Abacavir and risk of myocardial infarction in HIV-infected patients on highly active antiretroviral therapy: a population-based nationwide cohort study

Authors


Dr Niels Obel, Department of Infectious Diseases, Copenhagen University Hospital, Copenhagen, Denmark. Tel: +45 35 45 77 30; fax: +45 35 45 77 50; e-mail: niels.obel@rh.regionh.dk

Abstract

Objective

The aim of the study was to examine whether exposure to abacavir increases the risk for myocardial infarction (MI).

Design, setting and subjects

This was a prospective nationwide cohort study which included all Danish HIV-infected patients on highly active antiretroviral therapy (HAART) from 1995 to 2005 (N=2952). Data on hospitalization for MI and comorbidity were obtained from Danish medical databases. Hospitalization rates for MI after HAART initiation were calculated for patients who used abacavir and those who did not. We used Cox's regression to compute incidence rate ratios (IRR) as a measure of relative risk for MI, while controlling for potential confounders (as separate variables and via propensity score) including comorbidity.

Main outcome

Relative risk of hospitalization with MI in abacavir users compared with abacavir nonusers.

Results

Hospitalization rates for MI were 2.4/1000 person-years (PYR) [95% confidence interval (CI) 1.7–3.4] for abacavir nonusers and 5.7/1000 PYR (95% CI 4.1–7.9) for abacavir users. The risk of MI increased after initiation of abacavir [unadjusted IRR=2.22 (95% CI 1.31–3.76); IRR adjusted for confounders=2.00 (95% CI 1.10–3.64); IRR adjusted for propensity score=2.00 (95% CI 1.07–3.76)]. This effect was also observed among patients initiating abacavir within 2 years after the start of HAART and among patients who started abacavir as part of a triple nucleoside reverse transcriptase inhibitor (NRTI) regimen.

Conclusions

We confirmed the association between abacavir use and increased risk of MI. Further studies are needed to control for potential confounding not measured in research to date.

Introduction

The prognosis of HIV-infected patients has improved dramatically since the introduction of highly active antiretroviral therapy (HAART) [1]. At the same time, evidence is strong that the risk of myocardial infarction (MI) in HIV-infected patients on HAART is twice as high as in the general population [2]. The biological mechanisms underlying the association remain controversial [2,3]. One hypothesis is that the increased risk of MI is caused by HAART-induced dyslipidaemia. However, the risk of MI increases immediately after initiation of HAART, suggesting that factors other than changes in blood lipids are operative [2,4].

A recent paper from the Data Collection on Adverse Events of Anti-HIV Drugs (DAD) study showed that treatment with protease inhibitors (PIs) increased the risk of MI by 16% with each year of exposure [5]. In a further exploratory analysis of the same data, the authors unexpectedly found that MI risk among patients with recent abacavir use was 1.90 times higher than among patients receiving HAART without abacavir [6]. The results were later confirmed in a paper from the SMART study [7].

Using a Danish nationwide cohort of HIV-infected patients, we estimated the impact of abacavir treatment on the risk of hospitalization with MI. This nonrandomized cohort study may be subject to the same confounders as those potentially affecting the results of the DAD study. For this reason we used several approaches to control for confounding, including propensity score adjustment.

Methods

Setting

Denmark has a population of 5.4 million and the estimated prevalence of HIV infection in the adult population is 0.07% [8]. Denmark's tax-funded health care system provides antiretroviral treatment free of charge to all HIV-positive residents. Treatment of HIV infection is restricted to eight specialized medical centres, where patients are seen on an out-patient basis at intended intervals of 12 weeks. During our study period, national criteria for HAART initiation were any of the following: presence of an HIV-related disease, acute HIV infection, pregnancy, CD4 cell count <300 cells/μL, and, until 2001, plasma HIV-RNA >100 000 copies/mL. Structured treatment interruptions have generally not been recommended in Denmark.

