Abstract
We followed 205 HIV-infected adults on antiretroviral therapy for at least 12 weeks in a Gambian clinic, where routine viral load monitoring was performed. The 1- and 4-week self-reported adherence and timeliness in keeping to scheduled appointments were recorded at each visit. Seventy patients had measurable viremia between the 12th week and the 3rd year of therapy. Survival analysis of the first detectable viral load on therapy demonstrated an association with 4-week (hazard ratio [HR] 2.6, 95% confidence interval [CI] 1.5-4.3, P = .001) and 1-week (HR 1.9, 95% CI 1.1-3.3, P = .024) self-reported suboptimal adherence and with 1 to 15 days of late presentation for appointments (HR 1.6-1.8, P .027-.109). In a multiple regression model, only 4-week self-reported adherence remained as a significant predictor of viremia.
Introduction
Achieving and maintaining viral suppression is the core aim of antiretroviral therapy (ART) in HIV infection. It promotes CD4 recovery and reduces morbidity and mortality. Typically problems with medication adherence precede viremia and eventually immunologic and clinical failure in people taking ART. When HIV viremia is detected in the course of ART, this permits interventions by health care workers to regain viral control, such as counseling or regimen switches. Frequent viral load monitoring however adds substantial costs. Adherence monitoring, which is essentially free, can provide information prior to the development of viremia upon which health care workers can act.
Of the various ways of monitoring medication adherence, self-reported adherence of one form or another is used most commonly. 1,2 This is despite some studies reporting poor results with self-reported adherence. 3–5 While self-report is biased toward overestimating adherence, it generally performs reasonably well as a screening tool. Lower reported adherence is predictive of viremia on ART 6–8 and may be superior to pharmacy refill data. 9 The optimal reporting interval remains in question; however, 7 or more days may be superior to shorter intervals despite the potential for more accurate recall over shorter intervals. 8,10
Missing appointment is clearly a risk factor for loss of viral control, as it leads to interruptions in ART. Further evidence suggests that missed appointments in and of themselves predict later HIV viremia, 11–13 and it has been suggested that this holds in particular for missing appointments by more than 7 days. 14
Effective real-time adherence monitoring will permit “preemptive interventions” 15 potentially addressing adherence problems prior to the development of viral resistance and a loss of therapeutic options in HIV. Its optimization and implementation alongside such interventions is therefore one of the key priorities currently in the field of ART.
We sought to compare self-reported medication adherence with appointment adherence as a way to predict viremia.
Methods
Setting
The study was conducted at the Genito-Urinary Medicine (GUM) clinic at the Gambia Unit of the Medical Research Council (UK). The clinic provided routine care and follow-up to a population of over 1500 HIV-infected people identified through a combination of STI services, tuberculosis, and general medical care and referrals. Antiretroviral therapy–based care for HIV-infected individuals was available from October 2004 through June 2010. During the time frame of the study, the Gambia had an overall HIV prevalence of approximately 2% based on antenatal clinic surveillance (1.6% HIV-1 and 0.4% HIV-2 in 2008). 16
First-line ART was comprised of zidovudine (ZDV), lamivudine (3TC), and nevirapine (NVP) for those infected with HIV-1 and ZDV, 3TC, and ritonavir-boosted lopinavir (LPV/r) for those with HIV-2 or dual infections. Where clinically indicated, stavudine (d4T) was substituted for ZDV, and efavirenz (EFV) was substituted for NVP. HIV type was determined through a multistep algorithm that has been described previously: initial testing is based on tests performed on 2 specimens drawn at different times, and in the event of a dually reactive specimen or differences between the 2 results, a Pepti-Lav test is used. In instances where the Pepti-Lav result is ambiguous, final diagnosis is made on the basis of polymerase chain reaction. 17 Patients were monitored clinically, with CD4 counts and with routine viral loads. CD4 counts were obtained with an FACScan (Becton-Dickinson, Oxford, UK). Viral loads were performed using a validated locally developed method with internal calibration, external quality control, and an estimated lower limit of detection of 100 copies/mL. For dually infected individuals, generally the HIV-1 viral load was used for the study. In the event of 2 consecutive viral loads over 5000 copies/ml, patients were switched to second-line ART.
