Combination antiretroviral therapy (cART) has dramatically decreased the morbidity and mortality associated with HIV infection and has led to prolonged survival overall.1 As a result of declines in AIDS-related mortality and the increasing age of the HIV-infected population, non-AIDS causes of death such as cardiovascular disease (CVD), viral hepatitis, and malignancy now account for most deaths among HIV-infected persons receiving cART.2,3 Although traditional cardiovascular risk factors contribute to myocardial infarction (MI) risk in HIV-infected persons as in the general population, the risk of MI in HIV-infected persons on cART seems to be significantly higher.4–6 Both HIV infection itself and cART may contribute independently to this increased cardiovascular risk.7
The Data Collection of Adverse events of Anti-HIV Drugs (D:A:D) study, a large, prospective, observational study with an international cohort of 33,347 HIV-1–infected individuals, was initiated to explore the association between cART and risk of MI. The D:A:D study has reported a 26% increase in the relative risk of MI per year of exposure to cART in general, and a 16% increase with the protease inhibitor drug class.8,9 Unexpectedly, the D:A:D study also reported an increased risk of MI associated with current or recent (within 6 months) use of abacavir (ABC) [relative risk 1.9, 95% confidence interval (CI): 1.47 to 2.45, P = 0.0001] and didanosine (relative risk 1.49, 95% CI: 1.14 to 1.95, P = 0.003) after a median follow-up of 5.1 years as compared with other antiretroviral drugs.10
Because the D:A:D data are observational, other researchers have sought to replicate the ABC association using independent data sets. The findings have been conflicting. Several observational studies seem to support the results of the D:A:D study.11,12 Also, the Strategies for Management of Anti-Retroviral Therapy (SMART) trial, a randomized clinical trial (RCT) evaluating treatment strategy, found an association between ABC and increased risk of CVD.13 On the other hand, analysis of pooled data from 52 GlaxoSmithKline (GSK) sponsored clinical trials with at least 24 weeks of cART found no excess risk of MI with ABC therapy.14 Among the GSK sponsored trials, 36 were RCTs: 12 randomized with respect to ABC therapy, 14 randomized with respect to other antiretrovirals and used ABC as a background medication, and 10 permitted ABC as a background medication or did not include ABC at all. Sixteen GSK trials were single-arm trials: 13 included ABC as a component of cART and 3 allowed ABC as background medication. Similarly, a recent analysis of data from 6 RCTs of initial cART regimens in the AIDS Clinical Trials Group (ACTG) found no significant association between recent ABC use and risk of MI.15 Because the pooled analysis of GSK data did not stratify by trial, it cannot be considered a proper meta-analysis. The ACTG pooled analysis has the important limitation that only 3 of the 6 trials included in the analysis had a randomized ABC arm.
Given the conflicting results from the observational studies, the SMART trial, and pooled analyses by GSK and ACTG, the US Food and Drug Administration (FDA) set out to conduct a meta-analysis of RCTs in which ABC use was randomized as part of cART to estimate the effect of ABC use on the risk of MI. The meta-analysis of RCTs was undertaken to reduce potential biases that may not be controlled for in analyses of observational studies and to preserve randomization within a trial, which was not accounted for in the GSK and ACTG analyses.
The FDA conducted a literature search in March of 2009 for all clinical trials that included a randomized ABC treatment arm. In the literature search, the term “abacavir” was queried in either the Subject or Title for articles about human studies using 3 databases: IPA, Inteleos, and Embase. GSK conducted a similar search on the Scopus database. Based on the queries by both FDA and GSK, the literature search produced 544 nonunique references. The FDA removed duplicate articles and screened articles meeting the following prespecified inclusion and exclusion criteria:
1. Include parallel-arm randomized trials, where ABC treatment was randomized.
2. Exclude pharmacokinetic trials.
3. Exclude pediatric trials.
4. Exclude trials with fewer than 50 subjects.
5. Exclude prematurely terminated or unfinished trials.
6. Exclude trials conducted in Africa.
Trials with fewer than 50 subjects randomized to ABC were generally pharmacokinetic, phase 1 or phase 2 studies. Terminated or unfinished trials were excluded from the meta-analysis because of the difficulty of obtaining data and possible poor or incomplete adverse event records. Trials conducted in Africa were excluded because of differing cardiovascular risks, comorbidities, and access to care in that subject population.
