Risk of cardiovascular events from current, recent, and cumulative exposure to abacavir among persons living with HIV who were receiving antiretroviral therapy in the United States: a cohort study

The risk of CVDe was assessed among PLWH who started ARV drugs in the U.S. between October 1, 2009 and December 31, 2014. Data were obtained from medical and prescription claims data included in the IMS’ PharMetrics Plus database. October 1, 2009 was the earliest possible date for complete availability of relevant data; ART prescription history prior to this date was not available. PharMetrics Plus is a large health plan insurance claims database in the U.S., and is comprised of adjudicated claims for more than 150 million unique enrollees from across four regions of the U.S. [25]. The data undergo a series of quality checks to minimize errors. This study used a pre-defined algorithm (Fig. 1) to extract and define the study population of PLWH exposed to any ART in the database. The study population was restricted to those ≥18 years of age. The baseline time point was defined as the date of ART initiation in the database and individual follow up time was censored at the first of three events after baseline: 1) first occurrence of CVDe, 2) last recorded date of ART receipt, 3) December 31, 2014. The study was approved by the Committee for Protection of Human Subjects at the University of California, Berkeley.

Exposures to specific ARV agents were identified by generic product identifier (GPI) codes. Person-time of exposure to abacavir was compared to exposure to ARV agents other than abacavir. Any two prescriptions for an ARV agent separated by <30 days were combined to represent a single continuous exposure; gaps ≥30 days were not combined and this person-time was not included in the analysis. These data are longitudinal, and each subject’s follow up time was divided into consecutive one-month periods during which treatment was allowed to vary. The values of covariates were updated at the start of each month. The outcome of CVDe for an individual was defined as the first occurrence of an AMI or receipt of a coronary artery intervention procedure (i.e. percutaneous coronary intervention or coronary artery bypass graft) after baseline. AMI and coronary artery intervention procedures were ascertained using the International Classification of Disease, 9th Revision, Clinical Modification (ICD-9-CM) or Current Procedural Terminology (CPT) codes, respectively (Additional file 1: Table S1). The ICD-9 code used for AMI (410.xx) has been previously validated in another claims database [26].

The temporal ordering of covariate, treatment, and outcome allowed for a time-varying analysis, and the opportunity for causal interpretations. The first observation of a time-dependent covariate corresponded to its baseline value and once a health condition developed, an individual was assumed to have the condition for the remainder of the study. Current exposure to abacavir was defined as exposure (yes/no) during each one-month observation period. Recent exposure was defined as exposure (yes/no) in the previous six months (inclusive of the current month). Cumulative exposure was defined as the total duration of exposure an individual had received at a particular time point in one-month increments, updated monthly. Duration of exposure ceased to accumulate upon discontinuation of the drug but resumed if the drug was restarted. HIV-infection status and covariates were ascertained using the ICD-9-CM or CPT codes (Additional file 1: Table S1).

The risk of CVDe from a current, recent, and cumulative exposure to abacavir was estimated by the parameters of pooled logistic regression marginal structural models using stabilized inverse probability of treatment weights (sIPTW) [27]. The sIPTW was generated from four treatment models – two each for the numerator and the denominator of the weight [16]. For the denominator, the time point specific probability of exposure initiation was first estimated by fitting a main term pooled logistic regression to data up to the individual’s first month of receiving the exposure or end of follow up for those who were never exposed. The probability of exposure continuation was then estimated by fitting the model to data after the first month of starting the exposure. The denominator was modelled as a function of baseline covariates: gender, tobacco use/smoking (ever), substance or alcohol abuse (ever), serologic evidence of hepatitis B and C infections, history of stroke, cancer or old myocardial infarction, and time-dependent covariates: age, year of ART initiation, body weight, receipt of hypoglycemic agents (i.e. sulfonylureas, biguanides, insulin, thiazolidinedione) or medications for CVDe (i.e. aspirin, beta-blocker, angiotensin converting enzyme inhibitor, angiotensin receptor blocker, calcium channel blocker, statins) or diagnoses of: chronic kidney disease (CKD), dyslipidemia, heart failure, cardiac dysrhythmia, atherosclerosis, diabetes mellitus, and hypertension. The exposure continuation model additionally contained a variable for past month’s exposure status. The probabilities for the numerator of the sIPTW were similarly modelled but as a function of baseline covariates only. The follow-up time was modeled as a function of natural cubic splines with three internal knots placed at 25th, 50th and 75th percentiles. The marginal structural model was adjusted for the sIPTW and the baseline covariates. Same treatment weights were used for estimation of CVDe risk from current, recent, and cumulative exposure to abacavir. In order to assess the change in risk over time, the adjusted and marginal models were fit as a function of categories of cumulative exposure, i.e., never exposed, 1–6, 7–12, 13–18, 19–24, and >25 months of exposure. In sensitivity analyses, the study population was restricted to individuals free of CVD at baseline, and to individuals without a history of alcohol and substance abuse at baseline. Sensitivity analyses were additionally carried out to assess if the risk of CVDe from abacavir exposure differed after adjusting for other anti-retroviral agents. Using the same sIPTW models, we tested for interaction to see whether risk of CVDe from current abacavir exposure is modified in the presence of 13 different risk factors (Additional file 1: Table S5). In addition to the marginal structural results, corresponding results from unadjusted and adjusted Cox models were calculated. This study assumes uninformative censoring. Data were extracted and processed from the main claims databases using TERADATA (Dayton, OH), SAS version 9.1 (SAS Institute, Cary, NC), and STATA version 13.1 (StataCorp, College Station, TX). The marginal structural models were implemented in STATA version 13.1, based on Fewell et al. [28]. The rationale, definition, and implementation of the marginal structural models are described in Additional file 1: Appendix 1.

