Author + information
- Received January 13, 2017
- Revision received April 24, 2017
- Accepted April 25, 2017
- Published online July 31, 2017.
- John R. Teerlink, MDa,
- Min Qian, PhDb,
- Natalie A. Bello, MD, MPHb,
- Ronald S. Freudenberger, MDc,
- Bruce Levin, PhDb,
- Marco R. Di Tullio, MDb,
- Susan Graham, MDd,
- Douglas L. Mann, MDe,
- Ralph L. Sacco, MD, MSf,
- J.P. Mohr, MDb,
- Gregory Y.H. Lip, MDg,
- Arthur J. Labovitz, MDh,
- Seitetz C. Lee, MD, PhD, MPHb,
- Piotr Ponikowski, MD, PhDi,
- Dirk J. Lok, MDj,
- Stefan D. Anker, MD, PhDk,
- John L.P. Thompson, PhDb,
- Shunichi Homma, MDb,∗ (, )
- WARCEF Investigators
- aSection of Cardiology, San Francisco Veterans Affairs Medical Center and School of Medicine, University of California San Francisco, San Francisco, California
- bColumbia University Medical Center, New York, New York
- cLehigh Valley Hospital, Allentown, Pennsylvania
- dState University of New York at Buffalo, Buffalo, New York
- eWashington University, St. Louis, Missouri
- fUniversity of Miami, Miami, Florida
- gInstitute of Birmingham Centre for Cardiovascular Sciences, Birmingham, England, United Kingdom
- hUniversity of South Florida, Tampa, Florida
- iMilitary Hospital, Wroclaw, Poland
- jDeventer Hospital, Deventer, the Netherlands
- kInnovative Clinical Trials, Department of Cardiology & Pneumology, University Medical Center Göttingen (UMG), Göttingen, Germany
- ↵∗Address for correspondence:
Dr. Shunichi Homma, Margaret Milliken Hatch Professor of Medicine, Columbia University Medical Center, PH 3-342, 630 West 168th Street, New York, New York.
Objectives The aim of this study was to determine whether aspirin increases heart failure (HF) hospitalization or death in patients with HF with reduced ejection fraction receiving an angiotensin-converting enzyme (ACE) inhibitor or angiotensin receptor blocker (ARB).
Background Because of its cyclooxygenase inhibiting properties, aspirin has been postulated to increase HF events in patients treated with ACE inhibitors or ARBs. However, no large randomized trial has addressed the clinical relevance of this issue.
Methods We compared aspirin and warfarin for HF events (hospitalization, death, or both) in the 2,305 patients enrolled in the WARCEF (Warfarin versus Aspirin in Reduced Cardiac Ejection Fraction) trial (98.6% on ACE inhibitor or ARB treatment), using conventional Cox models for time to first event (489 events). In addition, to examine multiple HF hospitalizations, we used 2 extended Cox models, a conditional model and a total time marginal model, in time to recurrent event analyses (1,078 events).
Results After adjustment for baseline covariates, aspirin- and warfarin-treated patients did not differ in time to first HF event (adjusted hazard ratio: 0.87; 95% confidence interval: 0.72 to 1.04; p = 0.117) or first hospitalization alone (adjusted hazard ratio: 0.88; 95% confidence interval: 0.73 to 1.06; p = 0.168). The extended Cox models also found no significant differences in all HF events or in HF hospitalizations alone after adjustment for covariates.
Conclusions Among patients with HF with reduced ejection fraction in the WARCEF trial, there was no significant difference in risk of HF events between the aspirin and warfarin-treated patients. (Warfarin Versus Aspirin in Reduced Cardiac Ejection Fraction trial [WARCEF]; NCT00041938)
Recommendations for antithrombotic therapy in patients with heart failure (HF) in sinus rhythm have fluctuated widely over recent decades with varying roles for aspirin and warfarin. On the basis of recent clinical trials (1–3), 2 major official guidelines have concluded that, in patients with chronic HF without atrial fibrillation, a prior thromboembolic event, or a cardioembolic source, there is no evidence that an oral anticoagulant reduces mortality or morbidity compared with placebo or aspirin (4,5). The emergence of novel oral anticoagulants has rekindled interest in the prospect of improving outcomes in HF through an antithrombotic strategy (6–8).
