MR-SPLIT: A novel method to address selection and weak instrument bias in one-sample Mendelian randomization studies

Area of Science:

• Epidemiology
• Statistical Genetics

Background:

• Mendelian Randomization (MR) uses genetic variants as instrumental variables (IVs) to infer causality.
• Two-stage least squares (2SLS) is common in MR but vulnerable to bias from weak IVs and the winner's curse in one-sample analyses.

Purpose of the Study:

• To develop a novel method, MR-SPLIT, to mitigate bias from instrumental variable selection and weak instruments in one-sample MR.
• To improve the efficiency and robustness of causal effect estimation in MR analyses.

Main Methods:

• Introduced Mendelian Randomization with adaptive Sample-sPLitting with cross-fitting InstrumenTs (MR-SPLIT).
• Employed a 2SLS IV regression framework with adaptive sample-splitting and cross-fitting techniques.
• Utilized multiple sample-splitting for enhanced robustness.

Main Results:

• MR-SPLIT demonstrated superior performance compared to existing methods in simulation studies.
• The method effectively reduced bias, controlled Type I error rates, and increased statistical power.
• MR-SPLIT showed practical utility in a real-world data application.

Conclusions:

• MR-SPLIT offers a robust solution for addressing bias in one-sample MR analyses.
• The method is crucial for reliable causal inference when dealing with weak instrumental variables and potential selection bias.