A1103
Title: Debiased multivariable Mendelian randomization
Authors: Ting Ye - University of Pennsylvania (United States) [presenting]
Abstract: Mendelian randomization (MR) is a method that uses genetic variants as instrumental variables to infer the causal effect of a modifiable exposure on an outcome. Multivariable MR is an extension of standard MR by simultaneously studying multiple exposures. It has two key strengths: it is an effective way to account for horizontal pleiotropy as it can include traits on other pathways as additional exposures, and it can estimate the direct effect of each exposure on the outcome that is not mediated by the other exposures. However, robust multivariable MR faces major statistical and computational challenges. A robust and scalable multivariable MR method, MVMR-dIVW, is proposed which effectively removes the weak instrument bias of the popular multivariable inverse-variance weighted method and can account for balanced horizontal pleiotropy. In conclusion, the results are demonstrated in simulated and real datasets.
