University of California, San Francisco Logo

University of California, San Francisco | About UCSF | Search UCSF | UCSF Medical Center

Evaluating adherence to highly active antiretroviral therapy with use of pill counts and viral load measurement in the drug resources enhancement against AIDS and malnutrition program in Mozambique
Global Health Sciences Literature Digest
Published January 20, 2009
Journal Article

San Lio MM, Carbini R, Germano P, et al. Evaluating adherence to highly active antiretroviral therapy with use of pill counts and viral load measurement in the drug resources enhancement against AIDS and malnutrition program in Mozambique. Clin Infect Dis 2008 May15;46(10):1609-16.


To assess the validity of a computerized pill count against the gold standard of viral load measurement and to explore factors associated with treatment adherence


An HIV treatment program at the Machava public hospital in Matola, Mozambique


A total of 531 patients (222 male; 309 female) aged ≥15 years who had been receiving first-line antiretroviral therapy (ART) for at least 3 months at the Machava treatment center were eligible for inclusion and were enrolled in the study between March 10 and 30, 2005. Of the 531 patients, 137 left the program or discontinued first-line therapy during the follow-up period, leaving 394 patients in the final analysis. At enrollment, the mean duration of exposure to highly active antiretroviral therapy (HAART) was 14.7 months for all 531 enrolled patients and 14.1 months among the 394 patients included in the follow-up analysis.

Study Design

Prospective observational cohort study with 12 months of follow-up


The Drug Resource Enhancement against AIDS and Malnutrition (DREAM) program was launched in Mozambique in 2002. It provides free treatment and testing, measurement of CD4 count and viral load, clinical examinations, nutritional support, and home-based care to HIV-infected patients at 13 centers in Mozambique. Participants at the Machava DREAM center received a fixed-combination drug regimen of two nucleoside analog reverse-transcriptase inhibitors (NRTIs; lamivudine and zidovudine or stavudine) and one non-nucleoside analog reverse-transcriptase inhibitors (NNRTIs; nevirapine), given as two pills per day.

Primary Outcomes

Primary outcomes included viral load ≥1000 copies/mL at enrollment (within 90 days) and at 12 months after enrollment. Adherence was based on software-recorded pill counts every 30 days and compared the percentage of pills prescribed and consigned for the period between appointments with the number of pills returned at the following appointment. For patients who left the study before the end of follow-up, pill counts were calculated for the time that they participated, but comparison with a 12-month viral load measurement was not possible.


Of the 137 patients observed for less than 12 months (mean, 5.8 months, SD ±2.1 months), 8 (5.8%) died, 17 (12.4%) quit the program, 41 (29.9%) transferred to a different health center, 44 (32.1%) switched regimens due to toxicity, and 27 (19.7%) switched regimens due to virologic failure.

Of the 394 patients (161 male and 233 female; mean age 37.0 years) who completed follow-up, the mean treatment adherence (±SD) was 95.2% ± 6.9%. Two hundred eighty-four (72.1%) had adherence of >95%, 66 (16.7%) of 91% to 95%, 22 (5%) of 86% to 90%, and 22 (5%) had adherence of <80%. At 12 months, of patients with >95% adherence, 274 (96.5%) had a final viral load of <1000 copies/mL, whereas of the 110 patients with adherence of ≤95%, 92 (83.6%) had a viral load of <1000 copies/mL. When adjusting for the duration of HAART exposure, the Mantel-Haenszel adjusted risk of patients with ≤95% adherence having a viral load ≥1000 copies/mL at 12 months of follow-up was 5.11 times greater (95% confidence interval [CI], 1.26-11.54) than that of patients who had >95% adherence. Cox proportional hazard regression showed that other than treatment adherence, age of <30 years (hazard ratio [HR] 2.18; 95% CI 1.02-4.68), and a baseline CD4 cell count of <338 cells/mm3 (HR 2.98; 95% CI, 1.26-7.07) were significantly associated with having a viral load of ≥1000 copies/mL. The sensitivity and specificity of adherence of >95% for a viral load of ≥1000 copies/mL were 61.9% and 77.4%, respectively, among patients with 15 to 27 months of HAART, and 75.0% and 77.1%, respectively, for patients with >27 months of HAART.


The authors concluded that data from the pill-count method agreed with data from viral load measurement and that pill counting can be recommended as a reliable tool that is replicable and economical.

Quality Rating

This study was of adequate quality. Although the authors clearly described participant recruitment and attrition, and although the sample size and observation period were sufficient, the strength of the evidence is limited due to the study's observational nature and the patient exclusions. The generalizability in particular is greatly limited by the inclusion only of patients who stayed on first-line ART regimens. As 71 of the 137 patients who did not complete 12 months on a first-line ART regimen moved to a different regimen due to failure to control viral load or to drug toxicity, the 394 patients whose results were analyzed were selected for those who were successfully controlling viral load. This limitation makes it difficult to assess how well pill count can monitor all patients who are failing first-line therapy. The authors say a separate study will be dedicated to this group of patients, but it would have been much more useful to have both groups analyzed in one report. Additionally, as this study included only patients enrolled at the Machava DREAM center, its results may not be generalizable to other populations.

In Context

No gold standard has been established for monitoring HAART adherence, and the various current methods lack homogeneity. Although self-reported adherence may be helpful for identifying factors associated with noncompliance, this report may overestimate treatment adherence.(1) Pill counts and the Medication Events Monitoring System (MEMS) are considered to be more useful than self-report, although research comparing the agreement of these methods with viral load measurement is lacking, particularly in African settings.(2) This report contributes data on pill counts, although not on the MEMS system and, as noted above, it probably overestimates the correlation of pill counts with viral load measures by excluding patients who failed first-line therapy.

Programmatic Implications

Monitoring viral load is the preferred method of evaluating treatment adherence and the therapeutic success of prescribed regimens, but recording pill counts may be a reliable, economic, and timely tool for monitoring adherence in resource-limited settings. In this report, the authors offer little guidance on using pill counts in lieu of viral load monitoring. If, for example, pill counts were to be used in conjunction with viral load measuring to make decisions about ad hoc viral load testing (rather than regular monitoring), in their data <95% adherence gave a false-positive rate of 23% for increased viral load in both durations of treatment groups. This might be acceptable predictive accuracy, if also generalizable, for deciding when to measure viral load, but its implications should be explored. The use of odds ratios and relative hazards are not readily interpreted in the setting of prediction and clinical decision making.


  1. Vervoort SC, Borleffs JC, Hoepelman AI, Grypdonck MH. Adherence in antiretroviral therapy: a review of qualitative studies. AIDS 2007; 21:271-81.
  2. Calmy A, Ford N, Hirschel B, et al. HIV viral load monitoring in resource-limited regions: optional or necessary? Clin Infect Dis 2007; 44:128-34.
  3. Sarna A, Luchters S, Giebel S, et al. Short- and long-term efficacy of modified directly observed antiretroviral treatment Mombasa, Kenya: A randomized trial. J Acquir Immune Defic Syndr 2008;48:611-19.