Patients were provided sufficient antiretroviral (ARV) drugs for the entire interval through their next scheduled appointment, with extra pills in case they were unable to make it to their scheduled appointment for any reason. The number of days of extra medicines varied, typically at least one-half week, although patients were known on occasion to accumulate larger surpluses.
Data Collection
Baseline information on patients was collected at the time of confirmation of the HIV diagnosis. This information included HIV type, CD4 count, age, and sex. Upon initiation of ART, a new CD4 count was obtained if the prior one had been drawn more than 3 months previously. Patients on ART were seen by protocol at weeks 2, 4, 8, and 12 after initiating ART and at least every 12 weeks thereafter, more frequently if requested by the patient or the physician. CD4 counts and HIV viral loads were performed at weeks 12, 24, and every 24 weeks thereafter or as clinically indicated.
At every visit after the initiation of ART, patients were asked how many times they had missed taking their ARV drugs or taken their ARV drugs late by more than 2 hours in the preceding 7 days and in the preceding 4 weeks (following Mannheimer et al. 18 ) For the 1-week self-report, these were initially recorded as a number (out of 14 doses), while the 4-week self-report was initially categorized as 0, 1 to 3, 4 to 6, 7 to 14, 15 to 28, or more than 28 doses missed “out of about 56 possible.” These data were transcribed using double entry from the written progress note.
Follow-up appointments were provided to patients in writing. Dates of clinic attendance and scheduled follow-up dates were recorded electronically (single data entry).
Patients
Upon enrollment in the clinic, patients gave informed consent for analysis of their anonymized clinical data. Of the 310 treatment-naive adults with a recorded date of starting ART, 205 had viral load data between the 12th week and 3rd year of ART, both 1- and 4-week self-reported adherence data, as well as recorded appointments and visits in each of the intervals between their viral load measurements (total of 1684 visits). Of these, there were 70 with detectable viremia (>100 copies/ml) within 3 years of starting ART.
Statistical Analysis
We obtained hazard ratios (HRs) of viremia on the basis of survival analysis with Cox regression using STATA 10. Patients without a detectable viremia were right censored at their last recorded viral load determination within 3 years of ART initiation, including those that died, were transferred, or were lost to follow-up (LTFU). We tested the proportional hazards assumptions for the time-dependent variables using the scaled Schoenfeld residuals at the 5% significance level. 19
Both 1- and 4-week self-reported adherence at each visit were coded as binary variables (any suboptimal versus perfect adherence). For the survival analyses, self-reported adherence from time of ART initiation up to and including the visit at which the vertical line was drawn was a time-varying factor categorized as 0, 1, or more than 1 episodes of suboptimal adherence. For regressions where the time-varying factor was the number of days by which a patient had ever presented late to an appointment since ART initiation, thresholds of 1, 2, 3, 4, 7, 15, and 30 days were used. This was based on the appointment made at whichever of the 3 prior clinic visits gave the least apparent number of days late, as acute care and follow-up for intercurrent illness occasionally took place between routinely scheduled ART monitoring visits. For the combined analysis, a multiple regression model was used for incorporating 4-week self-reported adherence and days late. For analysis by age, patients were categorized as up to 30, 30 to 45, and older than 45 years. For baseline CD4 counts, patients were categorized as having up to 50, 50 to 250, and >250 cells/mm3 in the 3 months prior to starting ART.
