Correcting for Residual Population Substructure in Family-Based Tests of Gene-Environment Interactions with Application to Gene-Virus Interactions in Non-Diabetic Nephropathy

• Activity Number: 668
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Jasmin Divers*+ and Barry I Freedman and Carl D Langefeld
• Companies: Wake Forest School of Medicine and Wake Forest School of Medicine and Wake Forest School of Medicine
• Keywords: GxE interaction ; Admixed population ; Family-based association tests ; G-estimation ; Confounding ; Type I error inflation

Abstract

Family-based (FB) association studies were designed to be robust against the confounding effect of population stratification and admixture. This approach has since been extended to study gene-environment (GxE) interaction assuming the robustness against the confounding effect still holds in this framework. However, it has recently been shown that FB GxE tests may still result in type I error inflation when population structure involves the exposure. Existing methods for addressing this confounding effect include the collection of additional exposure data from siblings who were not included in the original study, G-estimation and conditional likelihood estimation. We evaluated the performance of ancestry proportion estimates as a control variable in the context of a FB GxE interaction test when parental data is not available and ascertainment is based on a phenotypic variable evaluated in first degree relatives. Comparisons are based on simulated data, which we use to assess both the type I error and power using GEE and conditional likelihood estimation. The best performing method is then used to test for gene-by-virus interaction effects on measures of kidney function.