Séminaire de statistique: «Constrained Instruments and their Application to Mendelian Randomization with Pleiotropy»

by Lai Jiang (Postdoc, McGill)

Abstract: In Mendelian randomization, genetic variants are used to construct instrumental variables to estimate the causal effect of a phenotype of interest on a disease outcome. However, Pleiotropy occurs when a genetic variant influence the response though multiple phenotypes of interest, and therefore renders it an invalid instrumental variable in MR studies. We propose a novel Smoothed Constrained Instrumental Variable (CIV_smooth) method to construct valid instrumental variables while correcting for pleiotropic phenotypes. In a series of simulation studies, we show that our method leads to causal effect estimation with reduced bias compared with popular methods. Finally, we analyzed Alzheimer’s disease data from Alzheimer’s disease (AD) Neuroimaging Initiative (ADNI) study, and identified significant causal effect of Amyloid beta on AD progression using CIV_smooth method. However, in view of the limitations of this study, we understand further work are still needed to make substantive causal statements.