Impact of Individual Antiretroviral Drugs on the Risk of Myocardial Infarction in Human Immunodeficiency Virus–Infected Patients:  A Case-Control Study Nested Within the French Hospital Database on HIV ANRS Cohort CO4 FREE

Cumulative exposure to protease inhibitors (PIs) has been associated with risk of myocardial infarction (MI) in human immunodeficiency virus (HIV)–infected patients,^1- 4 but the risk associated with individual PIs has not been widely reported, to our knowledge. More recently, specific nucleoside reverse transcriptase inhibitors (NRTIs), particularly abacavir, were incriminated.^5- 6 However, Brothers et al⁷ found no increase in the risk of MI associated with abacavir use in a pooled analysis of 12 randomized clinical trials. The results have raised a lot of debate because abacavir is 1 of the 2 most-used NRTIs to initiate therapy in the developed world in the recent period.^8- 9 Since then, many studies^10- 13 have explored the potential mechanisms for such an effect, with conflicting results. After the Data Collection on Adverse Events of Anti-HIV Drugs (D:A:D) study results were presented at the Conference on Retroviruses and Opportunistic Infections in 2008, the European Medicines Agency asked us whether the ongoing case-control study, nested within the French Hospital Database on HIV (FHDH) Agence Nationale de Recherches sur le SIDA et les hépatites (ANRS CO4), could help settle the issue of abacavir. We therefore wrote an analysis plan to evaluate the association between the risk of MI and cumulative exposure to NRTIs and recent or past exposure to NRTIs. In addition, we explored the role of specific non-NRTIs (NNRTIs) and PIs.

The HIV-infected patients were selected from the FHDH, an ongoing, prospective, observational, nationwide, hospital-based cohort. The only FHDH inclusion criteria are HIV type 1 or 2 infection and written informed consent. Data are collected prospectively by trained research assistants using standardized forms. Clinical events are coded using the International Statistical Classification of Diseases, 10th Revision.¹⁴ A follow-up form is completed at least every 6 months or at each visit or hospital admission during which a new illness is diagnosed, a new treatment is prescribed, or a noteworthy change in biological markers is noted. In July 2007, the database contained information on 74 958 patients who had been seen at least once between January 1, 2000, and December 31, 2006, which is 57% of all HIV-infected patients under care in France,¹⁵ corresponding to a total follow-up duration of 298 156 patient-years.

Cases were patients who had a first prospectively reported MI between January 1, 2000, and December 31, 2006. Patients with a history of MI were excluded. The International Statistical Classification of Diseases, 10th Revision code used to define MI was I21. The diagnosis of MI was confirmed by a cardiologist masked to antiretroviral treatment (ART) history who was provided with cardiac signs and symptoms; troponin, creatine kinase, or both levels; and electrocardiographic findings as recorded in the medical records. We used the American College of Cardiology/European Society of Cardiology definition.¹⁶ Only definite and probable cases of MI and possible death from MI were included. The index date was the date of MI diagnosis.

Control subjects were HIV-infected patients with no history of MI and no diagnosis evoking MI who had already been enrolled in the database when MI was diagnosed in the corresponding case (±6 months). They were matched with cases for age (±3 years) at MI diagnosis, sex, and clinical center. In a previous article,¹⁷ a nested case-control study using incidence density sampling with the same matching factors provided results similar to those of the published cohort analysis.³ We used a case-control approach rather than a cohort approach for the efficiency of this design because we needed to collect cardiovascular risk factors in the medical records for cases and controls. With the goal of having 3 controls per case and using incidence density sampling, we randomly selected up to 5 controls per case from the list of patients fulfilling the matching criteria. Cases were eligible as controls up to the onset of MI.

