Abstract
OBJECTIVES: Self-report antiretroviral therapy (ART) adherence has been consistently associated with clinical outcomes. This study aims to compare the accuracy of self-report ART adherence measures with varying recall timeframes or item contents to predict virological response. METHODS: Data from a cross-sectional study among 2146 participants on ART in Guangxi, China were used. Detectable viral load was defined as viral load > 50 copies/ml. Adherence was measured using the number of days on which all doses were taken in the past month (i.e., the "one-month days taken" measure), the number of days on which any dose was missed in the past month (i.e., the "one-month days missed" measure), missed doses over the past 3 days, and missed days over the past weekend. Each adherence measure was dichotomized at an empirically determined cut-off to determine poor vs. good adherence. Accuracy of using each dichotomized adherence measure to predict detectable viral load was assessed by sensitivity, specificity, and the area under the receiver-operating characteristic (AUROC) curve. Logistic regressions were used to calculate the association between poor adherence and detectable viral load. RESULTS: All four measures had sensitivity below 10.0%, specificity above 90.0%, and AUROC slightly above 0.50. In univariate logistic regression, detectable viral load was statistically significantly associated with poor adherence determined by the one-month days taken measure (OR = 1.98, 95% CI 1.15-3.42), the 3-day measure (OR = 2.18, 95% CI 1.10-4.34), and the weekend measure (OR = 2.86, 95% CI 1. 54-5.34). After adjusting for covariates, statistically significant association persisted only for the weekend measure (OR = 2.57, 95% CI 1.33-4.99). CONCLUSIONS: Adherence measures asking about days on which all doses were taken might work better than items asking about days on which respondents missed their doses, and weekend measures should be included to comprehensively capture adherence behaviors. Further studies looking at intermediate timeframes are also needed to capture patients' dose-missing patterns that may better predict detectable viral load.