Results
Most patients (75.6%) were infected with HIV-1; 16.1% were infected with HIV-2, and 8.3% were dually infected. Compared to those with HIV-1, the HRs of viremia for those with HIV-2 or dual infection were 1.35 and 2.73, respectively. The majority of the patients (66.3%) were female. The HR of viremia for men was 1.52. At ART initiation, median age was 39 years (range 15-77, interquartile range [IQR] 33-46). Compared to those up to age 30, the HR of viremia was 1.30 for those aged 31 to 45 and 1.23 for those older than 45 years. A baseline CD4 count was available for 104 patients, and the median was 135 cells/mm3 (range 10-470, IQR 50-245). Compared to those with a CD4 count over 250, the HR of viremia was 1.03 for those with a CD4 count between 50 and 250 and 1.61 for those with a CD4 count <50; the HR for the 101 patients without a baseline CD4 count was 1.16, a nonsignificant difference (results not shown). Median duration of follow-up was 227 days (range 84-1084, IQR 126-761). Of the baseline patient characteristics, only dual HIV-1/-2 infection was associated with the development of viremia (see Table 1).
Baseline Characteristics and Hazard Ratios for Viremia.
Abbreviations: CI, confidence interval; HR, hazard ratio.
Compared to intervals with no reports of suboptimal adherence in the preceding week, the HR for viremia at the end of an interval was 1.89 with 1 report of suboptimal adherence and 2.48 with 2 or more such reports. For suboptimal self-reported adherence in the preceding 4 weeks, the HR of viremia, compared to no reports of suboptimal adherence, was 2.55 for 1 report and 2.64 for 2 or more. While suboptimal adherence in recall periods of either 1 week or 4 weeks is associated with subsequent viremia, reporting on the 4 weeks preceding the clinic visit yields stronger associations than a 1-week recall period. The HRs of viremia are larger for multiple reports of suboptimal adherence than for single reports, although this is not as striking for the 4-week recall period. The results for the HRs of viremia based on self-reported adherence in the interval preceding the viral load determination are presented in Table 2.
Hazard Ratios for Viremia by Instances of Suboptimal Adherence Self-Report.
Abbreviations: CI, confidence interval; HR, hazard ratio.
Of the 1684 clinic visits, patients presented late in 1431 (85%) visits by at least 1 day to that or to a prior appointment. There were 1045 visits to which patients had presented late by up to 1 week, to that or to a prior appointment, representing 62% of all visits.
Survival analysis based on the number of days by which one was late for one’s appointment produced similar effect sizes across the range from 1 to 15+ (HR 1.6-1.8), with no effect seen for visits 30+ days after the appointment (HR 0.99). Despite the similarity in effect size, only delays between 7 and 29 days were significant. These results are presented in more detail in Table 3.
Hazard Ratios for Viremia by Days Late for Appointment.
Abbreviations: CI, confidence interval; HR, hazard ratio.
a Any visit late versus no visits late by the specified amount.
In a multiple regression model including both 4-week self-reported adherence and days late, only the self-reported adherence remained significant (results not shown).
Discussion
Both 1- and 4-week self-reported adherence to ART were predictive of HIV viremia, with the 4-week interval showing a stronger association with viremia. Having a history of presenting late for appointments was also associated with viremia. The HR of viremia was relatively stable across the range of time frames studied, with a more robust relationship for delays of 7 to 29 days. As expected, repeated reports of nonadherence were more predictive of viremia than were single reports, although single reports appear to be a reasonable predictor in this cohort.
Importantly, for the interpretation of both survival analyses patients are categorized based upon their cumulative history that is being late for a visit or reporting suboptimal adherence affected the coding of all subsequent visits. This form of coding was chosen to better reflect the cumulative nature of the risk of missed or late doses across a patient’s treatment history, while retaining the longitudinal structure of the data.
Evidence has been conflicting on the value of self-reported nonadherence as well as the optimal method of questioning and the reporting interval. 8 In comparing our results to those of others, our low event threshold for viremia (>100 copies/mL) would tend to increase the odds of small factors producing measurable effects. Moreover, we applied a fairly strict definition of taking one’s ARV drugs on time, within 2 hours, both to reinforce the importance of close adherence and to lower the threshold for reporting adherence difficulties. Whether this tight reporting standard had either of those effects or limits its comparability to other studies involving self-report is unknown.
