Author + information
- Received March 12, 2014
- Revision received January 4, 2015
- Accepted January 22, 2015
- Published online April 14, 2015.
- Dhruv S. Kazi, MD, MSc∗,†,‡∗ (, )http://twitter.com/kardiologykazi,
- Thomas K. Leong, MPH§,
- Tara I. Chang, MD, MS‖,
- Matthew D. Solomon, MD, PhD§,‖,
- Mark A. Hlatky, MD‖,¶ and
- Alan S. Go, MD‡,§,¶
- ∗Division of Cardiology, San Francisco General Hospital, San Francisco, California
- †Department of Medicine (Cardiology), University of California San Francisco, San Francisco, California
- ‡Department of Epidemiology & Biostatistics, University of California San Francisco, San Francisco, California
- §Division of Research, Kaiser Permanente Northern California, Oakland, California
- ‖Department of Medicine, Stanford University, Stanford, California
- ¶Department of Health Research and Policy, Stanford University, Stanford, California
- ↵∗Reprint requests and correspondence:
Dr. Dhruv S. Kazi, Department of Medicine, and Department of Epidemiology and Biostatistics, University of California San Francisco, Division of Cardiology, San Francisco General Hospital, 1001 Potrero Avenue, Room 5G1, San Francisco, California 94110.
Background Platelet inhibition after percutaneous coronary intervention (PCI) reduces the risk of myocardial infarction (MI) but increases the risk of bleeding. MIs and bleeds during the index hospitalization for PCI are known to negatively affect long-term outcomes. The impact of spontaneous bleeding occurring after discharge on long-term mortality is unknown.
Objectives This study sought to examine, in a real-world cohort, the association between spontaneous major bleeding or MI after PCI and long-term mortality.
Methods We conducted a retrospective cohort study of patients ≥30 years of age who underwent a PCI between 1996 and 2008 in an integrated healthcare delivery system. We used extended Cox regression to examine the associations of spontaneous bleeding and MI with all-cause mortality, after adjustment for time-updated demographics, comorbidities, periprocedural events, and longitudinal medication exposure.
Results Among 32,906 patients who had a PCI and survived the index hospitalization, 530 had bleeds and 991 had MIs between 7 and 365 days post-discharge. There were 4,048 deaths over a mean follow-up of 4.42 years. The crude annual death rate after a spontaneous bleed (9.5%) or MI (7.6%) was higher than among patients who experienced neither event (2.6%). Bleeding was associated with an increased rate of death (adjusted hazard ratio [HR]: 1.61, 95% confidence interval [CI]: 1.30 to 2.00), similar to that after an MI (HR: 1.91; 95% CI: 1.62 to 2.25). The association of bleeding with death remained significant after additional adjustment for the longitudinal use of antiplatelet agents.
Conclusions Spontaneous bleeding after a PCI was independently associated with higher long-term mortality, and conveyed a risk comparable to that of an MI during follow-up. This tradeoff between efficacy and safety bolsters the argument for personalizing antiplatelet therapy after PCI on the basis of the patient’s long-term risk of both thrombotic and bleeding events.
Contemporary percutaneous coronary intervention (PCI) involves arterial cannulation, periprocedural anticoagulation, and dual antiplatelet therapy. This combination of factors leads to procedure-related major bleeding in 2% to 5% of patients (1–4). An additional 1% to 5% of patients have spontaneous major bleeding in the first year after PCI, depending on the clinical definition used (5,6). Procedural bleeding is associated with increased mortality during the index hospitalization and in the first year after discharge (1,3,7–13). The effect, if any, of spontaneous bleeds that occur after hospital discharge (i.e., nonprocedural bleeds) on long-term outcomes is unclear.
The potential impact of spontaneous bleeding on long-term outcomes is important to assess for 2 reasons. First, although strategies to reduce periprocedural bleeding are receiving increasing attention (e.g., use of radial access for PCI), no such systematic efforts are in place to reduce bleeding that occurs after hospital discharge. Second, if spontaneous bleeding were indeed associated with higher late mortality, it would motivate careful consideration of the net clinical benefit of new antiplatelet and antithrombotic agents that reduce myocardial infarctions (MIs), but increase bleeding. This tradeoff between efficacy and safety may bolster the argument for personalizing the choice and duration of antiplatelet therapy after PCI on the basis of the patient’s long-term risk of both thrombotic and bleeding events.
To address this issue, we examined the association between major spontaneous bleeding or MI and long-term risk of death in a large, community-based cohort of patients receiving PCI within usual care clinical settings. We hypothesized that spontaneous bleeds in the first year after PCI would be associated with increased all-cause mortality that is comparable to the risk associated with a spontaneous MI.