There were 72,733 participants contributing 114,470 person-years of exposure to antiretroviral agents. On average, participants were exposed to ART for 1.5 years. The mean age of the study population was 46 years and 82% were males. The characteristics of the study population at baseline and summary of exposure to various antiretroviral drugs are described in Tables 1, 2 respectively. Overall, 714 CVDe occurred at an incidence rate of 6.23 (95% CI: 5.80, 6.71)/1000 person-years. Of the 714 outcomes, 137 were observed over 14,060 person-years of current exposure to abacavir at an incidence rate of 9.74 (95% CI: 8.24, 11.52)/1000 person-years, as compared to 577 outcomes over 100,410 person-years with an incidence rate of 5.75 (95% CI: 5.30, 6.24)/1000 person-years for those currently exposed to other ARV drugs. The incidence rate was highest for those exposed to abacavir between 13 and 18 months (11.32/1000 person-years) (Table 3). Of the 714 CVDe, 548 were cases of AMI. The overall incidence rate of AMI was 4.78 (95% CI: 4.39, 5.19)/1000 person-years (Additional file 1: Table S2). We calculated a population attributable risk (PAR) associated with abacavir exposure as:

At baseline, abacavir recipients had a higher prevalence of essential hypertension, diabetes mellitus, chronic kidney disease (CKD), dyslipidemia, lipodystrophy, heart disease, and use of cardiovascular medications (Table 1). In the pooled logistic regression model, older age, a diagnosis of CKD, symptomatic HIV infection, and presence of lipodystrophy were associated with an increased probability of receiving abacavir (Additional file 1: Table S3).

The sIPTW models showed the risk of CVDe (HR; 95% CI) was increased for current (1.43; 1.18, 1.73), recent (1.41; 1.16, 1.70) and cumulative (1.18; 1.06, 1.31) exposure (per year) to abacavir (Table 4). Separate models were run for each of current, recent, and cumulative exposure. The unadjusted and adjusted Cox models also showed increased risk for these exposures (Table 4). On further assessment of the risk from cumulative exposure, the HR varied with the duration of exposure in an inverted U-shaped pattern (Table 5 and Fig. 2 ); the relative hazard continued to increase up to 24 months of exposure, after which it decreased to non-significant levels but remained elevated compared to those never exposed to abacavir. Older age, male sex, tobacco use, other heart diseases, prior AMI, use of CVD-related medications, diabetes mellitus, and dyslipidemia were each associated with increased hazard of CVDe in the adjusted Cox model (Additional file 1: Table S4). We also assessed whether the risk was reversible after six months of stopping abacavir by comparing those with any abacavir exposure prior to but not in the last six months including the current month to those never exposed and found that the risk (HR; 95% CI) remained elevated (sIPTW model: 1.69; 0.89, 3.20; adjusted Cox model: 2.08; 1.17, 3.71). In tests of interactions, we observed that the risk of CVDe associated with abacavir use was more pronounced for age < 45 years (interaction p-value: 0.028) and for people without prior heart disease (interaction p-value: 0.016) (Additional file 1: Table S5).

In a sensitivity analysis, we observed a 53% higher risk of CVDe (sIPTW model) for current exposure to abacavir among individuals without a prior AMI or heart disease at baseline (Additional file 1: Table S6). This relationship was also assessed by excluding other heart diseases (heart failure, cardiac arrhythmia, atherosclerosis, or receipt of cardiovascular medications) from the adjustment set of covariates for both the marginal and the adjusted Cox model, with similar results. The risk also remained elevated by 41% when the study population was restricted to individuals not using illicit substances or alcohol at baseline (Additional file 1: Table S6). We further tested for CVDe risk from abacavir use after adjusting for cumulative exposure to other antiretroviral agents (tenofovir, emtricitabine, zidovudine, lamivudine, lopinavir, atazanavir, darunavir, efavirenz, nevirapine, rilpivirine, and raltegravir) using sIPTW models and found elevated risk (HR; 95% CI) for current (1.38; 1.12, 1.68), recent (1.34; 1.09, 1.64), and cumulative exposure (1.16; 1.03, 1.31). We then replicated the D:A:D model [5] for cumulative exposure by including recent exposure in the same model as cumulative exposure and observed that although our hazard ratio estimate for risk from cumulative exposure (per year) remained elevated (HR:1.08; 95% CI: 0.89, 1.30), it decreased to a non-statistically significant level. When we modelled the risk by partitioning the cumulative exposure into various durations, we observed a similar increased risk [HR (95% CI)] pattern as observed in our primary analysis (Table 4): 1–6 months: 1.91 (0.95–3.83); 7–12 months: 2.58 (1.16–5.71); 13–18 months: 2.68 (1.17–6.11); 19–24 months: 2.90 (1.37–6.17); and ≥25 months: 2.13 (0.93–4.88).