- Top of page
Highly active antiretroviral therapy (HAART) has resulted in dramatic declines in progression to AIDS and AIDS mortality in HIV-infected persons [1–6]. In some reports, death from nonAIDS-related causes exceed those attributed to AIDS in this patient population [2,7–9], and cardiovascular deaths are increasing in these patients [1,2]. Recent reports suggest a shift in the relative cause of death among HIV-infected individuals, with cardiovascular deaths accounting for fewer than 4% of all deaths pre-1997 [1,10], and for 7–10% in more recent years [8,10,11].
Reports from large observational studies demonstrate that considerable controversy exists over the association of HAART, particularly protease inhibitor (PI) therapies, with increased cardiovascular disease (CVD) risk [12–17]. PIs have been associated with alterations in surrogate markers of CVD, including coronary calcium scores and endothelial function [18,19], as well as with metabolic complications such as hyperlipidaemia, fat redistribution, insulin resistance, hypertension and diabetes mellitus [19–24]. Also, HIV-infected patients may have a higher prevalence of traditional CVD risk factors such as smoking than the general population [20,21]. In addition, as the mean age of the HIV-infected patient population has increased as a result of longer life expectancy with the disease, the consequent cardiovascular risk has also increased.
These changes in patient demographics and the increasing prevalence of traditional CVD risk factors such as smoking, along with the increasing prevalence of PI-associated metabolic complications, have clearly increased the risk of CVD in HIV-infected individuals. In this paper, we use a large longitudinal database to address the question: does PI exposure increase the risk of CVD in HIV-infected patients after adjusting for known CVD risk factors?
- Top of page
A total of 7542 patients met the enrolment criteria. The median duration of follow-up was 3.5 years (mean 3.5 years; maximum 7.4 years) and 2 years (mean 2.5 years; maximum 7.4 years) for the PI and nonPI groups, respectively. The median PI exposure time was 1.7 years (mean 2 years). Over 95% of the study patients in the PI arm had a minimum of 1 month of PI exposure, and over 75% had greater than 6 months of PI exposure. In the nonPI group, 67% were exposed to nonnucleoside reverse transcriptase inhibitors (NNRTIs), and the others were treated with nucleoside reverse transcriptase inhibitors (NRTIs) only.
As shown in Table 1, significant demographic differences were observed between the exposure groups. Significant differences were also found for the following risk factors: current smoking (more common among the nonPI group); hypertension (more common among the nonPI group); pre-existing hyperlipidaemia (more common in the PI group).
Table 1. Demographics and risk factor distribution
|Characteristic||PI group (n=5787)||NonPI group (n=1755)||P-value|
|Median duration of follow–up (years) (IQR)||3.5 (1.67–5.46)||2.0 (0.75–3.89)||<0.0001|
|Age (years) [mean (SD)]||39.4 (8.30)||38.7 (8.90)||0.0012|
|Age (years) [n (%)]|
| 18–34||1679 (29%)||604 (34%)||<0.0001|
| 35–49||3461 (60%)||951 (54%)||<0.0001|
| 50–64||600 (10%)||188 (11%)||NS|
| 65+||47 (1%)||12 (1%)||NS|
|Gender [n male (%)]||5078 (88%)||1402 (80%)||<0.0001|
|Race [n (%)]|
| White||3501 (61%)||883 (50%)||<0.0001|
| African American||1430 (25%)||587 (33%)||<0.0001|
|Smoking status [n (%)]|
| Current||1953 (34%)||678 (39%)||0.0002|
| Past||787 (14%)||230 (13%)||NS|
|Weight (kg) [mean (SD)]||169.1 (32.8)||170.1 (33.1)||NS|
|IVDU [n (%)]||42 (1%)||17 (1%)||NS|
|Cocaine use [n (%)]||110 (2%)||39 (2%)||NS|
|Hypertension [n (%)]||271 (5%)||109 (6%)||0.01|
|Diabetes mellitus [n (%)]||65 (1%)||16 (1%)||NS|
|Pre-existing CVD [n (%)]||7 (<1%)||1 (<1%)||NS|
|Pre-existing hyperlipidaemia [n (%)]||553 (10%)||83 (5%)||<0.0001|
A total of 127 CVD events were observed, with 112 in the PI group for an adjusted event rate of 9.8/1000 PYFU, and 15 in the nonPI group for an adjusted event rate of 6.5/1000 PYFU (P=0.0008). Among patients in the 35 to 65-year-old subset, CVD event rates were also higher in the PI group (11.5/1000 PYFU vs. 7.9/1000 PYFU; P=0.01).
