Defining self-reported and prescription measures of adherence to ART: multi-cohort analysis

Background: Early identification of poor adherers to ART could facilitate interventions to prevent treatment failure. However, cohorts collect adherence data in diverse ways and defining levels of adherence is not straightforward. We examined how different measures may be used to identify those at risk of viral failure.

Methods: 6/19 cohorts collaborating in the ART Cohort Collaboration (ART-CC) contributed adherence data: 3 from prescription refills (3 North American cohorts) and 3 from self-report questionnaires (2 European, 1 North American cohort). For prescription data, we derived 1-year percentage adherence and viral suppression (≤500 copies/ml) using the viral load closest to the 1 year time point. For self-report data, we derived percentage adherence in the last 28 days and viral suppression using the closest viral load measure after but within 6 months of the questionnaire. We plotted Receiver Operating Characteristic (ROC) curves to assess discrimination of percentage adherence for diagnosing viral suppression, and estimated the area under the ROC curve (AUROC): 0.5 corresponds to no and 1 to perfect discrimination.

Results: Adherence and viral load data were available from 13276 patients: 9591 and 3685 with prescription and self-report data respectively. Greater proportions of patients were virally suppressed and had ≥95% adherence in cohorts with self-report compared with prescription data. AUROCs varied from 0.56 to 0.85 between cohorts and were systematically higher in cohorts with prescription data, likely due to the categorical rather than continuous nature of the self-report data.

Conclusions: Cohorts were heterogeneous in terms of viral suppression and adherence. Prescription and self-report data measure different aspects of adherence. Self-report items may provide a better snapshot of current adherence enabling identification of inadequate adherence before viral loads begin to climb. This real-time reporting advantage of self-report questionnaires may constitute an intervention which may explain the apparent lower discriminatory power compared with prescription data.