The answer to another major question has remained elusive; in patients treated with an angiotensin-converting enzyme (ACE) inhibitor or angiotensin receptor blocker (ARB), does aspirin increase HF-related events in patients with heart failure with reduced ejection fraction (HFrEF) compared to those not receiving these medications? This question evolved from recognition that the cyclooxygenase inhibiting properties of aspirin, which reduce prostaglandins and nitric oxide, could be detrimental in HF by counteracting the beneficial effects of ACE inhibitors which increase prostaglandins through inhibition of their degradation (9). Multiple small studies provide some support for this perspective (9,10). In the WATCH (Warfarin and Antiplatelet Therapy in Chronic Heart Failure) trial, there was a significant increase in HF hospitalizations in patients treated with aspirin compared to warfarin (2). Given the high incidence of coronary artery disease and coronary artery stenting in the HF population (11,12), the potential adverse impact of aspirin on HF hospitalizations continues to be relevant to the daily care of these patients (13).
We previously reported the results of the WARCEF (Warfarin Aspirin Reduced Cardiac Ejection Fraction) trial in 2,305 patients enrolled and followed for up to 6 years (1). The result showed that warfarin had significant benefit compared with aspirin with respect to the prevention of ischemic stroke throughout the follow-up period among patients with HFrEF. However, the increase in the incidence of major bleeding offset the benefit of warfarin. Although warfarin reduced ischemic stroke, we found no difference between warfarin- and aspirin-treated groups with regard to all-cause death or stroke (time to first event) and the composite secondary endpoint of death, ischemic stroke, intracranial hemorrhage, myocardial infarction, or HF hospitalization (time to first event). We also reported, in an unadjusted secondary analysis, no increase in time to first hospitalization for aspirin compared to warfarin, but did not focus in depth on the effect of aspirin compared to warfarin specifically on HF-related events (1). Furthermore, no large randomized trial has addressed this clinically relevant issue. The purpose of these analyses was to investigate the effect of aspirin compared to warfarin specifically on HF-related events. Given that patients with HFrEF frequently experience multiple HF hospitalizations (14), herein we provide detailed analyses of both time to first and time to recurrent events.
Details of the WARCEF trial have been published previously (1). In this randomized, double-blind trial, 2,305 patients with left ventricular ejection fraction ≤35% in sinus rhythm were randomly assigned to warfarin (target international normalized ratio 2.75, with acceptable target range of 2.0 to 3.5) or aspirin (325 mg/day). Patients were enrolled at 168 centers in 11 countries between October 2002 and January 2010. The mean follow-up time was 3.5 ± 1.8 years. Patients who had a clear indication for warfarin or aspirin were not eligible. Additional eligibility criteria were a modified Rankin score of 4 or less (on a scale of 0 to 6, with higher scores indicating more severe disability), and planned treatment with a beta-blocker, an ACE inhibitor (or, if the side-effect profile with ACE inhibitors was unacceptable, with an ARB), or hydralazine and nitrates. Patients were ineligible if they had a condition that conferred a high risk of cardiac embolism, such as atrial fibrillation, a mechanical cardiac valve, endocarditis, or an intracardiac mobile or pedunculated thrombus.
Assessment of outcomes and major adverse events
In WARCEF, an independent end-point adjudication committee, whose members were unaware of the treatment assignments, adjudicated all primary and secondary outcomes and major hemorrhages. Heart failure hospitalizations were defined as hospital admissions for HF or hospitalization for which HF was a major contributing factor for admission and which met all of the following criteria: 1) signs and symptoms of HF on admission; 2) admission to the hospital for at least 24 h, excluding time in an emergency room or observation unit; and 3) the use of intravenous diuretic, vasodilator, or inotropic therapy for the purposes of treating HF. All deaths were first classified as cardiovascular or noncardiovascular and cardiovascular deaths were further classified into other types. HF deaths were defined by the presence of at least 1 of the following at the time of death: 1) cardiogenic shock; 2) pulmonary edema; or 3) refractory HF (patient requiring continuous positive inotropic therapy or mechanical circulatory assistance or experiencing HF symptoms at rest or requiring confinement to bed or a chair).
The current analyses are for the 489 WARCEF patients who experienced at least 1 HF-related event (HF hospitalization, HF death, or both).