In univariate analyses, cumulative PI therapy for ≥60 days, current or past smoking, age 35–49, 50–64 or ≥65 years, and prior history of hypertension, diabetes, CVD or hyperlipidaemia were significantly associated with higher CVD event risk (Table 2). In the multivariate regression model adjusting for all risk factors, cumulative PI therapy for ≥60 days was associated with an increased risk of CVD events (P=0.03) (Table 2). Other independent risk factors for CVD events were: current smoking (P<0.0001), past smoking (P=0.03), age 35–49 years (P=0.004), age 50–64 years (P=<0.0001), age ≥65 years (P<0.0001) (reference group: age 18–34 years), hypertension (P=0.03), diabetes mellitus (P=0.0006) and pre-existing CVD (P<0.0001). Gender, pre-existing hyperlipidaemia, race, cocaine use, IVDU and weight were not significantly associated with CVD events.
Table 2. Cox proportional hazards regression model: significant predictors of time to first cardiovascular disease event for all patients (n=7542)
|Risk factor||CVD events [n (%)]||Univariate analysis HR (95% CI)||Multivariate analysis HR (95% CI)|
|Yes (n=127)||No (n=7415)|
|PI exposure≥60 days|
| Yes||105 (82.7)||4707 (63.5)||1.69 (1.07–2.68)||1.71 (1.07–2.74)|
| No||22 (17.3)||2708 (36.5)||Reference||Reference|
| Current||56 (44.1)||2575 (34.7)||1.62 (1.14–2.30)||2.40 (1.59–3.64)|
| Past||29 (22.8)||988 (13.30||1.80 (1.18–2.69)||1.74 (1.06–2.84)|
| Never||42 (33.1)||3852 (52)||Reference||Reference|
| 35–49||65 (51.2)||4347 (58.6)||2.90 (1.53–5.49)||2.57 (1.35–4.88)|
| 50–64||42 (33.1)||746 (10.1)||10.93 (5.63–21.23)||8.09 (4.04–16.19)|
| ≥65||9 (7.1)||50 (0.7)||38.97 (16.14–94.07)||32.04 (12.94–79.34)|
| < 35||11 (8.7)||2272 (30.6)||Reference||Reference|
| Yes||19 (15)||361 (4.9)||3.68 (2.26–5.99)||1.80 (1.07–3.03)|
| No||108 (85)||7054 (95.1)||Reference||Reference|
| Yes||5 (3.9)||76 (1)||5.25 (2.15–12.86)||3.59 (1.44–8.95)|
| No||122 (96.1)||7339 (99)||Reference||Reference|
|Evidence of pre-existing CVD|
| Yes||8 (6.3)||0 (0)||69.17 (33.7–141.94)||19.88 (8.68–45.55)|
| No||119 (93.7)||7415 (100)||Reference||Reference|
|Evidence of hyperlipidaemia|
| Yes||21 (16.5)||615 (8.3)||1.63 (1.02–2.60)||1.11 (0.69–1.80)|
| No||106 (83.5)||6800 (91.7)||Reference||Reference|
In a model defining age as a continuous variable, increasing age was associated with a HRadj of 1.09 (95% CI 1.08–1.11) for CVD events. PI exposure was still associated with risk of CVD event with a HRadj of 1.75 (95% CI 1.09–2.78). When all nonsignificant variables were removed from the model, there was no change in the hazards ratio.
Several sensitivity analyses were conducted to test the robustness of the findings (Fig. 1). In a multivariate regression model for the subset of patients aged 35–65 years (n=5200), adjusting for all risk factors, cumulative PI exposure ≥60 days was also associated with an increased risk of CVD (HRadj 1.90; 95% CI 1.13–3.20). Follow-up time for the PI group was truncated in two different analyses to achieve the same median and mean follow-up times as in the nonPI group. As shown in Fig. 1, PI exposure ≥60 days was still associated with increased risk of CVD events [(analysis with similar median follow-up times HRadj 1.53; 95% CI 0.79–2.95) and (analysis with similar mean follow-up times HRadj 2.07; 95% CI 1.18–3.66)].
Figure 1. Cox proportional hazards regression models sensitivity analyses of protease inhibitor (PI) exposure ≥60 days and adjusted risk of cardiovascular disease (CVD) events. Reference groups for all analyses were the non-PI-exposed groups. All models were adjusted for age, gender, ethnicity, smoking status, weight, cocaine use, intravenous drug use, hypertension, diabetes mellitus, pre-existing CVD and hyperlipidaemia. f/u, follow up.
Download figure to PowerPoint
The PI exposure variable was redefined to determine whether increasing exposure duration had an impact on CVD risk (Fig. 2). Three exposure categories were defined as follows: PI exposure of 1 to <180 days (group A); 180 to <365 days (group B); ≥365 days (group C). In the multivariate regression model adjusting for all risk factors, a dose effect was observed, with only group C patients having an increased risk of CVD events (HRadj 1.51; 95% CI 0.98–2.32). Applying this model to the 35 to 65-year-old subset resulted in similar results, with group C patients having an increased risk of CVD events (HRadj 1.63; 95% CI 1.01–2.63).
