Divergent adherence estimates with pharmacokinetic and behavioural measures in the MTN-003 (VOICE) study

Introduction: In the Microbicide Trial Network MTN-003 (VOICE) study, a Phase IIB pre-exposure prophylaxis trial of daily oral or vaginal tenofovir (TFV), product adherence was poor based on pharmacokinetic (PK) drug detection in a random subsample. Here, we sought to compare behavioural and PK measures of adherence and examined correlates of adherence misreporting. Methods: We included participants with PK and behavioural data from VOICE random subsample. Behavioural assessments included face-to-face interviews (FTFI), audio computer-assisted self-interviewing (ACASI) and pharmacy-returned product counts (PC). TFV concentrations <0.31 ng/mL in plasma (oral group) and <8.5 ng/swab in vaginal group were defined as "PK non-adherent." Logistic regression models were fit to calculate the combined predictive ability of the behavioural measures as summarized by area under the curve (AUC). Baseline characteristics associated with over-reporting daily product use relative to PK measures was assessed using a Generalized Linear Mixed Model. Results: In this random adherence cohort of VOICE participants assigned to active products, (N=472), PK non-adherence was 69% in the oral group (N=314) and 65% in the vaginal group (N=158). Behaviourally, ≥10% of the cohort reported low/none use with any behavioural measure and accuracy was low (≥43%). None of the regression models had an AUC >0.65 for any single or combined behavioural measures. Significant (p<0.05) correlates of over-reporting included being very worried about getting HIV and being unmarried for the oral group; whereas for the vaginal group, being somewhat worried about HIV was associated with lower risk of over-reporting. Conclusions: PK measures indicated similarly low adherence for the oral and vaginal groups. No behavioural measure accurately predicted PK non-adherence. Accurate real-time measures to monitor product adherence are urgently needed.