The source population was from Kaiser Permanente Northern California, a large integrated healthcare delivery system with >3.3 million members in the San Francisco and greater Bay Area. The health plan membership is representative of the regional and statewide population, apart from slightly lower representation at the extremes of income and age.
We used health plan electronic health records to identify all members 30 years of age and older who received PCI between January 1, 1996, and December 31, 2008. We excluded patients who had incomplete demographic data, or <12 months of continuous membership and pharmacy benefit before the index procedure. We also excluded patients who died either during the index hospitalization or within 7 days after discharge because the outcome of interest was long-term mortality. All patients were followed through December 31, 2008 (the latest date for which complete mortality data were available at the time of analysis), or until disenrollment from the health plan, defined as a continuous gap in membership of 90 days or longer.
The primary outcome was death from any cause through December 31, 2008, which was identified from a combination of health plan hospitalization and administrative records, California state death certificate files, and Social Security Administration vital status files (14,15).
Exposures of interest
We defined spontaneous bleeding or MI as events occurring between days 7 and 365 after discharge from the index hospitalization. We treated events occurring during the initial hospital stay for PCI or within the first 7 days after discharge as procedural events because of the limited temporal resolution of available electronic data. On the basis of previously validated methods (16–19), we used International Classification of Diseases-Ninth Edition, codes to identify subsequent hospitalizations for MI (defined as a primary discharge diagnosis of an acute MI) or bleeding (a primary or secondary discharge diagnosis of intracranial bleeding, or a primary discharge diagnosis of extracranial bleeding). This was on the basis of our previous work demonstrating a low positive predictive value of codes for extracranial bleeds in the secondary position (17). We did not evaluate the use of blood transfusions because procedure codes are not sufficiently accurate to identify them in our data (20,21).
We obtained demographic and clinical data from health plan electronic health records and other associated databases. We identified comorbid conditions up to 4 years before the date of the index procedure and throughout follow-up using previously validated approaches from health plan hospitalization, ambulatory, laboratory, and pharmacy databases (22–24). We obtained a time-updated history of heart failure, diabetes mellitus, hypertension, dyslipidemia, mitral or aortic valvular heart disease, peripheral arterial disease, stroke or transient ischemic attack, dementia, systemic cancer, hypothyroidism, and chronic lung disease. We also identified previous hospitalizations for bleeding. Because kidney dysfunction and anemia are predictors of both major bleeding and mortality, we extracted baseline and longitudinal estimated glomerular filtration rates and hemoglobin concentrations from the health plan laboratory database (25–27). From health plan pharmacy records, we obtained data on the longitudinal use of the following prescription medications: angiotensin-converting enzyme inhibitors; angiotensin II receptor blockers; beta-blockers; diuretic agents, digoxin; statins, nonstatin lipid-lowering agents; nitrates; calcium-channel blockers; diabetes medications; anti-inflammatory drugs; warfarin; and antiplatelet agents (clopidogrel and ticlopidine). We then applied validated algorithms to characterize longitudinal exposure to these agents on the basis of information from dispensed prescriptions and the timing of refills (23,28). We could not reliably ascertain aspirin therapy because the majority of aspirin use in our population is over-the-counter and not captured in health plan pharmacy-dispensing databases. However, aspirin use among patients who have undergone a PCI in our population is nearly universal (29).
We compared baseline characteristics using the Student t test or Wilcoxon rank sum test for continuous variables and the chi-square test for categorical variables. We used an extended Cox regression model to evaluate the independent association of a spontaneous bleed or MI with death, after adjustment for baseline and time-updated confounders. The use of a time-updated model allowed each patient to contribute unexposed person-time (before a spontaneous bleed or MI) as well as exposed person-time (e.g., after the spontaneous bleed or MI). Similarly, the time-updated model allowed us to adjust for longitudinal changes in potential confounders (e.g., development of diabetes or chronic kidney disease during follow-up). Because few patients experienced both an MI and a bleed during follow-up, a separate hazard ratio for death was not estimated for this subgroup. We repeated the analysis, stratified by whether the index PCI was urgent/emergent (i.e., performed in the setting of acute coronary syndrome) or elective. Finally, because patients who bleed may discontinue their antiplatelet therapy afterwards, we conducted a mediation analysis that adjusted for antiplatelet agent use and timing during follow-up to explore whether doing so attenuated any observed association between occurrences of bleeding or MI with death. All analyses were performed using SAS statistical software, version 9.1.3 (Cary, North Carolina). A.G. and T.F. had direct access to the data, and all authors take full responsibility for the integrity of the study results.
