|
Activity Number:
|
76
- Contributed Poster Presentations: Section on Statistics in Epidemiology
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2020 : 10:00 AM to 2:00 PM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
|
Abstract #313171
|
|
|
Title:
|
Latent Mixture Model for Mendelian Randomization
|
|
Author(s):
|
Daniel Iong* and Yang Chen and Qingyuan Zhao
|
|
Companies:
|
University of Michigan and University of Michigan and University of Cambridge
|
|
Keywords:
|
mixture models;
mendelian randomization;
monte-carlo ;
EM algorithm;
causal inference;
epidemiology
|
|
Abstract:
|
Mendelian Randomization (MR) methods use genetic variants as instrumental variables to estimate causal effects of risk exposures on public health outcomes in epidemiology. Existing MR methods assume genetic variants that are valid instruments identify a common causal effect and interpret heterogeneity in causal effect estimates among individual genetic variants as evidence for violations of the instrumental variable assumptions. However, there has been recent evidence that heterogeneity may arise due to distinct biological mechanisms that affect the outcome through the risk exposure even when all of the genetic variants are valid instruments. In this paper, we will discuss a novel mixture modeling approach to identify groups of instruments that yield similar causal effects. We developed a Monte-Carlo EM algorithm to fit our proposed model and derived approximate confidence intervals for uncertainty quantification. We demonstrate our method on simulated data and on GWAS data to study the causal relationship between HDL cholesterol and coronary heart disease
|
Authors who are presenting talks have a * after their name.
