|Home | About | Journals | Submit | Contact Us | Français|
Prescription or pill-based methods for estimating adherence to antiretroviral therapy (ART), pharmacy adherence measures (PAMs), are objective estimates calculated from routinely collected pharmacy data. We conducted a literature review to evaluate PAMs, including their association with virological and other clinical outcomes, their efficacy compared with other adherence measures, and factors to consider when selecting a PAM to monitor adherence. PAMs were classified into 3 categories: medication possession ratio (MPR), pill count (PC), and pill pick-up (PPU). Data exist to recommend PAMs over self-reported adherence. PAMs consistently predicted patient outcomes, but additional studies are needed to determine the most predictive PAM parameters. Current evidence suggests that shorter duration of adherence assessment (≤6 months) and use of PAMs to predict future outcomes may be less accurate. PAMs which incorporate the number of days for which ART was prescribed without the counting of remnant pills, are reasonable minimum-resource methods to assess adherence to ART.
Since the introduction of combination antiretroviral therapy (ART) in the mid-1990s, human immunodeficiency virus (HIV)–1 infected patients have experienced decreasing levels of morbidity and mortality in both high-income countries (HICs) and low- and middle-income countries (LMICs) [1–3].
Successful HIV treatment largely depends on patient adherence to ART. Suboptimal adherence predicts virological failure [4–7], the development of HIV drug resistance [8–10], and death [11–13]. Standardized, simple, and routine cost-effective monitoring of adherence is necessary to identify patients at risk of poor outcomes who would benefit from targeted adherence support . Two simple methods for assessing adherence are patient self-report or prescription- or pill-based adherence measures, referred to in this review as “pharmacy adherence measures” (PAMs). Unlike patient self-reported adherehence, which can be affected by recall or social desirability bias, PAMs are objective and may be calculated from information routinely available in medical and pharmacy records .
The World Health Organization (WHO) recommends the assessment of adherence to ART with every patient contact . Despite these recommendations, there is no consensus regarding the optimal method to estimate individual- and population-level adherence to ART [15, 16]. This review summarizes currently available knowledge on PAMs, identifies their strengths and limitations, proposes factors to consider when selecting a PAM to monitor adherence and predict treatment outcomes, and identifies areas for future research.
PAMs are prescription- or pill-based adherence estimates calculated using dates of prescription refills and/or pill counts performed during routine clinic visits. Importantly, PAMs do not include self-reported measures, PCs performed outside of routine clinic visits (eg, unannounced PCs), monitoring of antiretroviral drug levels or monitoring with electronic devices (eg, electronic pill bottle [MEMS] caps). For purposes of clarity, we define the period of time over which individual patient adherence is estimated as “the duration of adherence assessment.” In addition, we identify 3 broad categories of PAMs: MPR, PC, and PPU. Definitions and formulae used to calculate these adherence estimates are provided in Table 1.
We searched PubMed, the Cochrane Database of Systematic Reviews, and the ISI Scientific Citation Index databases, using the terms “HIV” and “adherence” or “compliance,” together with “pharmacy,” “prescription,” “pill count,” “medication possession,” or “pick-up,” for articles published from inception until April 2010. We also searched reference lists of all included studies. All English-language publications investigating associations between PAMs and the following outcomes of interest were included: virological failure or suppression (ie, viral load greater or less than a defined threshold), change in viral load, immunological failure, HIV drug resistance, or mortality. Studies in which the outcome of interest occurred before the estimation of patient adherence or in which estimates were calculated by combining a PAM with an additional adherence measure were excluded. If different adherence analyses were published using data from a single cohort, we selected the publication that provided the most information. Because of marked study heterogeneity, meta-analyses were not performed.
In total, we identified 36 studies that met our inclusion criteria: 12 from LMICs (Table 2) [4, 13, 17–26] and 24 from HICs (Table 3) [10, 11, 27–48]. All LMIC studies were from sub-Saharan Africa. Eight LMIC studies used MPR [4, 13, 17, 19–23], 3 used PC [18, 25, 26], and 1 used PPU . HIC studies included 18 studies from North America, 5 from Europe, and 1 from Australia. Sixteen studies from HICs estimated adherence using MPR [10, 11, 27, 29–32, 34, 36–39, 41, 45, 47, 48], 5 used PC [28, 33, 35, 40, 43], and 3 used PPU [42, 44, 46].
