MR Dictionary

Modification of the inverse variance weighted (IVW) method, which aims to provide an unbiased causal effect estimate that is robust to many weak instrumental variables (IVs).

This method is a modification of the IVW method used in the two-sample MR setting with summary-level data, which aims to create an unbiased estimate that is robust to weak instrument bias. It does this by creating a bias correction factor and multiplying the obtained IVW estimate by this factor, which amplifies the obtained IVW estimate that is biased by weak IVs towards zero. Traditionally, MR studies will select IVs from a genome-wide association study (GWAS) of a single study or multiple studies as those genetic variants that reach a genome-wide significance p-value threshold (traditionally a p-value <5e-08). Some studies then choose to only include those IVs that meet the same (or prespecified) p-value threshold in a second and independent GWAS of a single study or multiple studies, as a screening method (sometimes called three-sample MR to select the "best" IVs. This debiased IVW method creates unbiased estimators even without performing this screening to select truly strong IVs.


Other terms in 'Weak instrument-robust two-sample MR methods':