A method that includes a Lasso-style process for identifying and selecting invalid instrumental variables (IVs) (i.e., those single nucleotide polymorpisms (SNPs) where there may be an effect on the outcome independent of the exposure) and estimating causal effects, even without any prior knowledge about which IVs are valid.
For MR, this method can be used with one-sample MR using individual-level data to identify invalid IVs and estimate the causal effect of an exposure on an outcome. However, one key consideration is that the Lasso selection process may characterise valid IVs as invalid if the invalid IVs are strong in their association with the exposure and this is also dependent on the correlation structure of the IVs. Therefore, the method also recommends the use of a median-based estimator (e.g., the weighted median method), where less than 50% of the IVs are invalid. This then does not depend on the relative strength of the IVs or their correlation structure.
References
Other terms in 'Pleiotropy-robust one-sample MR methods':
- constrained IVs
- MR for gene-environment interactions (MR-GxE)
- MR using G-Estimation under No Interaction with Unmeasured Selection (MR-GENIUS)
- Over-identification tests
- Pleiotropy-robust MR (PRMR)
- Some Invalid Some Valid Instrumental Variable Estimation (SISVIVE)
- Two-stage hard thresholding (TSHT) with voting