Twenty-seven (75%) of 36 studies reported virological outcomes; 19 were from HICs [27–37, 39–45, 48], and 8 were from LMICs [4, 17, 18, 22–26]. PAMs predicted virological failure in 14 (88%) of 16 studies, virological suppression in 8 (89%) of 9 studies, and viral load change in 3 (60%) of 5 studies. Studies conducted in LMICs generally assessed ART-naive populations receiving nonnucleoside reverse transcriptase inhibitor (NNRTI)–containing regimens and demonstrated that PAMs were predictive of either virological failure or virological suppression. In contrast, all studies demonstrating no association between PAMs and virological outcome were conducted in HICs and assessed ART-experienced patients using smaller sample sizes (≤115 subjects; range, 40–115 subjects) [30, 31, 43, 44] or estimated adherence over shorter durations of assessment (4 weeks) .
Given the importance of PAMs in predicting virological failure or suppression, we considered 2 important sources of study heterogeneity: the duration of adherence assessment and the temporal relationship between the adherence assessment and viral load testing.
The duration of adherence assessment was most commonly the first 6 months [17, 22, 24, 31–33, 39, 40, 44] or 12 months [17, 27, 41, 48] after ART initiation but ranged widely (range, 1–44 months). All studies with a duration of adherence assessment greater than 6 months demonstrated association with virological failure or suppression [4, 17, 22, 23, 25, 27, 34, 36, 41, 48]. However, only 11 (79%) of 14 studies demonstrated an association when the duration of adherence assessment was 2–6 months. Notably, associations were maintained over shorter durations of assessment (2–6 months) when larger sample sizes were used (>115 subjects) [17, 18, 24, 26, 28, 32, 36, 42, 46, 49], suggesting that studies that involved shorter durations of adherence assessment or smaller sample sizes lack power to detect statistically significant associations.
The time at which viral load was assessed varied and occurred either at the end of the period of adherence assessment or at a future time point. Fourteen (88%) of 16 studies demonstrated an association between PAMs and virological failure or suppression at the end of adherence assessment [4, 17, 18, 22–28, 30, 31, 33, 34, 36, 40], whereas in 5 (63%) of 8 studies, PAMs were predictive of future outcomes (range, 1–55 months after adherence assessment) [17, 22, 32, 39, 41, 42, 44, 48]. In 2 studies, PAMs were found to be more predictive of virological outcomes at the end of the period of adherence assessment than at a future time point. However, the duration of adherence assessment used to predict the future outcome was shorter in both studies, making it difficult to draw further conclusions [17, 22].
All studies that assessed the association between PAMs and CD4 cell count response demonstrated that lower levels of adherence were associated with a poorer CD4 cell count responses [19–22, 27, 31, 39, 42, 46, 47]. Of the 6 studies documenting association between PAMs and mortality [11, 13, 19, 20, 38, 39], all but 1  demonstrated increasing mortality with lower levels of adherence. In addition, 2 large studies that assessed African treatment programs showed an association between lower individual adherence and subsequent classification as lost-to-follow-up during the first 12 months after ART initiation  or after a median of 7.7 months . Importantly, authors of the study in which PAMs were associated with loss to follow-up and not with mortality acknowledge that many subjects who were lost to follow-up were likely to have died .
Data regarding the association between poor adherence to ART, as estimated by PAMs and HIV drug resistance, were limited. However, 2 studies that involved ART-naive patients receiving NNRTI- or protease inhibitor (PI)–based regimens [10, 28] demonstrated an association between adherence and acquired HIV drug resistance.
Ten studies documented PAMs and self-reported adherence and their associations with virological outcomes. Both PAMs and self-report measures were associated with virological outcomes in 3 studies; however, the superiority of one measure over the other could not be inferred [26, 34, 37]. In 4 studies, PAMs predicted virological outcomes, whereas self-reported adherence did not [23, 24, 26, 37]. In 1 study that compared PAMs with self-reported adherence using receiver operating characteristic curves, the PAM was superior to self-reported adherence (P < .001) . In contrast, self-reported adherence was superior to a PAM in 1 study  in which a 4-week PC assessment failed to predict change in viral load 12 weeks later, whereas improved self-reported adherence measured at the later time point was predictive of a viral load reduction. In 3 additional studies, neither the PAMs nor self-reported adherence predicted virological outcome [30, 43, 44].
Three studies compared PAMs with use of MEMS caps [35, 40, 43]. Better adherence by both measures predicted virological suppression in 1 study , neither predicted viral load change in a second study , and only use of MEMS cap predicted viral load change in the third .
