Genetic Association Test Based on Nonparametric Stratification of Propensity Scores

• Activity Number: 360
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Yaji Xu*+ and Yuan Jiang and Chi Song and Heping Zhang
• Companies: Yale University and Oregon State University and Yale Univeristy and Yale University
• Keywords: Association Test ; Nonparametric Classification Tree ; Propensity Score Matching/Stratification ; Logistic Regression

Abstract

Propensity scores are usually estimated using parametric models such as logistic regression and then used as a matching or stratification criterion in an observational study. However, propensity score matching can be viewed as a classification problem and machine learning algorithms such as decision tree can be applied. In this work, we built a nonparametric tree model to achieve propensity score stratification for genetic association studies that may involve many confounding factors, such as population stratification and environmental variables, which either can be easily misinterpreted as genetic effects, or can weaken the genetic association signals. The leaves of the classification tree serve naturally the strata of propensity scores. Mantel-Haenszel test is then employed to detect the disease associations after the stratification procedure. Simulation studies show that the tree-based stratification method is more robust and powerful than the parametric approach especially when numbers of confounding covariates are presented and the minor allele frequency at the disease locus is low. We applied our method to Study of Addiction: Genetics and Environment (SAGE) data.