Mendelian randomization incorporating uncertainty about pleiotropy.

Area of Science:

• Genetics
• Biostatistics
• Epidemiology

Background:

• Mendelian randomization (MR) relies on genetic instruments, but pleiotropy poses a significant challenge to justifying these assumptions.
• Existing MR methods (fixed effects, random effects, MR-Egger) make different assumptions about pleiotropy, risking misrepresentation of causal evidence if misapplied.

Purpose of the Study:

• To address uncertainty regarding the appropriate pleiotropy model in two-sample MR analyses.
• To evaluate the performance of Bayesian model averaging (BMA) when distinguishing between different pleiotropy models.

Main Methods:

• Utilized a Bayesian framework for MR analysis.
• Applied Bayesian model averaging (BMA) to account for uncertainty in pleiotropy models.
• Conducted simulations to assess the ability of large sample sizes to distinguish between pleiotropy models.

Main Results:

• Even large genome-wide meta-analysis sample sizes may be insufficient to definitively distinguish between pleiotropy models based on data alone.
• Simulations indicated that BMA provides a reasonable trade-off between bias and precision in MR analyses.
• BMA demonstrated effectiveness in handling uncertainty about the nature of pleiotropy.

Conclusions:

• Bayesian model averaging is recommended for Mendelian randomization studies when there is uncertainty about the underlying pleiotropy assumptions.
• BMA offers a robust strategy to balance bias and precision, enhancing the reliability of causal inference in MR.
• The choice of pleiotropy model significantly impacts MR results; BMA provides a principled way to navigate this uncertainty.