Heritable confounding of exposure and outcome in Mendelian randomization studies

Abstract Mendelian randomization (MR) leverages genetic variants to infer causal effects of exposures on outcomes, assuming these variants influence outcomes solely through the exposure. However, genetic instruments associated with heritable confounders of the exposure and outcome can violate this assumption, undermining gene-environment equivalence and biasing MR effect estimates. With increasing sample sizes in genome-wide association studies genetic instruments with smaller effect sizes are being identified as associated with a trait. Here we use simulations and an applied example estimating the effect of C-reactive protein on type 2 diabetes to demonstrate that variants with smaller effect sizes are more prone to heritable confounding, leading to biased causal estimates across common MR methods. This bias acts in the same direction as the confounded associations between the exposure and outcome observed in linear regression, but often with greater magnitude and acts in the same direction across a number of commonly used MR estimation methods, potentially leading to misleading confidence in the results. We show that incorporating known or suspected heritable confounders via multivariable MR or applying Steiger filtering can mitigate this bias, which is sometimes not seen with methods aimed at dealing with correlated pleiotropy. These findings highlight the importance of assessing and adjusting for heritable confounding in MR analyses to improve causal inference reliability.