Despite having a low long-term risk of thromboembolism, patients treated with aortic valve bioprostheses carry a modest early (<90 days) risk of embolic events (1). The extent to which this early risk is modifiable has been debated (2), and 2 small randomized trials (n = 268) have failed to demonstrate a benefit of vitamin K antagonism when compared with aspirin alone in this setting (5). Nevertheless, the use of antithrombotic agents early after bioprosthetic valve implantation remains common (2).

In this analysis, we used a contemporary cohort of patients from the Society of Thoracic Surgeons Adult Cardiac Surgery Database (STS ACSD) to evaluate: 1) contemporary patterns of post-discharge anticoagulation after bioprosthetic aortic valve replacement (AVR); 2) the absolute risk of early (<90 days) thromboembolic and bleeding events in this setting; and 3) the relative safety and effectiveness of both warfarin plus aspirin and warfarin-only versus aspirin-only in the early post-operative interval.

The STS ACSD collects detailed in-hospital data on adult patients undergoing cardiac surgery at more than 1,000 participating institutions throughout the United States. Patient follow-up is accomplished for older individuals (≥65 years of age) through a deterministic linkage process with Medicare inpatient insurance claims files (9). For purposes of this analysis, STS ACSD records were linked to Part A Medicare Provider and Analysis Review (MedPAR) files with indirect patient identifiers (i.e., age and sex, as well as dates of birth, admission, and discharge; linkage rate, 75.5%). Compared with patients in the linked population, eligible patients who were not linked to Medicare records were (on average) slightly younger, less often female, less often Caucasian, and more often from the Mid-Atlantic and Pacific Northwest regions. Otherwise, linked and unlinked patients were similar across demographic data and comorbidities of interest.

From the Medicare-linked ACSD cohort, we identified a group of 29,395 fee-for-service patients without a history of prior valve surgery, undergoing elective or urgent isolated AVR with a stented bioprosthesis (with or without concomitant coronary artery bypass grafting) at 800 centers from January 1, 2004, to December 28, 2006. We excluded patients in whom clinical equipoise was unlikely, including those with a pre-operative indication for warfarin, an indwelling mechanical heart valve, a pre-discharge contra-indication to warfarin therapy, a complication related to anticoagulation, or an adverse post-operative gastrointestinal event. Additionally, we excluded patients who died before hospital discharge and those missing data on sex, age, or discharge medications. The Duke University School of Medicine Institutional Review Board granted a waiver of informed consent and authorization for this study.

Analyses were stratified by age (65 to 74 years of age, ≥75 years of age), sex, and thromboembolic risk status. Consistent with American College of Cardiology/American Heart Association guidelines for the treatment of patients with valvular heart disease (10), patients were considered to be at an increased risk for early thromboembolic events if they had a pre-discharge history of atrial fibrillation, thromboembolism, or depressed ejection fraction (ejection fraction <30%).

Because this analysis was designed to investigate the safety and effectiveness of discharge anticoagulation strategies, all endpoints were evaluated from the time of hospital discharge. Primary endpoints included death or repeat hospital stay for embolic (including cerebrovascular accident, transient ischemic attack, and noncerebral arterial thromboembolism) or bleeding events (including hemorrhagic stroke, gastrointestinal bleeding, other bleeding). Secondary endpoints included repeat hospital stay for individual embolic and bleeding events. Death after hospital discharge was identified with the Medicare Denominator File, whereas nonfatal endpoints were identified with primary hospital diagnosis International Classification of Diseases-Ninth Revision-Clinical Modification codes (Online Appendix). Repeat hospital stay for hemopericardium was identified with STS records.

Baseline patient characteristics were stratified by treatment strategy and summarized as percentages for categorical variables and means with SD for continuous variables. The Mantel-Haenszel test was used to compare the distribution of categorical variables between groups, whereas the Wilcoxon rank sum test was used to compare continuous variable distributions. The SAS statistical software (version 9.1, SAS Institute, Cary, North Carolina) was used for all calculations.

Propensity scores estimated the probability of patients receiving either warfarin-only (vs. aspirin-only) or warfarin plus aspirin (vs. aspirin-only) at discharge as a function of 73 observed covariates (Online Appendix) (11). Overlap of the propensity scores for each comparison validated the statistical appropriateness of this method, and comparability of baseline characteristics after propensity score weighting (Online Appendix) suggested that the approach was successful.

