| Relevance assumption |
The chosen genetic variant(s) must be robustly associated with the exposure of interest, typically at a genome-wide significance level (p<5x10-8). |
| Independence assumption |
The genetic variant must be independent of other factors that affect the outcome. In other words, there must be no confounding pathways between the variant and the outcome. |
| Exclusion restriction assumption |
The genetic variant must only affect the outcome through the exposure under investigation. |
| Horizontal pleiotropy |
When a genetic variant influences the outcome independently of the exposure under investigation, that is, violation of the exclusion restriction assumption. This assumption is the focus of most sensitivity analysis methods (e.g., MR-Egger, weighted median/mode). However, these methods should not be used as a replacement for careful selection of variants (or otherwise application of critical reasoning). |
| Vertical pleiotropy |
When a genetic variant is associated with multiple traits on the same causal pathway to the outcome. This chain of traits to the outcome is what MR sets out to estimate and not a source of bias. For example, a missense variant in the IL6R gene influences IL-6 receptor level/function which in turn affects C-reactive protein levels and rheumatoid arthritis. |
| F statistic |
The strength of association between the genetic variant and exposure, and an indicator of the relative (weak instrument) bias that is likely to occur in estimating the exposure-outcome association. |
| Two-sample MR |
MR analysis where gene-exposure and gene-outcome associations are derived from different non-overlapping samples from the same underlying population. Two-sample MR analyses are typically performed using summary level data (e.g., beta and standard error for each genetic variant). The ability to mix and match exposure and outcome genetic association data has made two-sample MR by far the most common method used in MR studies. |
| One-sample MR |
MR analysis using data on the genetic variants, exposure, and outcome from the same sample. One-sample MR analyses are typically performed using individual-level data. |
| Population stratification |
When there are subgroups in the population that have both different phenotypic distributions and different allele frequencies for genetic variants that might be used in MR. This can result in spurious associations between genotype and phenotype. |
| Assortative mating |
When individuals choose their partners non-randomly, for example, taller women tending to partner with taller men. |
| Dynastic effects |
Dynastic or intergenerational effects occur when parental genotype affects offspring outcomes causal pathways independent of the offspring phenotype. For example, when more educated parents support their children’s education. |
| Linkage disequilibrium |
Correlation (non-random association) between genetic variants due to their physical proximity on a chromosome. MR estimates from each genetic variant are pooled analogously to a meta-analysis of clinical trials; therefore, naïvely combining estimates from variants that are not independent (in linkage) is analogous to including a trial in the meta-analysis more than once. Independent genetic variants are often selected by LD ‘clumping’, whereby the most significant variant is selected to represent the locus. |