After removing duplicates, 47 unique articles met the inclusion and exclusion criteria for trial selection. Of the 47 articles, 34 referenced at least one of the 5 ACTG trials or 16 GSK trials. The remaining 13 articles referenced 11 trials conducted by academic centers. The FDA review team contacted GSK and the investigators responsible for these trials to assess their relevance to this meta-analysis and to request subject-level data. A standardized data request was sent to the owners of these data sets.
MI was the prespecified endpoint of interest. Although all available subject-level data was obtained for most clinical trials, the FDA was able to obtain only subject-level data for adverse event reports of MI for 5 trials. For trials in which the FDA was not able to obtain additional subject-level data, summary-level data, such as treatment arm, demographics, and baseline characteristics, were requested from each sponsor/investigator to obtain consistent trial-level information.
All reported MI events were included in this analysis. Events were reported by trial investigators but were not adjudicated by FDA because of lack of access to additional subject-level data for laboratory values, electrocardiograms, patient narratives, and medical records. As such, we were not able to independently determine the number of MI events for each trial.
Statistical analysis methods were specified in advance and documented in a statistical analysis plan (SAP). The primary analysis method was Mantel–Haenszel (M-H) risk difference and the associated 95% CI. This method used all trials, including trials with no MI events. For trials with more than 2 arms, arms that were part of the same comparison group (ABC versus non-ABC) were combined. Calculation of the M-H odds ratio and associated 95% CI was also planned; this method only used trials with at least one event of interest. Additional alternative analysis were planned to assess the robustness of the results, including a random effects model and exact approaches.
As shown in Figure 1, a total of 26 randomized controlled clinical trials were included in the meta-analysis. The FDA procured subject-level data from all 16 GSK trials and 5 academic trials but was unable to get any data for 6 other academic trials, as shown in Table 1. Trials included in the meta-analysis are listed in Table 2. The 5 ACTG trials provided trial-level data after the FDA data request. Within this 26 trial database, 5028 subjects were randomized to ABC-containing cART regimens and 4840 subjects were randomized to non-ABC cART regimens. The mean follow-up for the 26 trials was 719 person-years with a minimum of 42.2 person-years and a maximum of 1257.3 person-years. This resulted in an average duration of follow-up of 1.62 person-years for each subject with a minimum of 0.49 person-years per subject (COL30305) and a maximum of 4.72 person-years per subject (ACTG 372A).
Baseline subject characteristics were not provided to the FDA for any of the 26 trials. If available, these data were obtained from publications. Table 3 depicts a summary of baseline subject characteristics grouped by trial sponsor (GSK, ACTG, and other academic). As shown in Table 3, important baseline covariates, including gender, age, body mass index, CD4 count, and HIV viral load are comparable between the ABC group and the non-ABC group.
Of the 9868 subjects included in the analysis, a total of 46 (0.47%) MI events were reported, including 24 (0.48%) MI events from subjects randomized to an ABC-containing regimen and 22 (0.46%) MI events from subjects randomized to a non-ABC regimen. Table 4 depicts a summary of overall results and results by trial sponsor (GSK, ACTG, and other academic). Overall, no statistically significant difference in MI events was detected between subjects receiving ABC-containing regimens and non-ABC regimens: risk difference = 0.008% with a 95% CI of −0.26% to 0.27% and a corresponding odds ratio of 1.02 with 95% CI of 0.56 to 1.84. Separate analyses by the trial sponsors (GSK, ACTG, and other academic) also did not show statistically significant difference in the MI risk between the ABC-treated subjects and the non-ABC treated subjects. Figure 2 depicts a forest plot of the 26 trials sorted by average duration of follow-up (longest duration at the top to shortest duration at the bottom). No trends regarding total person-years of follow-up were seen in the meta-analysis. No single trial showed a statistically significant increased risk of developing MI between subjects treated with ABC and subjects treated with non-ABC regimens.
In the previous section, results based on M-H risk difference and M-H odds ratio are presented. The M-H risk difference method uses information from all the trials, including those with zero events; in contrast, the M-H odds ratio method excludes trials without events.
Based on the exact and efficient inference procedure, the risk difference between ABC-containing regimen and non-ABC regimen is 0.03%, with a 95% CI of −0.44% to 0.50%.27 This result is consistent with the primary result based on the M-H risk difference. This method also uses data from trials with zero events. Similar to many exact approaches, this exact risk difference tends to provide conservative CIs.