We collected the cardiovascular risk factors listed in the French National Guidelines published in 2005,¹⁸ namely, age older than 50 years in men and 60 years in women, family history of premature coronary artery disease before age 55 years in the father or age 65 years in the mother, current smoker or smoking cessation within the previous 3 years, and hypertension, diabetes mellitus, or hypercholesterolemia. We also recorded current use of cocaine, intravenous drugs, or both as stated in the medical records¹⁹ and body mass index²⁰ because these also affect cardiovascular risk. These data were extracted from the medical records by trained research assistants experienced in HIV infection using a predefined case report form. Data on HIV infection, plasma HIV type 1 RNA load, CD4 and CD8 cell counts, the CD4 nadir, a detailed history of prescribed ART, and AIDS status²¹ before the index date were validated. All biological measurements were collected within 3 months of the index date.

Conditional regression models (TPHREG, SAS version 9.1; SAS Institute Inc, Cary, North Carolina) were used to quantify the relation between exposure to each ART drug and risk of MI. Lipid variables and diabetes mellitus, which may lie on the causal pathway between exposure to some ART drugs and risk of MI, were excluded from the main analysis. Smokers included current smokers and smokers who had quit less than 3 years before the index date. Obesity was defined by a body mass index greater than 30 (calculated as weight in kilograms divided by height in meters squared). Hypertension was defined as use of an antihypertensive medication or as hypertension reported with a diagnosis date in the medical record. We also studied the potential effect of HIV-related variables on risk of MI. The continuous variables were modeled in class or after log transformation. When there was missing data, a “missing” category was created so that all the patients were included in the analyses.

The first model included cumulative exposure to each ART drug. Ritonavir used as another PI booster was counted with the other PIs, whereas ritonavir alone was counted for itself. The second model included exposure to each ART drug plus, for each NRTI, a 3-class variable consisting of no exposure, recent exposure (current or in the previous 6 months), and past exposure (>6 months previously). In these 2 models, potential confounding factors were included one by one to determine whether they changed the odds ratio (OR) for any drug by at least 10% in any of the models. Subsequent models were adjusted on the selected confounders: hypertension; smoking status; family history of premature coronary artery disease; cocaine, intravenous drug use, or both; plasma HIV type 1 RNA level of 50 copies/mL or less or not; CD4 to CD8 cell ratio less than 1 or at least 1; and exposure to each ART drug. The ORs are reported for ART drugs to which at least 100 patients were exposed.

We conducted sensitivity analyses to assess the robustness of the results. We made an analysis restricted to patients who received their first ART regimen after inclusion in the cohort to detect a potential selection bias. We also examined whether the impact of PIs was the same when they were boosted or not with ritonavir. We also included in the models hypercholesterolemia, defined as a low-density lipoprotein cholesterol level of at least 160 mg/dL (to convert to millimoles per liter, multiply by 0.0259) or as the use of lipid-lowering drugs (statins or fibrates); hypertriglyceridemia, defined as triglyceride levels of at least 150 mg/dL (to convert to millimoles per liter, multiply by 0.0113); high-density lipoprotein cholesterol level, defined by a 4-category variable (<40 mg/dL, 40 to <60 mg/dL, ≥60 mg/dL, and value missing) (to convert to millimoles per liter, multiply by 0.0259); and diabetes, defined as use of an antidiabetic drug or as fasting glucose levels greater than 126 mg/dL (to convert to millimoles per liter, multiply by 0.0555) or diagnosis of diabetes reported with a diagnosis date in the medical record to determine whether these variables lay on the causal pathway linking ART to risk of MI. To explore a potential channeling bias, we compared cases exposed to short-term/recent abacavir therapy with the other cases and in view of the results conducted an analysis restricted to patients who did not use cocaine or intravenous drugs.

Of 74 958 patients, 423 MI cases were identified in the database; 38 of these cases (9.0%) were excluded from the analysis because MI was not confirmed, 31 because they did not meet the inclusion criteria, and 32 because they had a recurrent MI. In addition, 39 cases were excluded because their medical records were unavailable. Six cases corresponded to patients who were selected as controls but whose medical records mentioned an MI. Based on the 360 cases (including 289 cases in the case-control study: 32 patients with recurrent MI and 39 patients with missing medical records), the incidence of MI in the database was estimated to be 1.24 per 1000 patient-years (95% confidence interval [CI], 1.11-1.36). Overall, 289 cases and 884 controls were included in the analysis; 246 cases had 3 controls, 29 had more than 3, and 14 had fewer than 3. The diagnosis was definite in 74.4% of cases and probable in 16.6%, and 9.0% had a possible death due to MI.

