levels using data from 11 studies.
• Provided methods for assessing heterogeneity of causal effects across studies.

Conclusions:

• The proposed methods offer a flexible and efficient framework for causal inference in genetic epidemiology.
• The Bayesian hierarchical model accommodates varying genetic marker data across studies.
• These advancements improve the reliability of estimating causal effects from complex genetic data.