MR Dictionary

No effect modification assumption

This is one of a set of additional (to the core) MR assumptions that are required for estimating a well-defined causal parameter. 

If the genetic instrumental variable (IV) does not modify the effect of the exposure on the outcome within levels of the exposure and for all levels of the exposure, then this assumption holds. In that case, the IV estimate is consistent with the average causal effect (ACE), even if the effect of the exposure on the outcome is heterogeneous. When there is biological understanding, a gene-environment (G×E) interaction study can provide further evidence of causality. In general, MR studies using genetic variants with known biological function may provide evidence against effect modification being plausible. However, it is difficult in practice to know how important violation of these additional assumptions are in MR studies. Large genome-wide association study (GWAS) collaborations increasingly combine results from many studies (e.g., those across multiple European samples) and show consistency of association between genetic variants and traits across these studies at genome-wide significance (i.e., those variants used in most MR studies). This therefore suggests that homogeneity in effect estimates may exist for several MR studies in European populations and effect modification may not be likely. Non-parametric methods that provide bounds of causal effect estimates requiring only the core MR assumptions to be met may be applicable for some MR studies where violations of any of these additional assumptions is possible. See Homogeneity assumption.

References

Other terms in 'Sources of bias and limitations in MR':