Testing for Gene-Environment Interactions Under Environment Misspecification

• Activity Number: 540
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ryan Sun* and Xihong Lin
• Companies: Harvard University and Harvard School of Public Health
• Keywords: Gene-environment interaction ; Genome-wide ; Model misspecification

Abstract

Complex relationships between genetic and environmental risk factors characterize the etiology of many diseases, but there exist very few validated gene-environment interaction (GxE) results in literature. The standard GxE study repeatedly fits a single-marker generalized linear model which contains a genetic, environmental, and interaction term in the linear predictor. This model is often misspecified because we generally do not know the correct functional form of the environment term in the true data-generating mechanism. We study the impact of misspecification on inference for the interaction effect by investigating the asymptotic bias and variance of estimates from the misspecified working model. We assume the link function to be identity for continuous outcomes and logistic for binary outcomes. When naïve model-based variance estimates are incorrect, we provide an alternative testing procedure that has better small sample properties than the standard model-robust sandwich estimator. Performance of our methods is demonstrated through simulation studies and analysis of a GxE investigation for effect of metal exposures on neurodevelopment outcomes.