1565Identifying Optimal HIV Viral Load (VL) Thresholds for Predicting Antiretroviral Treatment Failure (TF) Using ROC Curve Analysis

Abstract

Background: Guidelines on HIV treatment differ globally in recommended viral load (VL) thresholds indicative of treatment failure (TF). Improvements in VL assay sensitivity and worldwide scale-up of VL monitoring make it increasingly important to determine optimal VLs thresholds for guiding treatment changes. Receiver operating characteristic (ROC) curves can identify optimal VL thresholds which maximize sensitivity and specifi city for predicting TF. The BRAVO study was a retrospective study, using the Roche COBAS AmpliPrep/COBAS TaqMan HIV-1 Test, v2.0, to explore thresholds of VL that might predict subsequent TF. Methods: Longitudinal patient specimens were collected through the VERxVE study, a randomized, double-blind (DB) study in treatment-naive patients treated with nevirapine immediate-release or extended-release (VIRAMUNE IR or XR), plus emtricitabine and tenofovir DF. Stored plasma samples from up to 24 timepoints (pre-treatment through 144 weeks of DB treatment), were tested with the Roche COBAS AmpliPrep/COBAS TaqMan HIV-1 Test, v2.0. The BRAVO patient endpoints of TF (which differed from the primary VERxVE endpoints) were determined by two independent HIV physicians based on VL trajectories. Guideline-based defi nitions of TF were also examined. A simple logistic regression model and ROC curves were created to determine the optimal VL thresholds for predicting TF at key study timepoints. Results: Of 563 evaluable patient-series, 74 patients (13%) had TF. By ROC curve analysis, a VL threshold of 95 copies/ml at week 24 and 33 copies/ml at week 48 maximized sensitivity and specifi city of predicting eventual TF [Area Under Curve: 0.75 and 0.73, respectively]. Overall sensitivity and specifi city were similar when compared to guidelines-based VL thresholds of 50 or 200 copies/ml and when using thresholds from an endpoint-based TF defi nition. Conclusions: ROC curve analysis identifi ed a VL threshold that maximized sensitivity and specifi city in predicting TF. Viral load thresholds of 50 or 200 copies/ml can also be used without a signifi cant decrease in overall sensitivity and specifi city. As VL assays may differ in sensitivity, ROC curve analysis can help optimize the clinical utility of VL monitoring.