Sequential Antiretroviral Adherence Measurement Using Electronic Bottle Cap Monitors in a Cohort of HIV-Infected Adults

Abstract

Most studies employing electronic bottle monitors to measure antiretroviral adherence are limited to 24 weeks of duration, providing a snapshot of adherence from a treatment course that may be lifelong. The stability of these measures in individual patients over time has not been previously described. We measured antiretroviral adherence using Medication Event Monitoring System (MEMS) caps in a patient cohort in 2004 and 2005 and repeated the measurement in 2008 and 2009. Forty-eight participants completed both monitoring periods. Mean adherence rates in the first and second periods were 74.2% and 68.9%, respectively. Adherence rates from the 2 periods for individual participants were highly correlated (Spearman rho = .66, P < .001).

Introduction

Excellent adherence to antiretroviral therapy (ART) remains a mainstay of HIV care. Electronic monitoring of adherence, using bottle cap monitors, is a rigorous method of quantifying adherence, and studies conducted using these devices have been influential in understanding the adherence–virologic outcome relationship. ^{1 –3} Many of the studies exploring this relationship have been limited to 24 weeks of observation, ^{3 –6} whereas the need for ART adherence is long term. It would be useful to know whether a quantitative assessment of adherence to ART at 1 time point is predictive of adherence rates in the future. We are not aware of any studies that have used electronic monitors to compare adherence to ART in the same individuals at 2 separate time points. We therefore conducted a follow-up ART adherence study approximately 4 years after the completion of an initial adherence study.

Methods

Between 2004 and 2005, we conducted a study (henceforth referred to as study I) in the Montefiore Medical Center AIDS Center exploring the relationship between 24-week adherence to lopinavir/ritonavir (LPV/r) in highly active antiretroviral therapy (HAART) recipients, as measured by Medication Event Monitoring System (MEMS; Aardex, Inc, Sion, Switzerland) monitors (or caps), and virologic outcomes. Sixty-four participants completed study I, which has been previously published. ⁴ Between August 2008 and March 2009, we attempted to contact all 64 of the study I completers in order to re-enroll them in a 24-week follow-up study of adherence (henceforth referred to as study II). The eligibility criteria for study II were completion of study I, current receipt of HAART, willingness to use the MEMS cap properly, and willingness to grant informed consent.

Participants who agreed to participate received a MEMS cap for the LPV/r component of their HAART, if they remained on this agent, or for their new anchor drug (as determined by the principal investigator, J.S.) if they had changed regimens. Study participants were reinstructed in the proper use of the MEMS caps, for example, open the cap once, and only once, at each dose time, do not take doses out in advance for later use or to load a pillbox, time the switchover of the cap to dose time when a new bottle of medication is obtained from the pharmacy. Participants were to complete 2 additional study visits after the initial visit, at 12 weeks and 24 weeks. Data collected via MEMS caps were downloaded onto the study computer at each visit and then uploaded (stripped of all patient identifying data) to a server at Aardex, Inc. Study staff administered a standardized questionnaire at weeks 12 and 24 on MEMS cap usage and time periods during which MEMS cap usage was suspended while the participant continued taking HAART (eg, hospitalization and incarceration). The results of these questionnaires were transmitted to Aardex, so that periods of improper or lapsed usage of MEMS caps could be excluded from analysis. All participants had an HIV-1 RNA assay (viral load [VL]) measured at enrollment, and another VL measurement on or about week 24 (a 2-week window period on either side of the week 24 time point was permitted) of the study. Both HIV-1 VLs were measured in the Montefiore Medical Center Clinical Laboratory using the Bayer Versant branched DNA (bDNA) Version 3.0 assay. Participants who completed the baseline and week 24 visits and who had VL measurements at both time points were entered into the final analysis.

Demographic and clinical information were collected from each study participant during a brief interview and medical record review at study enrollment. Medication Event Monitoring System data were compiled and cleaned by Aardex and transmitted directly to the principal investigator, independent of the data transfer to the study sponsor. Medication adherence was defined as (number of bottle openings/number of prescribed doses) × 100, excluding all time periods that participants reported taking their medications but not using the MEMS caps. Viral load measurements were accessed through Montefiore Medical Center’s clinical information system. The study protocol was reviewed and approved by the Montefiore Medical Center Institutional Review Board.

For the present analyses, data were extracted from the original study I databases but were restricted to those participants who completed study II (ie, study I participants who failed to complete study II were excluded). Since the adherence rate variable did not meet normality assumptions, comparison of means was accomplished using Wilcoxon signed rank test and bivariate correlation was accomplished using Spearman rank correlation. All statistical tests were 2-tailed, and P < .05 was the threshold for statistical significance.

