Guidelines for performing Mendelian randomization investigations: update for summer 2023

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 ten 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), extensions and additional analyses, 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 24 months.

Keywords: Mendelian randomization; causal inference; genetic epidemiology; guidelines.

Figure 1.. Flowchart highlighting some of the key analytic choices in performing a Mendelian randomization (MR) analysis.

Figure 2.. Generic analytic pipeline for Mendelian randomization (MR).

Figure 3.. Checklist of questions to consider when reviewing a Mendelian randomization investigation.

Figure 4.. Directed acyclic graphs illustrating validity and invalidity of instrumental variable assumptions in different scenarios.

a) Mediator is on causal pathway from exposure to outcome. b) Mediator is on causal pathway from genetic variants to exposure. c) Genetic variants influence the exposure, which has downstream effect on a related variable which does not affect the outcome. d) Genetic variants influence a related variable, and the related variable affects the outcome and exposure of interest. We note that the related variable may be known or unknown. e) Genetic variants influence the exposure and outcome via different causal pathways. f) Genetic variants influence the outcome primarily, and only influence the exposure via the outcome. In scenarios a, b, and c, as there is no alternative pathway from the genetic variants to the outcome, the instrumental variable assumptions are satisfied. In scenario d, the pathway from the genetic variants to the outcome does not pass via the exposure, and so the instrumental variable assumptions are not satisfied for the exposure (although they are satisfied for the related variable). Scenarios a, b, and c are examples of “vertical pleiotropy” (also called “indirect pleiotropy”) that do not invalidate the instrumental variable assumptions. Scenario d reflects a situation where the causal risk factor has been incorrectly identified – it is not the exposure, but the related variable. Scenario e reflects “horizontal pleiotropy” (also called “direct pleiotropy”) that violates the instrumental variable assumptions. Scenario f reflects a reverse causation situation where the genetic variant has been incorrectly identified as primarily affecting the exposure.

Figure 5.. Scatter plot of genetic associations with the outcome (vertical axis) against genetic associations with the exposure (horizontal axis).

Examples illustrated are: (left) no heterogeneity in the variant-specific causal estimates (effect of LDL-cholesterol on coronary heart disease risk using 8 variants associated with LDL-cholesterol); and (right) heterogeneity in the variant-specific causal estimates (effect of C-reactive protein on coronary heart disease risk using 17 genome-wide significant predictors of C-reactive protein). As indicated by differences in estimates, not all genetic variants are valid instrumental variables for C-reactive protein, and so a causal interpretation is not appropriate. Taken from Burgess et al., 2018 .