Association Between HIV Infection, Antiretroviral Therapy, and Risk of Acute Myocardial Infarction: A Cohort and Nested Case–Control Study Using Québec's Public Health Insurance Database

Abstract

Background: 

Morbidity associated with cardiovascular disease is increasing in the HIV-infected population. We aimed to study the impact of HIV and of antiretrovirals on acute myocardial infarction (AMI).

Methods: 

We performed a cohort and a nested case-control study using the dataset of the Régie de l'Assurance Maladie du Québec. HIV-positive patients were identified using ICD-9 diagnostic codes and matched to HIV-negative patients. Within the HIV-positive cohort, cases of AMI were identified and matched to HIV-positive patients without AMI. The coprimary outcomes were the risk of AMI associated with HIV exposure in the cohort study and that associated with exposure to antiretrovirals in the case-control study. Data were analysed using Poisson and conditional logistic regression.

Results: 

About 7053 HIV-positive patients were matched to 27,681 HIV-negative patients. Incidence rates of AMI in the HIV+ cohort was 3.88 95% confidence interval (CI) (3.26 to 4.58) per 1000 patient-years, compared to 2.21 95% CI (1.93 to 2.52) per 1000 patient-years in the HIV cohort. The adjusted incidence ratio of AMI for HIV-infected patients was 2.11 95%CI (1.69 to 2.63). Among HIV+ patients, 125 AMI cases were matched with 1084 HIV+ patients. We found increased odds ratio (95% CI) of AMI associated with any exposure to abacavir 1.79 (1.16 to 2.76), P = 0.02, efavirenz 1.83 (1.21 to 2.76) P = 0.004, lopinavir 1.98 (1.24 to 3.16) P = 0.004, and ritonavir 2.29 (1.48 to 3.54) P < 0.001.

Conclusions: 

HIV+ individuals were at higher risk of AMI than the general population, and several antiretrovirals were associated with an increased risk of AMI. Results should be interpreted with caution in absence of data on smoking and HIV clinical status.

INTRODUCTION

The clinical picture of HIV infection has changed greatly since the introduction of highly active antiretroviral therapy. From progression to a rapidly fatal disease, HIV infection has become a chronic condition. Metabolic and cardiovascular comorbidities cause increasing concerns as the HIV-positive population ages. Cardiovascular disease (CVD) is now recognized as an important cause of death in HIV-positive subjects more than 55 years of age.^1-4 HIV infection itself has been associated with an increased risk of CVD, possibly due to ongoing activation of the immune system.³ Some studies have shown an association between antiretroviral (ARV) use and increased risk of CVD.^5-11 Protease inhibitors (PIs) were the first class of ARVs shown to be associated with CVD risk.^12-16 However, more recently, large cohort studies have reported conflicting results for the association of some nucleoside reverse transcriptase inhibitors (NRTIs) and CVD.^5,17-26 We sought to characterize the burden of CVD amongst Quebec's HIV-positive population compared with the HIV-negative population and to determine if the use of ARVs was associated with an increased risk of acute myocardial infarction (AMI) in HIV-infected population. We performed a cohort study and a nested case-control study using the routine dataset of the Régie de l'Assurance Maladie du Québec.

METHODS

Study Design and Data Source

We conducted a cohort study and a nested case-control study using data obtained from the linkage of 2 administrative databases: the Régie de l'assurance-maladie du Québec (RAMQ) and Med-Echo database. The RAMQ database provides information related to medical services and pharmaceutical services. The medical services information includes physician-based diagnostic codes according to the International Classification of Diseases, ninth revision (ICD-9),²⁷ therapeutic procedures, and types of institutions where the medical procedures were performed. The medical coverage is universal in Québec so medical services data is available for the entire population. The pharmaceutical information includes data on all prescriptions dispensed: prescribing physician, drug name, dosage, formulation, quantity dispensed, date, and duration of the dispensation. The pharmaceutical information is available for publicly insured Quebec residents only. Publicly insured population represents approximately 41% of Quebec residents,²⁸ which includes all senior (65 years or older), welfare recipients and their children, and all workers and their families who do not have access to a private drug plan insurance program. Demographic information on age, sex, date of death, and date of coverage by the drug plan are also available in the RAMQ databases. Only patients with adequate public drug insurance coverage were studied.