Cases and controls were well matched for age and sex (Table 1). All the cardiovascular risk factors except obesity were more frequent in cases than in controls, and cases therefore had more cardiovascular risk factors (P < .001). Cases were less likely to have a controlled viral load (P = .006) and a normal CD4 to CD8 cell ratio (P = .001). The CD4 cell counts on the index date were not different in cases and controls. At enrollment in the cohort, 76% of patients had never received ART. On the index date, 95% of patients had been exposed to ART, and only 5% of cases and 7% of controls were not receiving therapy. The median length of ART exposure was 6.6 years in cases and 7.0 years in controls, and the median number of different ART drugs received was 7 in cases and 6 in controls. The proportions of patients exposed to each ART drug are given in Table 2. Most patients had been exposed to thymidine analogues (92.6%) and to PIs (77.7%).

In model 1, no association was found between cumulative exposure to abacavir and risk of MI (Table 3). In model 2 (Table 3), which also included for each NRTI a 3-class exposure variable (none, recent, and past), there was evidence of an interaction between recent/past exposure and cumulative exposure to abacavir. Whereas the OR for cumulative exposure to abacavir decreased from 0.97 in model 1 to 0.88 in model 2, the OR for recent exposure was 1.60 and for past exposure was 1.62. This interaction was not observed to that extent with any other NRTIs (Table 3). This finding prompted us to build an additional model in which exposure to abacavir was defined using a 5-class variable in which duration of exposure was combined with time of use (no exposure, <1 year of exposure and recent use [ie, short-term/recent exposure], <1 year of exposure and past use; >1 year of exposure and recent use, and >1 year of exposure and past use). Table 4 gives the results of univariate and multivariate analyses of this final model. Patients with short-term/recent exposure to abacavir had a significantly increased risk of MI (OR, 2.01). Although not significant, the risk of MI tended to increase with cumulative exposure to zidovudine (OR, 1.09 per year of exposure) and stavudine (1.11 per year of exposure). In a post hoc analysis, cumulative exposure to thymidine analogues (zidovudine and stavudine) was associated with an increased risk of MI (OR, 1.09 [95% CI, 1.00-1.19] per year). No effect was found with didanosine, lamivudine, tenofovir, or zalcitabine.

In the final model, no association was found between risk of MI and cumulative exposure to efavirenz (OR, 1.01) or nevirapine (1.00).

In the final model, the ORs of MI were 1.07 per year of exposure to indinavir with or without ritonavir (P = .32) and 1.10 per year of exposure to nelfinavir (P = .15). The risk was significant with lopinavir with ritonavir (OR, 1.33 per year) and with amprenavir/fosamprenavir with or without ritonavir (1.53 per year). There was no increased risk associated with exposure to saquinavir with or without ritonavir. In a post hoc analysis, cumulative exposure to any PI except saquinavir was associated with an increased risk of MI (OR, 1.15 [95% CI, 1.06-1.26] per year).

Similar results were obtained when the analysis was restricted to patients who were naive at inclusion in the cohort (ie, 61% of the full sample), with a slight difference, however, for abacavir. For short-term/recent abacavir use, the univariate OR was estimated to be 3.77 (95% CI, 1.86-7.64) and the adjusted OR to be 1.79 (95% CI, 0.74-4.27). Ritonavir boosting did not significantly change the association between PI exposure and risk of MI (eTable 1). The association between PI exposure and the risk of MI was not changed when metabolic variables were considered (eTable 2).

