icon-folder.gif   Conference Reports for NATAP  
  ID Week
Oct 8-12 2014
Back grey_arrow_rt.gif
ROC Curves Pinpoint ART Failure Threshold at 95 Copies at Week 24
  IDWeek 2014, October 8-12, 2014, Philadelphia
Mark Mascolini
Receiver operating characteristic (ROC) curve analysis pinpointed the virologic failure threshold of nevirapine regimens at 95 copies/mL at week 24 and 33 copies/mL at week 48, in an analysis based on VERxVE trial data [1]. The study also confirmed that 50 and 200 copies are reasonable virologic failure thresholds, but the newly identified thresholds are more sensitive and specific.
Roche and Boehringer Ingelheim investigators who conducted this study noted that antiretroviral treatment guidelines differ in recommended viral load cutoffs indicating treatment failure. They suggested that improved sensitivity of viral load assays and worldwide scale-up of viral load monitoring add to the urgency of refining failure thresholds to help guide clinicians in changing failing regimens. Because ROC curves are widely used to help determine more sensitive and specific assay thresholds, the researchers used this method to refine virologic failure thresholds.
The analysis involved viral load determinations from samples stored at up to 24 points per patient through 144 weeks of the VERxVE trial, a double-blind, randomized comparisons of nevirapine immediate-release versus extended-release in previously untreated people [2]. The viral load assay was the Roche COBAS AmpliPrep/COBAS TaqMan HIV-1 Test, version 2.0. Two independent HIV physicians determined clinical treatment failure by analyzing viral load trajectories, absolute viral load measurements, and presence or absence of intermittent viremia (blips) or persistent viremia. The endpoint definition of treatment failure was the FDA snapshot method of a single viral load below 50 copies at 144 weeks of treatment or the last viral load after at least week 48.
The investigators created a simple logistic regression model and ROC curves to identify viral load thresholds at weeks 24 and 48 that had maximized sensitivity and specificity for evaluating virologic response through up to 144 weeks of follow-up. Finally the investigators compared the ROC curve-identified viral thresholds with standard failure thresholds of 50 and 200 copies, and they compared clinically determined failure with failure as defined by the FDA snapshot.
Clinical evaluation by the two HIV physicians determined that treatment failed in 74 of 563 evaluable trial participants (13%). A failure threshold of 95 copies/mL at treatment week 24 determined by ROC analysis had a sensitivity of 88% and a specificity of 56% for identifying clinical failure and a combined sensitivity + specificity of 44% (95% confidence interval [CI] 25% to 60%). Sensitivity, specificity, and combined sensitivity + specificity with a 50-copy failure threshold were 78%, 63%, and 41% (95% CI 21% to 57%). Respective values with a 200-copy failure cutoff were 94%, 49%, and 43% (95% CI 26% to 60%).
At 48 weeks of treatment ROC analysis determined optimal sensitivity and specificity with a viral load threshold of 33 copies/mL (sensitivity 82%, specificity 59%, sensitivity + specificity 41%, 95% CI 19% to 60%). Respective values with a 50-copy cutoff were 88%, 44%, and 32% (95% CI 13% to 54%) and with a 200-copy cutoff 98%, 33%, and 31% (95% CI 16% to 51%).
Among 539 trial participants with enough data available for snapshot analysis, 101 (19%) had clinically determined treatment failure. Sensitivity, specificity, and sensitivity + specificity with ROC-identified failure thresholds of 95 and 33 copies/mL at weeks 24 and 48 were as detailed above in predicting clinical failure through 144 weeks. With the FDA snapshot method, sensitivity, specificity, and sensitivity + specificity at week 24 were 79%, 53%, and 32% (95% CI 15% to 47%) and at week 48 82%, 51%, and 33% (95% CI 17% to 47%).
The investigators concluded that failure thresholds of 95 copies/mL at week 24 and 33 copies/mL at week 48 identified by ROC curve analysis maximize sensitivity and specificity in predicting treatment failure through 144 weeks. Standard failure thresholds of 50 and 200 copies/mL yield slightly worse combined sensitivity and specificity "but can also be used without a clinically important decrease in overall sensitivity and specificity."
In the same way, the analysis revealed "no clinically important difference . . . in overall sensitivity and specificity" between clinically determined treatment failure and the standard 50-copy failure endpoint used in the FDA snapshot analysis, "although there was a trend towards greater predictive value using the clinical determination."
The investigators suggested that ROC curves can be useful in determining optimal thresholds for individual viral load assays, for particular patient populations, or for novel antiretroviral regimens. Because viral load assays may differ in sensitivity, the researchers called for further studies to identify optimal thresholds to predict treatment failure.
1. Luo R, Aslam S, Robinson P, Quinson AM, Tri Do T. Identifying optimal HIV viral load thresholds for predicting antiretroviral treatment failure using ROC curve analysis. IDWeek 2014. October 8-12, 2014, Philadelphia. Abstract 1565.
2. Gathe J, Andrade-Villanueva J, Santiago S, et al. Efficacy and safety of nevirapine extended-release once daily versus nevirapine immediate-release twice-daily in treatment-naive HIV-1-infected patients. Antivir Ther. 2011;16:759-769. http://www.ncbi.nlm.nih.gov/pubmed/21817198