Aspirin Use and All-Cause Mortality Among Patients Being Evaluated for Known or Suspected Coronary Artery Disease:  A Propensity Analysis FREE

Aspirin has been shown to be associated with decreased cardiovascular morbidity in multiple clinical trials^1- 2 but the association between aspirin use and all-cause mortality has been less well defined except in the setting of acute myocardial infarction.³ Although a few observational analyses have suggested a longer-term survival benefit,^4- 6 it is not clear whether this benefit persists after accounting for treatment selection biases as well as established predictors of survival in patients with known or suspected coronary artery disease, in particular impaired exercise capacity, left ventricular dysfunction, and myocardial ischemia.

In this study we sought, based on an a priori hypothesis, to determine if aspirin use was associated with a reduction in all-cause mortality among stable patients referred for stress echocardiography. Because the validity of observational studies of treatment effects may be limited by selection biases and confounding factors, we performed a propensity analysis.⁷

The study sample was derived from 9954 consecutive adult patients undergoing stress echocardiography at the Cleveland Clinic Foundation between 1990 and 1998. Patients were excluded if they had significant valvular heart disease, prior cardiac transplantation, congenital heart disease, were younger than 30 years, if they were referred for arrhythmia evaluation, for consideration of cardiac transplantation, or solely as part of a research protocol. We also excluded patients with documented contraindications to aspirin, including peptic ulcer disease, renal insufficiency, and concurrent use of nonsteroidal anti-inflammatory drugs. A total of 3780 patients were excluded. If patients had more than 1 stress echocardiogram, only the first was considered.

All patients gave informed consent before undergoing exercise testing. Approval was obtained from the Cleveland Clinic Foundation institutional review board to perform research analyses based on prospectively obtained stress laboratory databases in our institution.

Data on baseline demographics, medical history, cardiovascular risk factors, and medication use (including regular use of aspirin) were collected prospectively at the time of testing.⁸ All data were entered online prior to the start of the stress test by either a physician or a trained exercise physiologist, with those entering data into the database blinded to the hypothesis of this study as well as to the results of the subsequent stress test and stress echocardiogram. Quality control has been ensured as described elsewhere.⁹

Aspirin use was confirmed by a pretest patient interview or by a physician's note in the patient's chart. Regular aspirin use was defined as use of aspirin daily or every other day. The timing of the last dose of all medications was prospectively recorded. Among aspirin users, 93% had taken their last dose within 24 hours while 98% had done so within 48 hours.

Resting heart rate was based on a 30-second recording of pulse, while blood pressure was measured to the nearest 1 mm Hg using indirect mercury column sphygmomanometry. Height and weight were directly measured and body mass index was calculated as weight in kilograms divided by the square of height in meters. Diabetes was considered present if insulin or oral hypoglycemic medications were being used or if the patient had been prescribed a diabetic diet. Hypertension was defined as a resting systolic blood pressure of 140 mm Hg or greater, a resting diastolic pressure of 90 mm Hg or greater, or use of medications to reduce blood pressure.¹⁰ Prior coronary artery disease was defined as prior myocardial infarction, coronary artery revascularization, or the presence of at least 1 coronary stenosis (50% or greater diameter) on a prior coronary angiogram. Congestive heart failure was coded if a diagnosis was noted in the patient's record. For a global description of risk, the Mayo Risk Index was used; this index has a score from 0 to 5 with 1 point each for male sex, prior myocardial infarction, diabetes, insulin use, and typical angina pectoris. This score has been shown to correlate well with left main- or 3-vessel coronary artery disease.¹¹

Symptom-limited exercise testing was performed according to Bruce, modified Bruce, or Cornell protocols as previously described.¹² Exercise capacity was estimated in metabolic equivalent tasks (METs) (1 MET = oxygen consumption of 3.5 mL/kg per minute) and classified as being impaired if measured as fair or poor for age and sex, according to a validated classification scheme.¹³ An abnormal ST-segment response was considered present if, in the absence of an abnormal resting electrocardiogram or digitalis use, there was at least 1 mm of horizontal or downsloping ST-segment depression 80 milliseconds after the J-point. Chronotropic response was assessed by proportion of heart rate reserve used.¹⁴ Heart rate recovery was defined as the difference between heart rate at peak exercise and 1 minute thereafter.¹⁵

Details of the echocardiographic techniques used in our laboratory have been described in detail elsewhere.⁸ Briefly, images were obtained with the patient in the left lateral decubitus position. Parasternal long and short as well as apical 2- and 4-chamber images were obtained at baseline and immediately after exercise. Images were recorded on videotape diskette after online digitization. Images were reviewed and interpreted by 2 physician echocardiographers on the same day of examination regardless of image quality and in a blinded fashion with respect to clinical data, exercise data, and the hypothesis of this study. Ischemia and scarring were graded by a standard 16-segment model of the left ventricle. Myocardial ischemia was considered present if a new or progressive wall-motion abnormality was present on the postexercise images. Myocardial scarring was diagnosed by resting wall-motion akinesis or dyskinesis that was unchanged with stress.