Only 1 study compared different PAMs by investigating 2 different PC measures—one incorporating time into the denominator and the other without. Lower estimates of adherence by both PC measures predicted virological failure at 6 and 12 months . The ability of the 2 PAMs to predict virological failure was not directly compared; thus, superiority could not be established. Interestingly, the PC measure using time in the denominator classified more individuals as nonadherent and provided greater variability in adherence estimates.
To identify patients at risk for suboptimal clinical or virological response using PAMs, an understanding of the relationship between adherence and outcomes, including potential adherence thresholds or cutoffs, is essential. Studies commonly report adherence estimates dichotomously or across ≥3 strata. All studies that stratified adherence estimates were reviewed to identify potential threshold effects.
Historically, >95% patient adherence to ART has been cited as the threshold to achieve virological suppression. This threshold was based on a single study of ART-experienced patients receiving unboosted-PI regimens . Subsequent studies have suggested that adherence levels of <95% are associated with virological suppression in a considerable proportion of patients receiving NNRTI or boosted-PI regimens [6, 28, 50]. Seven studies using PAMs with stratified adherence estimates [4, 23, 28, 32, 36, 39, 41] failed to detect a threshold effect. Interestingly, in 2 studies, when 100% adherence was used as the highest stratum, no significant difference in the rate of viral rebound was observed, compared with levels of adherence of 95%–99%  and 90%–99% ; however, both studies reported decreased risk of viral rebound for every 10% increase in adherence across all strata. The ability of some studies to detect a threshold effect may have been limited by the fact that patients received different ART regimens [32, 36, 39, 41]. However, 2 studies that assessed only NNRTI-based regimens [4, 23] reported virological failure rates of 29% at 80%–95% adherence  and of ~25% at 80%–99% adherence . These observations are consistent with studies using self-reported adherhence, unannounced PCs, and use of MEMS caps for patients receiving NNRTI regimens in which the majority of individuals had virological suppression in adherence strata below 95% [6, 50].
Four studies reported adherence across ≥3 strata and observed a threshold effect for mortality. For individuals receiving predominantly NNRTI [13, 19] or unboosted-PI regimens [38, 39], ≥2 adherence strata above 70% , 75% , or 80% [13, 19] did not differ in their ability to predict mortality, but lower adherence strata did predict increased mortality. Importantly, investigators attempted to account for “reverse causation” (ie, cessation of ART because of reasons related to poor survival) by using prolonged durations of adherence assessment before observing subjects for survival outcomes. On the basis of these data, a threshold effect predicting increased mortality among patients with a level of adherence of <80% noted by use of PAMs may serve as a potential target for adherence interventions, especially if available resources are limited.
Two studies described entirely  or predominantly (74% of subjects)  ART-naive populations who had received unboosted-PI regimens, with those who had adherence rates of 75%–90% having the highest risk for developing resistance. Because of these limited findings, we were unable to draw conclusions about adherence thresholds for the emergence of HIV drug resistance. Importantly, no studies examined the relationship between PAMs and drug resistant HIV in patients exclusively receiving NNRTI or boosted-PI regimens.
PAMs are ideally suited to monitoring adherence because they are objective and can be easily derived from data routinely collected for other purposes, such as clinical care, medication billing, fulfillment of legal requirements, or drug supply management. Importantly, PAMs may overestimate actual pill taking if individuals discard or share pills and, therefore, estimate maximum possible adherence. In addition, PAMs do not provide information on patterns of nonadherence known to be associated with the development of resistance to NNRTIs [51, 52].
Despite their limitations, in settings in which frequent routine viral load monitoring is not available, PAMs can play an important role in monitoring individual and population-level adherence to ART. Although prospective studies of adherence interventions and viral load testing targeted at patients with lower levels of adherence, as determined by PAMs, have not been reported, findings from 2 studies are optimistic [17, 36]. In a study conducted in sub-Saharan Africa, PAMs were superior to CD4 cell count criteria in predicting virological failure, and when PAMs were performed before determinations of viral load and CD4 cell count, PAMs were as accurate as CD4 cell count changes in predicting virological failure. These results support the use of PAMs for potential identification of patients at risk of future virological failure . In a second study, which was from Canada, analysis of repeated measures of adherence, which account for changes in adherence over time, predicted future viral rebound , suggesting that routine surveillance of patient adherence with PAMs can be used to alert clinicians to possible future virological failure.
