Defining the optimal cut-point of self-reported ART adherence to achieve viral suppression in the era of contemporary HIV therapy: a cross-sectional study

Abstract

When considering adherence to antiretroviral therapy (ART) for HIV, many different cut-points are used. The primary goals of this study were to identify a level of self-reported medication adherence that best distinguished HIV viral suppression from non-suppression, and to compare the ability of a single-item and a 3-item adherence questionnaire to predict HIV viral suppression. This cross-sectional analysis included 380 persons with HIV (PWH) from the Florida Cohort study who completed a self-reported ART adherence measure within 30-days of having an HIV viral load test. We used Receiver Operating Characteristic (ROC) curve analyses and ROCContrast to compare the ability of a single-item and a 3-item self-reported adherence measure to predict HIV viral suppression (defined as ≤ 200 copies/mL). We used the Youden index and chi square statistics to assess specific cut-points, and repeated the analysis with a different definition of HIV viral suppression (≤ 1000 copies/mL). The mean percent adherence was 92.4% using the single-item score and 90.4% using the 3-item score; 81.6% had viral suppression. The areas under the curve for the single-item and 3-item adherence measures were generally poor overall and not significantly different from each other (0.589 and 0.580, p = 0.67). The Youden index identified cut-points of 93% and 89% as maximizing the sensitivity and specificity for the single-item and 3-item measures, respectively, whereas a cut-point of 80% on the single-item measure was best able to discriminate those with viral suppression (58% vs. 84%, p < 0.001). Results were similar with viral suppression defined as ≤ 1000 copies/mL. In this sample of PWH, a single question on medication adherence was as good as a 3-item questionnaire in predicting HIV viral suppression, although neither had good discriminatory ability. A cut-point close to 90% adherence maximized sensitivity and specificity, although viral suppression was very similar for nearly all measures above 80%.