Guidelines for performing Mendelian randomization investigations: update for summer 2023

best practices
causal inference
genetic epidemiology
guidelines
mendelian randomization
statistical methods
  • Core Goal: This paper provides updated guidelines for Mendelian Randomization (MR) investigations, stressing the need for a rigorous, multi-method approach to inferring causality between an exposure and an outcome using genetic variants.
  • Validity Checks: The guidelines mandate systematic testing of the core IV assumptions, particularly the Exclusion Restriction (no pleiotropy), using multiple sensitivity analyses such as MR-Egger, Weighted Median/Mode, and Steiger Filtering.
  • Best Practices: Recommended practices for Two-Sample MR include proper variant selection (clumping), careful data harmonization, and comprehensive reporting of all sensitivity analysis results, including visual plots, to ensure robustness and transparency.
Published

23 January 2026

PubMed: 32760811 DOI: 10.12688/wellcomeopenres.15555.3 Overview generated by: Gemini 2.5 Flash, 28/11/2025

Key Findings and Purpose

This article provides an updated set of guidelines and best practices for conducting and reporting Mendelian Randomization (MR) studies. As MR methodology rapidly advances, this iteration addresses new statistical methods, common methodological challenges, and the need for standardized reporting. The core purpose is to help researchers rigorously apply MR to infer causality between an exposure and an outcome using genetic instruments.

The Core Assumptions and Validity Checks

MR relies on three key Instrumental Variable (IV) assumptions. The guidelines emphasize that MR studies must systematically test and address potential violations of these assumptions, which are often the main sources of bias:

  1. Relevance: The genetic instrument must be associated with the exposure. (Checked via F-statistic, strong instruments are required).
  2. Independence (No Confounding): The genetic instrument must not be associated with confounders of the exposure-outcome relationship. (Addressed by excluding associations with measured confounders and assuming sufficient randomization by meiosis).
  3. Exclusion Restriction (No Pleiotropy): The genetic instrument must affect the outcome only through the exposure. (Violated by pleiotropy, where a single variant affects multiple traits independently).

Falsification and Sensitivity Analyses

The guidelines strongly recommend using multiple sensitivity analyses to test the untestable Exclusion Restriction assumption, including:

  • MR-Egger: Used to detect and adjust for directional pleiotropy.
  • Weighted Median and Weighted Mode: Provide consistent causal estimates even if up to 50% of the information comes from invalid variants.
  • Cochran’s Q Statistic: Tests for heterogeneity, suggesting potential pleiotropy or violation of the “no confounding” assumption.
  • Steiger Filtering: Confirms that the genetic instrument influences the intended exposure more strongly than the outcome, testing the causal direction.

Two-Sample MR and Data Requirements

The guidelines detail best practices for Two-Sample MR (2SMR), the most common form of MR, which uses genetic summary statistics from two separate (but ancestrally matched) populations: one for the instrument-exposure association and one for the instrument-outcome association.

  • Data Selection: Genetic variants must be carefully selected to be robustly associated with the exposure (typically \(P < 5 \times 10^{-8}\)) and independent (clumped).
  • Harmonization: Data harmonization (aligning the effect allele and its effect size across the two datasets) is a critical step to ensure variant estimates are comparable.
  • Sample Overlap: The guidelines discuss the risk of bias due to sample overlap (when the two samples share individuals) and recommend specific correction methods (e.g., using robust standard errors) when overlap is unavoidable.

Reporting and Transparency

The paper stresses the importance of clear, comprehensive reporting to ensure reproducibility and critical appraisal:

  • Clarity on Assumptions: Researchers must explicitly state the assumptions being tested and which sensitivity analyses were used to address them.
  • Reporting Results: All sensitivity analysis results should be reported, not just the primary MR result. Visual tools, such as scatter plots, funnel plots, and forest plots, should be included to aid interpretation and illustrate heterogeneity/pleiotropy.

Conclusion and Future Directions

The updated guidelines reflect the growing sophistication of MR, emphasizing the move from simple, single-method MR to a multi-method approach where a range of sensitivity analyses are mandatory to demonstrate the robustness of causal findings. Future research is encouraged to develop methods for dealing with non-linear relationships, time-varying exposures, and the use of multi-omic data in MR frameworks.