The use of parental genotypes in Mendelian randomization

Mendelian randomization method has been used in genetic epidemiological studies to estimate the causal effect of a biomarker on clinical outcomes. Genotypes are used as instrumental variables (IV) to remove the bias due to unmeasured confounding in observational data. A valid inference relies on the assumption that IV is not correlated with unmeasured confounders. While this assumption may be violated in many scenarios, it can be difficult to assess potential violations. To relax this assumption, we proposed a framework of 2-stage least square method stratified on parental genotypes when available. This method can effectively reduce confounding and provide an unbiased estimate of the causal effect, even when unmeasured confounders are correlated with the IV. We extended the method for scenarios when only partial parental genotypes are available. As a demonstration, we applied the method to a study of triglyceride response to drug and diet. The result was consistent with the known negative effect of high-density lipoprotein on insulin resistance.

Keywords

Mendelian randomization, instrumental variable, causal inference, 2-stage least square, parental genotypes, unmeasured confounding, population stratification

2010 Mathematics Subject Classification

Primary 62J12, 62Pxx. Secondary 92-xx.

Full Text (PDF format)

Received 2 December 2022

Accepted 1 April 2023

Published 19 July 2024