What are the assumptions under Mendelian randomization?

[le_chang]

• Relevance (IV1): the instruments must be associated with the exposure risk factor. This means that the genetic variants are associated with exposure at a genome-wide significance level (i.e., 5*10^-8) and replicated in a few independent studies.
• Exclusion restriction (IV2): the only way the IVs influence the outcome is through the relevant exposure risk factor.
• Independence (IV3): the IVs must not associate with confounders.

References:

• Lawlor, Debbie A et al. Statistics in medicine vol. 27,8 (2008)
• Davies, Neil M et al. BMJ (Clinical research ed.) vol. 362 k601. 12 Jul. (2018)
• Hemani, Gibran et al. eLife vol. 7 e34408. 30 May. (2018)