Study population and data collection

The Danish HIV Cohort Study, which has been described elsewhere, includes all HIV-infected patients treated in the eight specialized HIV centres since 1 January 1995 [8,9]. The current study included the 2952 Danish residents in the Danish HIV Cohort who were (1) diagnosed with HIV infection before 1 January 2005; (2) lived in Denmark on the date of HIV diagnosis; and (3) were older than 15 years of age at the time of HAART initiation.

HAART was defined as a treatment regimen of at least three antiretroviral drugs or a treatment regimen including a combination of a nonnucleoside reverse transcriptase inhibitor (NNRTI) and a boosted PI.

Death and emigration status were ascertained from the Danish Civil Registration System, which has tracked the vital status of all Danish residents since 1968 [10]. Almost 14% of the patients in The Danish HIV Cohort Study are included in the DAD study.

Hospitalization with acute MI

Hospitalization data for cohort members were obtained from the Danish National Hospital Registry (DNHR), which was established in 1977 and covers hospitalizations at all acute care hospitals in the country [10]. The registry maintains a record of all in-patient diagnoses [coded according to the International Classification of Diseases 8th revision (ICD-8) until the end of 1993, and according to ICD-10 thereafter] [11]. Out-patient contacts and emergency room visits were added on 1 January 1995. We defined the study endpoint as a first-time hospital diagnosis of MI (code 410.09 or 410.99 in ICD-8; codes I21.0 to I22.9 in ICD-10). We also extracted data from the DNHR on diagnoses of heart diseases other than the study outcome and on comorbidities known to be risk factors for ischemic heart disease: diabetes, alcoholism, chronic obstructive lung disease, hypertension, liver disease and kidney disease.

Covariates

The following covariates were considered for inclusion in the regression models described below: age at start of HAART (grouped in quartiles: <32, 33–38, 39–46 and >46 years), gender, year of HIV diagnosis (before vs. after 1 January 1995), year of HAART initiation (before vs. after 1 January 1998), CD4 count at start of HAART (≤200 vs. >200 cells/μL), viral load at start of HAART (>100 000 vs. ≤100 000 HIV-1 RNA copies/mL), Caucasian race (yes/no) and route of infection (injecting drug use vs. other). Dates of initiation of the following antiretroviral drugs (widely used in Denmark) were treated as time-dependent variables: zidovudine, stavudine, didanosine, lamivudine, tenofovir, efavirenz, nevirapine, ritonavir, saquinavir, indinavir, lopinavir, and atazanavir. Also, a variable indicating the presence of each comorbidity prior to HAART initiation was included in the models.

Exposure of interest

We aimed to investigate whether use of abacavir was associated with increased risk of MI. The presence of abacavir treatment was introduced as a time-dependent variable thereby classifying observation time into exposed and unexposed to abacavir.

Statistical analysis

We calculated person-years (PYR) of follow-up from the date of HAART initiation until the recorded date of death, emigration, loss to follow-up, 1 May 2007, or first diagnosis of MI, whichever came first. Patients diagnosed with MI before HAART initiation were excluded.

The analysis was conducted in four steps. First, we calculated the incidence [with 95% confidence intervals (CIs)] of the first hospitalization with MI, comparing periods before and after first initiation of abacavir treatment. We then fitted a Cox's regression model to compute the incidence rate ratio for the first hospitalization with MI, as an estimate of relative risk controlling for confounding. We assessed the proportional-hazards assumption with plots and tests based on smoothed-scaled Schoenfeld residuals. In these analyses exposure to abacavir treatment was introduced as a time-dependent variable from date of first exposure to abacavir until end of study.

Secondly, we performed an analysis in which time on and time off abacavir were included in the same model. For abacavir-exposed patients, time on this medication was calculated as the period from the initiation of abacavir until 6 months after its discontinuation, and time off abacavir was calculated from 6 months after its discontinuation until either reinitiation of abacavir therapy or the end of the observation period (in accordance with the DAD study). All treatment periods were included in these analyses.

Thirdly, we undertook an analysis in which the start date of abacavir therapy was introduced as two time-dependent variables: (1) date of initiation of abacavir therapy as a part of a triple nucleoside reverse transcriptase inhibitor (NRTI) regimen (mainly trizivir) not containing a PI or an NNRTI; and (2) date of initiation of abacavir therapy as part of a PI- or an NNRTI-containing regimen. These analyses were performed because PI-sparing HAART regimens may have been preferred for treatment of HIV-infected patients with increased risk of heart disease.