The funding organization had no role in participant recruitment or data collection. Data analyses were completed at Montefiore Medical Center, with statistical results independently verified by an employee of the funding organization (R.A.R., a biostatistician). Members of the funding organization were permitted to make suggestions during the manuscript preparation, but final decisions regarding manuscript content and submission were made at the sole discretion of the principal investigator.

Results

Of the 64 participants who completed study I (in 2004-2005), 4 had died and 8 were lost to follow-up and could not be contacted despite several attempts by telephone and mail over the course of the enrollment period for study II. Of the remaining 52 participants, all remained on HAART and all agreed to participate in study II (in 2008-2009). Forty-eight of these 52 completed study II, 1 discontinued HAART during the study, 2 died, and 1 was lost to follow-up. The final data set is comprised of these 48 participants. The mean adherence rate in study I of the 16 participants who did not complete study II did not differ significantly from that of the 48 participants described in this article (68.7% ± 21.6% versus 74.2% ± 22.4%, P = .39). The mean duration of adherence monitoring in study II was 24.1 weeks, and the mean time elapsed between enrollment in study I and completion of study II was 1552 ± 171 days or 4.25 years.

The mean age of the cohort was 49.5 ± 6.6, 48% were male, 50% female, 2% transgender, 46% Latino, 48% African American, and 6% white. Three quarters had Centers for Disease Control and Prevention (CDC)-defined AIDS, and the mean time since initial diagnosis of HIV infection was over 13 years. Thirty-six of the 48 remained on LPV/r-based regimens. The remaining 12 had switched to a new anchor drug—5 to atazanavir (ATV), 5 to darunavir, and 2 to efavirenz (EFV). The switch from LPV/r occurred a mean of 18 months before enrollment in study II (range: 2-40 months). Eight of the participants had a detectable VL on the measurement preceding the switch.

Correlation of Adherence Rates in Study I and Study II

There was a highly significant direct correlation of adherence rates observed in study I and study II (Spearman rho = .66, P < .001). Figure 1 provides a pictorial representation of the relationship of adherence rates in the 2 studies. The correlation was significant for the 36 participants who remained on LPV/r in study II (Spearman rho = .75, P < .001) but not for the 12 participants who had switched to different anchor drugs (Spearman rho = .40, P = .20).

Figure 1.

Scatter plot of participants’ adherence rates in study I and study II.

In order to explore the characteristics of those participants who experienced substantial changes in adherence between the 2 studies, we stratified the group into adherence quartiles for both studies. Study I adherence rate ranges according to quartile, from lowest to highest, were 27.2% to 50.6%, 53.6% to 74.9%, 79.1% to 92.7%, and 93.0% to 100%. Study II adherence rate ranges according to quartile, from lowest to highest, were 5.9% to 43.7%, 45.3% to 76.1%, 77.6% to 93.2%, and 94.2% to 100%. Twenty-four participants (50%) remained in the same quartile for both studies, and 43 (89.6%) differed by ≤1 quartile. Three participants moved up by 2 or more quartiles, and 2 participants moved down by 2 quartiles. All 3 of the former participants had switched from twice per day LPV/r in study I to a fully once per day regimen in study II. Visual inspection of their bottle-opening schedule revealed that 2 of the participants were, for the most part, taking 1 dose per day of their twice-daily (BID) regimens in study I and essentially doubled their adherence rates by continuing to take 1 dose per day of their new once-daily (QD) regimen. The third patient had a haphazard adherence profile on the BID regimen in study I and had a significant improvement in adherence to a QD regimen in study II. The 2 participants who moved down 2 quartiles remained on their same BID LPV/r-based regimens throughout both studies. Visual inspection of their bottle-opening schedules did not reveal any specific pattern to account for their deterioration in adherence rates.

Virologic Outcomes

The overall mean adherence rate in study II was 68.9% ± 28.0%. The mean adherence of the 38 participants with week 24 VL < 75 copies/mL was significantly higher than the 10 with VL ≥ 75 copies/mL (74.1% ± 27.3% versus 49.3% ± 22.2%, P = .01). There was a trend toward decreasing adherence rates from study I to study II among the 48 participants (74.2% ± 22.4% versus 68.9% ± 28.0%, P = .09), yet the proportion of participants with VL < 75 copies/mL increased significantly from study I to study II (60.4% versus 79.2%, P = .05). The proportion of participants with VL < 75 copies/mL according to study II adherence quartile (from lowest to highest adherence) was 90.9%, 50.0%, 83.3%, and 92.3%.

