Evaluating New Matrix Pooled Testing Methods for Detecting HIV Treatment Failure with and without Covariate Information

Thesis (Master's)--University of Washington, 2016-08

University of Washington Abstract Evaluating New Matrix Pooled Testing Methods for Detecting HIV Treatment Failure with and without Covariate Information Adam Brand Chair of the Supervisory Committee: Susanne May, Associate Professor Department of Biostatistics Antiretroviral treatment for HIV has proven to lengthen and improve patients’ lives, and greatly reduce the risk of transmission. Patients receiving antiretroviral treatment for HIV must be monitored for treatment failure, so the treatment regimen can be altered to maintain low HIV viral load levels. Monitoring/testing treated HIV patients requires resources not available to resource-limited regions most in need. It has been shown that matrix pooled testing can be efficient compared to individual testing under certain conditions however, further improvements are needed. The current best-performing matrix pooled testing method for detecting HIV treatment failure does not use covariate information. Other methods which do use covariate information have not yet been shown to outperform this method in settings with realistic, skewed viral load values. We propose new methods of matrix pooled testing, some of which use covariate information, and evaluate method performance with respect to relative efficiency, sensitivity and number of testing rounds. Based on simulation results we identify a method which incorporates covariate information and outperforms the current best-performing method in settings using realistic, skewed viral load values when we have access to a predictor(s) of virologic failure.