The Kaiser Foundation Research Institute and Stanford University Institutional Review Boards approved the study. Due to the nature of the study, waiver of informed consent was obtained.
The study population consisted of 33,393 patients who underwent PCI between 2002 and 2008 (Figure 1). Overall, 487 died during the index hospitalization or within 7 days after discharge and were excluded from this analysis. Among the 32,906 patients discharged alive after PCI, 14% had a PCI for the treatment of an acute coronary syndrome, while the remainder had an elective PCI procedure. During a mean 4.42 years of follow-up, 530 patients were hospitalized for a spontaneous bleed (crude rate: 2.14 per 100 person-years; 95% confidence interval [CI]: 1.98 to 2.31), which occurred at a median of 100 (interquartile range [IQR]: 40 to 209) days post-discharge. Patients who experienced a bleed were older, more likely to be women, and had a greater comorbidity burden than patients who did not have a bleed (Table 1). They were also more likely to receive warfarin, but antiplatelet therapy use was similar in the 2 groups. The majority (77%) of spontaneous bleeds during follow-up were gastrointestinal (Table 2) and 15% were intracranial, of which one-third were secondary to trauma.
During follow-up, 991 patients were hospitalized for an MI (crude rate: 4.26 per 100 person-years; 95% CI: 4.03 to 4.50), which occurred a median of 127 (IQR: 64 to 222) days post-discharge. Patients who had an MI during follow-up were older, more likely to be women, and had a greater comorbidity burden compared with patients who did not have an MI (Table 1). Patients who had an MI during follow-up had lower estimated glomerular filtration rates and hemoglobin levels at baseline, and received more medications during follow-up.
There were 4,048 deaths during follow-up. Among the 530 patients who had a spontaneous bleed, 164 (31%) died during follow-up, at a median of 424 (IQR: 66 to 1,276) days after the bleed. Among the 991 patients who had a spontaneous MI, 315 (32%) died during follow-up, at a median of 572 (IQR: 75 to 1,504) days after the MI. Twenty-one deaths occurred among 39 patients who had both an MI and a bleed. The crude annual death rate among patients who had a spontaneous bleed (9.4 per 100 person years) and patients who had a spontaneous MI (7.6 per 100 person years) was higher than among patients with neither event (2.6 per 100 person years; Table 3).
After adjustment for demographic characteristics, comorbid conditions, procedural events, longitudinal laboratory results, and medication use, spontaneous bleeding was associated with increased mortality (adjusted hazard ratio [HR]: 1.61; 95% CI: 1.30 to 2.00) compared with the reference group of patients who experienced neither event during follow-up (Figure 2). The higher adjusted relative rate of death associated with spontaneous bleeding was similar to that observed for a spontaneous MI (HR 1.91; 95% CI: 1.62 to 2.25; Figure 2). The results were similar when stratified by indication for the index PCI (elective vs. urgent). Time-updated adjustment for the use of antiplatelet agents did not change the associations of spontaneous bleeding (HR: 1.61; 95% CI: 1.30 to 2.01) or spontaneous MI (HR: 1.92; 95% CI: 1.63 to 2.27) with all-cause death.
In this large, diverse, community-based sample of adults, spontaneous bleeding episodes between days 7 and 365 after discharge for a PCI were associated with a 61% higher rate of all-cause mortality during follow-up. The higher death rate associated with bleeding was similar to the 91% increase in the relative death rate associated with experiencing an MI during follow-up. The higher mortality after a spontaneous bleed was not explained by the subsequent lower use of antiplatelet agents.
Several investigators found that periprocedural major bleeds (either during the index hospitalization or within the first 30 days after a PCI) are associated with increased mortality in both “real-world” populations (1,3,4,7) and randomized trial cohorts (8–13). Periprocedural bleeding in these studies was associated with significantly increased in-hospital, 30-day, and 1-year mortality, similar to that associated with a periprocedural MI. For instance, 57,246 periprocedural bleeds among 3,386,688 procedures in the American College of Cardiology National Cardiovascular Data Registry CathPCI Registry in the United States between 2004 and 2011 were associated with higher in-hospital death across all strata of pre-procedural bleeding risk (4). The authors estimated that 12% of the adjusted population-attributable risk for in-hospital death was related to major bleeding. Similarly, an analysis of 5,384 patients from 4 randomized controlled trials of abciximab noted that major hemorrhage in the first 30 days post-procedure was associated with a similar death risk at 1 year as a periprocedural MI (9).