Discussion

Electronic monitoring of antiretroviral adherence using MEMS caps is a rigorous approach to adherence measurement, and it has been critically important in understanding the relationships between medication ingestion, virologic suppression (or lack thereof), and virologic resistance. ^{1 –3} Because such monitoring is expensive and the bottle cap monitors are cumbersome, the typical study employing MEMS caps is short in duration; generally 24 weeks for antiretroviral research. ^{3 –6} The need for ART is, however, long term. Little is known about the predictive value of MEMS cap-derived adherence rates at 1 time point for adherence rates at future time points. Pariente et al reported a significant correlation of individual adherence rates measured during 2 phases of a study over a continuous 7-month study period. ⁷ We are not aware of any prior studies that have employed MEMS caps to compare individual adherence rates at separate time points separated by years.

In the current study, we assessed the degree of correlation of adherence to HAART in a cohort of patients at 2 points in time, approximately 4 years apart. Not surprisingly, we found that adherence rates in study I were highly predictive of adherence rates in study II. The graphical presentation of the data in Figure 1 illustrates the particularly close correlation in those with the highest adherence rates. With few exceptions, patients who were outstandingly adherent in study I remained outstandingly adherent over 4 years later in study II. Only 5 patients changed their adherence behaviors enough to move 2 quartiles from their original adherence quartile assignment. Examination of the adherence profiles of these patients suggests that 1 common reason for significantly improved adherence was a change in dosing schedule from BID to QD.

Viral load measurements were collected in the course of both studies. After 24 weeks of monitoring in study I, 60.4% of the cohort had VL < 75 copies/mL, with a mean adherence rate of 74.2%. After 24 weeks of monitoring in study II, 79.2% of the cohort had VL < 75 copies/mL, with a mean adherence rate of 68.9%. This rate of virologic suppression compares favorably with the 78.3% rate of virologic suppression, defined by the less stringent threshold of VL < 400 copies/mL, in those with adherence rates ≥95% in the landmark study by Paterson et al. ¹ Although we do not have adherence data for the time period elapsing between the 2 studies, the close correlation of adherence rates measured at 2 time points several years apart, taken together with the virologic outcomes, suggests that modern boosted protease inhibitor regimens can produce virologic suppression over the long term despite chronically suboptimal adherence. This is consistent with recent observations that the risk of virologic failure in those achieving complete virologic suppression on HAART decreases over time at all adherence strata ⁸ above 50%.

Our study has limitations that deserve mention. The sample size was small, and a significant number of participants could not be located to re-enroll in study II (N = 12) or failed to complete it (N = 4). Analysis of these participants’ data from study I did not reveal any significant difference in adherence rates from the remainder of the study sample. Although MEMS cap monitoring is an established strategy for quantitating adherence, it is not foolproof. Medication Event Monitoring System caps may underestimate adherence in some patients, largely due to taking “pocket doses” of medications and filling pillboxes. ⁹ Some patients in this study with very low adherence rates and VL < 75 copies/mL, including 1 with an adherence rate of 5.9%, were likely taking doses without opening their MEMS caps, although they denied this at study visits. Detailed inspection of their MEMS data (not shown) revealed that most of these patients had prolonged periods without any MEMS cap openings and only 1 had a dramatic increase in adherence immediately prior to the week 24 study visit, that is, “white coat compliance.” ¹⁰ The original study enrolled a cohort of patients almost all of whom were already receiving and tolerating LPV/r-based HAART and were planning to continue on it for at least 6 months. The large majority remained on LPV/r-based HAART in the current study. Our findings may not be generalizable to less successful LPV/r recipients. The data presented in this study suggest that many participants were able to maintain virologic suppression despite imperfect adherence over approximately 4 years. Whether virologic failure, caused by the slow acquisition of resistance mutations, would occur over longer periods of follow-up is not known. Finally, the study was conducted in 1 center in the Bronx, New York. It is not known whether our findings are applicable to other HIV-infected populations in other locations.

Medication Event Monitoring System cap adherence studies have generally provided a brief snapshot of adherence patterns in patients receiving lifelong therapy. In this study, we repeated 24 weeks of MEMS cap monitoring in a cohort of patients approximately 4 years after the original study in order to assess the degree of correlation of the adherence rates measured over time. Our findings indicate that the antiretroviral adherence rates of individual patients, measured by MEMS caps, at 2 time points, several years apart, are closely correlated.