Guidelines for performing Mendelian randomization investigations

Abstract

This paper provides guidelines for performing Mendelian randomization investigations. It is aimed at practitioners seeking to undertake analyses and write up their findings, and at journal editors and reviewers seeking to assess Mendelian randomization manuscripts. The guidelines are divided into nine sections: motivation and scope, data sources, choice of genetic variants, variant harmonization, primary analysis, supplementary and sensitivity analyses (one section on robust statistical methods and one on other approaches), data presentation, and interpretation. These guidelines will be updated based on feedback from the community and advances in the field. Updates will be made periodically as needed, and at least every 18 months.

Genome-wide association studies and Mendelian randomization analyses for leisure sedentary behaviours

“…Furthermore, we found evidence for heterogeneity and thus potential pleiotropy in the multivariable MR corrected for education, suggesting that unobserved confounders could play a role in the relationship between television watching and CAD. As far as verifiable using currently available methods 37 , all results point to the same direction and therefore seem to support the rationale that interventions targeting television watching may reduce CAD risk [12][13][14] , especially considering the high prevalence 2 and non-occupational characteristics of television watching. We Fig.…”

“…Several other sources of bias may be at play in the MR that should be acknowledged to correctly interpret the results. To address the assumptions of MR and control for different types of biases, we performed several sensitivity analyses according to the latest guidelines 37 , each with their own strengths and weaknesses that are more extensively described in the Supplementary Discussion. It is important to recognize these limitations and strengths in light of the potential complicated relationship between education and disease 38 .…”

Evaluating the relationship between circulating lipoprotein lipids and apolipoproteins with risk of coronary heart disease: A multivariable Mendelian randomisation analysis

“…Importantly, such univariable MR approaches that are more robust to horizontal pleiotropy notably do not, with a few exceptions [56,70], allow simultaneous statistical adjustment for multiple traits. The repertoire of robust univariable MR approaches [51] that seek to act as sensitivity analyses for potential unbalanced horizontal pleiotropy each make a different series of assumptions [53]. In the situation when horizontal pleiotropy is present in a dose-response manner (i.e., on average, SNPs that associate with higher levels of the exposure of interest also associate with higher/lower degrees of horizontal pleiotropy), this violates the 'InSIDE' assumption [55], and the MR analyses yield biased estimates.…”

“…In sensitivity analyses, we conducted univariable MR analyses robust to some forms of potential unbalanced horizontal pleiotropy [51] (horizontal pleiotropy being the process by which genetic variants used to instrument an exposure also associate with other traits that influence the outcome, a form of violation of the exclusion restriction assumption of instrumental variable analyses [52]) using weighted median [53], weighted mode [54], and MR-Egger [55] approaches. To further account for potential unbalanced horizontal pleiotropy in the multivariable MR analysis, we also conducted multivariable MR-Egger analyses [56,57].…”

Open Targets Genetics: An open approach to systematically prioritize causal variants and genes at all published human GWAS trait-associated loci

“…To demonstrate the value of colocalization evidence, we examined coding variants that were finemapped to single-variant resolution, and which colocalized with a molecular QTL for the same gene (729 variants, Supplementary Table 6). Such cis-variants make good genetic instruments for testing the causal effect of the molecular phenotype on disease 28 , and the ratio of coefficients for the cis-variants is an estimate of the effect size of the molecular phenotype on disease. Using this approach we identified several known gene-trait associations.…”