Figure 2. Cox proportional hazards regression models sensitivity analyses of protease inhibitor (PI) exposure for duration 1 to <180 days (group A), 180 to <365 days (group B) and ≥365 days (group C), and adjusted risk of cardiovascular disease (CVD) events. Reference groups for all analyses were the nonPI-exposed groups. All models were adjusted for age, gender, ethnicity, smoking status, weight, cocaine use, intravenous drug use, hypertension, diabetes mellitus, pre-existing CVD and hyperlipidaemia.
Download figure to PowerPoint
With all nonsignificant variables removed and age defined as a continuous variable, additional analyses were run with the outcome variably defined as (1) only AMI (31 events) or (2) a composite endpoint of CHD (including AMI, PTCA, CABG, angina and CAD) (93 events). In these analyses, the HRadj (95% CI) for PI exposure was 1.76 (0.66–4.64) for AMI and 1.88 (1.07–3.31) for CHD. The distribution of all CVD events is shown in Table 3.
Table 3. Distribution of cardiovascular disease events
|CVD event [n (%)]||PI group (n=112)||NonPI group (n=15)|
|AMI||25 (22)||5 (33)|
|Angina pectoris||32 (29)||3 (20)|
|CAD||23 (21)||3 (20)|
|CVA||17 (15)||2 (13)|
|TIA||9 (8)||1 (7)|
|PVD||4 (4)||1 (7)|
|PTCA||0 (0)||0 (0)|
|CABG||2 (<1)||0 (0)|
Additional sensitivity analyses were performed to test whether measures of HIV-infection duration and disease history, such as nadir CD4 T-cell count, duration of known positive HIV test and cumulative NRTI exposure, could account for the observed effect. Adding all three variables to the base model, the estimated HRadj (95% CI) for PI exposure was 2.7 (1.4–5.1). Additionally, we tested whether calendar year of index might modify the observed effect, and this did not change the results.
- Top of page
The results of this analysis suggest an increased CVD event rate in HIV-infected patients exposed to PI therapies. In multivariable analyses adjusting for most major CVD risk factors, there was an increased risk of CVD events in patients exposed to PI therapies. The observed increased risk was statistically significant in the primary and subgroup analyses.
The sensitivity analyses conducted showed similar increased risk estimates for the PI-exposed group, even after follow-up was truncated in the PI group to achieve similar median and mean follow-up times. However, the 95% CI of the risk estimate from the model truncating follow-up time to the same median included 1. Also, when measures of disease duration and disease history (such as nadir CD4 T-cell count, duration of known HIV-positive test and cumulative NRTI exposure) were tested in the model, there was no change in the direction and significance of the PI effect.
As expected, the major CVD risk factors were significant independent risk factors for CVD in this population of patients. The strongest risk was associated with increasing age category and pre-existing CVD events. The observed increased risk of CVD events associated with PI exposure is probably clinically meaningful as most major CVD risk factors, with the exception of family history, were adjusted for in this analysis.
We cannot fully explain the absence of a gender effect in this analysis. Besides the relatively low proportion of females in this analysis, an explanation for this might be the more pronounced metabolic derangement associated with HAART therapy seen in female patients when compared to male patients . Also, in the United States, HIV-infected male patients with hyperlipidaemia may be more likely to receive lipid-lowering agents than infected female patients . These factors might combine to alter the relative risk of a CVD outcome in women compared to men; hence the CVD risk differences between the genders might not be as evident in treated HIV-infected persons as is seen in the general population.
We have not accounted for the subsequent development of diabetes and hyperlipidaemia after the index date, as the development of these abnormalities has been linked to the use of PI agents, and is expected to be in the causal pathway for eventual CVD disease. Not adjusting for these abnormalities is not therefore likely to introduce any bias in the analyses performed.
Compared to some other recent analyses [14,16], a greater number of major CVD risk factors were accounted for. We also defined the outcome events as would be done in a clinical practice setting, with all abstracted events verified by independent medical abstractors from the sponsoring agency for the database using discharge records and clinic records. Although other studies have shown a relationship between antiretroviral drugs and MI, to our knowledge there are very few studies on other cardiovascular outcomes in this patient population. We feel that this is a more clinically relevant analysis, as events such as CVA, TIA and PVD confer considerable morbidity upon the patient. As shown in the sensitivity analyses, it is likely that models based on aggregate endpoints will provide different estimates of risk from models based upon myocardial infarction alone. Removal of all noncoronary events (CVA, TIA and PVD) from the analyses yielded similar risk estimates to the primary analysis. When only MI was modelled, there were too few events to allow any conclusions to be drawn from this study.