The 31 cases with short-term/recent exposure to abacavir were not significantly different from the other cases except for cocaine or intravenous drug use, time receiving ART, and AIDS status before MI (Table 5). In view of this result, we conducted an analysis that included only cases (n = 250) and their matched controls (n = 704) who were not cocaine or intravenous drug users (eTable 3). The OR for short-term/recent exposure to abacavir was 1.27 (95% CI, 0.64-2.49). In contrast, for the other associations, the ORs remained similar (eTable 3). There were not enough patients to repeat this analysis in cocaine or intravenous drug users.

We conducted a case-control study nested within a large database of HIV-infected patients to study the association between ART and risk of MI. We found that the risk of MI was increased by cumulative exposure to any studied PI except saquinavir and particularly to lopinavir with ritonavir and amprenavir/fosamprenavir with or without ritonavir. Cumulative exposure to thymidine analogues was also associated with an increased risk of MI. Abacavir initiation was associated with an increased risk of MI, whereas longer exposure to abacavir was not. These associations persisted when the analysis was restricted to nonusers of cocaine and intravenous drugs, except for abacavir. All NNRTIs and NRTIs other than abacavir and thymidine analogues were not associated with risk of MI.

The use of a nested design allowed us to avoid the main drawback of the case-control design, namely, classification bias on exposure, while allowing us to fully validate the treatment histories prospectively recorded in the database. We did not include recurrent MI because it would have been difficult to control for the selection bias by analysis for these cases given that the ART drug prescribed to them was likely to have been chosen differently. We also excluded 39 potential cases whose medical records were lacking. Although this could represent a small selection bias, it would have been impossible to adjust for confounding for these cases. Three-quarters of the cases and controls had never received ART before being enrolled in the cohort, a feature that would tend to limit the selection bias; the analysis restricted to these patients gave the same results as the main analysis. There was a small amount of missing data for the cardiovascular risk factors except for family history of premature coronary artery disease, in which the proportion of missing data was 31% for cases and 62% for controls, suggesting that the report of this item in the medical record may have occurred after the MI was diagnosed and most often was unknown by the physician when prescribing the treatment. Therefore, although it is a limitation of this study, it is unlikely to have played a major confounding role. Additional variables could have been accounted for as potential confounders, such as renal function.²² However, no study to date, to our knowledge, has reported different results when accounting only for the main MI risk factors compared with when accounting for the main MI risk factors and renal function. Because creatinine level was not measured regularly for all the patients throughout the study period, we could not account for this variable in this study. Given the association between traditional cardiovascular risk factors and renal function, we do not believe that renal function could be a major confounder in this study population because most patients exposed to abacavir in this study were not naive patients but were highly preexposed patients. This situation is different from deciding which NRTI to prescribe to a naive patient nowadays, a decision that could certainly be influenced by renal function. Observational studies, such as this one, cannot demonstrate the causal nature of an association. However, the ORs for PIs and NNRTIs were very close in the univariate and multivariate models, indicating that the association observed with all the PIs except saquinavir with or without ritonavir is unlikely to be explained by remaining confounders.²³ The situation was different in the case of NRTIs, however, particularly for tenofovir and abacavir. For example, the univariate OR was 1.19 for tenofovir, whereas in the final model it was 1.00; in addition, the OR declined from 2.76 to 2.01 for short-term/recent abacavir exposure. In the analysis restricted to cases and their matched controls included as naive in the cohort, the OR declined from 3.77 to 1.79 for short-term/recent abacavir exposure, and in the analysis restricted to nonusers of cocaine and intravenous drugs, it declined from 2.00 to 1.27. It follows that the present results are likely to be more robust for PIs and NNRTIs than for NRTIs.²³