The Med-Echo database contains discharge summaries for all acute care hospitalizations. Patient records were linked across the databases of RAMQ and Med-Echo by use of unique health insurance numbers. Data recorded in the RAMQ medication database and in the Med-Echo database have been formally compared with data in patient's charts and physical reviews and found to be comprehensive and valid.^29,30 RAMQ and Med-Echo databases have been used in the past for epidemiological research leading to scientific articles published in peer-reviewed medical journals.^31-35

Ethics

This study was approved by the CHUM ethics committee. Permission to use the data was obtained from the Government of Quebec Ethics Committee, the Commission d'Accès à l'information.

Study Population

HIV-positive patients within the RAMQ database were identified as follows. Patients were included either if they had a positive serology for HIV (Québec-specific code 795.8) or if they had and HIV ICD-9 code (042.0-044.9) and were using ARVs (excluding lamivudine monotherapy) was included. The ARV criterion was added to ensure maximal specificity in HIV status definition, (even though this could lead to over-representation of treated patients.) These patients formed the HIV-positive cohort. The selection of the HIV-positive patients was between January 1, 1985, and December 31, 2007. The date of the first HIV diagnostic was termed the “entry date”. Exclusion criteria from the HIV-positive cohort were inadequate coverage with drug insurance (defined as less than 6 months of coverage before and 2 months after entry date or age younger than 20 years.) For each patient in the HIV-positive cohort, up to 5 HIV-negative patients were randomly selected, matched on age (within 5 years), sex, entry date, and adequate period of pharmaceutical insurance coverage. HIV-negative patients had to have active pharmaceutical coverage at the time of entry date of their matched HIV-positive counterpart, and their entry date was defined as that of their matched HIV-positive counterpart. The matching on entry date was necessary to account for secular trends in incidence and treatment of CVD. These patients formed the HIV-negative cohort.

Definition of Outcome

AMI was identified by the Canadian Classification of Health Interventions CIM-9-QC codes (410.0-410.9x before March 2006 and CIM-10 codes: I210-I213, I2140-I2149, I219-I221, and I228-I229 after March 2006.) The diagnoses of AMI were retrieved from the hospitalisation database (Med-Echo). A previous study on diagnosis accuracy of AMI using hospital discharge summaries found a sensitivity of 62.1%, specificity of 99.9%, positive predictive value of 94.7 and negative predictive value of 99.1%.³⁶ AMI occurring before HIV-positive status was known were not considered. The date of AMI was termed the “index date”.

Nested Case-Control Study Within the HIV-Positive Cohort

Cases were defined as HIV-positive patients with a diagnostic code of AMI. Controls were selected within the HIV-positive cohort amongst patients that did not have an AMI up to the case's AMI date. Up to 10 controls were selected for each case, matched on age (within 5 years), sex, and entry date (controls had to be in active follow-up at the time of their case's entry date). The control's index date was that of their corresponding case.

Definition of Exposure

In the cohort study, the exposure of interest was HIV. In the case-control study, patients were considered exposed to ARV if they received at least 1 ARV dispensation between the entry date and the index date. The following ARV alone or in combination were considered; the PIs [amprenavir, atazanavir, darunavir, fosamprenavir, indinavir, lopinavir, nelfinavir, ritonavir (boosting or treatment doses), saquinavir and tipranavir]; the NRTIs (abacavir, didanosine, emtricitabine, lamivudine, stavudine, tenofovir, zalcitabine and zidovudine); the nonnucleoside reverse transcriptase inhibitors (NNRTIs) (delavirdine, efavirenz, and nevirapine). For each analysis, the reference group was composed of patients not exposed to the specific ARV studied at any time between their entry date and their index date. Recent exposure was defined as having received a dispensation in the 6 months before the index date, in keeping with previous studies.³⁷

Adjustment for Confounders and Effect Modifiers

We adjusted the results for traditional risk factors by including in the multivariate model the use of cardiovascular-related drugs as markers for these risk factors. We adjusted for prior use of antihypertensive drugs, antidiabetic drugs, lipid-lowering drugs, and antiplatelets, or anticoagulation drugs. We also adjusted for prior history of CVD (myocardial infarction or stroke), chronic renal failure or hemodialysis, hepatitis C infection, illicit drug use, and alcohol abuse. We looked at effect modification by sex and by calendar year.