Our study extends these observations to nonprocedural bleeding after hospital discharge from the index PCI. We found that spontaneous, late bleeding was significantly associated with higher mortality, comparable to the adverse effect of a late, spontaneous MI. Of note, our cohort enrolled a broad spectrum of patients undergoing PCI within typical clinical care, including low-risk patients undergoing elective procedures, and high-risk patients undergoing urgent or emergent revascularization for acute coronary syndrome. Stratification by whether the procedure was elective or urgent did not materially change the results. Finally, this study included a large, community-based patient cohort; thus, our findings likely reflect “real-world” outcomes.
There are 2 plausible explanations for the association between bleeding and mortality: 1) a major bleed directly increases long-term mortality; or 2) a major bleed is a marker of overall poor prognosis (i.e., the sickest patients are at a higher risk for both bleeding and death). These alternative hypotheses cannot be tested in a randomized trial, so we must rely on experimental and observational data to provide key insights. Substantial theoretical, basic, and clinical evidence suggests a causal relationship between bleeding after PCI and subsequent mortality (Central Illustration). Bleeding produces a prothrombotic milieu by activating the coagulation cascade and up-regulating cytokines such as plasminogen activator inhibitor-1. Large bleeds produce hypovolemia and anemia, which reduce oxygen delivery and cause reflex tachycardia, which increase oxygen demand and can lead to myocardial ischemia in patients with obstructive coronary disease (30). Blood transfusions produce secondary biochemical changes that adversely impact microcirculatory flow, hypoxia-induced vasodilation, oxygen delivery (31,32), and survival (33,34). Some of these adverse effects are accentuated in the setting of endothelial dysfunction, making ischemic myocardium particularly vulnerable. Additionally, the development of anemia may also adversely affect long-term clinical outcomes due to increased neurohormonal activation and ventricular remodeling, which result in greater wall stress and increased oxygen consumption (35–37). However, randomized studies of interventions that reduce periprocedural bleeding (e.g., a transradial approach for access or use of bivalirudin) have produced mixed results, with some demonstrating reduced short-term and long-term risk of death (38,39), and others showing no effect (40–42).
Our findings, particularly if reproduced in other cohorts and in the setting of the newer antiplatelet agents, have 3 key implications. Firstly, they underscore the need for systematic evaluation of bleeding avoidance strategies after hospital discharge (34,43,44). Although gastric acid suppression with proton pump inhibitors or histamine-2 receptor antagonists significantly reduces the risk of bleeding in patients on antiplatelet therapy, no randomized trials have compared the 2 groups of agents in patients on dual antiplatelet therapy (45). Observational data suggest that proton pump inhibitors offer greater protection from upper gastrointestinal bleeding than histamine-2 receptor antagonists in this setting (46), but whether these agents also improve long-term thrombotic outcomes should be explored. Secondly, because the reduced incidence of MI and stent thrombosis comes at the expense of an increased risk of bleeding (47–49), the net clinical benefit of long-term use of potent antiplatelet and antithrombotic drugs after PCI should be re-examined. Validated tools to predict the risk of bleeding and MI after PCI may help physicians personalize treatment plans for patients on the basis of relative risk for bleeding and thrombotic events. Although routine use of risk prediction tools has been challenging, the rapid expansion of electronic health records nationally may facilitate integration of effective clinical decision support systems within the clinical workflow. Third, our findings argue that comparative effectiveness studies of coronary revascularization should focus on both harms and benefits (9). Although studies of long-term outcomes of PCI have traditionally targeted recurrent MI, the similar prognostic impact of thrombotic and bleeding events on long-term mortality suggests that bleeding should also be considered when determining the net clinical benefit of treatment. This is an especially important consideration because major bleeds outnumbered MIs in the intervention arms of pivotal trials leading to the approval of both ticagrelor and prasugrel (50,51).
We did not have data on selected variables (e.g., coronary anatomy or left ventricular systolic function) that may have influenced long-term outcomes in this cohort. We could not systematically capture data on minor bleeds that did not lead to hospitalization, asymptomatic MIs, and MIs resulting in sudden cardiac death prior to hospitalization. The low observed rate of concurrent bleeding and MI in our dataset may result from the specificity of the algorithm used to identify these events in claims data. We could not grade the severity of bleeding from the available data. However, we adjusted for longitudinal data on a wide range of comorbidities, drug therapies, and laboratory values—many of which were not accounted for in prior studies—in a large community-based sample of patients, enabling us to mitigate the impact of major known confounders. Although the observed association between bleeding and death persisted after extensive adjustment for baseline and longitudinal confounding variables, we cannot completely rule out the effects of unmeasured or residual confounding. This analysis includes patients undergoing PCI from 1996 through 2008, during which PCI evolved considerably. However, given the rarity of the events studied, further stratification (e.g., by stent type) would have resulted in an unacceptable loss of statistical power. Given that this study examines the impact of spontaneous bleeding or MI occurring several months post-discharge, this association is unlikely to be modified by technical details of the procedure. Nevertheless, this question needs to be examined in a more contemporary cohort, particularly in the context of newer antiplatelet and novel anticoagulant agents unavailable during the study period to determine whether the observed safety-effectiveness tradeoffs extend to patients receiving these agents. Finally, we did not have access to cardiovascular versus noncardiovascular causes of death, which might be instructive because MI and bleeding may be associated with all-cause death through different mechanisms. Future studies should examine the association of spontaneous bleeding and MI after PCI with cause-specific mortality.