Fourthly, because abacavir is used as a second-line drug in many settings, we performed an analysis in which the start date of abacavir therapy was introduced as two other time-dependent variables: (1) start date of abacavir therapy in cases in which it was initiated <2 years after the start of HAART; (2) start date of abacavir in cases in which it was initiated 2 or more years after the start of HAART. The cut-off of 2 years was chosen because most HAART-naïve patients who were due to initiate the recommended regimen in Denmark (abacavir, lamivudine and efavirenz) were first started on zidovudine and subsequently switched to abacavir. This was done in an attempt to lower the risk of hypersensitivity reactions. We calculated the number of patients initiating abacavir treatment within 2 years after starting HAART vs. the number initiating abacavir treatment >2 years after staring HAART, as well as the observation times represented by these treatment periods (calculated as cumulative time from the initiation of abacavir treatment until the end of the observation period). For comparison, we conducted the same analysis for zidovudine, stavudine, didanosine and lamivudine.

As a supplementary analysis we used the propensity-score matching method [12]. Using logistic regression, we calculated each patient's predicted probability of being treated with abacavir based on the patient's covariate pattern. Covariates included in this model were age at the start of HAART (grouped in quartiles: <32, 33–38, 39–46 and >46 years), gender, year of HIV diagnosis (before vs. after 1 January 1995), year of HAART initiation (before vs. after 1 January 1998), CD4 count at start of HAART (≤200 vs.>200 cells/μL), viral load at start of HAART (>100 000 vs. ≤100 000 copies/mL), Caucasian race (yes/no), route of infection (injecting drug use vs. other), heart diseases other than the study outcome, and the presence of comorbidities at HAART initiation (diabetes, alcoholism, chronic obstructive lung disease, hypertension, liver disease and kidney disease). Model fit was assessed using goodness-of-fit statistics (Pearson χ2, P=0.4; Hosmer and Lemeshow test, P=0.07). We found 1761 abacavir users and 1191 nonusers. We were able to match 1126 abacavir users to appropriate nonusers (94.5% of possible pairs), thereby eliminating differences in the propensity score between the users and nonusers. For most covariates the standardized difference in percentage between abacavir users and nonusers was reduced after matching. After matching there were no statistically significant differences between users and nonusers for route of infection, age and viral load at start of HAART. Finally, we repeated the Cox regression analyses for the seven models in Table 2.

Table 2.   Relative risk (RR) of myocardial infarction in HIV-infected patients exposed to abacavir
 Unadjusted RR and 95% CIAdjusted RR and 95% CI (Cox regression model)§Adjusted RR and 95% CI (propensity score model)§
  • *

    Time measured from start of abacavir therapy to end of observation period.

  • Time measured from initiation of abacavir to 6 months after cessation of abacavir or end of observation period.

  • Time measured from 6 months after cessation of abacavir to next exposure or end of observation period.

  • §

    § The following covariates were included in the multivariate analyses: age at start of HAART, gender, year of diagnosis, year of HAART initiation, CD4 cell count and viral load at start of HAART, race, injecting drug use, and use of other antiretroviral drugs and comorbidity at start of HAART.

  • CI, confidence interval; HAART, highly active antiretroviral therapy; NRTI, nucleoside reverse transcriptase inhibitor; NNRTI, nonnucleoside reverse transcriptase inhibitor; PI, protease inhibitor.