Adjusted risk ratios (RRs) were estimated with generalized linear models (for binomial distribution) with a log link function, including a single covariate for treatment strategy and weighting each observation by the inverse of the estimated propensity score (12). Robust sandwich variance estimates were used to obtain 95% confidence intervals (CIs) (14) to account for statistical dependence of patients within sites. Adjusted risk differences were obtained by the same method, with an identity link function. Probabilities are interpreted as risks in this closed cohort with complete follow-up and a low occurrence of death as a competing risk. The adjusted numbers-needed-to-treat (NNT) or -harm (NNH) were calculated as the inverse of the risk difference in the inverse probability weighted populations with similar distributions of measured patient characteristics and are interpreted as an estimated average number of patients needed to treat with either warfarin-only (vs. aspirin-only) or warfarin plus aspirin (vs. aspirin-only) to result in 1 fewer (NNT) or 1 more (NNH) associated event.

Due to a clinical concern that patient frailty might have biased our treatment comparisons, we analyzed the treatment effect for each of the 2 comparisons within a “discharge-to-home” cohort that was expected to have a low proportion of frail patients. Although the incidence of adverse events was low in this cohort, the treatment effects were similar to that observed in the overall study cohort (Online Appendix).

The study cohort included 25,656 patients who survived to hospital discharge after isolated AVR (± coronary artery bypass grafting) with a stented bioprosthesis at 1 of 797 centers between 2004 and 2006 (Figure 29_gr1). Within this patient cohort, the median age was 77 years (interquartile range: 72 to 81 years), and 39.4% were women. A prior history of at least 1 of the following risk factors for early thromboembolism was present at hospital discharge in 13,458 patients (risk factor positive: 52.5%): atrial fibrillation (41.1%), thromboembolism (13.6%), or ejection fraction <30% (4.8%).

At hospital discharge, the 3 most prevalent anticoagulation strategies included aspirin-only (n = 12,457, 48.6%), aspirin plus warfarin (n = 5,972, 23.3%), and warfarin-only (n = 2,999, 11.7%). No anticoagulant was recommended in 6.5% (Figure 29_gr2). Institutional variation was high in both: 1) the proportion of patients treated with warfarin at hospital discharge; and 2) the proportion of warfarin-treated patients receiving aspirin plus warfarin (Online Appendix). Baseline characteristics for patients receiving each of the 3 most prevalent anticoagulation strategies are presented in (Table 1).

The 3-month incidence of death after hospital discharge was low (aspirin-only, 3.0%; aspirin plus warfarin, 3.1%; warfarin-only, 4.0%). After risk adjustment, the addition of warfarin to aspirin was associated with a 0.6% absolute and 20% relative risk reduction for mortality at 3 months (adjusted RR: 0.80, 95% CI: 0.66 to 0.96; NNT 153) (Table 2)—this trend was consistent across strata of age, sex, and thromboembolic risk (Figure 29_gr3A). Warfarin-only was not associated with a reduction in the risk of mortality (vs. aspirin-only) (Table 2).

The 3-month incidence of embolic events after hospital discharge was low (0.9%); however, the majority of events (71.5%) were neurologic, and repeat hospital stay for stroke was associated with a 3.9% unadjusted risk of mortality within the subsequent 2 weeks. The 3-month incidence of embolic events was higher among patients with (vs. without) characteristics thought to be associated with an early risk of thromboembolism (1.30% vs. 0.81%). The addition of warfarin to aspirin was associated with a 0.4% absolute reduction (48% relative reduction) in the 3-month risk of embolic events (p = 0.006) in the overall cohort, an association that was preserved after risk adjustment (adjusted RR: 0.51, 95% CI: 0.50 to 0.97; NNT 212) (Table 2). This effect was most prominent among the oldest (≥75 years of age) patients (adjusted RR: 0.44, 95% CI: 0.27 to 0.72; NNT 143) (Figure 29_gr3B). In the overall cohort, warfarin-only was not associated with a reduction in the risk of embolic events at 3 months (vs. aspirin-only) (Table 2), a result that was consistent across most subgroups.

The incidence of bleeding requiring repeat hospital stay by 3 months after hospital discharge was low (1.6%), and the majority of bleeding events (77.5%) involved the gastrointestinal tract. Hemopericardium was rare among patients treated with aspirin-only (0.36%) but was more common among those treated with warfarin plus aspirin (0.77%) or warfarin-only (1.0%). Likewise, hemorrhagic stroke was extremely rare in the aspirin-only cohort (0.11%) but was slightly more common among patients treated with warfarin plus aspirin (0.16%) or warfarin-only (0.21%). Overall, repeat hospital stay for bleeding was associated with a 3.5% unadjusted risk of mortality within the subsequent 2 weeks. The unadjusted 3-month incidence of repeat hospital stay for a bleeding event was higher among those in whom warfarin was added to aspirin at hospital discharge (2.8% vs. 1.0%, p < 0.0001). This effect persisted after risk adjustment (RR: 2.80, 95% CI: 2.18 to 3.60, NNH 55) ((Figure 29_gr3)C, Table 2). Warfarin-only was not associated with a higher risk-adjusted incidence of repeat hospital stay for bleeding compared with aspirin-only (Table 2).