Statistical Analysis

Crude and adjusted rates, rate ratios, and their 95% confidence interval (CI) were modelled using univariate and multivariate Poisson regression models. Within the case-control study, odds ratios associated with exposure to each ARV were obtained using conditional logistic regression. Statistical analyses were performed using the SAS software (SAS Institute) release 9.1 and Stata version 11.

RESULTS

Cohort Study

Amongst the RAMQ database, 7494 patients were identified as being HIV positive. Of these, 441(5.9%) patients were excluded (174 had inadequate insurance coverage and 267 were less than 20 years of age). The 7053 remaining patients constituted the HIV-positive cohort, representing 35,851 patient-years of follow-up. They were matched to 27,681 HIV-negative patients, constituting the HIV-negative cohort. Table 1 shows the baseline characteristics and comorbidities during follow-up of HIV-positive and HIV-negative patients. In the HIV-positive cohort, 78% were male, and the median age was 37 years (range 22-72 years.), which was comparable with the HIV-negative cohort. Before their diagnosis of HIV positive (before entry date), the patients from the HIV-positive cohort had comparable prevalence of hypertension, diabetes, dyslipidemia, and use of antiplatelet drugs than their HIV-negative counterparts (data not shown). However, during their follow-up, they were more at risk of being treated for diabetes (7.1 vs. 5.4%, P < 0.001), hypertension (24.7 vs. 17.1, P < 0.001), dyslipidemia (15,6 vs. 11.3%, P < 0.001), or use antiplatelet drugs (14.2 vs. 11.3%, P < 0.001).

TABLE 1: Baseline Characteristics of Patients and Prevalence of Covariates During Follow-Up (n = 34,734)

Among the HIV-positive cohort, 139 AMI cases occurred, representing an incidence of 3.88 95% CI (3.26 to 4.58) per 1000 person-years. Among the HIV-negative cohort, 226 MI cases occurred, representing an incidence rate of 2.21 per 1000 person-years 95%CI (1.93 to 2.52). Crude and adjusted hazard ratio (HR) for AMI according to HIV status are presented in Table 2. HIV infection was associated with a crude HR of AMI of 1.72 95% CI (1.39 to 2.13) and an adjusted HR of 2.11 95% CI (1.69 to 2.63). Table 3 presents genders-specific hazards ratio for AMI associated with HIV infection. Although the effect of HIV was greater in females [3.77 95% CI (1.79 to 7.96], vs. 2.04 (1.62 to 2.57) in males], the difference in effect size was not statistically significant (P = 0.17). The presence of an AIDS-defining condition did not modify the risk of AMI associated with HIV infection (Table 3).

TABLE 2: Crude and Adjusted Hazard Ratio of AMI for Each Covariates

TABLE 3: Adjusted* Incidence of Acute Myocardial Infarction According to Gender and AIDS Status: Subgroup Analysis

Time Trend Analysis

There was a significant interaction with calendar year and the HR of AMI associated with HIV infection, even after adjustment for age, gender, diabetes, dyslipidaemia, renal failure, and past cardiovascular events (P for interaction 0.001). Figure 1 shows the incidence rate of AMI according to calendar year in HIV-positive versus HIV-negative individuals. AMI incidence was stable in the HIV-negative cohort throughout the study period, but increased steadily in the HIV-infected cohort [crude HR: 1.08 95% CI (1.03 to 1.13) per year] The increase became nonsignificant when adjusted for age and the other factors mentioned above. [adjusted HR: 1.05 95% CI (1.00 to 1.10] per year].

FIGURE 1: Trends in incidence of acute myocardial infarctions according to HIV serostatus, per 1000 person-years.