In the prespecified SAP, fixed effect models were chosen over a random effect model because of the small number of events expected. To evaluate the heterogeneity in the data, the Cochran Q statistic and the I2 statistic were calculated based on all the trials with at least one event. The Q statistic was 20.51 and the corresponding P value for test of heterogeneity was 0.75, suggesting that no statistically significant heterogeneity was found. Similarly, the I2 statistic was 0.17, suggesting that only 17% of the between-trial variation was because of heterogeneity rather than to random chance. Therefore, the choice of fixed effects model is appropriate in this meta-analysis. However, a random effects model that incorporates trials with zero events, the modified DerSimonian and Laird (D-L) method, yielded a risk difference of 0.0003% with a 95% CI of −0.26% to 0.26%.28,29 The result is very similar to the primary results based on a fixed effect model (the M-H risk difference).
Stratified odds ratios based on the exact method and the Peto method were also conducted as alternative analyses. The stratified odds ratio based on the exact method was 1.02 with 95% CI of 0.54 to 1.92, and the Peto stratified odds ratio was 1.02 with 95% CI of 0.56 to 1.83. These results are consistent with the primary analysis based on the M-H odds ratio. Note that trials with zero events do not contribute to any of the analyses based on odds ratios.
Duration of Follow-Up
If a differential duration of follow-up were to exist between the ABC and non-ABC groups (eg, ABC subjects might have tended to drop out of a trial earlier than non-ABC subjects), the results of this meta-analysis might be biased. As shown in Table 5, for all 21 trials with available information, the average duration of follow-up was similar between the ABC and the non-ABC groups. Overall, the average duration of follow-up was 1.43 years for the ABC group and 1.49 years for the non-ABC group.
As shown in Table 5, the duration of follow-up was well balanced between the ABC and non-ABC groups within each trial. The duration, however, varied noticeably across the different trials. The interpretation of the risk difference in a trial-level meta-analysis is challenging when trials have different durations. In our meta-analysis, though, results were consistent regardless of the measure of risk; that is, the risk difference or odds ratio.
Recent observational studies have suggested increase in risk of MI for patients with current or recent exposure to ABC. Because residual confounding is not completely controllable in observational studies and because of concerns of multiple testing, we conducted a meta-analysis with a prespecified primary endpoint and SAP of prospective controlled trials in which ABC use was randomized and in which MI risk was moderate (0.45%).
Through a process of literature search, trial identification, and data acquisition, the FDA conducted a meta-analysis based on 26 RCTs in which ABC was randomized as part of cART to estimate the effect of ABC use on MI risk. We originally intended to obtain subject-level data for each trial to conduct a subject-level meta-analysis that used a consistent definition of MI, as this would provide a greater level of evidence than a trial-level meta-analysis. However, subject-level data were not procured for the 5 ACTG trials, thereby not allowing a subject-level meta-analysis. Realizing that a trial-level meta-analysis has some limitations as discussed below, we felt that obtaining key trial characteristics, such as a prespecification of the primary safety endpoint, MI, and a SAP, would provide meaningful information for the research question of interest.
Based on our trial-level meta-analysis, no statistically significant association between the use of ABC and increased risk of developing MI was found. The M-H risk difference between ABC-containing cART regimens and non-ABC cART regimens was 0.008% with a 95% CI of −0.26% to 0.27%. The M-H odds ratio of ABC compared with non-ABC was 1.02 with a 95% CI of 0.56 to 1.84. These results were robust to various alternative analyses.
The major strengths of our meta-analysis are the minimization or elimination of confounding and selection bias through maintaining the ABC randomization within each trial and preserving the study-level randomization. In addition, our analysis was based on the prespecification of a single hypothesis that does not have the weakness of multiple testing associated with it.
One weakness in our analysis is that MI events were not adjudicated and were reported as part of adverse event reporting in clinical trials. In our analysis, MI was based on an FDA request to the trial sponsor/investigator and not on events reported in the literature. In addition, protocols were provided for most studies to confirm event ascertainment similarities. While such steps were meant to limit errors in event ascertainment, absent a consistent and thorough adjudication process from all trials, we were left to rely on event ascertainment as reported by trial investigators.
Another weakness of our meta-analysis is that it is based on trial-level information and not subject-level information because we were not able to procure the subject-level data for all trials included in our meta-analysis. A subject-level analysis with adjudicated event ascertainment would be considered the highest standard in the assessment of MI risk with ABC use. This would allow, for example, an assessment of the timing of events, assessments of data quality, application of consistent methods for event determination, and assessments of informative censoring. Given these limitations, it is still worth noting that discontinuation rates were low overall in these HIV trials, minimizing the possible effects of informative censoring.
