The aim of the study was to evaluate the impact of different patterns of nonadherence on treatment outcomes in patients with long-term follow-up.
The aim of the study was to evaluate the impact of different patterns of nonadherence on treatment outcomes in patients with long-term follow-up.
This cohort study included patients who began highly active antiretroviral therapy during 1996–1999, with the last follow-up in 2007. Adherence was evaluated every 2 months by monitoring of pharmacy refills and by using self-reports. Patients were considered nonadherent at a specific visit when less than 90% of the prescribed drugs had been taken. Adherence was categorized as follows. (A) Continuous adherence: a patient had to be adherent in all of the evaluations throughout the period of follow-up. (B) Treatment interruption: drugs were not taken for more than 3 days, for any reason. Treatment failure was defined as viral load >500 HIV-1 RNA copies/mL or death. Cox proportional risk models were used to calculate adjusted relative hazards (ARHs) of treatment failure.
A total of 540 patients were included in the study, with a median follow-up of 8.3 years. Only 32.78% of patients achieved and maintained continuous adherence, and 42.78% of patients had treatment interruptions. Noncontinuous adherence [ARH 1.48; 95% confidence interval (CI) 1.02–2.14] and treatment interruptions (ARH 1.39; 95% CI 1.04–1.85) were associated with treatment failure for the overall cohort; however, for patients with more than 3 years of follow-up, only treatment interruptions were independently associated with treatment failure.
Only one-third of patients managed to achieve continuous adherence, and almost half of the patients had treatment interruptions, which have a particularly marked effect on treatment outcomes over the long term.
The goals of HIV infection treatment are to suppress viraemia, to improve immune function and to delay disease progression over the long term [1,2]. Different factors affect the effectiveness of antiretroviral therapy (ART), such as the level of plasma HIV RNA, the degree of immunodeficiency and drug resistance; however, nonadherence may be the most important challenge to achieving the goals of ART.
Previous studies have emphasized the effect of adherence on virological, immunological and survival endpoints [3–8].
A high level of adherence to combined ART is essential to minimize the risk of treatment failure. However, the level of adherence necessary to obtain and maintain virological suppression may vary according to the type of ART [9,10].
Nonadherence is a dynamic process; it varies over time, and may be expressed in different forms, such as a ‘missed dose phenomenon’ [11,12] or the interruption of therapy. Many studies on adherence in HIV-infected patients have focused on quantitative and global measures, and only a few studies have reported data beyond 3 years of follow-up [13–15].
The aim of this study was to evaluate the impact of different patterns of nonadherence (missed doses and treatment interruptions) on the risk of treatment failure in HIV-infected patients with long-term follow-up.
This cohort study analysed data of HIV-infected patients older than 18 years of age who began highly active antiretroviral therapy (HAART) between 1996 and 1999 at the HIV Unit of the Hospital del Mar in Barcelona, Spain. The last follow-up was in July 2007. Participating patients were required to have had at least one adherence evaluation to be included in the study.
Patients on ART were given appointments to obtain antiretroviral drugs at the hospital pharmacy department every 2 months. All patients obtained their drugs at a single dispensing site. The hospital pharmacist in charge of antiretroviral drug delivery carried out the adherence study. Two pharmacists were sequentially involved throughout the study. A computer-assisted pharmacy dispensing system was used to calculate pharmacy claim adherence. Adherence was also assessed at each pharmacy visit by self-report, by asking the patients how many doses had been missed and how the missed doses occurred, as occasional omissions or treatment interruptions, in the month before the visit. To obtain this information, semi-structured interviews of patients were conducted by pharmacists specializing in HIV therapy. A composite measure of adherence was used, whereby patients were considered nonadherent at a specific visit when they declared that they took <90% of the total dose of prescribed antiretroviral drug(s) and/or the pharmacy claim percentage was <90%. An overall adherence measure was obtained for the first year of antiretroviral therapy and for the year previous to the last visit.
The adherence assessment was categorized as follows. (A) Continuous evaluation, which assessed adherence throughout the follow-up period; to be considered continuously adherent, a patient had to be adherent in all of the evaluations. (B) Interruption of treatment, which was reported when a patient had not taken any antiretroviral therapy for more than 3 days, for any reason, at any time during the follow-up; interruption was divided into two groups: interruption as a result of nonadherence and interruption as a result of medical advice. The above-mentioned categories were analysed as a dichotomic variable.