Abacavir exposed*2.22 (1.31–3.76)2.00 (1.10–3.64)2.00 (1.07–3.76)
Actual abacavir use2.30 (1.34–3.97)1.95 (1.05–3.60)1.94 (1.01–3.72)
Earlier exposed to abacavir, but no actual use1.84 (0.74–4.58)2.37 (0.88–6.36)2.37 (0.82–6.85)
Abacavir initiated as part of triple NRTI regimen and no NNRTI or PI1.85 (0.91–3.75)1.91 (0.88–4.17)1.76 (0.73–4.28)
Abacavir initiated with an NNRTI or PI2.48 (1.38–4.46)2.06 (1.06–4.01)2.14 (1.07–4.30)
Abacavir initiated within 2 years after HAART initiation1.81 (0.90–3.61)1.77 (0.82–3.82)1.73 (0.76–3.92)
Abacavir initiated more than 2 years after HAART initiation2.71 (1.40–5.25)2.66 (1.31–5.39)2.76 (1.30–5.87)

We analysed data using sas software, version 9.1.3 (SAS Institute, Cary, NC, USA).

The study was approved by the Danish Data Protection Agency.

Results

The study cohort consisted of 2952 HIV-infected patients, of whom 2257 were men (Table 1). Twenty-two patients with an MI prior to HAART initiation were excluded. Nearly 60% of patients exposed to HAART initiated abacavir during the study period of 19 124 PYR and almost one-third of the observation time was obtained from individuals after first exposure to abacavir. 97.4% of the patients had complete follow-up.

Table 1.   Characteristics of the study population
  1. HAART, highly active antiretroviral therapy; IQR, interquartile range.

Number of patients in the study2952
Person-years of follow-up19 124
Person-years of follow-up before initiation of abacavir12 808
Person-years of follow-up after initiation of abacavir6317
Ever exposed to abacavir in HAART [n (%)]1761 (59.6)
Deceased during follow-up [n (%)]511 (17.3)
Emigrated during follow-up [n (%)]71 (2.4)
Lost to follow-up [n (%)]7 (0.2)
Age at initiation of HAART (years) [median (IQR)]39.1 (33.0-46.6)
Male patients [n (%)]2257 (76.4)
Infection mode [n (%)] 
 Homosexual contact1394 (47.2)
 Heterosexual contact1072 (36.3)
 Injecting drug use315 (10.7)
 Other/unknown171 (5.8)
Caucasians [n (%)]2417 (81.9)
Diagnosed with HIV before 1 January 1995 [n (%)]1336 (45.3)
Start of HAART treatment before 1 January 1998 [n (%)]1182 (40.0)
First MI after HAART initiation (n)67
First MI after HAART initiation and before initiation of abacavir (n)31
First MI after initiation of abacavir (n)36
Time from HIV-positive diagnosis to HAART initiation (years) [median (IQR)]3.3 (0.2-8.2)
Viral load at HAART initiation (log10 copies/mL) [median (IQR)]4.7 (4.1-5.4)
CD4 count at HAART initiation (cells/μL) [median (IQR)]190 (78-290)

We observed 67 MIs in the study period, of which 36 occurred after initiation of abacavir. The overall hospitalization rate for MI was 3.5/1000 PYR (95% CI 2.8–4.5). Prior to initiation of abacavir the incidence rate was 2.4/1000 PYR (95% CI 1.7–3.4) and after initiation of abacavir it was 5.7/1000 PYR (95% CI 4.1–7.9). In the unadjusted analysis, the relative risk of hospitalization with MI after initiation of abacavir was 2.22 (95% CI 1.31–3.76) (Table 2). The relative risk adjusted for potential confounders was 2.00 (95% CI 1.10–3.64) and the relative risk adjusted for confounding using propensity scores was also 2.00 (95% CI 1.07–3.76) (Table 2).

The estimated relative risk of MI did not substantially change when time on abacavir was introduced into the model as separate periods from the initiation of abacavir therapy until 6 months after its discontinuation. The risk of MI also remained elevated after cessation of abacavir (Table 2). Further, we found no major difference in estimates between patients who initiated abacavir therapy in the first 2 years after the start of HAART and patients starting abacavir as part of a triple NRTI regimen (Table 2). Almost two-thirds of patients initiated abacavir therapy 2 or more years after initiation of HAART, a marked difference from use of other NRTIs (Table 3).

