The GENIUS Approach to Robust Mendelian Randomization Inference

Mendelian randomization (MR) is a popular instrumental variable (IV) approach, in which one or several genetic markers serve as IVs that can sometimes be leveraged to recover valid inferences about given exposure-outcome causal association subject unmeasured confounding. A key IV identification condition known the exclusion restriction states cannot have direct effect on outcome not mediated by exposure view. In MR studies, such an assumption requires unrealistic level of prior knowledge mechanism causally affect outcome. As result, possible violation seldom ruled out practice. To address this concern, we introduce new class estimators are robust under data generating mechanisms commonly assumed literature. The proposed approach named "MR G-Estimation No Interaction with Unmeasured Selection" (MR GENIUS) improves Robins' G-estimation making it both additive confounding and assumption. certain settings, GENIUS reduces estimator Lewbel (2012) widely used econometrics but appears largely unappreciated More generally, generalizes Lewbel's practical including multiplicative models for binary outcome, odds ratio models, case control study design censored survival outcomes.