Mendelian randomization (MR) is a method by which genetic variants are leveraged as instrumental variables (IV) to investigate causal relationships between an modifiable exposure and a clinically relevant outcome. The application of standard IV methods in MR analysis faces two main challenges. First, most genetic variants are not relevant for studying a particular exposure of interest. Second, not all relevant genetic variants may serve as valid instruments due to pleiotropy, a phenomenon that one genetic variant may influence many seemingly unrelated traits. In this talk, we describe a data-driven method for Mendelian randomization that addresses these two challenges simultaneously. A key component of our proposal is a novel resampling method that construct pseudo variables to address selection bias. Such selection bias commonly arises from the application of standard MR methods to an ultra-high dimensional genetic dataset. Theoretical and synthetic data analyses show that the proposed method performs favorably compared to existing MR methods. We illustrate our approach through estimating the causal effect of obesity on health-related quality of life using data from the Wisconsin Longitudinal Study. Event in person (M1020 SPHII) and online at: https://umich.zoom.us/j/97795640076 Meeting ID: 977 9564 0076 Passcode: 123330
Mandi Larson, larsoma@umich.eduFighting Noise with Noise: Mendelian Randomization Studies with Pseudo Variables
Biostatistics Seminar with Linbo Wang, PhD Assistant Professor of Statistics University of Toronto
November 4, 2021
3:30 pm - 4:30 pm
M1020
Contact Information: Mandi Larson, larsoma@umich.edu
Mendelian randomization (MR) is a method by which genetic variants are leveraged as instrumental variables (IV) to investigate causal relationships between an modifiable exposure and a clinically relevant outcome. The application of standard IV methods in MR analysis faces two main challenges. First, most genetic variants are not relevant for studying a particular exposure of interest. Second, not all relevant genetic variants may serve as valid instruments due to pleiotropy, a phenomenon that one genetic variant may influence many seemingly unrelated traits. In this talk, we describe a data-driven method for Mendelian randomization that addresses these two challenges simultaneously. A key component of our proposal is a novel resampling method that construct pseudo variables to address selection bias. Such selection bias commonly arises from the application of standard MR methods to an ultra-high dimensional genetic dataset. Theoretical and synthetic data analyses show that the proposed method performs favorably compared to existing MR methods. We illustrate our approach through estimating the causal effect of obesity on health-related quality of life using data from the Wisconsin Longitudinal Study. Event in person (M1020 SPHII) and online at: https://umich.zoom.us/j/97795640076 Meeting ID: 977 9564 0076 Passcode: 123330