Subjects and Methods

Study settingThe distribution and the population-based monitoring of antiretroviral therapy in British Columbia has been extensively described in the literature [3, 8, 9]. Since 1986, 6305 HIV-1–infected men and women have ever received antiretroviral therapy. Prescriptions are distributed through several designated pharmacies or through one of 1200 or more physicians who have ever prescribed antiretroviral therapy. Although the majority of HIV-1–infected men and women who have received therapy are from Vancouver or the surrounding region, eligible treatment recipients have come from all regions in the province

Since October 1992, the distribution of antiretrovirals has been the responsibility of the HIV/AIDS Drug Treatment Program of the British Columbia Centre for Excellence in HIV/AIDS (hereafter, “the Centre”). The Centre distributes antiretrovirals on the basis of guidelines generated by the Therapeutic Guidelines Committee, which is made up of physicians, pharmacists, virologists, health service researchers, and economists [10]. Because of the anonymous administrative nature of the database, patient consent for inclusion is not required. The Centre’s HIV/AIDS Drug Treatment Program has received approval from the University of British Columbia Ethics Review Committee at its St. Paul’s Hospital site. The program also conforms with the province’s Freedom of Information and Protection of Privacy Act

Data collectionAll antiretroviral therapy recipients in the province are entered into an Oracle-based monitoring and evaluation reporting system that uses standardized indicators to prospectively track the antiretroviral use and clinical and health status of HIV-1–infected individuals. Physicians enrolling an HIV-1–infected individual into the system must complete a drug request enrollment prescription form, which compiles information on the applicant’s address, HIV-1–specific drug history, CD4 cell counts, plasma HIV-1 RNA levels, current drug requests, and follow-up physician data. Typically, persons receiving antiretroviral therapy are assessed by physicians at least once every 3 months, at which time prescriptions are renewed or modified. At the time of the first antiretroviral refill, subjects are asked to provide informed consent for accessing electronic medical records and to complete a participant survey, which elicits information on sociodemographic characteristics and clinical and health status. At the same time, the treating physicians are asked to complete a form using the World Health Organization (WHO) clinical staging system [11]

The Centre recommends that plasma HIV-1 RNA levels and CD4 cell counts be monitored at baseline, 4 weeks after starting antiretroviral therapy, and every 3 months thereafter. Plasma HIV-1 RNA levels were determined using the Roche Amplicor Monitor assay (versions 1.0 and 1.5; Roche Diagnostics) via either the standard method or the ultrasensitive adaptation. CD4 cell counts were measured by flow cytometry, followed by fluorescent monoclonal antibody analysis (Beckman Coulter)

Study subjectsAll HIV-1–infected subjects in the current study were entered into the Centre’s monitoring and evaluation system when they were first prescribed antiretrovirals. Eligible study subjects were ⩾18 years of age, achieved at least 2 consecutive HIV-1 viral loads <500 copies/mL, and were receiving antiretroviral therapy with prescriptions filled at least 3 times since initial viral load suppression. Subjects were first given triple combination therapy between 1 August 1996 and 30 September 2003 and had their CD4 cell counts and plasma HIV-1 RNA levels measured within 6 months before the first antiretroviral therapy start date. Study data from eligible subjects were extracted from the Centre’s HAART (highly active antiretroviral therapy) Observational Medical Evaluation and Research cohort and were followed up through 30 September 2004

Outcome measureThe primary end point in this analysis was time to virological failure, defined as the time to plasma viral load ⩾1000 copies/mL on at least 2 consecutive measurements. This virological cutoff was chosen because it is used clinically to define whether an individual is maintaining virological response [12] and because there is ongoing debate as to the significance of lower quantifiable viral loads [13–16]. The time to virological failure was defined as the interval of time between the first of 2 consecutive viral loads <500 copies/mL after initiating therapy and the time of the second of 2 consecutive viral loads ⩾1000 copies/mL

The event date was chosen to be the date of the second of 2 consecutive viral loads ⩾1000 copies/mL because a single viral load is insufficient evidence to confirm treatment failure; that is, the time of the second viral load ⩾1000 copies/mL is the time at which a provider identifies an individual as clinically having experienced treatment failure