The primary end point was all-cause mortality. As we have discussed elsewhere, the use of "cardiac" or "cardiovascular" mortality as an end point has a number of inherent limitations, including incorrect or biased documentation by treating physicians and inaccurate clinical assessments in an environment characterized by low autopsy rates.¹⁶ We used the Social Security Death Index, which has been shown to be highly specific (>99.5%) and unbiased.^17- 18 We have reported elsewhere on the high sensitivity (approximately 97%) of this index among Cleveland Clinic stress laboratory patients.⁹ Follow-up was for a median of 3.1 years.

Differences between aspirin users and nonusers were compared using χ² statistics for categorical variables and t or Wilcoxon rank sum tests, as appropriate, for continuous variables. Aspirin use was related to all-cause mortality using univariable and multivariable Cox proportional hazard regression analyses¹⁹ with consideration of clinically plausible interactions. The proportional hazards assumption was confirmed by inspection of log (−log [survival]) curves and by examination of time-dependent covariates. Survival curves were constructed using Kaplan-Meier estimates²⁰ with comparisons between curves based on the log-rank χ² statistic.

Because aspirin use was not randomly assigned in this patient population, potential confounding and selection biases were accounted for by developing a propensity score for aspirin use. The rationale and methods underlying the use of a propensity score for a proposed causal exposure variable have been previously described.⁷ The propensity for aspirin use was determined without regard to outcome, using multivariable logistic regression analysis.²¹ A full nonparsimonious model was developed that included 34 covariates, some of which are listed in Table 1. This model yielded a c statistic of 0.83, indicating a strong ability to differentiate between aspirin users and nonusers. A propensity score for aspirin use was then calculated from the logistic equation for each patient. This score ranged from 0.03 to 0.98 and, in effect, represented the probability that a patient would be using aspirin.

Using a macro (available at: http://www2.sas.com/proceedings/sugi26/p214-26.pdf), we used the propensity scores to match aspirin users to unique control patients. Specifically, we sought to match each aspirin user to a non–aspirin-using patient who had a propensity score that was identical to 5 digits. If this could not be done, we then proceeded to a 4-, 3-, 2-, or 1-digit match. Once this threshold was exceeded, that aspirin-using patient was excluded. We were able to match 1351 aspirin-using patients to 1351 unique non–aspirin-using patients.

To determine which patient characteristics predicted maximum absolute benefit from aspirin, we derived multivariable nonproportional hazard equations for each individual patient's predicted survival using a wholly parametric method.²² For the propensity-matched patients, each patient-specific equation was solved twice, once as if the patient had been taking aspirin and once as if he/she had not been taking aspirin; this approach is similar to another analysis we have described analyzing the potential benefits of bilateral mammary artery grafting.²³ The logarithm of the difference in predicted survivals with and without aspirin was then treated as the dependent variable for a linear regression analysis that sought to identify those patient characteristics most strongly associated with a large beneficial difference in predicted mortality. Appropriate regression diagnostics, including examination of residuals and testing for outliers, excessively influential observations, and multicollinearity, were performed to confirm the validity of these analyses.

All analyses were performed using SAS version 8.1 (SAS Institute, Cary, NC). Parametric survival analyses were performed using PROC HAZARD and PROC HAZPRED (available at: http://www.clevelandclinic.org/heartcenter/hazard).

Among 6174 adult patients eligible for analysis, 2310 (37%) were taking aspirin at the time of stress echocardiography. Baseline and exercise characteristics according to aspirin use are summarized in Table 1. Aspirin users were older and more likely to be men; they were also more likely to have hypertension, diabetes, and prior histories of coronary artery disease, coronary artery bypass grafting, and percutaneous coronary intervention. Patients taking aspirin were also more likely to be taking β-blockers, lipid-lowering drugs, and angiotensin-converting enzyme (ACE) inhibitors. They were more likely to have ischemic ST-segment changes during stress and echocardiographic evidence of stress-induced ischemia. The patients not using aspirin had higher left ventricular ejection fraction and were more likely to be smokers.

During 3.1 years of follow-up, 276 patients (4.5%) died. Table 2 summarizes mortality outcomes based on aspirin use. In a crude analysis, there was no association between aspirin use and mortality (4.5% vs 4.5%). After adjusting only for age and sex, an association between aspirin use and reduced mortality became evident, with an even stronger association noted after further adjusting for a history of known coronary artery disease. This association was not materially affected by adjusting for other confounders (Table 2).

