Adjustment for Population Stratification in Sequencing Association Studies and Model Averaged Matching Estimator

Abstract

In chapter 1, we develop a Markov random field-embedded linear mixed model to correct for population stratification induced by a combination of sharp spatial phenotypic distribution and geographically localized rare genetic variants in sequencing association studies. We then derive variant-set association tests under the proposed model, and illustrate the effectiveness of the method using both simulated genotypes and real genotypes from the 1000 Genomes Project. In chapter 2, we continue the investigation of population stratification in sequencing association studies due to nonlinear spatial variation in the mean of the phenotype. We propose to model such nonlinear spatial variation using a natural thin plate spline constructed from the top two principal components, and prove that such a spline can be incorporated in a linear mixed model as a combination of fixed and random effects. Variant-set association tests are developed under the spline-embedded linear mixed model. We then demonstrate the effectiveness of our method via simulations, and illustrate its application using sequencing data from the UK10K study. In chapter 3, we consider the problem of accurately estimating the effect of a treatment, intervention, or exposure on an outcome of interest in observational studies with a large number of potential confounders. We devise a Bayesian model averaged matching estimator to perform confounding adjustment in such studies, which simultaneously accounts for confounder selection uncertainty and obviates model dependence. We examine the effectiveness of our method in a simulation study, and illustrate its application in the assessment of comparative efficacy and safety of two anticoagulant medications using patient data extracted from two health insurance databases with nationwide coverage.