Case-Control Study

The 139 cases of AMI in HIV-positive patients were considered as cases for the nested case-control study. Of these, 14 were excluded (12 could not be matched with at least 1 control and 2 had HIV diagnosis and AMI on the same day). The 125 remaining cases were matched to 1084 HIV-positive controls who did not have AMIs during follow-up. Table 1 shows the basic characteristics of patients included in the case-control study. Of the 12 cases that could not be matched, 6(50%) were females. Their median age was 72 years, which differed from the median age of the included 125 cases (47 years) (P value for difference in age 0.001, P value for difference in sex <0.001). The first AMI in an HIV-positive patient occurred in 1989, and 92% of the events took place in the last 10 years (1997-2007).

Table 4 shows the results of the case-control study. The drugs that were associated with an increased risk of AMI for any exposure were abacavir [odds ratio (OR): 1.79 95% CI (1.16 to 2.76)], lopinavir [OR: 1.98 95% CI (1.24 to 3.16)], ritonavir (either as treatment or boosting doses) [OR: 2.29 95% CI (1.48 to 3.54)] and efavirenz [OR: 1.83 95% CI (1.21 to 2.76)]. Results for the current exposure were similar for abacavir, lopinavir, and ritonavir, but efavirenz was no longer associated with increased risk, and delavirdine was, but with a very wide CI.

TABLE 4: Number of Acute Myocardial Infarctions, Percentage Exposed, Crude and Adjusted* Odds Ratio Associated With any Exposure to ARV Drugs in 125 Patients With Acute Myocardial Infarction and 1084 Matched Controls

TABLE 4: (continued)Number of Acute Myocardial Infarctions, Percentage Exposed, Crude and Adjusted* Odds Ratio Associated With any Exposure to ARV Drugs in 125 Patients With Acute Myocardial Infarction and 1084 Matched Controls

Sensitivity Analysis

We performed a post hoc analysis to assess whether these medications were more often prescribed to patients with high cardiovascular risk. We assessed the use of antihypertensive, lipid-lowering, antidiabetic, and antiplatelet drugs amongst the patients receiving drugs associated with increased AMI risk. There was no difference between the groups in these markers of higher cardiovascular risk (data not shown).

DISCUSSION

In this population-based cohort study, we found an increased incidence of AMI in HIV-positive individuals, with an adjusted incidence ratio of 2.11 95%CI (1.69 to 2.63). This is consistent with other studies that have evaluated this risk. Klein et al³⁸ reported an increase in admission for coronary heart disease for HIV infected versus HIV uninfected of 6.5 versus 3.8 per 1000 patient-years. A higher risk of coronary artery disease admission was found in younger HIV infected compared with age-matched HIV-uninfected patients.³⁹ In a study that used a Massachusetts administrative database containing 3851 HIV-infected patients and more than 1 million HIV-negative controls, Triant et al² found a mean incidence rate of AMI of 11.13 versus 6.98 per 1000 person-years, respectively. These rates are slightly higher than ours (3.88 vs. 2.21 per 1000 person-years), but these differences can be explained by the use of different populations and case definitions. We also found an increasing incidence of AMI in HIV-positive patients for each calendar year throughout the study although incidence remained stable in HIV-negative patients, illustrating the growing public health concern of CVDs in HIV-infected individuals.

The main strength of the cohort study are its size, long follow-up over time, and the presence of an HIV-negative comparator group.

However, there are some significant weaknesses. In the cohort study, the exposure to HIV was determined using administrative codes, so there could be some exposure misclassification. However, as the outcomes were ascertained after the exposure to HIV, this cannot be differential according to outcome status, so any bias will be toward the null, leading to an underestimation of the effect of HIV on rate ratios of AMI. Another potential source of bias is that sudden cardiac death was not taken into account. This could lead to bias if rates of sudden cardiac death were different in regards to HIV infection or to exposure to specific medications. However, no data from previous literature suggest that sudden cardiac death is differential in those populations.

The outcomes of AMI were ascertained using a hospitalisation database and routine codes, so some misclassification of outcome status is possible as well. However, myocardial infarction is a diagnosis with firmly established diagnostic criteria. It is unlikely that, faced with a clinical syndrome compatible with AMI, a physician would diagnose AMI differently between HIV-positive and HIV-negative patients. Therefore, any misclassification of outcome would only attenuate the effect found. Sudden cardiac death was not included, which decreases the power of our study by decreasing the number of events. Yet, we have no reason to believe that this will lead to bias, as the proportion of AMI presenting as sudden cardiac death should not differ between groups.