Prespecified stratified bivariable analyses were performed according to age, sex, diabetes, smoking, prior myocardial revascularization procedures, use of β-blockers or ACE inhibitors, left ventricular systolic function, and echocardiographic evidence of ischemia. The only possible noted interactions with aspirin in bivariable analyses were a left ventricular ejection fraction of 40% or less and a prior history of coronary artery bypass grafting. However, these interaction terms were not significant after multivariable adjustment. Stratified multivariable analyses showed reduced mortality associated with aspirin use irrespective of left ventricular function or history of prior coronary artery bypass grafting (Table 2).

Based on systematically collected data for 34 variables including baseline demographics, medical risk factors, and the interactions between them, a logistic regression model was used to generate a propensity score for aspirin use. Major independent correlates of aspirin use included prior percutaneous or surgical myocardial revascularization, male sex, lipid-lowering therapy, nitrate use, and history of coronary artery disease.

Baseline characteristics comparing the propensity-matched aspirin users and aspirin nonusers are shown in Table 3. As opposed to the entire population, these propensity-matched patients were well matched; the only significant difference was that men who used aspirin had a slightly higher functional capacity than men who did not. During follow-up, 153 (6%) patients died. Aspirin use was associated with a lower risk of death (4% vs 8%, P = .002) (Figure 1 and Table 4). Aspirin use was significantly associated with reduced mortality by univariable analysis and multivariable analysis. We found no interactions between aspirin use and older age, impaired left ventricular systolic function, diabetes, smoking, history of coronary artery disease, prior coronary intervention, and echocardiographic evidence of myocardial ischemia.

Based on wholly parametric-derived patient-specific survival equations, a predicted absolute mortality difference from aspirin use was derived for each propensity-matched patient. The 3 strongest correlates of a large absolute mortality benefit were age, impaired exercise capacity, and a history of known coronary artery disease. A linear regression equation relating these 3 variables to the logarithm of the absolute survival difference associated with aspirin demonstrated that 74% of the variability in survival difference could be explained (Figure 2). Older patients who had either impaired exercise capacity or known coronary artery disease appeared to derive the greatest absolute benefit from aspirin use.

In the main cohort of 6174 patients, there were 2228 (36%) women, among them 531 (24%) regular users of aspirin. During 3.1 years of follow-up 77 women died, with no difference noted between aspirin users and nonusers (3.8% vs 3.4%). After adjusting for age, Mayo Risk Index, ejection fraction, history of prior coronary artery bypass surgery, and functional capacity, aspirin use was associated with a lower mortality rate (adjusted hazard ratio, 0.59; 95% confidence interval [CI], 0.35-1.00; P = .05).

In the propensity-matched cohort of 2702 patients, there were 777 women; 400 (51%) regularly used aspirin. There were 36 deaths, with aspirin use associated with a lower risk (3.5% vs 5.8%). After adjustment for age, propensity score, ejection fraction, and functional capacity, aspirin use remained predictive of a lower risk of death (adjusted hazard ratio, 0.50; 95% CI, 0.25-1.00; P = .05).

Among consecutive patients referred for stress echocardiography to evaluate known or suspected coronary artery disease, aspirin use was associated with a substantial reduction of all-cause mortality. When we assessed mortality risk using a standard Cox regression analysis among all patients, a 33% reduction in mortality was found. Subsequently, we performed a rigorous propensity analysis, limiting analyses to 2702 propensity-matched patients. The results were essentially unchanged, with aspirin associated with a substantial reduction in risk of death.

We estimated the absolute benefit of aspirin based on specific patient characteristics, thus predicting which patients might benefit most from aspirin treatment. We showed aspirin to be particularly beneficial among patients who were older, who had impaired exercise capacity, or who had a history of coronary artery disease. Sedentary patients subjected to strenuous exercise have been shown to have increased platelet activation and hyperreactivity compared with physically fit subjects.²⁴ Thus, aspirin may "treat" poor physical fitness by attenuating the associated increased platelet activation. To the best of our knowledge, this is the first study suggesting aspirin to be beneficial in patients with impaired exercise capacity—one of the most powerful predictors of mortality in patients with known or suspected heart disease. We were only able to demonstrate this association because we specifically analyzed a population of patients undergoing exercise testing.