In the aspirin and warfarin groups, baseline characteristics for patients who experienced at least 1 HF-related event (HF hospitalization, HF death, or both) were compared using 2-sample t tests for continuous variables and chi-square tests for categorical variables. Comparisons of the percentages of patients in the 2 treatment groups who experienced at least 1 HF-related event used the exact test of 2 independent proportions, and comparisons of the rates used the exact conditional binomial test for 2 independent Poisson variables. We used Cox models to assess the effect of treatment on time to the first HF-related event, and extended these Cox models to accommodate time to recurrent events analysis using 2 modeling techniques. One, the total time conditional or Prentice-Williams-Peterson counting process (PWP-CP) model is conditional: only patients who have already experienced a prior HF-related event are included in the risk set for a next HF-related event (15). Thus, the hazard ratio (HR) gives information on the effectiveness of treatment on the kth HF-related event among patients who have undergone a previous event. In the other technique, the total time marginal model (Wei, Lin, and Weissfeld [WLW]), marginal Cox models are estimated for each event time. In the Cox model for the kth HF-related event, all patients remain in the analytic risk sets whether or not they previously experienced such events, until they either have a kth event or are censored. Thus, the HR associated with the kth event from a marginal model is based on information from a larger group of patients than that from a conditional model because the marginal model does not exclude patients without prior events. Robust sandwich estimators for standard errors were used to produce the p values for the WLW models (16).
All of the above analyses were adjusted for variables found in univariable models to be predictive of at least one of the 5 events (first HF hospitalization, HF death, first HF hospitalization or death, recurrent HF hospitalization, recurrent HF hospitalization or HF death) (Online Table 1). The variables identified as predictive were age, continent from which patients were enrolled, body mass index, non-Hispanic white, systolic blood pressure, diabetes mellitus, hypertension, myocardial infarction, ischemic cardiomyopathy, peripheral vascular disease, prior stroke or transient ischemic attack, alcohol consumption, education, New York Heart Association (NYHA) functional class III or IV, ejection fraction, 6-min walk, baseline Mini-Mental State Examination score, baseline Minnesota Living With Heart Failure questionnaire (MLWHF) score, already on warfarin or other oral anticoagulant, diuretics, implantable cardioverter defibrillator, blood, urea, nitrogen level, estimated glomerular filtration rate (eGFR), hemoglobin, sodium, and white blood cell count. Multiple imputation was used to account for missing covariate data. We created 5 datasets using a sequential regression imputation method (17), performed Cox regression analyses on each data set, and subsequently combined the results to produce the reported HRs and p values using the method described by Rubin (18). The very few ties in the data were handled by replacing the continuous-time Cox partial likelihood function with Cox’s discrete-time partial likelihood function. The analysis was performed in SAS with the TIES=DISCRETE option in the PROC PHREG procedure. All statistical analyses were performed using SAS software version 9.4 (SAS Institute, Cary, North Carolina).
Descriptive statistics are presented in Table 1. With the exception of a borderline difference in age (p = 0.05), there were no significant differences between treatment groups in the baseline characteristics of patients experiencing at least 1 HF-related event.
Analysis of first events
In unadjusted analyses of the 489 first-HF hospitalization events, a significantly lower proportion of patients treated with aspirin had either at least 1 HF hospitalization or HF death (aspirin: n = 224, 19.3%; warfarin: n = 265, 23.2%; odds ratio [OR]: 0.79; p = 0.022); and aspirin also resulted in a significant risk reduction in time to first HF-related events (Table 2) (unadjusted HR: 0.82 [95% confidence interval (CI): 0.69 to 0.98]; p = 0.031). However, after adjustment for baseline covariates, there was no significant difference between the treatment groups in time to HF event (Table 2) (adjusted HR: 0.87 [95% CI: 0.72 to 1.04]; p = 0.117).
Similarly, in unadjusted analyses of first HF hospitalization alone (451 events), fewer patients treated with aspirin (n = 208, 17.9%) had at least 1 HF hospitalization, compared to 243 (21.3%) warfarin-treated patients (OR: 0.81; p = 0.041), and time-to-first HF hospitalization was marginally improved by aspirin compared to warfarin (Table 2) (unadjusted HR: 0.83 [95% CI: 0.69 to 1.00]; p = 0.052). After adjustment, however, the difference between the treatment groups was not significant (adjusted HR: 0.88 [95% CI: 0.73 to 1.06]; p = 0.168).
There were 70 (6.0%) HF deaths in the aspirin-treated patients compared to 76 (6.7%) in the warfarin group (OR: 0.90; p = 0.550), with no significant difference in time to HF death (Table 2, Figure 1) (unadjusted HR: 0.93 [95% CI: 0.67 to 1.28]; p = 0.639; adjusted HR: 0.99 [95% CI: 0.71 to 1.38]; p = 0.952). This suggests no differential treatment effect on HF mortality, although the number of events is relatively small.
