Subjects were recruited from the Bronx HIV Epidemiologic Research on Outcomes (HERO) cohort, which began at Montefiore Medical Center's Substance Abuse Treatment Program in 1985, and is comprised of current and former opiate users. HERO cohort members were eligible for the adherence study if they had current prescriptions for combination antiretroviral therapy and were willing to use MEMS caps for each antiretroviral medication. Subjects using other medication dispensing devices, such as pillboxes, were ineligible. MEMS caps fit standard size medication bottles, and record the time and date of each opening as a presumptive dose.
Subjects remained in the adherence study for 6 months, attending research visits at 4-week intervals. At the first visit, the purpose of the MEMS caps was explained, and subjects were assisted with transferring their medications to pill bottles fitted with MEMS caps. They were also instructed to open the MEMS bottles only to withdraw doses at the time of ingestion, not to transfer medications to other containers, and to carry the MEMS bottles with them if they anticipated taking medication while away from home.
At each follow-up visit, adherence data were downloaded using MEMS software (MEMS View, version 161; Aprex Corporation), and blood was drawn for viral load quantification (b-DNA Quantiplex assay [version 3.0; Bayer Diagnostics, Emeryville, Calif]). Sociodemographic and laboratory data were obtained at research visits performed contemporaneously as part of the HERO protocol. Antiretroviral history and selected psychosocial variables were assessed at baseline, and use of drugs and alcohol was assessed at every visit. Subjects received monetary reimbursement for each visit and a cash incentive for returning the MEMS caps at study end.
All data downloaded from the MEMS caps were exported into the SAS system (SAS Proprietary Software, Release 6.11; SAS Institute Inc., Cary NC). The period of analysis was defined as beginning at 2:00 am on the day after the baseline interview, and ending at 2:00 am on the day of the sixth follow-up interview (or final interview if the subject withdrew). Data were combined for subjects who were given replacement caps for the same medication, and excluded for subjects who did not use MEMS caps for defined periods of time (e.g., while hospitalized or incarcerated).
An overall mean adherence rate for each study subject was calculated by averaging the adherence rate for each medication. The adherence rate for each medication was calculated by dividing the number of MEMS cap openings by the number of doses prescribed during the entire period of MEMS cap use. In addition, dose interval adherence was estimated by calculating the percent of days on which at least 1 dose was taken, the percent of days on which the correct number of doses was taken, and the percent of days on which all medication doses were taken within 25% of the correct dosing interval (e.g., within 9 to 15 hours of the previous dose for a twice-per-day medication).
For each of the adherence indices described above, adherence was analyzed as a continuous variable. HIV viral load determinations were performed at each visit and, because a significant minority (15%) of study subjects were antiretroviral naïve and experienced precipitous decreases in viral load during the study's first 2 months, mean HIV viral load (log-transformed) was calculated by excluding the first 2 viral load measurements and including the last 5. Associations between continuous variables were assessed by Spearman correlation coefficients, associations of drug use behaviors and other categorical variables with adherence were assessed using the Kruskal-Wallis test, and associations of categorical variables with viral suppression were assessed using Fisher's exact test. Multivariate linear regression analysis using a stepwise selection procedure was performed to determine the best predictive model for adherence, and stepwise logistic regression was used to determine the best predictive model for viral suppression; both regression procedures used a P value of .15 for model entry.