Associations of Self-reported and Prescription-based Adherence with Viral Suppression in Five HIV Cohort Studies

Background: Adherence to ART is crucial for successful treatment of HIV, but measurement of adherence is difficult. Methods of measurement and their use to identify patients at risk of treatment failure vary across settings. Methods: We combined data from 5 cohort studies (ART Cohort Collaboration). Adherence data were either from prescription refills or self-report questionnaires. For prescription data, we used logistic regression to examine associations of percentage adherence during the 1st year of ART with viral suppression (defined as ≤500 copies/ml) at 1 year. For self-report data, we examined associations of adherence in the last 28 days with viral suppression based on the closest viral load measure after but within 6 months of the questionnaire date. The distribution of adherence differed markedly between cohorts and measurement methods, so cut-points for adherence levels were defined separately for prescription (lower ≤70%, medium 71-90%, higher >90%) and self-report (lower ≤95%, medium 96-99%, higher 100%) data. Results: Prescription data were available for 7850 patients (3 cohorts) and self-report data for 3417 patients (2 cohorts). The table shows the distribution of adherence ORs for suppression in each adherence category. Despite dramatic differences in the distribution of adherence between cohorts, associations of adherence suppression were consistent for prescription data. Overall ORs (95% CI) for suppression in lower and medium compared to higher adherence groups were 0.07 (0.06-0.08) and 0.38 (0.33-0.43) respectively. Conclusion: Adherence varies both by cohort and by data collection method. Adherence was consistently associated with viral suppression, but prescription-based assessment appeared more strongly associated than assessment based on self-report. A limitation is that choice of cut-points was data driven. Research on the effect of ART adherence on mortality will require harmonised data from large cohort collaborations.