Background
Cardiovascular disease (CVD) accounts for approximately 16% of deaths among persons living with HIV (PLWH) [
1]. Risk factors for CVD are more prevalent among PLWH [
2], and use of various antiretroviral (ARV) drugs has been shown to be associated with an increased risk of CVD [
3]. With rapid expansion of antiretroviral therapy (ART) coverage both domestically and abroad, researchers and clinicians have become increasingly aware of potential ARV drug-related adverse events. Whether the commonly used ARV drug abacavir is associated with an increased risk of CVD has been intensely debated. Abacavir sulfate is a guanosine analog nucleoside reverse transcriptase inhibitor that possesses retroviral suppressive properties similar to tenofovir [
4], and is a commonly prescribed backbone ARV agent. However, the writing of prescriptions of abacavir declined after the Data Collection on Adverse Events of Anti-HIV Drugs (D:A:D) study group reported in 2008 an increased risk of acute myocardial infarction (AMI) among PLWH exposed to abacavir [
5‐
7]. Independent investigations that were subsequently carried out have both supported [
7‐
17] and refuted [
18‐
23] the D:A:D study group’s findings.
While studies conducted more recently have generally suggested an increased risk of CVD from abacavir exposure [
8,
10,
14,
17], they were limited by few outcomes, with results occasionally underpowered [
8,
17]. Failure to identify a clear underlying biological mechanism to explain the epidemiologic findings has added to the deliberation [
24]. Furthermore, there has also been a lack of consensus regarding whether the risk of CVD from exposure to abacavir reverses within a few months of stopping the drug [
5,
16] and a lack of understanding on how the risk varies as exposure accumulates. In this study, we have sought to address these limitations by investigating the risk of CVD events (CVDe) from current, recent, and cumulative exposure to abacavir among PLWH using conventional and causal statistical methods.
Discussion
In a large database claims-based study, we found an increased risk of CVDe associated with exposure to current, recent, and cumulative exposure to abacavir using both adjusted Cox and marginal structural models estimated with inverse probability treatment weights. The overall incidence rate of AMI in this study was 4.78/1000 person-years, which compares to 3.3/1000 person-years in the 2008 D:A:D study. AMI incidences of 1.41/1000 people and 1.2/1000 people were seen in the general population in Olmstead county in Minnesota in 2006 and in men 35–65 years of age in the Framingham study population, respectively [
29,
30]. This relatively higher incidence of AMI in the PLWH could be due to HIV infection [
31‐
33], ART use [
3], or both; PLWH have been shown to have more risk factors for CVD as compared to the general population [
31‐
33]. The incidence rates of AMI associated with exposure to abacavir in this study (6.9/1000 person-years) and in the D:A:D study (6.1/1000 person-years) were ~4–5 fold higher than the general population estimates and approximately 2-fold higher than in the general population of PLWH [
2,
31‐
34]. Some of the difference in results between this study and the D:A:D study including higher incidence rate of AMI in this study could be because participants in this study were all exposed to ART whereas the D:A:D study included individuals who had not yet started ART, as well as those who had discontinued ART totally. We calculated a population attributable risk of 8%. This means 8% (
n = 57) of the total CVDe risk in the PLWH could be prevented if abacavir was not used, assuming a causal relationship between abacavir use and CVDe risk.
In an attempt to characterize an underlying biological mechanism for the increase in CVDe risk associated with abacavir use, we assessed how the risk varied with duration of exposure. The relative hazard of AMI increased with increasing duration of exposure in an inverted U-shaped pattern, peaking between 13 and 24 months of exposure and leveling off thereafter, suggesting a dose response relationship between cumulative time exposed to abacavir and risk, up to 24 months. This result agrees with earlier finding by Young et al. in which they first showed that the risk of CVD increased with increasing duration of exposure, with greatest risk between 6 and 36 months and exposure beyond 36 months adding little to the risk, suggesting a dose-response pattern. We observed the dose-response relationship for various durations of cumulative exposure after controlling for recent exposure as well in addition to other variables in the model; the D:A:D study group [
5] had reported that the observed risk for cumulative exposure disappeared after adjusting for recent exposure, meaning that the CVD risk existed only up to first 6 months of exposure, after which the risk reversed. In a separate model, we tested the risk reversibility and found a 69% increased risk of CVDe among those who had stopped abacavir prior to last six months, suggesting a risk not reversible within six months of stopping the drug. We did not formally test whether the inverted U-shaped curve described for cumulative exposure provides a better fit to the observed risk estimates than a simple linear association.
Whereas this and Young et al.’s study results do not support an underlying mechanism related to immediate exposure to abacavir, the results are not consistent with an atherogenic mechanism, in which an ongoing or increasing risk would be expected with an increasing duration of exposure, without necessarily reaching a peak effect and leveling off after 24 months. The finding of an early peak in the increased risk of AMI in the course of abacavir treatment is helpful in understanding how risk may change with continuing versus changing therapy. The study results presented here suggest a reversible but more gradual underlying mechanism with a longer lasting impact that regresses more slowly after removal of the exposure.
