Heterogeneity among studies in rates of decline of antiretroviral therapy adherence over time: results from the multisite adherence collaboration on HIV 14 study

Abstract

Objective: To use electronic drug monitoring to determine if adherence to HIV antiretroviral therapy (ART) changes over time, whether changes are linear, and how the declines vary by study.

Design: We conducted a longitudinal study of pooled data from 11 different studies of HIV-infected adults using ART. The main outcome was ART adherence (percent of prescribed doses taken) measured by electronic drug monitoring. We modeled and compared changes in adherence over time using repeated measures linear mixed effects models and generalized additive mixed models (GAMMs). Indicator variables were used to examine the impact of individual studies, and the variation across studies was evaluated using study-specific parameter estimates calculated by using interaction terms of study and time.

Results: The mean age of the subjects was 41 years, 35% were female, most had high school education or less, and 46% were African American. In GAMMs, adherence declined over time. The GAMMs further suggested that the decline was nonlinear, and in both sets of models, there was considerable study-to-study variability in how adherence changed over time.

Limitations: Findings may not be generalizable to non-US populations or to patients not in clinical studies.

Conclusions: Although overall ART adherence declined with time, not all studies showed declines, and a number of patterns of change were seen. Studies that identify clinical and organizational factors associated with these different patterns are needed. Models of changes in adherence with time should take account of possible nonlinear effects.

Figures

Figure 1

Unadjusted adherence levels by study month (minimum, 25th percentile, median, 75th percentile, and maximum) for all 11 studies.

Figure 2

Unadjusted adherence levels by study month (minimum, 25th percentile, median, 75th percentile, and maximum) for each study individually.

Figure 3

Fitted values for adherence from GAMM by time. The y-axis shows the contribution of the smoother to the fitted values along with the 95% confidence band.

Figure 4

Smoothed adherence curves over time for the 11 study sites, with 95% confidence bands. P-values for the interaction between study site and time were significant for study 1 (p=0.013), study 2 (p=0.010), study 7 (p<.0001 study and>