Estimating the prevalence of transmitted HIV drug resistance using pooled samples. (April 2016)

• Record Type: Journal Article
• Authors: Finucane, Mariel M; Rowley, Christopher F; Paciorek, Christopher J; Essex, Max; Pagano, Marcello
• Is Part Of: Statistical methods in medical research. Volume 25:Number 2(2016:Apr.)
• Page Start: 917
• Page End: 935
• Publication Date: 2016-04
• Subjects: Pooling -- HIV -- primary drug resistance -- Bayesian modeling -- measurement error; Medicine -- Research -- Statistical methods -- Periodicals; Research -- Periodicals; Review Literature -- Periodicals; Statistics -- methods -- Periodicals; Médecine -- Recherche -- Méthodes statistiques -- Périodiques; 610.727
• DOI: 10.1177/0962280212473514
• Languages: English
• ISSNs: 0962-2802

Abstract

In many resource-poor countries, hiv -infected patients receive a standardized antiretroviral cocktail. In these settings, population-level surveillance of drug resistance is needed to characterize the prevalence of resistance mutations and to enable antiretroviral therapy programs to select the optimal regimen for their local population. The surveillance strategy currently recommended by the World Health Organization is prohibitively expensive in some settings and may not provide a sufficiently precise rendering of the emergence of drug resistance. By using a novel assay on pooled sera samples, we decrease surveillance costs while simultaneously increasing the accuracy of drug resistance prevalence estimates for an important mutation that impacts first-line antiretroviral therapy. We present a Bayesian model for pooled-testing data that garners more information from each resistance assay conducted, compared with individual testing. We expand on previous pooling methods to account for uncertainty about the population distribution of within-subject resistance levels. In addition, our model accounts for measurement error of the resistance assay, and this added uncertainty naturally propagates through the Bayesian model to our inference on the prevalence parameter. We conduct a simulation study that informs our pool size recommendations and that shows that this model renders the prevalence parameter identifiable in instances when an existing non-model-based estimator fails.