In this observational study of HIV-infected drug users, adherence was stable over time and across medications. Self-reported adherence was higher than MEMS adherence, but a strong relationship was observed between both measures and HIV load. MEMS, however, was found to be a more sensitive measure of clinically significant nonadherence.
We did not observe a significant increase in adherence in the day preceding the research visit, measured by either self-report or MEMS. One-day adherence estimates were comparable to 1-week estimates, and both were predictive of virologic suppression. This is the first study to show that 1-day adherence is correlated both with adherence over time and with concurrent virus load. This suggests that, for some populations, a careful assessment of adherence for the day preceding the visit may provide an efficient and reliable adherence measure.
We found that adherence did not vary during the 6 months of the study. Because most subjects were antiretroviral experienced, this may not reflect the natural history of adherence during antiretroviral initiation. In addition, although our population was characterized by high rates of ongoing drug use, all subjects were receiving comprehensive HIV care, often on site at their methadone clinics. Although our results may therefore not be generalizable to drug users without a usual source of care, they suggest that adherence is stable when drug users have access to treatment for both substance abuse and HIV.
Although we found that adherence was not related to the number, class, or dosing frequency of medications, these results must be interpreted with caution. In the present study, medications were not randomly assigned, and the treating physicians may have chosen complex regimens only for patients they deemed “good adherers.” In addition, study patients may have accepted prescriptions for complex regimens only if they felt able to adhere. However, among this group of antiretroviral-experienced patients who were engaged in HIV primary care, complex medication regimens were not associated with poor adherence.
As has been found elsewhere, self-reported adherence was higher than MEMS adherence [14
]. The self-report rates we observed are consistent with the median self-reported adherence rate of 89% found by Bangsberg et al. [6
] in a cohort of homeless and marginally housed subjects, and they were somewhat lower than the rates reported in clinical trials [4
]. Self-reported adherence was highly correlated with concurrent HIV load (r
= .4–.5); these correlations are also similar to that reported by Bangsberg et al. (r
= .6) [6
]. These data demonstrate that self-report, although it overestimates adherence, is valid and reliable for use in research settings. To our knowledge, no other studies have demonstrated a strong relationship between self-reported adherence and virologic outcomes in drug users.
In contrast to self-report, our estimates of MEMS adherence were lower than were those that others have reported. Bangsberg et al. [6
] found median MEMS adherence of 67% after adjustment for “pocket doses” (doses not taken directly from the MEMS bottle). An even higher MEMS adherence rate (82%) was observed in a study by Kastrissios et al. [27
] that was nested within an AIDS Clinical Trials Group protocol. We believe that there are 2 reasons for the lower MEMS adherence rates we observed. First, we did not adjust the MEMS data for “pocket doses.” Because we observed that the percentage of days on which ≥1 MEMS dose was taken (mean, 64%; median, 75%) was higher than the percentage of days on which the correct number of doses was taken (mean, 39%; median, 27%), we concluded that subjects removed “pocket doses” on some days when they opened the bottle only once. However, both of these measures were similarly correlated with HIV load (r
= .53 and .60, respectively), which indicates that the higher adherence estimate does not more-accurately predict virus load. We believe that the true mean adherence rate is somewhat higher than we found but not higher than 64% (the percentage of days on which ≥1 dose was taken).
The other reason for the lower MEMS rates we observed is that our study was conducted in a “real world” population with high rates of poverty, unemployment, and active drug use. Despite this, the MEMS adherence rate we observed is comparable to the average adherence of 50% that has been reported in association with many chronic diseases [28
]. This suggests that, with appropriate access to care, adherence among drug users is similar to that of other community-based populations.
MEMS adherence was strongly correlated with concurrent virus load. Ours is one of the few studies to date to have examined this relationship, and the correlations we observed are consistent with observations by Paterson et al. (r
= .55) [7
] and are slightly lower than those reported by Bangsberg et al. (r
= .81) [6
]. Because MEMS may underestimate adherence and self-report may overestimate adherence, we expect that the true relationship between adherence and virus load is stronger than we observed using either MEMS or self-report. Nonetheless, the likelihood of achieving virologic suppression was greater at high levels of MEMS adherence than it was at high levels of self-reported adherence. For future research, particularly studies of adherence-enhancing interventions, we recommend that MEMS continue to be used to augment self-reported adherence measures.