Adherence and other exposure measuresThe adherence measure was based on medications dispensed, not prescribed; this is also commonly referred to as “refill adherence/compliance.” Subjects received new prescriptions at various time intervals ranging from monthly to, at most, every 3 months. Adherence was assessed over multiple time intervals per subject. An adherence interval was defined as the duration between 3 refills, as depicted in figure 1. The 3 refills represent 2 contiguous refill periods (figure 1), and this period was chosen because of concerns that shorter periods may not reflect a sufficiently stable measure of an individual’s pill–taking behavior [5]

The intended duration (days supplied) varied from subject to subject and within subjects over time; that is, prescriptions early in therapy (i.e., first months) generally provided 30 days’ supply. Refills later in therapy were generally extended to 60 days’ and then 90 days’ supply if the subjects returned appropriately for refills and evidenced a virological response

Adherence was defined as days’ supply of medications dispensed divided by days between prescription fills (expressed as a percentage) [17] . Although patients with undetectable viral loads less commonly have medications changed, the regimens can change because of toxicity or other medical reason. These changes can result in staggering of refills, which would stagger the measurement of adherence. Therefore, no index drug was used. Rather, any prescription for antiretroviral medication filled >10 days after a prior antiretroviral drug prescription was filled was considered a refill. If any antiretroviral was dispensed within 10 days after a prior fill, only the earlier fill was used to represent the start of that refill period. Furthermore, a window period of 30 days was allowed before classifying a subject as <100% adherent, so as to account for potential stockpiles at home or provider-prescribed temporary medication stoppage (e.g., for adverse effects, medical procedures, etc.); that is, each subject was each subject was allowed a 30-day grace period for which a <30-day gap between refills was considered continuous therapy. In secondary analyses, this grace period was removed in the calculation of adherence, so as to determine the impact that it had on the results. Of note, although refill-derived adherence can be >100% if subjects obtain refills before the expected refill dates, we truncated the adherence measure at 100%

The following baseline explanatory variables were also measured for inclusion as potential confounders of the relation between adherence and virological failure: age, sex, CD4 cell count, plasma HIV-1 RNA level, prior AIDS diagnosis, protease inhibitor use, current or past injection drug use, subject’s physician’s experience with HIV-1 treatment (defined as the number of HIV-1–infected patients the physician had previously treated at the time the study subject was first prescribed antiretrovirals by the HIV/AIDS Drug Treatment Program), and start of antiretroviral therapy after August 1997 (the time at which therapeutic guidelines changed so that triple combination HAART was available to all with baseline viral loads >5000 copies/mL or CD4 cell counts <500 cells/mm³)

Statistical analysesThe analyses included 2 definitions of the outcome: whether the subject had a virological failure at the end of the study period and time to virological failure. Time zero in the time to failure analysis was the first of at least 2 viral loads <500 copies/mL. The event date was the second of at least 2 consecutive viral loads ⩾1000 copies/mL. Event-free subjects were right censored at the last viral load <1000 copies/mL until 30 September 2004

The analysis included adherence in 4 different ways: (1) dichotomized at ⩽95% or >95% [2] and included only for the subject’s final observed adherence period; (2) categorized as >95%, 70%–95%, and <70% [4] and included only for the subject’s final observed adherence period; (3) dichotomized as ⩽95% or >95% and included as a time-varying variable for each of the subjects’ observed adherence periods; and (4) as a continuous measure in 10% decrements as a time-varying variable for each of the subjects’ observed adherence periods

Cox proportional hazards models were constructed for the association between adherence as a time-varying variable and time to virological failure. The relation between adherence category during the final interval (figure 1, interval N) and the proportion of subjects maintaining suppression was assessed using the test for trend

Analyses were performed using SAS software (version 8.02; SAS Institute). All tests of significance were 2-sided, with P<.05 used as the threshold of statistical significance