This analysis is the first large-scale evaluation of anticoagulation strategies at hospital discharge after bioprosthetic AVR in the United States. In contemporary practice, there exists considerable variation in anticoagulation strategies, and aspirin alone remains a common regimen. Although the risk of early embolic events is relatively low in patients treated with aspirin-only, those treated with warfarin plus aspirin at hospital discharge experience a lower adjusted incidence of early embolic events—a benefit that is balanced by an increased risk of repeat hospital stay for bleeding.

Considerable disagreement persists with regard to the appropriate anticoagulant strategy in the first 3 months after bioprosthetic AVR. Based primarily on expert consensus (Class IIa, Level of Evidence: C), the American College of Cardiology/American Heart Association guidelines have recommended the addition of warfarin to aspirin at hospital discharge for all patients receiving a bioprosthetic AVR (10); however, 2 major international surveys have demonstrated that many practicing cardiac surgeons do not endorse this recommendation (2). Our data demonstrate diverse practice patterns within the United States, with warfarin use observed in a minority of these patients (35%). The diversity of discharge strategies is a testament to the limitations of available evidence.

To date, evidence guiding the selection of anticoagulants has been limited (15). Single-center observational studies evaluating the addition of warfarin to aspirin at hospital discharge have failed to detect a difference in either bleeding or embolic outcomes (16), and 2 small randomized pilot studies evaluating the use of vitamin K antagonism in this setting have also not detected a treatment difference (total n = 268) (5). Our analysis is the first to address this question with a cohort that is adequately powered to detect a difference in these serious but relatively rare outcomes. Significantly, our analysis further evaluates the consistency of these results in important patient subgroups. Although observational in nature, this analysis suggests a beneficial treatment effect of antithrombotic agents for the prevention of early embolic events when added to antiplatelet agents at hospital discharge in patients receiving bioprosthetic AVR. The addition of warfarin to aspirin at hospital discharge would be a reasonable treatment option, on the basis of these results, with an expected number needed to avert 1 death of 153 patients and 1 embolic event of 212 patients. The therapeutic benefit observed with the addition of warfarin to aspirin was not without risk in this elderly cohort, and 1 additional bleeding event was observed at 3 months for every 55 patients treated with warfarin. The majority of these additional bleeding events involved readmission for gastrointestinal bleeding, and no difference in the incidence of hemorrhagic stroke was observed across the 2 treatment strategies. Nevertheless, these events were not benign, and readmission for bleeding was associated with an absolute 3.5% increase in the risk of mortality over the subsequent 2 weeks. This finding highlights the complexity associated with this therapeutic decision and the need for a balanced discussion of alternative treatment strategies.

First, although our study is an exploration of the relative safety and effectiveness of discharge anticoagulant strategies in a real-world setting, and these data reflect real-world practice, they are still observational in nature. As a result, treatment comparisons might be subject to certain biases, including treatment selection bias. Additionally, subtle differences between the Medicare-linked and Medicare-eligible STS ACSD cohorts might limit the generalizability of these results. Second, this analysis was performed with inpatient insurance claims for outcome ascertainment and might underestimate the risk of both bleeding and embolic events in this population, especially for minor “nuisance” events that are less likely to provoke a hospital admission. Finally, although most repeat hospital stay events of Medicare fee-for-service patients are captured with MedPAR billing data, a small subset might occur at hospitals that do not file Medicare claims. This is not expected to have biased the observed treatment effect, but it might have led to an underestimation of the incidence of repeat hospital stay in this cohort.

The early risk of embolic events is low in patients receiving bioprosthetic AVR, and considerable practice variation exists in the selection of discharge anticoagulants in the United States. Although a warfarin plus aspirin strategy was associated with a higher risk of bleeding (vs. aspirin only), the early risk of embolic events was lower among patients treated with this strategy. Given the clear trade-off between thromboembolic and bleeding events, we would recommend the use of warfarin plus aspirin for bioprosthetic aortic valve patients thought to be at low risk for bleeding complications, with an aspirin-only strategy reserved for those thought to be at high risk for bleeding.