Overlapping-sample Mendelian randomisation with multiple exposures: a Bayesian approach.

Area of Science:

• Biostatistics
• Genetic Epidemiology
• Causal Inference

Background:

• Mendelian randomization (MR) is crucial for causal inference in medical research, utilizing genetic variants as instrumental variables (IVs).
• Traditional MR methods primarily focus on two-sample settings, overlooking one-sample or overlapping-sample scenarios which are common in practice.

Purpose of the Study:

• To develop a unified Bayesian method for Mendelian randomization (MR) applicable to one-sample, two-sample, and overlapping-sample settings.
• To enhance causal inference by addressing limitations of traditional MR methods in different sample overlap scenarios.

Main Methods:

• A novel Bayesian approach was proposed, converting two- or overlapping-sample MR to a one-sample MR with imputation of missing data using Markov chain Monte Carlo.
• The method was generalized to accommodate pleiotropy and multiple exposures.
• Performance was evaluated through simulations using metrics like mean, standard deviation, coverage, and power, and compared against classic MR methods.

Main Results:

• The proposed method demonstrated higher precision and power in estimating causal effects, particularly with increased sample overlap and instrument strength.
• Pleiotropy negatively impacted estimates, but overall model performance was robust across all sample settings with high coverage.
• Compared to classic MR, the Bayesian method yielded more precise estimates and consistently higher power when true causal effects were non-zero.

Conclusions:

• The developed Bayesian MR model offers significant flexibility, being applicable to all sample settings (one-, two-, and overlapping-sample).
• This represents a valuable advancement in MR methodology, extending beyond traditional one- or two-sample limitations.
• The Bayesian framework facilitates straightforward extensions for more complex causal inference analyses in medical research.