Table 3.   Number of patients initiating therapy with an antiretroviral drug during and after 2 years of highly active antiretroviral therapy (HAART), and the observation time represented by these patients
DrugNumber of patients initiating treatment with an NRTI in the first 2 years of HAART (% of total number of patients initiating the drug) (observation period in PYR, % of observation time)Number of patients initiating treatment with an NRTI more than 2 years after start of HAART (% of total number of patients initiating the drug) (observation period in PYR, % of observation time)
  1. NRTI, nucleoside reverse transcriptase inhibitor; PYR, person-years.

Abacavir591 (34%) (2521 PYR, 40%)1170 (66%) (3795 PYR, 60%)
Zidovudine2647 (98%) (18 815 PYR, 98.6%)64 (2%) (275 PYR, 1.4%)
Lamivudine2803 (98%) (18 381 PYR, 98.5%)64 (2%) (273 PYR, 1.5%)
Stavudine868 (84%) (6134 PYR, 86.5%)163 (14%) (960 PYR, 13.5%)
Didanosine617 (76%) (5125 PYR, 85%)196 (24%) (913 PYR, 15%)

Discussion

We conducted a cohort study of all Danish HIV-infected patients treated with HAART to examine the impact of abacavir treatment on risk of a first hospitalization with MI. We confirmed the finding of the DAD study of an increased risk of MI after initiation of abacavir therapy [6].

The major strengths of the study are its nationwide population-based design, combined with long and nearly complete follow-up. We were also able to follow the study patients from the time of HAART initiation.

The study has several potential weaknesses that merit discussion. We relied on registry-based discharge diagnoses to identify first-time hospital diagnoses of MI. While discharge diagnoses in general may not be entirely accurate, registration of MI has been shown to be valid [13]. Although we missed patients who died of MI before hospitalization, we assume that rates of pre-hospitalization death are not likely to differ by receipt of abacavir therapy and do not expect potential underreporting of MI-related deaths before hospitalization to bias our relative risk estimates. We also obtained data on comorbidity from the DNHR. This registry includes all in-patient and out-patient hospital contacts in Denmark. As almost all patients with serious diseases are treated in the Danish hospital system, we consider that it is reasonable to assume that this information gives reliable estimates of comorbidity. We lacked data on certain risk factors for ischemic heart disease, such as serum cholesterol and smoking, but had access to all hospital diagnoses registered in the DNHR and were able to adjust our estimates for several important confounders. We thus expect that our adjusted estimates of relative risk of MI associated with abacavir initiation are robust. Still, some unmeasured or residual confounding may have influenced our risk estimates [14]. It is also important to note that our study cohort was generated in the same era of HAART as the DAD cohort and thus may be subject to the same confounding. Previous reports on the effects of abacavir in observational and randomized studies have been conflicting. Abacavir has been linked to greater risk of lipoatrophy in two observational cohort studies [15,16], but this effect was not confirmed in subsequent randomized trials [17–21].

We found that abacavir was used mainly as a second-line therapy, in contrast with other NRTIs. In 2002 it was shown that substitution of zidovudine and stavudine with abacavir partly reversed lipoatrophy [21] (routine pre-emptive switching from thymidine analogues was first instituted later). Furthermore, abacavir is one component in the formulation of trizivir, which is often given to noncompliant patients [22]. Abacavir, as a new NRTI, was also frequently included in second-line regimens for virological failure. Therefore, in the first part of the study period, abacavir was used mainly in second-line regimens for patients with metabolic problems and adherence problems, factors that may be associated with increased risk of cardiovascular disease. This may have generated a scenario prone to confounding by indication, in which patients with an a priori higher risk of cardiovascular disease were prescribed abacavir.

In recent years, both Danish and international recommendations have included abacavir, efavirenz and a third NRTI as one of the preferred first-line regimens. Because efavirenz and abacavir increase the risk of skin reactions, patients needing HAART often start with other NRTIs and subsequently substitute them with abacavir. Thus, the group of patients in our cohort whose first HAART regimen contained abacavir was too small to allow a subgroup analysis of MI risk. As a surrogate analysis, we estimated MI risk in patients who started abacavir therapy in the first 2 years after initiation of HAART. We also found an increased risk of MI in this group.