Prior work has suggested that abacavir-induced platelet hyper-reactivity and aggregation could potentially lead to thrombosis and myocardial infarction [
35‐
37]. Specifically, abacavir may induce platelet hyper-reactivity by competitive inhibition of a nitric oxide-induced soluble guanylyl cyclase via its active metabolite, carbovir-triphosphate, leading to a decreased production of cyclic guanosine monophosphate, an inhibitor of platelet aggregation and secretion [
24,
35,
36,
38]. It is possible that abacavir may trigger an acute platelet response leading to endothelial injury with a longer lasting impact. It is also unclear whether abacavir may exert its effect on CVD risk through an increase in inflammatory biomarkers. While the SMART/INSIGHT study investigators [
15], Kristoffersen et al. [
39], and Hileman et al. [
40] showed evidence for a possible role of inflammatory biomarkers in causing CVD among abacavir users [e.g. increased levels of high sensitivity c-reactive protein (hsCRP) and interleukin-6], several other studies have shown that levels of inflammatory biomarkers such as hsCRP, interleukin-6, selectin P and E, D-dimer, vascular adhesion molecule-1, intercellular adhesion molecule-1, and tumor necrosis factor alpha are not elevated after exposure to abacavir [
41‐
54]. Future interdisciplinary studies may explore these areas by bridging basic, translational and clinical science to provide additional insights into the mechanisms underlying abacavir-associated cardiovascular risk. We have not established a clear reason for observing a higher risk of CVDe associated with abacavir use among the younger age-group and individuals without a pre-existing cardiac condition in the test of interactions. While we acknowledge the exploratory nature of the analyses for interaction testing with the possibility that the results could be due to chance, the observation of a higher CVDe risk in individuals without prior heart disease may stand to support the finding of an increased risk in younger age people. The increased CVDe risk in younger age people could also reflect a higher prevalence of cocaine and injection drug use among them [
19]. It would be important to test in other populations whether CVD risk associated with abacavir use differs by age.
We used the sIPTW approach because individuals with certain risk factors for CVD such as CKD, hypertension, diabetes mellitus, and dyslipidemia, may be preferentially channeled into (or away from) receiving abacavir based on its known toxicity in the presence of these conditions. The sIPTW approach may also be necessary because post-baseline values of these variables may simultaneously serve as confounders and causal intermediates; adjusting for these through traditional methods can lead to biased results [
55]. Under such settings, the use of inverse probability weights provides a valuable tool for balancing confounders across exposure groups without conditioning on variables affected by treatment [
55,
56]. Some of our results for various durations of cumulative exposure appreciably differed between conventional Cox models and marginal structural models. For example, the hazard ratios (95% CI,
p value) for 7–12 months, 13–18 months, and 19–24 months of cumulative exposure were 1.27 (0.87–1.86;
p = 0.219), 1.71 (1.10, 2.65;
p = 0.016), and 1.62 (0.98, 2.69;
p = 0.060), respectively, in adjusted Cox models. The corresponding hazard ratios from marginal structural models were 1.41 (0.97, 2.06;
p = 0.073), 1.78 (1.16, 2.72;
p = 0.009), and 1.90 (1.16, 3.11;
p = 0.011) (Table
4).
A key strength of this study is the application of conventional and robust methods to address key study questions while using a very large U.S. health-plan dataset containing longitudinal information on usage of ART in >70,000 PLWH receiving care across the U.S. The recency of the data is an asset. Most studies that showed an association between abacavir use and CVD risk so far were hospital based [
5,
9,
10,
14,
16‐
19] and hence may be subject to similar bias, such as channeling bias, that could arise from specific prescription behavior of physicians. Therefore, reproduction of the results in another representative population, such as that enrolled in the claims database, would be relevant and important. The similarity of these results to those from prior studies, the reproducibility of the results in the sensitivity analyses, and the finding of a background incidence rate of AMI comparable to that found in other studies are reassuring. A limitation of the study is that the ICD-9 and CPT diagnostic codes used may be prone to coding errors; however, such errors are likely to affect the exposure groups non-differentially and may not bias the study results. It is possible that information on covariates, such as body-weight, for which re-imbursement may not be sought could be under-reported in the database. Again, we expect this problem to exist non-differentially across exposure groups. This is an observational cohort study and is therefore subject to confounding from unmeasured factors and possible channeling bias; we have attempted to account for the latter by adopting an sIPTW-based analytic approach. Covariates that could be relevant but not available in the claims database and hence missing in our study are race/ethnicity, CD4 cell count, and HIV viral load. Adjustment for CD4 cell count and HIV viral load made little difference to the relative rate of AMI in a prior study [
5]. There is potential for bias in the study results from residual confounding that may arise from the binary categorization of certain variables in the study, rather than having a graded continuous response. We assumed uninformative censoring for the study because participants in both the exposure groups, i.e., PLWH receiving abacavir based ART regimen and PLWH receiving non-abacavir based ART regimen, may be at similar risk of adverse HIV–related life events that may cease their continued enrollment into the health-plan and hence representation in the database. We chose AMI and/or coronary artery interventions only to define CVDe so as to be as specific as possible with the study outcome’s representation of ischemic CVD; however, we might use a broader definition including other cardiac conditions or cerebrovascular events for the study outcome.