Using the global randomization test as a Mendelian randomization falsification test for the exclusion restriction...

Area of Science:

• Genetics and Epidemiology
• Statistical Genetics
• Causal Inference

Background:

• Mendelian randomization (MR) relies on instrumental variables, assuming they affect the outcome only through the exposure.
• Violation of the exclusion restriction assumption can lead to biased causal estimates in MR studies.
• Identifying and correcting for such violations is crucial for reliable genetic epidemiology research.

Purpose of the Study:

• To introduce and evaluate a global randomization test as a falsification method for the exclusion restriction assumption in MR.
• To compare the statistical power of the global randomization test against existing methods for detecting selection bias and horizontal pleiotropy.
• To demonstrate the practical application of the global randomization test using real-world data.

Main Methods:

• Simulations were conducted to assess the power of the global randomization test in detecting selection bias and horizontal pleiotropy.
• The test's performance was compared against Bonferroni correction, effective number of covariates correction, and an r² permutation approach.
• Proof-of-principle analyses were performed using UK Biobank data, with C-reactive protein (CRP) as the exposure and coronary heart disease (CHD) as the outcome.

Main Results:

• The global randomization test showed higher power than other methods for detecting selection bias when covariate correlations were low.
• It demonstrated comparable or superior power across all simulated horizontal pleiotropy scenarios.
• Applied analyses revealed significant evidence of selection bias, identified 51 pleiotropic variants for CRP, and showed attenuated CRP-CHD effects after their exclusion.

Conclusions:

• The global randomization test is a valuable tool for MR researchers to falsify the exclusion restriction assumption.
• It effectively detects selection bias and horizontal pleiotropy, enhancing the validity of MR causal estimates.
• This method contributes to more robust causal inference in genetic epidemiology.