Patients’ reserves of ARV drugs were not rigorously measured; they were provided ample margins of ARV drugs to minimize the risk of their running out in the event they were late for their scheduled appointment. Programs allowing less stockpiling would presumably force their clients to run out of medicines more frequently and show a stronger effect of the visit delay on viremia. However, as a result of our protocol, we were unable to determine the effect size that running out of pills had on patients’ adherence. Presumably this played a role in the more conclusive effect on adherence seen by patients presenting 7 to 29 days late for their appointment. The fact that the even longer delay of 30+ days has no predictive value is likely a result of such individuals stopping and eventually restarting therapy following additional adherence counseling, whereby their viral load monitoring would be delayed for 12 weeks or more. This is also a different and pharmacokinetically less hazardous pattern of poor adherence than routinely missing occasional doses. The possibility of missing interval visit data is also higher where the apparent visit delay is greater; however, this could not be assessed quantitatively.
The modest association between viremia and late presentation for clinic appointments, even for short delays when patients presumably had not run out of ARV drugs, combined with this effect’s dropping out of multivariate analysis with self-reported adherence, is consistent with missed appointments being a proxy measure of poor adherence, up to a point. This was also the conclusion of a 2-center study in East Africa, where Blacher et al 14 found that presenting late by more than 8 days for one’s appointment was still predictive of failure after controlling for self-report (odds ratio 2.0), 14 as well as a large single-center cohort in South Africa that distinguished between missing visits with a physician and missing prescription refill visits. 13
While these 2 methods of monitoring ART have the advantages of being nearly free and readily available, their predictive power is modest at best. Our data further suggest that the combination of these measures is no better than the self-reported adherence by itself. This could be explained by patients presenting substantially late for their appointments being more likely to report having missed doses. Assessing the number of days by which a patient was late for appointment also depended on accurate recording of appointment dates, which were directly entered by the clinician, as well as visit dates. Reliance on these 2 data points may have had the effect of increasing the uncertainty around timely appointment keeping.
The current study examines viremia rather than established viral failure. This implies that self-reported adherence problems or missing appointments could be used to trigger more in-depth interviewing, adherence interventions, or viral load testing. HIV clinic staff could respond to these markers by taking extra time to explore challenges to adherence and adaptive strategies with these individuals, before failure becomes established with concomitant resistance to therapy. Such “face time,” in addition to its specific benefits as an adherence intervention, might also provide nonspecific benefits in terms of the therapeutic alliance between the health care worker and his or her HIV-positive client. 20–24
Thirty-two percent of ART-eligible people seen at the GUM clinic died or were LTFU prior to ART. 25 Of those that started, 34% were not included in this study because of missing data: they died, were transferred, or were LTFU before having their first viral load determination on therapy, or did not have any self-reported adherence data. Our findings are therefore potentially representative only of those people remaining in care. Moreover, it may not be generalizable to clinics observing different protocols in terms of follow-up or with substantially different patient populations.
A second major limitation to the interpretation of the data is the potential for increased intensity of monitoring of individuals deemed by their physician to be “at risk” for suboptimal adherence to ART or scheduled visits. This potential selection bias is only partially minimized by our use of categorical coding of the key variables.
As monitoring tools, self-reported adherence and timeliness in keeping appointments can flag at-risk patients for adherence interventions. Indirect measures are insufficiently sensitive to replace laboratory monitoring, however, 26 making adoption of point-of-care viral load measures a priority as soon as such tests become available.
Footnotes
Acknowledgments
The authors would like to acknowledge the patients and staff of the GUM clinic, the Data Management staff, in particular Gilleh Thomas, and the virology laboratory, in particular Abraham Alabi and Matthew Cotten, at the MRC’s Gambia Unit, as well as Jef van den Ende for his support and guidance in the analysis of the data.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The care of these patients was made possible by the Medical Research Council and the United Nations Global Fund for AIDS, Tuberculosis, and Malaria. KP was employed during that time by the Medical Research Council and was supported during the write-up of this article by a SOFI-A grant funded by the Flemish regional government and administered by the Institute of Tropical Medicine.