A major concern is that the increased risk of cardiovascular disease found in abacavir-exposed patients results from a ‘channelling bias’ [23]. However, we still observed an increased risk of MI in patients who initiated abacavir within 2 years after initiation of HAART, arguing against such an effect. Also, patients who initiated abacavir as part of a treatment with three NRTIs had an increased risk of MI. In contrast to the DAD study, we saw an increased risk of MI in patients who were off abacavir for over 6 months. Although this estimate is imprecise, it may indicate that either the abacavir effect lingers for a long period after discontinuation of the drug or that the estimate remains substantially confounded, for example by ‘channelling bias’.

To further control for the effect of potential confounding, we supplemented our analyses with propensity score-based confounding adjustment. This step did not identify any factors explaining the increased risk of MI in abacavir-exposed patients. While safety analyses from randomized trials have not indicated effects of abacavir treatment on risk of MI, these studies were not designed to study potential cardiovascular effects of this drug [24].

The pathways by which abacavir may induce cardiovascular disease are unclear. In the DAD study abacavir had no association with the risk of stroke [25]. Potential mechanisms for putative abacavir toxicity include interaction with other drugs, as HIV-infected patients receive at least two additional antiretroviral drugs (and even more in second-line regimens). The abacavir regimens may increase inflammation, causing plaque instability. Metabolic products of abacavir, but not of other NRTIs, can bind to specific human leucocyte antigen molecules, mediating release of proinflammatory cytokines, resulting in a hypersensitivity reaction [26]. Perhaps a similar, more protracted mechanism is involved in a putative cardiotoxicity, although the timing clearly is inconsistent with a hypersensitivity reaction.

Abacavir is a key drug in modern HIV treatment and understanding of its potential toxicities is urgently needed. Markers of cardiovascular risk factors are improving in quality [27] and it would be helpful to test whether these markers predict increased risk of cardiovascular disease in patients randomized to abacavir arms in previously completed clinical trials.

In conclusion, the findings from this study and the DAD study suggest that abacavir is associated with an increased risk of MI. Further studies are needed to quantify the association and to control for potential, as yet unmeasured, confounding.

Acknowledgements

We thank the staff of our clinical departments for their continuous support and enthusiasm, Preben and Anna Simonsen's Foundation, and the Clinical Institute of Copenhagen University for financial support. No funding sources were involved in study design, data collection, analysis, report writing or decision to submit the paper.

Centres in the Danish HIV Cohort Study

Departments of Infectious Diseases at Copenhagen University Hospitals, Rigshospitalet (J. Gerstoft, N. Obel) and Hvidovre (G. Kronborg), Odense University Hospital (C. Pedersen), Aarhus University Hospitals, Skejby (C. S. Larsen) and Aalborg (G. Pedersen), Herning Hospital (A. L. Laursen), Helsingør Hospital (L. Nielsen) and Kolding Hospital (J. Jensen).

Conflicts of interest

N. Obel has received research funding from Roche, Bristol-Myers Squibb, Merck Sharp & Dohme, GlaxoSmithKline, Abbott, Boehringer Ingelheim, Janssen-Cilag and Swedish Orphan. C. Pedersen has received research funding from Abbott, Roche, Bristol-Myers Squibb, Merck Sharp & Dohme, GlaxoSmithKline, Swedish Orphan and Boehringer Ingelheim. J. Gerstoft has received research funding from Abbott, Roche, Bristol-Myers Squibb, Merck Sharp & Dohme, PharmAsia, GlaxoSmithKline, Swedish Orphan and Boehringer Ingelheim. H. T. Sørensen does not report receipt of fees, honoraria, grants or consultancies. However, the Department of Clinical Epidemiology, Aarhus University Hospital, is involved in studies funded by various companies (Amgen, Pfizer, Glaxo SmithKline and Centocor) in the form of research grants administered by Aarhus University. None of these studies overlaps with the present study. D. K. Farkas, G. Kronborg, C. S. Larsen, G. Pedersen, A. Riis, C. Pedersen and H. T. Sørensen report no conflicts of interest.

Ancillary