Skip to content

Mendelian Randomization

Summary

Mendelian Randomization (MR) is an epidemiological method that uses genetic variants as instrumental variables to estimate the causal effect of modifiable exposures on health outcomes from observational data. By leveraging the random assortment of alleles at conception, MR helps overcome confounding and reverse causation, mimicking the structure of a randomized controlled trial. However, its validity depends on strict instrumental variable assumptions that must be carefully evaluated using statistical and empirical triangulation.

Core Assumptions

For a genetic variant to serve as a valid instrumental variable, three core assumptions must be satisfied: 1. Relevance: The genetic instrument must be robustly associated with the exposure of interest. This is commonly evaluated using $F$-statistics to check for weak instrument bias. 2. Independence: The genetic instrument must share no uncontrolled common cause (confounding factors) with the outcome. This can be evaluated via within-family studies to control for population structure and dynastic effects. 3. Exclusion Restriction: The genetic instrument must affect the outcome solely through its effect on the exposure, and not via any alternative pathways. Violation of this assumption is known as horizontal pleiotropy.

Study Designs

  • Individual-level vs. Summary-level: Individual-level MR utilizes genotype, exposure, and outcome data from the same individuals, whereas summary-level MR uses GWAS summary statistics of the variant-exposure and variant-outcome associations. Summary-level analysis allows combining large datasets (e.g., from OpenGWAS) to dramatically increase sample sizes.
  • One-sample vs. Two-sample: One-sample MR evaluates genetic associations in a single cohort, whereas two-sample MR utilizes independent cohorts for the variant-exposure and variant-outcome estimations (often analyzed using packages like TwoSampleMR or MendelianRandomization).

Methodological Challenges

  • Horizontal Pleiotropy: The main threat to validity, which occurs when a variant affects the outcome directly. Methods like multivariable MR and outlier exclusion are used to address this.
  • Selection and Collider Bias: Arises in highly selected cohorts (e.g., UK Biobank volunteers) or disease-progression studies, and is addressed using inverse propensity weighting or g-estimation.
  • Temporal and Life-course Dynamics: MR estimates generally reflect the lifetime effect of a genetic predisposition, which may not align with the acute effect of a clinical intervention (e.g., drug treatment).
  • Evidence Triangulation: The integration of MR results with independent experimental designs (e.g., randomized controlled trials, animal models, observational studies) to establish robust causal claims.

Applications

Citations

  • Sanderson, E., Levin, M. G., Walker, V., Yuan, S., Badini, I., Dolce, J., Mahida, K. J., Nho, J. W., Pingault, J. B., Damrauer, S. M., Hemani, G., & Davies, N. M. (2026). Challenges and future directions for Mendelian randomization. Nature Genetics, 58(5), 984–994. DOI: 10.1038/s41588-026-02546-6. Source paper: s41588-026-02546-6.pdf