| The development of drug-resistant human immunodeficiency virus (HIV) is a hurdle to the long-term antiretroviral treatment of infected individuals. Resistant HIV is associated with high viral burden and diminished response to therapy, underscoring the need for statistical tools for interpreting genotypic resistance patterns, as well as for monitoring drug-resistant HIV within the infected population. The analysis of HIV genotype data is complicated by: (1) the high dimensionality of the measure, (2) the complex interactions between mutations at multiple residues, and (3) the viral heterogeneity within an infected individual. In this work, we present three papers which address these issues and allow for the modeling of HIV phenotype based on observations of genotype, as well as for the surveillance of resistance in an infected population.; In Paper 1, we model the relationship between genotype and phenotype via interpoint genotypic and phenotypic distance. Use of interpoint distances in regression models is hampered by their highly correlated and heteroscedastic nature. We show that the least squares estimators resulting from such models can be written as a new class of weighted U-statistic drawn from non-identically distributed data, We prove the asymptotic normality of these estimators and derive expressions for their variance. We apply the findings to an HIV dataset and demonstrate a positive linear relationship between mean phenotypic and genotypic distance.; Paper 2 presents a novel genotypic score for use in the prediction of phenotype. We extend methods developed for the biosurveillance of disease to the high-dimensional genotype space. The resulting score is a scalar measure of the dissimilarity between an observation of genotype and the genotypes in the drug-sensitive population. We show that the score is linearly related to phenotype with predictive value comparable to existing methods.; In the third paper, we examine the statistical considerations of using viral sequencing on pooled plasma samples for the estimation of the prevalence of drug-resistance in the HIV-infected population. We propose two likelihood-based approaches and show via simulation that sample pooling has the ability to obtain unbiased estimates of prevalence which are more precise than the same number of tests on individual subjects. |
