Variable Selection in Logistic Models for Testing SNP-SNP Interactions

• Activity Number: 245
• Type: Contributed
• Date/Time: Tuesday, July 31, 2007 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Huiyi Lin*+ and Renee Desmond and Yung-Hsin Liu and Seng-jaw Soong
• Companies: The University of Alabama at Birmingham and The University of Alabama at Birmingham and Pharmaceutical Product Development Inc. and The University of Alabama at Birmingham
• Keywords: single nucleotide polymorphism ; gene interaction ; logistic model ; variable selection

Abstract

Complex disease traits are associated with single nucleotide polymorphisms (SNPs) interactions. Logistic models are commonly used for variable selection in testing SNP-SNP interactions. No study has examined evidence-based variable selection for high order interactions in logistic models. This simulation study was designed to compare variable selection procedures in logistic models for testing SNP-SNP interactions. Data on 10 SNPs were simulated for 400 and 1000 subjects (case/control ratio=1). The simulated model included one main effect and two 2-way interactions. The variable selection procedures included automatic selection (stepwise, forward and backward), 2-step selection, AIC-based and BIC-based selection. The hierarchical rule effect was examined. We reported stepwise selection without hierarchical rule was preferred based on the percentages of true and false positive terms.