Association Between Use of Bleeding Avoidance Strategies and Risk of Periprocedural Bleeding Among Patients Undergoing Percutaneous Coronary Intervention FREE

Improving hospital safety is a recognized health care priority in the United States.¹ There is an important opportunity to improve the safety of percutaneous coronary intervention (PCI), a treatment performed approximately 1 million times a year in the United States alone.² Periprocedural bleeding is the most common noncardiac complication of PCI and is associated with risk of early mortality^3- 4 as well as higher costs of care.⁵ Moreover, the rate of periprocedural bleeding varies substantially across institutions and is modifiable through the use of bleeding avoidance strategies such as vascular closure devices, bivalirudin, and radial access. Underscoring the importance of bleeding complications and the opportunity for improvement, the Centers for Medicare & Medicaid Services has identified bleeding and hematoma after cardiovascular procedures to be quality indicators among centers participating in its Acute Care Episode demonstration.⁶

To assist clinicians in identifying patients' risks for periprocedural bleeding, a risk model was previously developed and validated using the National Cardiovascular Data Registry (NCDR) CathPCI Registry.⁷ This model uses preprocedural variables to predict risk of bleeding events after PCI. To date, however, whether this model is useful in identifying patients with greater potential to benefit from bleeding avoidance strategies is unknown. It is also unknown whether clinicians treat the highest-risk patients preferentially. Failure to treat the highest-risk patients with bleeding avoidance strategies would demonstrate the potential for risk-stratifying patients at the time of PCI to direct therapy.

To address these gaps in knowledge, we described peri-PCI bleeding rates associated with the use of manual compression, vascular closure devices, bivalirudin, or both strategies (vascular closure devices plus bivalirudin) in patients across a spectrum of preprocedural bleeding risk; we also examined current patterns of the use of these strategies as a function of bleeding risk.

The NCDR CathPCI Registry is a voluntary nationwide reporting system for diagnostic cardiac catheterization and PCI procedures jointly sponsored by the American College of Cardiology (ACC) and the Society for Cardiovascular Angiography and Interventions. Descriptions of the NCDR have been published.^8- 9 Demographic, clinical, procedural, and institutional data elements from diagnostic catheterization and PCI procedures are collected at more than 1100 participating centers. Data are entered via either a secure Web-based platform or software provided by ACC-certified vendors into a secure, centralized database (CathPCI version 3.04) stored at the ACC Heart House in Washington, DC. Data quality assurance includes automatic system validation and reporting of data completeness, random on-site auditing of participating centers, and education and training for site data managers.¹⁰ A comprehensive description of NCDR data elements and definitions is available at http://www.ncdr.com/WebNCDR/ELEMENTS.ASPX.

All data elements and definitions were prospectively defined by a committee of the ACC. Race/ethnicity data were reported by patients or family and recorded in the medical record. Trained chart abstractors recorded race/ethnicity on the standard NCDR CathPCI case report form. The options were white, black, Hispanic, Asian, Native American, and other. Glomerular filtration rate was calculated using admission serum creatinine values and the abbreviated Modification of Diet in Renal Disease formula.¹¹

This study was approved by the Saint Luke's Health System institutional review board and was determined to meet the definition of research not requiring informed consent.

Only patients who underwent PCI via the femoral artery approach were included in this analysis. Exclusion criteria consisted of patients with more than 1 PCI procedure during a hospitalization (since bleeding events could not reliably be attributed to a specific procedure), incomplete data for calculation of expected bleeding rates, PCI through access of a nonfemoral artery (ie, radial, brachial), cardiogenic shock, missing device data, death in the catheterization laboratory, or unknown data on bleeding events.

Candidate bleeding avoidance strategies consisted of vascular closure devices (Angio-Seal [St Jude Medical, St Paul, Minnesota], Perclose A-T [Abbott Vascular, Abbott Park, Illinois], or other) without bivalirudin; bivalirudin (Angiomax [The Medicines Company, Parsippany, New Jersey]) without vascular closure devices; or both strategies (vascular closure devices plus bivalirudin). Patients receiving manual compression did not receive vascular closure devices or bivalirudin and thus served as the reference group.

Bleeding rates were determined for all patients and within 3 clinically important subgroups, based on patients' pre-PCI bleeding risk scores derived using the NCDR CathPCI bleeding risk model.⁷ Risk scores were generated for each patient in this study based on the inverse logarithmic sum of the β coefficients for each of the following pre-PCI variables: ST-segment elevation myocardial infarction (MI), non–ST-segment elevation MI, female sex, previous congestive heart failure, no previous PCI, New York Heart Association/Canadian Cardiovascular Society Class IV heart failure, peripheral vascular disease, age, and estimated glomerular filtration rate.

In-hospital bleeding complications following PCI were ascertained and voluntarily reported by centers. The primary outcome for this study was periprocedural bleeding, which, as defined by the NCDR CathPCI data definition, required a blood transfusion or a prolonged hospital stay for management or was associated with a decrease (>3 g/dL) in hemoglobin level.