In a previous analysis of the D:A:D study, the relative rate of MI per year of PI exposure was 1.16 (95% CI, 1.10-1.23), a value close to the estimated OR of 1.15 (1.06-1.26) in the present study. Similarly, as in a recent analysis of the D:A:D study,²⁴ we found no association between risk of MI and exposure to saquinavir with or without ritonavir, whereas we found that lopinavir with ritonavir increased the risk of MI. In both studies, the risk of MI associated with PI exposure changed little regardless of whether the PIs were boosted by ritonavir or when lipid variables and diabetes were taken into account. Although this may be explained by uncertainties in lipid level measurement, it could also imply that mechanisms other than increased lipid levels, such as an effect on endothelial cells,²⁵ could be involved in the increased risk of MI associated with cumulative PI exposure. Saquinavir is known to offer a better triglyceride profile than the other PIs, in particular lopinavir with ritonavir.²⁶ This might perhaps explain the difference that we observed in the risk of MI for this PI compared with others. The 2 studies also gave similar results for NNRTIs. No effect of exposure to efavirenz or nevirapine was found in either study, although efavirenz has a more negative effect than does nevirapine on the lipid profile.²⁷

Available studies^24,28- 30 have given divergent results for NRTIs, most likely because prescription of this drug class has been associated with multiple confounding factors that were handled differently in the different studies. This might, for example, explain why an association between didanosine exposure and MI was found in the D:A:D study⁵ but not in the Strategies for Management of Anti-Retroviral Therapy (SMART) study⁶ or in the present study. The D:A:D study showed no increase in the risk of MI after exposure to thymidine analogues, although this was its main hypothesis.⁵ In contrast, we found that exposure to stavudine or zidovudine increased the risk of MI. This latter result is unlikely to be explained by confounding factors because the OR was 1.06 in the univariate model and 1.09 in the multivariate model. Further independent studies are needed to settle this issue because these drugs are still widely used in developing countries.³¹ The fat redistribution induced by these drugs might explain the observed effects.³² Without knowledge of the D:A:D study analysis of 2008,⁵ we would have examined only cumulative exposure to abacavir and would therefore have found no association with MI. Only because we were asked to confirm or refute the D:A:D study results did we explore current and past use of abacavir in addition to cumulative exposure. Although we found that recent abacavir treatment initiation was associated with an increased risk of MI in the full data set (OR, 2.01; 95% CI, 1.11-3.64), the association disappeared when we restricted the analysis to nonusers of cocaine and intravenous drugs (1.27; 0.64-2.49). Because this latter result was obtained in an analysis including 81% of the full sample, it is unlikely that the difference between the 2 analyses is explained mainly by a power issue. The estimated OR in the restricted analysis (1.27) was much smaller than that obtained in the full sample. Note that the D:A:D and SMART studies were not adjusted for exposure to cocaine or intravenous drugs but rather for transmission groups. In addition, most patients enrolled in these 2 studies had already received ART previously (73% in the D:A:D study³³ and 95% in the SMART study³⁴). This could induce a larger selection bias than the present study, in which only 24% of patients had received ART before enrollment. Moreover, recurrent MI was not excluded from the D:A:D and SMART studies. All these differences could explain why different results were obtained for NRTIs in the D:A:D, the SMART, and the present studies. In addition, the SMART study⁶ included only a few cases of MI (n = 19), and other studies^7,12,29- 30 with low numbers of events could not exclude a small increase in the risk of MI associated with abacavir. In a recently published cohort study³¹ involving 67 cases of MI, an association between exposure to abac avir treated as a time-dependant covariate (yes or no) and an increased risk of MI was found. However, the analysis was not adjusted for tobacco exposure or family history, the lipid profile was not available, and only hospitalization for MI was accounted for, not death due to MI. This result is slightly different from that of the D:A:D study or the SMART study, and, again, the differences could be explained by differences in definitions of the event or in the way of accounting for potential confounders.

