MR Dictionary

Homogeneity Assumption

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

The assumption is that either: - the association between the genetic IV and the exposure is homogeneous (i.e., is the same for everyone in the population) or - the effect of the exposure on the outcome is homogeneous. If this assumption holds then the IV estimate is consistent with the average causal effect (ACE) for the population to which inference is being made. Increasing biological knowledge of genetic variant functionality may provide evidence for some genetic instruments fulfilling this assumption. ‘Weaker’ assumptions (no effect modification and monotonicity) can alternatively be evoked. It is difficult in practice to know how important violation of these additional IV assumptions are in MR studies. Large GWAS collaborations increasingly combine results from many studies (though mostly from European original populations) and show consistency of association across these studies for variants defined as associated with the exposure at genome-wide significance (variants used in most MR studies), suggesting homogeneity may exist for several MR studies in European populations. Non-parametric methods that provide bounds of causal effect estimates requiring only the core IV assumptions may be applicable for some MR studies. Triangulation of results with other (non-MR) methods is likely to improve causal inference from MR studies. 

References

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