Use of PAMs to monitor adherence requires the following minimum data: ART regimen dispensed, date of dispensing, and number of days of ART dispensed. Selection of a PAM will depend on available resources at a site or in a program, as well as a local assessment of the strengths and limitations of different PAMs. MPR estimates are the most studied and incorporate time in the denominator (Table 1); thus, patients need to return to the dispensary before their medication finishes if taken as prescribed, to be considered 100% adherent. PC and PPU measures that do not incorporate time in the denominator (Table 1) may overestimate adherence, because patients may use all dispensed ART but do so over longer periods than intended. PC measures are limited by the increased resources required to routinely count and record remnant pills at each clinic or pharmacy visit. In addition, if patients do not bring all remnant pills for counting or share or lose pills, the rate of adherence will be overestimated. Although PC measures may provide a more accurate assessment of adherence by accounting for unused ART, to our knowledge, no data comparing PC to non-PC PAMs are available. PPU measures are the least studied PAM. Unlike MPR and PC measures, PPU estimates are dichotomous and, therefore, do not provide a range of adherence, limiting their ability to identify individuals in need of increased adherence support.
In the absence of data suggesting an advantage of PC over non-PC measures, and considering the extra resources required to count remnant pills, we do not suggest using PC measures. Furthermore, measures incorporating the number of days for which ART was prescribed in their definition, such as MPR and some PPU, measures are likely to be the most informative. Available data suggest that shorter durations of adherence assessment (≤6 months) may be less accurate at predicting virological outcome. Moreover, PAMs are more likely to accurately predict outcomes at the end of a period of adherence assessment than at future time points. Not surprisingly, the balance of studies suggests that PAMs are superior to self-reported adherence in predicting virological outcome. Finally, a threshold effect for mortality is observed at adherence levels of 70%–80%, in contrast to virological outcomes, for which no adherence thresholds were observed.
PAM-based adherence estimates can be used by pharmacists and other health care providers to promote ART adherence. Although the literature on pharmacist-directed interventions is limited, pharmacy-based adherence interventions have successfully combined adherence education [53-55], tailoring regimens to patient lifestyles [54, 56], and the management of adverse drug reactions [55, 56], resulting in improved adherence [53, 54, 56] and improved virological [53, 55] and immunological  response. Further investigation of these interventions is warranted in HICs and in LMICs where similar interventions have not been reported.
Although many studies have assessed various PAMs and their associations with clinical and virological outcomes, significant gaps in our understanding remain. Research is needed to compare different PAMs in the same population and against other adherence measures and biomarkers, such as antiretroviral levels in hair. In addition, the optimal duration of adherence assessment remains to be clarified for different clinical and virological outcomes. Also, because the relationship between adherence and virological outcomes varies over time  and by regimen [6, 50], studies should investigate the predictive value of PAMs in both ART-naive and ART-experienced patients receiving different regimens.
The potential benefit of PAMs includes the identification of individuals at risk for virological failure and undesirable treatment outcomes. Prospective studies incorporating PAMs with interventions designed to improve adherence, clinical outcome, and virological outcome have not been reported but are necessary if PAMs are to be used to optimize clinical care. Researchers attempting to design such studies will face multiple challenges, such as calculating accurate adherence estimates, devising tools for clinicians to easily interpret PAM results, and correctly applying interventions to at-risk patients.
Pharmacy-based methods for estimating adherence during routine clinical care are heterogeneous, yet they predict virological and other clinical outcomes in the majority of studies. Limited comparative data suggest that PAMs are likely superior to self-reported adherence measures. Nevertheless, additional studies are needed to clarify this finding and to identify which PAM parameters are most predictive of clinical or virological outcomes and which measure is best suited to each treatment setting. Available evidence suggests that PAMs are more accurate in predicting current rather than future outcomes and that PAMs applied over shorter durations of adherence assessment (≤6-months) are likely to be less predictive of outcome than PAMs estimated over longer durations. In conclusion, available data suggest that MPR and PPU estimates, which include the number of days for which ART was prescribed, are appropriate minimum-resource methods to assess patient adherence to ART.
Some of the authors are staff members of the World Health Organization. The authors alone are responsible for the views expressed in this publication and they do not necessarily represent the decision or stated policy of the World Health Organization.
Financial support.National Health and Medical Research Council (Postgraduate Scholarship to J.H.M.), National Institutes of Health (5K23AI074423-04 to M.R.J., 5T32AI007438-18 to S.Y.H.)
Potential conflicts of interest.All authors: no conflicts.