Results

Between 1 August 1996 and 30 September 2004, a total of 2217 antiretroviral-naive subjects ⩾18 years of age initiated triple combination therapy consisting of 2 nucleosides plus a protease inhibitor (boosted or nonboosted) or a nonnucleoside reverse-transcriptase inhibitor, and 1715 (77.4%) achieved confirmed suppression (i.e., at least 2 consecutive viral loads <500 copies/mL). Of the 1715 eligible subjects, 1634 (95.3%) met the minimum requirement of having 3 refill dates after suppression and were, therefore, included in the study

Subjects included achieved their first 2 consecutive viral load measures <500 copies/mL over a median interval of 93 days (interquartile range [IQR], 50–189 days). At the time of initiation of antiretroviral therapy, the median age was 39 years (IQR, 33–46 years), the median CD4 cell count was 200 cells/mm³ (IQR, 80–350 cells/mm³), the median plasma HIV-1 RNA level was >100,000 copies/mL (IQR, 48,000 to >100,000 copies/mL), and the subjects’ physicians had cared for a median of 63 patients with HIV-1 infection (IQR, 7–167 patients). There were 383 (23.4%) subjects who had a history of injection drug use. A total of 272 (16.6%) study subjects had a prior diagnosis of AIDS. The overall median follow-up time for study subjects was 29 months (IQR, 15–49 months). As of 30 September 2004, a total of 606 (37.1%) subjects experienced a virological rebound during the observation period, with a median time to rebound of 22 months (IQR, 13–39 months)

In table 1, we compare the baseline characteristics between subjects with >95% adherence at the final interval and those with ⩽95% adherence. Those with >95% adherence over the final interval were less likely to be injection drug users, were less likely to have received a protease inhibitor in the first regimen, were slightly older (median age, 39.0 vs 37.6 years; P=.023), and were more likely to have received follow-up from physicians with more HIV-1 treatment experience (median no. of HIV-1–infected patients under the physician’s care, 69 vs. 38; P<.001) than subjects with ⩽95% adherence at the final time point

Overall, adherence was high, with 1379 (84.4%) of the subjects having >95% adherence over the final interval. When the final adherence period was categorized as <70%, 70%–95%, or >95%, there was a statistically significant trend toward a higher proportion of subjects having virological failures as adherence decreased. There were 465 (33.7%) subjects with virological failure among those with >95% adherence, whereas 34 (40.5%) of the 84 with 70%–95% adherence had virological failure, and 107 (62.6%) of the 171 with <70% adherence had virological failure (P for trend<.001)

Table 2 displays the baseline characteristics and time-updated adherence measures and their associations with virological rebound, based on Cox proportional hazards models. After controlling for all the other factors in multivariate analyses, younger age, higher baseline CD4 cell count, starting treatment after August 1997, and being an injection drug user ever were all significantly associated with the likelihood of virological failure, irrespective of adherence. Time-updated adherence ⩽95% was strongly associated with an increased hazard of rebound, with a 66% increased risk (hazard ratio [HR], 1.66 [95% confidence interval {CI}, 1.38–2.01]) in multivariate models when compared against those with adherence >95%. When time-dependent adherence was expressed as a continuous variable in these models, there was an 11% increased risk of failure for every 10% decrease in adherence (HR, 1.11 [95% CI, 1.07–1.15]). In secondary analyses, the grace period was removed, and adherence for each interval was recalculated. Time-updated adherence ⩽95% remained associated with an increased hazard of rebound in fully adjusted models, although to a somewhat attenuated degree (HR, 1.42 [95% CI, 1.21–1.68])

It is important to note that the time-updated measure of adherence was not assessed at the exact time of virological rebound. Because the adherence measure was based on refill dates and refills were not linked to viral load measurements, there was a time lag between the adherence measure and the event/censor date (see figure 1). This lag was a median of 84 days (IQR, 41–161 days)