One potential limitation associated with the use of clinical-trial data relates to the possibility that subjects enrolled in these clinical trials may be at decreased risk of MI relative to the general population because of various exclusion criteria. This could lessen the likelihood of finding a positive association with MI. However, recent observational studies involving French and Kaiser–Permanente health record databases demonstrate a risk of MI among HIV-infected patients that is similar to that seen in our study.30,31 Furthermore, in the D:A:D study, the relative risk for MI was constant within all risk groups with current use of ABC. If this is true, our study's ability to identify a risk difference observed between ABC and non- ABC randomized subjects should not have been impacted by the particular magnitude of MI risk inherent to the subjects in our study. Although several limitations are noted with our trial-level meta-analysis, to our knowledge this represents the largest trial-level meta-analysis to date of clinical trials in which ABC use was randomized. When taken together with the results from other publications, our meta-analysis raises questions about an association between MI and ABC use, reaffirming that a clear determination of MI risk remains uncertain. For a more certain understanding of the cardiovascular safety profile of ABC use, an appropriately powered RCT with a prespecified analysis plan and adjudicated primary CVD endpoints would need to be conducted.
1. The CASCADE Collaboration. Survival after introduction of HAART in people with known duration of HIV-1 infection. Lancet. 2000;355:1158–1159.
2. Mocroft A, Brettle R, Kirk O, et al.. Changes in the cause of death among HIV positive subjects across Europe: results from the EuroSIDA study. AIDS. 2002;16:1663–1671.
3. Lewden C, Salmon D, Morlat P, et al.. Causes of death among human immunodeficiency virus (HIV)-infected adults in the era of potent antiretroviral therapy: emerging role of hepatitis and cancers, persistent role of AIDS. Int J Epidemiol. 2005;34:121–130.
4. D:A:D Study Group. Combination antiretroviral therapy and the risk of myocardial infarction. N Engl J Med. 2003;349:1993–2003.
5. Triant VA, Lee H, Hadigan C, et al.. Increased acute myocardial infarction rates and cardiovascular risk factors among patients with human immunodeficiency virus disease. J Clin Endocrinol Metab. 2007;92:2506–2512.
6. Currier JS, Taylor A, Boyd F, et al.. Coronary heart disease in HIV-infected individuals. J Acquir Immune Defic Syndr. 2003;33:506–512.
7. Friis-Moller N, Weber R, Reiss P, et al.. Cardiovascular disease risk factors in HIV patients—association with antiretroviral therapy. Results from the DAD Study. AIDS. 2003;17:1179–1193.
8. Friis-Moller N, Sabin CA, Weber R, et al.. Combination antiretroviral therapy and the risk of myocardial infarction. N Engl J Med. 2003;349:1993–2003.
9. The D:A:D Study Group. Class of antiretroviral drugs and the risk of myocardial infarction. N Engl J Med. 2007;356:1723–1735.
10. Sabin C, Worm S, Weber R, et al.. Use of nucleoside reverse transcriptase inhibitors and risk of myocardial infarction in HIV-infected patients enrolled in the DAD Study: a multi-cohort collaboration. Lancet. 2008;371:1417–1426.
11. Obel N, Farkas DK, Kronborg G, et al.. Abacavir and risk of myocardial infarction in HIV-infected patients on highly active antiretroviral therapy: a population-based nationwide cohort study. HIV Med. 2010;11:130–136.
12. Lang S, Mary-Krause M, Cotte L, et al.. 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. Arch Intern Med. 2010;170:1228–1238.
13. Strategies for Management of Anti-Retroviral Therapy/INSIGHT; DAD Study Groups. Use of nucleoside reverse transcriptase inhibitors and risk of myocardial infarction in HIV-infected patients. AIDS. 2008;22:F17–F24.
14. Brothers CH, Hernandez JE, Cutrell AG, et al.. Risk of myocardial infarction and abacavir therapy: no increased risk across 52 GlaxoSmithKline-sponsored clinical trials in adult subjects. J Acquir Immune Defic Syndr. 2009;51:20–28.
15. Ribaudo H, Benson C, Zheng Y, et al.; ACTG A5001/ALLRT Protocol Team. No risk of myocardial infarction associated with initial antiretroviral treatment containing abacavir: short and long-term results from ACTG A5001/ALLRT. Clin Infect Dis. 2011;52:929–940.
