| Genetic correlation |
This method estimates genetic correlation using GWAS summary statistics, using properties of linkage disequilibrium to allow for rapid screening for correlations among a diverse set of traits without the need for individual level data. However, this approach is still subject to genetic confounding (pleiotropy) and misclassification bias and requires larger samples than methods that use individual data. A well-powered null finding would argue against a causal association between exposure and outcome. However, direction of causation cannot be identified. |
| Polygenic risk score association |
Polygenic risk scores can be derived where there are multiple variants identified with genome-wide significance for a trait or disease. These can be weighted to represent the proportion of the variance in the risk factor that they explain, and used as a proxy for an exposure to investigate associations of interest. The use of a risk score allows for a larger proportion of the variance to be explained, although it is very likely it will increase the risk of pleiotropy. |
| Bidirectional Mendelian randomization with polygenic risk scores |
If polygenic risk scores are available for both the exposure and outcome of interest, associations can be investigated in both directions, which may provide evidence in support of an association in a particular causal direction. |
| Mendelian randomization sensitivity analyses |
Mendelian randomization Egger regression extends the basic Mendelian randomization method by meta-analysing the SNP outcome association from each individual SNP that is associated with the exposure. This treats each SNP as akin to a small study in a traditional meta-analysis. Regression analysis, allowing variation in the intercept, means it is able to provide an estimate of the extent to which genetic pleiotropy has an impact on the causal estimates from Mendelian randomization analyses. Kang median instrument analysis has been shown to identify causal effects as long as fewer than 50% of instruments are invalid, without requiring knowledge of which instruments are invalid. It also allows identification of when this 50% threshold is violated. |