Analysis of all events
In unadjusted analyses of all 1,078 HF events, patients assigned to aspirin had a lower rate of HF-related events (hospitalizations or HF deaths) than patients assigned to warfarin (Figure 2) (aspirin: 502 events, 12.4 per 100 patient-years; warfarin: 576 events, 14.2 per 100 patient-years; rate ratio: 0.87 [95% CI: 0.77 to 0.99]; p = 0.028). There was also a significant difference between the overall unadjusted HF hospitalization rate (Figure 3) in the 2 treatment groups (aspirin: 432 events, 10.7 per 100 patient-years; warfarin: 500 events, 12.4 per 100 patient-years; rate ratio: 0.87 [95% CI: 0.76 to 0.99]; p = 0.031).
Table 3 presents adjusted and unadjusted analyses of time to recurrent events from both the PWP-CP (conditional) and WLW (marginal) models, which give similar results. The conclusion for time to first HF event (hospitalization or death) stays the same: there is a significant benefit for aspirin before, but not after, adjustment for baseline covariates. There is no significant risk difference between aspirin- and warfarin-treated patients in subsequent events (second HF hospitalization or death and third HF hospitalization or death) in both unadjusted and adjusted analyses.
The results for HF hospitalization alone show a marginally significant difference (p = 0.051) in the unadjusted analysis for first event. Having noted this, there are no significant differences between warfarin and aspirin patients in either adjusted or unadjusted analyses of any of the recurrent HF hospitalization events, first or subsequent, under either the PWP-CP or the WLW model.
The WASH (Warfarin/Aspirin Study in Heart Failure) and WATCH trials raised concern about the safety of aspirin use in the advanced HF population (NYHA functional class III and IV) as both showed an increased risk of HF hospitalization rates for aspirin compared to warfarin (2,19). Although no mortality benefit was observed, in WASH there was a trend toward higher mortality in the aspirin group compared to the warfarin or placebo groups. Similarly, when analyzing the SOLVD (Studies of Left Ventricular Dysfunction) trial treatment and prevention arms, there was a strong interaction between the use of antiplatelet agents and all-cause mortality in patients with HF and a reduced ejection fraction. Although patients with left ventricular systolic dysfunction who received antiplatelet agents in addition to enalapril experienced a significant reduction in the combined end point of death or HF hospitalization compared to those randomized to placebo, this benefit was attenuated when compared to those patients who received enalapril in the absence of an antiplatelet agent (20).
There are mechanistic reasons and data to support the potentially detrimental effect of aspirin in patients with HF (21). The upregulation of prostaglandin synthesis and resulting vasodilatory effect may be an important mechanism to counteract various mediators of vasoconstriction in patients with HF (22). Aspirin has also been shown to reduce renal prostaglandin E2 and decrease renal sodium excretion as well as decreasing eGFR (23–25). By interfering with prostaglandin production, aspirin and other cyclooxygenase inhibitors may exert harm by blunting these vital compensatory responses.
On the other hand, in addition to causing platelet activation and aggregation, thromboxane A2 directly causes vasoconstriction and is thought to mediate, at least in part, the vasoconstrictive effect of angiotensin II (26). In this case, selective inhibition of thromboxane production by aspirin may be beneficial in patients with HF. As a result of these complex and potentially contradictory downstream effects of aspirin in patients with HF, the potential clinical consequence of its use in this population is an important concern.
In this post hoc analysis of the WARCEF data, which are much more extensive than WASH and WATCH combined (8,077 follow-up years for 2,305 patients in WARCEF compared to 3,383 follow-up years for 1,767 patients in the others combined), included both time to first event and all-event analyses. The results did not show an increased risk of HF hospitalization or HF death in patients receiving aspirin compared to those receiving warfarin in either analysis. Although the reason for the discrepancy between WARCEF and WATCH is not known, it bears emphasis that the results of WARCEF reflect a contemporary trial of patients with systolic HF, where >98%, 85%, 75% of patients were receiving an ACE inhibitor or ARB, a beta blocker, or a mineralocorticoid receptor antagonist, respectively. WARCEF provides important reassurance that the use of aspirin is not associated with an increase in clinically meaningful exacerbations of HF leading to hospitalization or an increase in death due to HF, when compared to patients who were receiving warfarin.