16. Podzamczer D, Ferrer E, Sanchez P, et al.; ABCDE (Abacavir vs. d4T (stavudine) plus efavirenz) Study Team. Less lipoatrophy and better lipid profile with abacavir as compared to stavudine: 96-week results of a randomized study. J Acquir Immune Defic Syndr. 2007;44:139–147.
17. Martinez E, Arnaiz JA, Podzamczer D, et al.; Nevirapine, Efavirenz, and Abacavir (NEFA) Study Team. Substitution of nevirapine, efavirenz, or abacavir for protease inhibitors in patients with human immunodeficiency virus infection. N Engl J Med. 2003;349:1036–1046.
18. Martin A, Bloch M, Amin J, et al.; STEAL Study Group. Simplification of antiretroviral therapy with tenofovir-emtricitabine or abacavir-lamivudine: a randomized, 96 week trial. Clin Infect Dis. 2009;49:1591–1601.
19. MacArthur RD, Novak RM, Peng G, et al.; Community Programs for Clinical Research on AIDS (CPCRA) 058 Study Team. A comparison of three highly active antiretroviral treatment strategies consisting of non-nucleoside reverse transcriptase inhibitors, protease inhibitors, or both in the presence of nucleoside reverse transcriptase inhibitors as initial therapy (CPCRA 058 First Study): a long-term randomized trial. Lancet. 2006;368:2125–2135.
20. Martinez E, Arranz JA, Podzamczer D, et al.; BICOMBO Study Team. A simplification trial switching from nucleoside reverse transcriptase inhibitors to once-daily fixed-dose abacavir/lamivudine or tenofovir/emtricitabine in HIV-1-infected patients with virological suppression. J Acquir Immune Defic Syndr. 2009;51:290–297.
21. García-Benayas T, Blanco F, Alcolea A, et al.. Short communication: benefits in the lipid profile after substitution of abacavir for stavudine: a 48-week prospective study. AIDS Res Hum Retroviruses. 2004;20:1289–1292.
22. Moyle G, Sabin CA, Cartledge J, et al.. A randomized comparative trial of tenofovir DF or abacavir as replacement for a thymidine analogue in persons with lipoatrophy. AIDS. 2006;20:2043–2050.
23. Phase II Comparator Study of Substitution of Tenofovir or Abacavir Receiving Thymidine Analogue as Part of HAART. ClinicalTrials.gov identifier: NCT00270556. www.clinicaltrials.gov
24. Opravil M, Hirschel B, Lazzarin A, et al.. A randomized trial of simplified maintenance therapy with abacavir, lamivudine, and zidovudine in human immunodeficiency virus infection. J Infect Dis. 2002;185:1251–1260.
25. Carr A, Workman C, Smith DE, et al.. Abacavir substitution for nucleoside analogs in patient with HIV lipoatrophy: a randomized trial. JAMA. 2002;288:207–215.
26. Gerstoft J, Kirk O, Obel N, et al.. Low efficacy and high frequency of adverse events in a randomized trial of the triple nucleoside regimen abacavir, stavudine and didanosine. AIDS. 2003;17:2045–2052.
27. Tain L, Cai T, Pfeffer M, et al.. Exact and efficient inference procedure for meta-analysis and its application to the analysis of independent 2x2 tables with all available data but without artificial continuity correction. Biostatistics. 2009;10:275–281.
28. Emerson J, Hoaglin D, Mosteller F. Simple robust procedures for combining risk differences in sets of 2×2 tables. Stat Med. 1997;15:1465–1488.
29. Laird N, Fitzmaurice G, Ding X. Comments on ‘Empirical vs natural weighting in random effects meta-analysis’. Stat Med. 2010;29:1266–1267.
30. Klein D, Leyden W, Xu L, et al.. Contribution of immunodeficiency to CHD: cohort study of HIV+ and HIV- Kaiser Permanente members. Presented at: the 18th Conference on Retroviruses and Opportunistic Infections, February 28, 2011; Boston, MA: Abstract 810.
31. Lang S, Mary-Krause M, Cotte L, et al.. Increased risk of myocardial infarction in HIV-infected patients in France, relative to the general population. AIDS. 2010;24:1228–1230.
32. Greenland S, Robins JM. Estimation of a common effect parameter from sparse follow-up data. Biometrics. 1985;41:55–68.
© 2012 Lippincott Williams & Wilkins, Inc.