Treatment failure was defined as death or viral load >500 HIV-1 RNA copies/mL at the last evaluation. The study addressed global failure at the end of follow-up for each patient (death or viral load >500 copies/mL); not only the first failure was evaluated.
In addition to treatment failure and adherence, the following variables were also included in the analysis: gender; age; HIV exposure [injecting drug user (IDU) and others]; CD4 count, categorized as <200, 201–350 and >350 cells/mL; and HIV RNA, categorized as >5, 3–5 and <3 log10 copies/mL.
All patients included in the study underwent the same type of ART, initiating HAART with a protease inhibitor plus two nucleoside reverse transcriptase inhibitors (NRTIs). In addition, the last HAART (when evaluation of failure was carried out) was classified as NRTI-based, nonnucleoside reverse transcriptase inhibitor (NNRTI)-based, boosted protease inhibitor-based, or protease inhibitor-based therapy, respectively; these categories were included in the analysis.
The dependent variable in this study was treatment failure. In univariate analyses, survival curves were obtained using the Kaplan–Meier method. Comparison of failure functions was carried out by estimating and testing associated relative hazards by proportional hazard regression.
In multivariate analyses, Cox proportional risk models were used to calculate relative hazards of treatment failure and 95% confidence intervals (CIs) were calculated. Predictor variables found to be significant (P<0.05) in univariate analysis were added to the multivariate model, as were those that were not significant because of their potential confounding effect.
The date of any patient death was censored. Patient survival was recorded on the date of the last follow-up visit.
A total of 590 patients initiated HAART during the period 1996–1999, and 540 patients were included in the study. The causes of exclusions were that the patient did not have at least one adherence evaluation, or that there were missing data on CD4 cell count or viral load at baseline. The median follow-up period was 8.3 years [interquartile range (IQR) 3.13–9.30]. The median number of adherence assessments per patient was 48 (IQR 18–54). The mean patient age was 36.25 years, and 69% of the patients were male. The median CD4 count was 233 cells/μL, and the median viral load was 4.32 log10 copies/mL. A total of 265 patients (49.07%) acquired HIV infection by injecting drug use. The HAART regimens used were as follows: NRTI-based, 65 (12%); NNRTI-based, 156 (28.9%); protease inhibitor-based, 172 (31.9%); and boosted protease inhibitor-based, 147 (27.2%). During follow-up, 196 patients (36.3%) remained on their initial regimen, 127 (23.5%) had one switch, 107 (20.4%) had two switches, and 110 (20.4%) had three or more switches for any reason.
There were 90 deaths (16.7%) and 221 patients (40.9%) had a viral load >500 copies/mL; a total of 250 patients (46.3%) had treatment failure. The number of patients lost to follow-up was 120 (22.2%).
The proportion of patients considered to have appropriate adherence (>90%) after the first year of combined ART was 58.1%, and for the last year of therapy this proportion was 62%. For patients with more than 3 years of follow-up, 29.3% of those initially considered adherent became nonadherent, and 43.3% of those considered nonadherent became adherent.
The relationship between type of HAART and appropriate adherence was as follows: 55.4% of patients on triple NRTI-based regimens, 67.9% of those on NNRTI-based regimens, 40.1% of those on unboosted protease inhibitor-based regimens, and 59.2% of those on boosted protease inhibitor-based regimens were adherent. The relative risks of nonadherence, taking boosted protease inhibitor-based therapy as the reference, were 0.86 (95% CI 0.47–1.58) for NRTI-based regimens, 0.54 (95% CI 0.33–0.89; P<0.01) for NNRTI-based regimens, and 2.09 (95% CI 1.39–3.27; P<0.001) for unboosted protease inhibitor-based regimens.
Only 177 patients (32.78%) fulfilled the definition of continuous adherence, and 231 patients (42.78%) had treatment interruptions; of the 231 patients with treatment interruptions, 93 (17.2%) were nonadherent and 138 (25.6%) interrupted treatment following their physician's advice.