Most PIs studied to date have been found to increase the risk of MI, and this increase is not solely mediated by an effect on lipid metabolism. The 10-year OR of the risk of MI was estimated to be 4 for exposure to all the PIs except saquinavir. To translate this result into practical terms, one can calculate the number of patients to treat for 10 years with a PI to observe an additional MI (number needed to harm). For a patient whose risk of MI is the risk observed in the French HIV-infected patients in this study, that is, 1.2 per 100 after 10 years, the number needed to harm is estimated to be 29, meaning that for 29 patients treated with a PI for 10 years with this level of risk of MI, there will be an additional MI. If one considers a patient whose 10-year risk is 20%, the number needed to harm is estimated to be 3, meaning that for 3 patients treated with a PI for 10 years there will be an additional MI. This means that long-term exposure to this drug class should be avoided if virologically possible in patients with multiple cardiovascular risk factors. There are currently no data sets, including our own, in which exposure to atazanavir with or without ritonavir or darunavir with or without ritonavir is sufficient to conclude on these 2 newer PIs.

We found no association between NNRTI exposure and risk of MI, and this result also seems to be robust. The results for NRTIs are more complex and are more likely to be affected by residual confounding. Although cumulative exposure to thymidine analogues seemed to increase the risk of MI, the observed association with short-term/recent exposure to abacavir disappeared when restricting the analysis to nonusers of cocaine or intravenous drugs. Together, these elements suggest that the relationship between exposure to abacavir and risk of MI cannot be considered causal.

Correspondence: Dominique Costagliola, PhD, INSERM, 56 Bd V Auriol, BP 335, Paris CEDEX 13, 75625 France (dcostagliola@ccde.chups.jussieu.fr).

Accepted for Publication: December 24, 2009.

Author Contributions:Study concept and design: Lang, Mary-Krause, Cotte, Gilquin, Partisani, Simon, Boccara, and Costagliola. Acquisition of data: Lang, Cotte, Gilquin, Partisani, Simon, and Boccara. Analysis and interpretation of data: Lang, Mary-Krause, Cotte, Gilquin, Partisani, Simon, Boccara, and Costagliola. Drafting of the manuscript: Lang, Mary-Krause, and Costagliola. Critical revision of the manuscript for important intellectual content: Lang, Mary-Krause, Cotte, Gilquin, Partisani, Simon, Boccara, and Costagliola. Statistical analysis: Lang, Mary-Krause, and Costagliola. Obtained funding: Mary-Krause and Costagliola. Administrative, technical, and material support: Mary-Krause and Costagliola. Study supervision: Mary-Krause and Costagliola.

Financial Disclosure: Dr Mary-Krause has received honoraria from GlaxoSmithKline. Dr Cotte has received travel grants, honoraria, or study grants from various pharmaceutical companies, including Abbott, Boehringer Ingelheim, Bristol-Myers Squibb, Gilead Sciences, GlaxoSmithKline, Merck Sharp & Dohme-Chibret, Roche, and Tibotec. Dr Gilquin has received travel grants, consultancy fees, or honoraria from various pharmaceutical companies, including Abbott, GlaxoSmithKline, Janssen-Cilag, Roche, and Tibotec. Dr Partisani has received travel grants from Boehringer Ingelheim, GlaxoSmithKline, and Tibotec. Dr Simon has received travel grants, honoraria, or consultancy fees from Boehringer Ingelheim, GlaxoSmithKline, and Tibotec. Dr Boccara has received lecture fees from Gilead Sciences. Dr Costagliola has received travel grants, consultancy fees, honoraria, or study grants from various pharmaceutical companies, including Abbott, Boehringer Ingelheim, Bristol-Myers Squibb, Gilead Sciences, GlaxoSmithKline, Janssen-Cilag, Merck Sharp & Dohme-Chibret, and Roche.

Funding/Support: This study was supported by the ANRS (French National Agency for Research on AIDS and Viral Hepatitis).

Role of the Sponsor: After approval of the protocol, the ANRS had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or in the preparation, review, and approval of the manuscript.

Previous Presentation: This work was presented in part at the 16th Conference on Retrovirus and Opportunistic Infections; February 9, 2009; Montreal, Quebec, Canada.

Additional Contributions: We thank the participants and research assistants of the FHDH. The FHDH is supported by the ANRS, INSERM, and the French Ministry of Health.