First, as in all clinical trials, patients enrolled in WARCEF are a selected population of individuals that may not be representative of community dwelling HF populations and can limit the generalizability of these findings. Second, our analysis does not have the benefit of a placebo control group. Although this would be ideal, it was not the design of the WARCEF trial. Given this, the comparison of aspirin to warfarin, which is randomized, is clinically meaningful, and suggests (but does not prove) that the use of aspirin is not associated with detrimental effects in terms of the clinical outcomes of patients with HFrEF. Third, it is useful to assess the precision of findings that are not statistically significant. Examining the upper confidence limits for first and second HF-related events in the marginal models in Table 3, we can rule out risk elevations above 5% (upper 95% CI for HR: 1.05) for first HF hospitalization or death (HFHD); 15% for second HFHD; 7% for first HF hospitalization (HFH); and 20% for second HFH. (The results of the conditional model and for third HF-related events are less reliable). Fourth, we did not correct for multiple comparisons because we were not trying to “prove” a positive post hoc finding, which would have required such adjustment. Given this, correction for multiple comparisons is unnecessary, and indeed, would only reinforce the null conclusions.
Among patients with HFrEF in the WARCEF trial, those who received aspirin experienced fewer HF events than those who received warfarin. After adjustment, however, there was no significant difference in risk between the aspirin and warfarin-treated patients.
COMPETENCY IN MEDICAL KNOWLEDGE: Despite mechanistic reasons and prior trial results raising a concern that aspirin may result in an increased risk of HF hospitalization and higher mortality in patients with HFrEF on an ACE inhibitor or ARB, the use of aspirin was not associated with an increased risk of HF hospitalization or death compared to warfarin in WARCEF.
TRANSLATIONAL OUTLOOK: Further randomized, placebo-controlled studies could address the potential hazard of aspirin use in addition to an ACE inhibitor or ARB in patients with HFrEF.
For a supplemental table, please see the online version of this paper.
Supported by grants from the National Institute of Neurological Disorders and Stroke to Dr. Homma (U01-NS043975) and Dr. Thompson (U01-NS-039143). Dr. Teerlink has received support from Actelion, Amgen, Astra-Zeneca, Bayer, Bristol-Myers Squibb, Celyad, Merck, Novartis, Relypsa, Stealth, St. Jude Medical/Abbott, Trevena, and ZS Pharma. Dr. Sacco has received research grants from National Institute of Neurological Diseases and Stroke, National Center for Advancing Translational Sciences, American Heart Association, the Evelyn McKnight Brain Foundation, and Boehringer-Ingelheim. Dr. Lip has received consultant fees from Bayer/Janssen, Bristol-Myers Squibb/Pfizer, Bioitronik, Medtronic, Boehringer-Ingelheim, Microlife, and Daiichi-Sankyo; and has also been on the speakers board for Bayer, Bristol-Myers Squibb/Pfizer, Medtronic, Boehringer-Ingelheim, Microlife, Roche, and Daiichi-Sankyo. Dr. Labovitz has received a research grant from Bristol-Myers Squibb/Pfizer for the AREST trial. Dr. Anker has received consultant fees from Boehringer-Ingelheim, Bayer, and Janssen; and was on the steering committee for COMMANDER-HF. Dr. Homma has received consultant fees from St. Jude Medical, Daiichi-Sankyo, Bristol-Myers Squibb, and Pfizer. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- angiotensin-converting enzyme
- angiotensin receptor blocker
- heart failure
- heart failure with reduced ejection fraction
- Received January 13, 2017.
- Revision received April 24, 2017.
- Accepted April 25, 2017.
- 2017 American College of Cardiology Foundation
- Massie B.M.,
- Collins J.F.,
- Ammon S.E.,
- et al.
- McMurray J.J.,
- Adamopoulos S.,
- Anker S.D.,
- et al.
- Yancy C.W.,
- Jessup M.,
- Bozkurt B.,
- et al.
- Shantsila E.,
- Lip G.Y.H.
- Shantsila E.,
- Lip G.Y.
- Mozaffarian D.,
- Benjamin E.J.,
- Go A.S.,
- et al.
- Masoudi F.A.,
- Wolfe P.,
- Havranek E.P.,
- Rathore S.S.,
- Foody J.M.,
- Krumholz H.M.
- Ross J.S.,
- Chen J.,
- Lin Z.,
- et al.
- Wei L.J.,
- Lin D.Y.,
- Weissfeld L.
- Raghunathan T.E.,
- Lepkowski J.M.,
- Van Hoewyk J.V.,
- Solenberger P.
- Rubin D.B.
- Al-Khadra A.S.,
- Salem D.N.,
- Rand W.M.,
- Udelson J.E.,
- Smith J.J.,
- Konstam M.A.
- Hall D.,
- Zeitler H.,
- Rudolph W.
- Dietz R.,
- Nagel F.,
- Osterziel K.J.
- Baur L.H.,
- Schipperheyn J.J.,
- van der Laarse A.,
- et al.