The proportion of patients with treatment failure according to the different factors analysed is shown in Table 1.
|>45 years (n=74)||24 (32.4)||50 (67.6)||0.02|
|31–44 years (n=336)||158 (47)||178 (53)|
|18–30 years (n=130)||68 (52.3)||62 (47.7)|
|>350 cells/mL (n=180)||72 (40)||108 (60)||0.01|
|200–349 cells/mL (n=120)||50 (41.7)||70 (58.3)|
|<200 cells/mL (n=236)||127 (53.8)||109 (46.2)|
|<3 log10 copies/mL (n=252)||110 (43.7)||142 (56.3)||0.05|
|3–5 log10 copies/mL (n=184)||85 (46.2)||99 (53.8)|
|>5 log10 copies/mL (n=104)||60 (57.6)||44 (42.3|
|Female (n=167)||83 (49.7)||84 (50.3)||0.29|
|Male (n=373)||167 (58.6)||206 (55.2)|
|Injecting drug use|
|No (n=279)||97 (34.5)||182 (65.5)||0.001|
|Yes (n=261)||153 (58.6)||108 (41.4)|
|Type of HAART|
|Boosted PI (n=147)||48 (32.7)||99 (67.3)||0.001|
|NRTI (n=65)||27 (41.5)||38 (58.5)|
|NNRTI (n=156)||50 (32.1)||106 (67.9)|
|PI (n=172)||125 (72.7)||47 (27.3)|
|Yes (n=177)||46 (26)||131 (74)||0.001|
|No (n=363)||204 (56.2)||159 (43.8)|
|No (n=309)||113 (36.6)||196 (63.4)||0.001|
|Yes (n=231)||137 (59.3)||94 (40.7)|
The crude relative hazard of treatment failure for all patients included in the study (n=540) is shown in Table 2. The factors significantly related to treatment failure were: younger age; CD4 count lower than 200 cells/μL; viral load greater than 5 log10 copies/mL; injecting drug use; and ART based on unboosted protease inhibitors. Nonadherence measures were related to treatment failure.
|>45 years (n=74)||1||1|
|31–44 years (n=336)||1.64 (1.06–2.51)||0.02||1.31 (0.83–2.09)||0.25|
|18–30 years (n=130)||1.82 (1.14–2.91)||0.01||1.47 (0.88–2.42)||0.13|
|>350 cells/mL (n=179)||1||1|
|200–349 cells/mL (n=121)||1.06 (0.74–1.52)||0.76||1.19 (0.82–1.73)||0.35|
|<200 cells/mL (n=236)||1.43 (1.07–1.92)||0.02||1.52 (1.10–2.09)||0.01|
|<3 log10 copies/mL (n=252)||1||1|
|3–5 log10 copies/mL (n=184)||1.12 (0.84–1.49)||0.43||1.08 (0.80–1.44)||0.62|
|>5 log10 copies/mL (n=104)||1.47 (1.06–2.06)||0.02||1.35 (0.95–1.91)||0.38|
|Male (n=373)||0.90 (0.69–1.17)||0.43|
|Injecting drug use|
|Yes (n=261)||1.89 (1.44–2.48)||0.001||1.41 (1.07–1.86)||0.01|
|Type of HAART|
|Boosted PI (n=147)||1||1|
|NRTI (n=65)||1.45 (0.90–2.34)||0.13||1.52 (0.94–2.47)||0.09|
|NNRTI (n=156)||1.12 (0.75–1.66)||0.58||1.41 (0.94–2.10)||0.09|
|PI (n=172)||5.95 (4.23–8.35)||0.001||6.17 (4.35–8.74)||0.001|
|No (n=363)||2.52 (1.82–3.49)||0.001||1.48 (1.02–2.14)||0.04|
|Yes (n=231)||1.54 (1.20–1.98)||0.001||1.39 (1.04–1.85)||0.03|
The factors related to treatment failure in the adjusted analysis (see Table 2) were CD4 count lower than 200 cells/μL; injecting drug use; therapy based on unboosted protease inhibitors; noncontinuous adherence; and treatment interruptions.
We identified 153 patients (28.3%) who had noncontinuous adherence but no treatment interruptions; in this group of patients, nonadherence was not associated with treatment failure. The crude relative hazards of treatment failure for patients with treatment interruptions were: no interruption, 1; interruption as a result of nonadherence, 3.37 (95% CI 2.51–4.53; P<0.0001); interruption as a result of medical advice, 0.96 (95% CI 0.70–1.32; P=0.81). The adjusted relative hazards (ARHs) were: no interruption, 1; interruption as a result of nonadherence, 3.62 (95% CI 2.37–5.52; P<0.0001); interruption as a result of medical advice, 1.36 (95% CI 0.91–2.04; P=0.13).
