A0710
Title: Nonlinear Mendelian randomization
Authors: Jinzhu Jia - Peking University (China) [presenting]
Abstract: Using the Mendelian randomization (MR) approach to explore the causal relationship between exposure and outcome can effectively avoid confounding bias. However, most of the current MR methods are only suitable for cases where the effect of exposure on the outcome is linear. Two approaches to nonlinear MR are proposed, either by fitting a linear model of exposure-instrument and quadratic model of outcome-instrument (QC method) or by fitting a linear model of exposure-instrument and quadratic model of outcome-predicted exposure (two-stage method), the quadratic causality of exposure could be identified and estimated on the outcome. A series of simulations showed that the QC method and two-stage method had high power and low type I error. In real data applications, the QC method and two-stage method found that body mass index had a J-shaped effect on basal metabolic rate and had an inverted J-shaped effect on the level of high-density lipoprotein cholesterol in UK Biobank participants.