We sought to expand the analytical database beyond the earlier published data from the HOPS investigators . This earlier publication showed an association between PI exposure and MIs, but results from the adjusted Cox model had a wide 95% confidence interval and were not statistically significant. By expanding the dataset and including nonMI events, we increased the observation time and statistical power of this study. We were also able to conduct additional analyses, such as the dose–response analyses, not performed in the earlier publication.
Our findings differ from those recently reported from a large study using data from the US Veteran's Administration (VA) . This might be a result of methodological differences as well as differences in the populations studied. In that study , CVD admission trends were analysed over time, with only 42% of the patients ever taking PIs compared to 77% in our study. The data source for that study was also different; in our study, CVD events were derived from abstracted medical records, whereas in the VA study CVD events were based on hospital diagnosis codes from administrative databases.
Our findings are consistent with those of some recently published studies [12,14,15,17,27]. Four of these studies reported specifically on MI as the outcome of interest. A French hospital database study of HIV-infected men found that men exposed to PI therapy had an increased risk of MI (HRadj 2.56; 95% CI 1.03–6.34) . The duration of PI exposure was associated with significantly increased MI incidence in a dose-dependent relationship. Patients exposed for more than 30 months had the highest MI incidence rate. In the HOPS study discussed above, MI incidence increased annually following the introduction of PIs in 1996 . PI exposure was associated with an adjusted hazards ratio of 6.51 (95% CI 0.89–47.8) for MI. Published data from the Data Collection on Adverse Events of Anti-HIV Drugs (DAD) study group showed an increased risk of MI for each year on combination antiretroviral therapy (RRadj 1.26; 95% CI 1.12–1.41) . The Kaiser hospitalization study has recently reported an association between PI exposure and risk of hospitalization for MI (HRadj 4.1; P<0.0001) and CHD (HRadj 3.37; P<0.0001) . The most recent data from the DAD study were for outcomes of cardiovascular and cerebrovascular events, and showed an association with increasing combination antiretroviral therapy use (RRadj per year of exposure: 1.26, 95% CI 1.15–1.38) .
There are limitations to the current analysis. First, we observed a difference in the follow-up times between exposure groups. Longer follow-up in the PI group could lead to a biased estimate of effect. We tested this bias by truncating the follow-up in the PI group to obtain similar follow-up for the exposure groups and repeated the primary analysis. As already discussed, the HR for this analysis was similar to that from the primary analysis, indicating that the adjusted incidence rate and time-dependent analysis accounted for the differential follow-up. Secondly, to the extent that clinicians are influenced by the association of PI therapies with metabolic disturbances, an ascertainment bias could explain the findings in this analysis. Conversely, clinicians may also fail to initiate or continue PI therapy in patients at high risk of CVD events, thereby introducing a bias towards the null.
Detailed historical information on exposure to antiretroviral therapy prior to enrolment in the database is not available for all the patients. This has the potential to bias the result towards the null, as patients with unknown prior PI exposure could be classified as never exposed, resulting in a misclassification. Incomplete baseline risk factor information is also of concern. However, any missing information is likely to be randomly distributed among all patients, which should not bias the relative risk estimates. Use of lipid-lowering therapy was not accounted for in this analysis. Lipid-lowering therapy is, however, more likely to be used in PI-treated patients [28,29], which is expected to exert a bias towards the null. Patients on PI drugs who develop metabolic complications may not only be prescribed a lipid-lowering drug, but be switched to a PI-sparing regimen, thus diminishing the risk for CVD events. Our analysis does not account for such switches in therapy, and these patients would remain classified as PI-exposed in our analysis, which would also bias the results towards the null. Lastly, the period of our analyses (1 January 1996 to 30 June 2003) limits the interpretation of our results to the antiretroviral medications available during that period.
In conclusion, our results suggest an association between PI exposure and increased risk of CVD. This association was shown in the overall population as well as in a subset of patients aged between 35 and 65 years. Although the effect of PI treatment on CVD risk was evident in this analysis, the demonstrated benefits of HAART therapy still outweigh the risk of subsequent cardiovascular events . However, a combination of several factors, such as longer life expectancy, higher smoking rates among HIV-infected persons and prolonged exposure to PI-based HAART regimens, may lead to greater CVD event rates, especially as this population continues to age. Further studies are needed to confirm this association as well as to identify the causal pathway. Finally, we believe that the evaluation of CVD risk profiles of patients initiating or currently on HAART should be part of clinical practice. Nonpharmacological and pharmacological steps should be taken to address modifiable risk factors such as smoking, hyperlipidaemia, hypertension and diabetes in patients at risk of future CVD events.