The ARHs of factors associated with treatment failure for patients with more than 3 years of follow-up (n=412) were: ART based on unboosted protease inhibitors, 3.40 (95% CI 2.19–5.30; P<0.0001); injecting drug use, 1.58 (95% CI 1.11–2.26; P<0.01); and treatment interruption, 3.45 (95% CI 2.25–5.30; P<0.0001). The ARHs of the different types of treatment interruption were: interruption as a result of nonadherence, 5.49 (95% CI 3.18–9.46; P<0.0001); interruption as a result of medical advice, 2.20 (95% CI 1.36–3.58; P<0.001). Continuous adherence did not reach statistical significance as an independent risk factor for treatment failure (AHR 1.21; 95% CI 0.74–1.98; P=0.44).
This study of a large cohort of HIV-infected patients treated with combined ART and followed for a long period (median 8.3 years) shows that adherence is a dynamic process, and that different patterns of nonadherence at different times may have different impacts on ART outcomes.
The proportions of adherent patients were 58.1 and 62% for the initial and last evaluations, respectively. These rates are lower than those reported in other studies in developed countries in which pharmacy records were used. Hogg et al.  reported that, among 1282 HIV-1-infected individuals, 74% had adherence >75 and 57% had adherence >95% in their first year on HAART, based on pharmacy refill data. In addition, in a 1-year retrospective review of pharmacy refill data for 100 patients in Canada, Ostrop et al.  reported >80% adherence in 75% of patients. In a large African cohort study, half of enrolled patients claimed more than 80% of their prescriptions, and one-third claimed >60% . Using electronic monitors for adherence evaluation the mean adherence reported ranged from 54 to 84% [3,18–20]. This discrepancy is likely to reflect differences in adherence measurement tools used or a selection bias in the different populations studied.
In the present study, only 32.8% of patients could be considered adherent in the continuous evaluation, and 42.8% of patients had treatment interruptions; 17.2% of patients were truly nonadherent. A French cohort study, in which longitudinal adherence was measured using self-report for 20 months, reported that 31.4% of patients were highly adherent in all of the measurements . Other studies have shown that the trend is for adherence to decrease over time [15,22,23]; these studies have also emphasized the importance of longitudinal evaluation. Treatment interruptions or so-called ‘drug holidays’ have not been analysed in depth in the adherence literature. Such ‘drug holidays’ have usually been defined as 1 or 2 days without taking any antiretroviral drug, with prevalence reported as 5.8% in the Swiss HIV cohort study , 15% in a large Spanish study (12), and 27% in a French study . In the present study, the proportion of treatment interruptions was higher than previously reported in other studies; the explanation for this may be that the measure included any patient who, for any reason (adherence, physician advice, and adverse event), interrupted therapy.
Concerning therapy, compared with boosted protease inhibitor regimens, unboosted protease inhibitor therapy was associated with a higher risk (2.09; 95% CI 1.39–3.27) and NNRTI-based therapy with a lower risk (0.54; 95% CI 0.33–0.89) of nonadherence. One explanation may be that NNRTI-based therapy is generally simpler than boosted protease inhibitor-based therapy.
Nonadherence patterns have differing impacts on treatment outcome, depending on the duration of follow-up. For the entire group, treatment interruptions (only those resulting from nonadherence) and noncontinuous adherence were independently associated with treatment failure in the adjusted analysis. For the group of patients with more than 3 years of follow-up, treatment interruptions, as a result of nonadherence and medical advice, were independently related to treatment failure. This may be because a longer exposure to antiretroviral therapy carries a greater risk of interruption, regardless of the cause, and a greater risk of resistance development. A low level of continuous adherence was an independent risk factor for treatment failure only if treatment interruption was excluded from the Cox regression model (AHR 2.13; 95% CI 1.36–3.33; P<0.001).
Our study has several limitations. First, adherence was assessed using data on pharmacy refills, which is not a ‘gold standard’ measure; it provides no certain evidence that pharmacy claim data reflect the number of pills taken correctly by a patient, however, in our study, pharmacy data was contrasted by patient interviews. Other studies have found that pharmacy records correlate well with other adherence measures, such as appointment keeping and medication consumption , use of electronic monitors , drug resistance , viral load suppression  and survival [7,8,17].
Secondly, because adherence data were reported only in aggregate form, we could not measure individual adherence in a time-dependent fashion. Finally, adherence was analysed as a dichotomic variable, and clearly it is a continuous one.
In conclusion, our qualitative study may have clinical implications. We found that only one-third of patients achieved a continuous high level of adherence, but failure to achieve a continuous high level of adherence had less impact on the outcome of ART than treatment interruptions.
The authors thank Laurie Covens for his assistance with the English version of the manuscript.