Extensive literature documents the cardiovascular benefits of aspirin therapy among adults without a cardiovascular history,^1,25- 27 patients with chronic stable angina,^28- 29 patients presenting with AMI,^3,6 and patients with unstable angina.^30- 33 Randomized trial evidence demonstrates that aspirin reduces all-cause mortality among patients with AMI.³ It is less clear if aspirin use reduces long-term all-cause mortality in stable patient populations. Two recent observational analyses of patients enrolled in the Bezafibrate Infarction Prevention Trial demonstrated reduced mortality rates among patients taking aspirin, irrespective of the presence or absence of diabetes or therapy with ACE inhibitors.^4- 5 Furthermore, the Collaborative Group of the Primary Prevention Project recently demonstrated in a randomized trial a similar, although not statistically significant, reduction in relative risk for all-cause mortality (0.81; 95% CI, 0.58-1.13).³⁴ These findings are similar to ours but did not reach statistical significance, most likely due to a small number of events.

The current study extends these previous findings in several important respects. First, we demonstrated that aspirin use is associated with a reduction in long-term all-cause mortality, which is a clinically relevant, objective, and wholly unbiased end point.¹⁶ Second, because we focused on patients referred for stress echocardiography we were able to account for several critical predictors of mortality, including left ventricular systolic function, stress-induced myocardial ischemia, and impaired exercise capacity. Third, unlike prior observational studies of aspirin use and outcome,^4- 5,26 we used propensity analysis, which has been argued to be a powerful means of accounting for baseline confounding and selection biases.⁷

Furthermore, we observed this mortality reduction in a large cohort of consecutive patients seen within a clinical practice, as opposed to a clinical trial. It has been argued that patients enrolled in clinical trials may not be representative of patients seen in practice.³⁵ The patients included in our study population may represent a more representative sample of "real world" patients referred for evaluation of known or suspected cardiovascular disease than those included in many of the randomized controlled trials that have previously evaluated aspirin use for mortality reduction. Among the patients included in the Physicians' Health Study, 84% had no history of cardiovascular disease.¹ Additionally, those patients and those evaluated in other primary prevention trials had low rates of cardiovascular risk factors.^27,34 The studies evaluating aspirin use by patients with unstable angina also enrolled comparatively few patients with multiple cardiac risk factors or positive histories of previous coronary intervention.^31,33,36 Thus, the lower-risk population enrolled in the previous randomized trials may have contributed to their finding no mortality benefit. Furthermore, in a follow-up report of the Physicians' Health Study evaluating posttrial self-selected aspirin use and subsequent mortality, self-selected aspirin use was associated with multiple cardiovascular risk factors and a decrease in all-cause mortality.³⁷

The mechanisms by which aspirin may reduce mortality include its platelet-blocking effects, its anti-inflammatory properties, or other as-yet unknown actions. Aspirin has been shown to be a powerful antiplatelet agent that acts by blocking the production of thromboxane A₂,³⁸ which may then reduce the risk of fatal cardiovascular events.^39- 40 Recently, increasing interest has focused on inflammation, as assessed by C-reactive protein levels and cardiovascular risk.^41- 42 Aspirin has been shown to reduce C-reactive protein levels.⁴¹ In the randomized Physicians' Health Study the reduction in cardiovascular risk associated with aspirin was most pronounced among men with elevated baseline C-reactive protein levels.⁴³

The major limitation of this study is that aspirin use was not based on a randomized assignment. Although the use of observational studies for assessment of treatment effects is controversial,⁴⁴ recent work has suggested that observational studies, when properly done, are not likely to produce misleading or biased results.^45- 47 Furthermore, we used propensity analysis to enable an even more rigorous adjustment for selection bias and confounding than would be possible with standard multivariable analysis.⁷ Nonetheless, it must be acknowledged that observational studies can only partially control for factors actually measured and can adjust for these factors only as well as the instrument used to measure them is capable. In contrast, randomization allocates both known and unknown confounding variables and avoids the introduction of bias from either the participants or their physicians. Other limitations of our study included lack of information about aspirin dose, aspirin allergy, or duration of treatment, as well as lack of data regarding medication adjustments made after stress testing.

Despite these limitations, the association between aspirin use and reduced mortality meets currently accepted criteria for likely causality.⁴⁸ The association was strong, with a greater than 30% reduction in risk of death. A temporal pattern is evident in Kaplan-Meier analyses. Biological plausibility is present, considering the known importance of increased platelet activity associated with coronary artery disease, aging,⁴⁹ and impaired physical fitness.²⁴ Our results are consistent with other observational non–propensity-adjusted analyses⁵ and with a recent randomized study,³⁴ and the association appears to be largely unaffected by possible bias and confounding, whether assessed by standard multivariable analyses or more rigorous propensity analyses. Thus, our findings provide additional support for recommending the routine use of aspirin in patients with, or at risk for, cardiovascular disease—not only for preventing morbid events but also for reducing all-cause mortality.