Discussion

This study demonstrates that, among individuals with prior virological suppression, high adherence, as estimated using a refill-based, time-updated measure, is associated with a lower risk of confirmed virological failure. Furthermore, our findings show that the high levels of adherence required for maintaining virological suppression are similar to the levels needed to achieve viral suppression in individuals newly starting therapy [1, 2, 18]. Importantly, as in prior work [1, 2], we did not identify a clear threshold effect for adherence; that is, greater adherence conferred a greater likelihood of maintained suppression even below the 95% threshold. As such, although the highest levels of adherence are an appropriate ideal, if not attainable, incremental increases in adherence at levels <95% should still be viewed as a worthwhile goal

Our results demonstrate that the dynamic measurement approach to adherence is associated with virological outcome. This use of the adherence measure is unique in that it is able to account for changes in an individual’s adherence pattern over time. This time-updated method overcomes the problem of other approaches that assign an individual’s adherence on the basis of a single time point or summarized over the entire study period, thereby failing to address the important dynamic of adherence change over time

The novel use of this time-varying adherence measure opens the possibility of innovative uses of this technique. By having a measure of adherence that is frequently updated, it is possible that clinicians can use this tool as an early warning system alerting them to their patients’ nonadherence before virological failure. In essence, this measure could operate as a form of “virtual” directly observed therapy. The mechanism by which directly observed therapy improves adherence has been well described [19]. Only testing this strategy in the setting of a randomized clinical trial will confirm whether this theoretical advantage is real

There are several features of the present study that should be highlighted. First, the present study was performed within a province-wide treatment program in which all subjects had access to medical attention, combination antiretroviral therapy, and laboratory monitoring free of charge. We are confident, therefore, that our results are not influenced by access-to-therapy-related issues, which have often compromised the interpretation of similar population or cohort-based studies. Second, despite the potential limitations that our measure of adherence is only a surrogate marker of actual pill taking and had not previously been used in a time-updated manner, surrogate measures of adherence have been validated in several studies to date [3–6], have been found to be associated with virological suppression to a magnitude similar to results obtained by use of microelectronic monitors, and have been correlated with untimed plasma antiretroviral concentrations [8] and survival [3]. Third, although the highest levels of adherence were associated with the highest rates of suppression, there were still substantial proportions of subjects who are able to adhere to lesser degrees and maintain suppression. Therefore, our conclusion that high levels of adherence are desirable is relevant only on a population basis. Fourth, despite allowing a 30-day grace period when categorizing a subject as <100% adherent, we still showed a strong association between the measure and outcome. If we were to eliminate this allowance, the measure may be even more sensitive to nonadherence. However, it would likely increase the misclassification of individuals who have been following prescribed orders to stop taking the medications temporarily or who have stockpiles at home. Such misclassification would explain the slightly attenuated results obtained when the grace period was removed in the secondary analysis. Fifth, the number of days supplied increased from 30 days to up to 90 days if subjects exhibited virological responses over time. These longer refill intervals are likely to result in detection of nonadherence later than shorter refill intervals would because the adherence calculation requires the refill dates. Therefore, our measure may be less sensitive to nonadherence for subjects who are switched to longer refill intervals. Sixth, when regimens changed, subjects would be considered adherent when in fact they were not taking the medications but had not notified the pharmacist that they had stopped taking the medications. These regimen changes unknown to the pharmacists may have resulted in overestimates of adherence. Seventh, because this study focused exclusively on subjects who had achieved virological suppression, the result cannot be generalized to the relation between adherence and virological control in subjects who never achieve suppression. Finally, despite a time lag between the date of adherence measurement and the event date, we were still able to show a relation between adherence and outcome. Because the suboptimal adherence in the present study began, on average, 2–3 months before confirmed virological failure, it is possible that this measure of adherence may be clinically useful as a data element for incorporation into strategies for improving adherence

In conclusion, the present study demonstrates that high levels of adherence result in the highest proportion of subjects maintaining suppression of viral load over time. Furthermore, our results support the conclusion that incorporating refill-based measures of adherence into clinical practice may allow for the early identification of subjects destined to experience virological failure because of poor adherence. Consideration should be given to automating this measure and providing it to the clinical care team so as to alert them to occult nonadherence, with the goal of intervening before virological failure. This strategy is ripe for testing in a clinical trial