Among a large, diverse cohort of patients receiving PCI, spontaneous bleeding after discharge was independently associated with significantly higher long-term mortality during follow-up, comparable to that of an MI. This argues for systematic evaluation of strategies to reduce bleeding beyond the index procedure, and emphasizes the need for careful long-term evaluation of the net clinical benefit of emerging antiplatelet and antithrombotic strategies in the management of coronary heart disease.
COMPETENCY IN MEDICAL KNOWLEDGE: Post-procedural bleeding and myocardial infarction in patients undergoing PCI are associated with increased mortality 1 year later.
TRANSLATIONAL OUTLOOK: Additional studies are needed to inform decisions about the intensity and duration of antiplatelet therapy on the basis of individualized assessments of the risks of thrombotic and bleeding events for patients undergoing PCI.
The authors are deeply grateful to Sue Hee Sung, Yvonne Walston, and Elaine Steel for their administrative support.
This study was funded in part by the American Heart Association (Grant Number 0875162N), Kaiser Permanente Northern California Division of Research, Stanford University, and the University of California-San Francisco. The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; or preparation, review, and approval of the manuscript. Dr. Go has received a research grant from Genentech. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- confidence interval
- hazard ratio
- interquartile range
- myocardial infarction
- percutaneous coronary intervention
- Received March 12, 2014.
- Revision received January 4, 2015.
- Accepted January 22, 2015.
- American College of Cardiology Foundation
- Moscucci M.,
- Fox K.A.,
- Cannon C.P.,
- et al.
- Mehta S.K.,
- Frutkin A.D.,
- Lindsey J.B.,
- et al.,
- for the National Cardiovascular Data Registry
- Mehta S.R.,
- Yusuf S.,
- Peters R.J.,
- et al.,
- for the Clopidogrel in Unstable angina to prevent Recurrent Events trial (CURE) Investigators
- Cannon C.P.,
- Harrington R.A.,
- James S.,
- et al.,
- for the PLATelet Inhibition and patient Outcomes Investigators
- Lindsey J.B.,
- Marso S.P.,
- Pencina M.,
- et al.,
- for the EVENT Registry Investigators
- Mehran R.,
- Pocock S.J.,
- Stone G.W.,
- et al.
- Ndrepepa G.,
- Berger P.B.,
- Mehilli J.,
- et al.
- Eikelboom J.W.,
- Mehta S.R.,
- Anand S.S.,
- et al.
- Manoukian S.V.,
- Feit F.,
- Mehran R.,
- et al.
- Lincoff A.M.,
- Bittl J.A.,
- Harrington R.A.,
- et al.,
- for the REPLACE-2 Investigators
- Newman T.B.,
- Brown A.N.
- Singer D.E.,
- Chang Y.,
- Fang M.C.,
- et al.
- Selby J.V.,
- Ray G.T.,
- Zhang D.,
- et al.
- Levey A.S.,
- Coresh J.,
- Greene T.,
- et al.,
- for the Chronic Kidney Disease Epidemiology Collaboration
- Go A.S.,
- Yang J.,
- Ackerson L.M.,
- et al.
- Masoudi F.A.,
- Ponirakis A.,
- Yeh R.W.,
- et al.
- Bassand J.P.
- Gutierrez A.,
- Rao S.V.
- Aronson D.,
- Suleiman M.,
- Agmon Y.,
- et al.
- Chase A.J.,
- Fretz E.B.,
- Warburton W.P.,
- et al.
- Stone G.W.,
- Witzenbichler B.,
- Guagliumi G.,
- et al.,
- for the HORIZONS-AMI Trial Investigators
- Rao S.V.,
- Hess C.N.,
- Barham B.,
- et al.
- Abraham N.S.,
- Hlatky M.A.,
- Antman E.M.,
- et al.
- Wiviott S.D.,
- Braunwald E.,
- McCabe C.H.,
- et al.,
- for the TRITON-TIMI 38 Investigators