Even though our findings are consistent with previous literature that HIV infection and its treatment are associated with increased rates of AMI, this type of study cannot be interpreted as causal. Concerns remain about the presence of confounders that could not be captured, in particular smoking, family history of CAD, creatinine, viral load, and CD4 levels. It is known that HIV-infected individuals smoke more than HIV-negative individuals. Therefore, in the cohort study, tobacco might be responsible for part of the association seen.

In our case-control study, we found that, among NRTIs, abacavir was associated with increased odds ratios for AMI with either any or current exposure. Among the PIs, lopinavir and ritonavir were also associated with increased OR of AMI. Our findings for the NRTIs and PIs are somewhat consistent with previous literature. The Data Collection on Adverse Events of Anti-HIV Drugs study is the largest prospective cohort study to look at the association between cardiovascular risk and ARV. They reported an association between recent exposure to abacavir and an increased risk of AMI.^26,40 They also found a positive association between lopinavir/ritonavir and an increased risk of AMI.¹⁶ The French Hospital Database study found an association between recent exposure to abacavir and an increased risk of MI [ORs: 2.01; 95% CI (1.11 to 3.64)]. They also found an association with PIs, particularly lopinavir/ritonavir.⁴¹

It has been hypothesised recently that some of the cardiovascular risk associated with abacavir was due to preferential prescriptions of abacavir over tenofovir in patients with renal failure.¹⁸ This is not likely to be the case in our study because tenofovir was not reimbursed by Quebec's insurance program up to 2007 and therefore was not easily available for our patients.

More surprisingly, we also found increased risks of AMI with any exposure to the NNRTI drug efavirenz, and with recent use of delavirdine, which has not previously been associated with adverse cardiac outcomes. Although efavirenz is known to have a small lipid effect, no other biological mechanism can be offered at this time to explain this finding. However, there might have been some confounding by indication. Because the adverse metabolic effects of PIs have been known for some time, it is possible that the population prescribed NNRTI in the province of Québec were those perceived by their physicians as having higher cardiovascular risk and that this has been incompletely accounted for in the adjusted measures (and even thought it was not apparent in our sensitivity analysis).

The major strengths of our case-controls study come from its nested nature, which removes the possibility of recall bias. Also, the exposure to medication was ascertained using comprehensive data on drug dispensations and is exempt of bias. This is superior to simple prescription data, as patients must copay for each of their medications, increasing the probability that patients take the medication they are dispensed. Therefore, misclassification of exposure to ARVs is unlikely. Exposure to medication before follow-up in the case-control study, however, would not have been captured by our definition of exposure. We do not believe this would have led to bias, as we were interested in exposure during the follow-up time and not cumulative life-long exposure.

Limitations of the case-control study include the use of multiple models, which increase the risk of type 1 error. Limited power is also an issue for some of the drugs studied. Lack of data on some important confounders, as in the cohort part of our study, is a limitation of the case-control study (discussed above).

The choice of exposure chosen for the case-control study can also be debated. We defined exposure as any or recent exposure to medication and AMI, which does not take into account cumulative drug doses. This was done because, contrary to PIs, the mechanism underlying a potential increased risk for other ARVs is unknown. If risk is mediated through interference with coagulation or through plaque, destabilization or increase in inflammation, cumulative exposure might not be necessary to increase risk.

Finally, our database only contains pharmaceutical information on 41% of Québec's population, which limits the generalizability of the findings, but does not diminish the internal validity of our study.

In summary, our study found an increased risk of AMI in HIV-positive compared with HIV-negative individuals and increasing rates of AMI over time in our HIV-positive population. We also found that any and recent exposure to some ARVs was associated with increased AMI risk. However, the level of increased risk is minimal compared with traditional risk factors such as smoking, family history, hypertension, diabetes, and dyslipidaemia. Although our study has shown an association between various individual drugs and an increased risk of AMI, causality cannot be established with this type of study. Given limitation about data on confounders, these data should be interpreted with caution and in the light of other data existing on the subject. Meta-analysis of existing observational data and long term randomized-controlled trials that focus on cardiovascular morbidity and mortality in HIV-infected patients are needed to guide therapeutic choices in these patients.