Demographic data were described across treatment groups as mean (SD) for continuous variables and number (%) for categorical variables. Patients with attempted use of vascular closure devices were included with the vascular closure devices group. Based on individual risk scores calculated using the NCDR CathPCI bleeding risk model,⁷ patients were categorized into 3 groups of risk for post-PCI bleeding events occurring during hospitalization: low (<1%), intermediate (1%-3%), and high (>3%).

To minimize confounding, a propensity score match analysis was implemented. Scores for each bleeding avoidance strategy (manual compression, vascular closure devices, bivalirudin, or vascular closure devices plus bivalirudin) were derived using a multinomial regression model.^12- 16 Variables used to derive these propensity scores were demographics (age, sex, race/ethnicity); clinical characteristics (body mass index, New York Heart Association heart failure classification); coronary artery disease risk factors (diabetes, hypertension, dyslipidemia, smoking, family history of coronary artery disease); coronary artery disease history (PCI, coronary artery bypass graft surgery, MI); other cardiovascular disease history (congestive heart failure, cerebrovascular disease, peripheral vascular disease, valve surgery, cardiac transplantation); other disease history (chronic obstructive pulmonary disease, renal failure); and presenting syndrome (no symptoms, atypical chest pain, stable angina, unstable angina, ST-segment elevation MI, and non–ST-segment elevation MI).

Matching was performed simultaneously on all strategy propensity scores within the bleeding risk groups using nearest-neighbor matching without replacement, with a caliper width of 0.5%.¹² Absolute standardized differences were computed to evaluate matching effectiveness and displayed graphically^17- 19; values less than 10% and closer to zero demonstrate a more balanced cohort.¹⁷ Hierarchical modeling was then performed on the matched cohort to account for hospital characteristics. From this model, odds ratios were obtained to estimate the number of patients needed to treat to prevent 1 bleeding event for each strategy compared with manual compression.^20- 22

Statistical significance was defined as P < .05. All statistical analyses were performed by the Saint Luke's Mid America Heart Institute Department of Biostatistics using SAS version 9.2 (SAS Institute, Cary, North Carolina).

From January 1, 2004, to September 30, 2008, 1 759 408 patients underwent PCI. After exclusions (Figure), there were 1 522 935 eligible patients at 955 centers. Patient demographics, clinical characteristics, risk factors, disease history, admission presentation, and selected hospital characteristics for the entire population and for subgroups according to bleeding avoidance strategy are shown in Table 1 and Table 2.

Overall, bleeding occurred in 30 429 patients (2%). Bleeding rates by candidate strategy for the overall study population are shown in Table 3 and eFigure 1). Manual compression was used in 35% of patients, vascular closure devices in 24%, bivalirudin in 23%, and vascular closure devices plus bivalirudin in 18%. Bleeding events were reported in 2.8% of patients who received manual compression, compared with 2.1% receiving vascular closure devices, 1.6% receiving bivalirudin, and 0.9% receiving both strategies (P < .001) (Table 3 and eFigure 1).

According to the NCDR CathPCI bleeding risk model, bleeding risk was classified as low (<1%) in 475 152 patients (31%), intermediate (1%-3%) in 746 727 (49%), and high (>3%) in 301 056 (20%). Observed rates of bleeding in these categories were 0.72%, 1.73%, and 4.69%, respectively. Bleeding rates associated with candidate strategies stratified by preprocedural risk category are shown in Table 3 and eFigure 1. In the low-risk group, manual compression was associated with a bleeding rate of 0.9%, vascular closure devices with a rate of 0.9%, bivalirudin with a rate of 0.6%, and vascular closure devices plus bivalirudin with a rate of 0.4% (P < .001). As preprocedural risk of bleeding increased, differences in actual bleeding rates between strategies became more pronounced. In the intermediate-risk group, manual compression was associated with a bleeding rate of 2.3%, vascular closure devices with a rate of 1.9%, bivalirudin with a rate of 1.4%, and vascular closure devices plus bivalirudin with a rate of 0.8% (P < .001); in the high-risk group, the corresponding rates were 6.1%, 4.6%, 3.8%, and 2.3%, respectively (P < .001).

In the overall group, 508 455 of 529 247 eligible patients receiving manual compression (96%) were propensity matched using multinomial regression modeling to patients who received bleeding avoidance strategies. Corresponding values were 144 594 of 146 557 (99%) in the low-risk group, 252 898 of 261 363 (97%) in the intermediate-risk group, and 110 963 of 121 327 (91%) in the high-risk group. The effectiveness of propensity matching in the total cohort is demonstrated in eFigure 2, which is a standardized difference plot for variables used in the propensity model. After matching, the absolute standardized difference between patients who received bleeding avoidance strategies and controls for each covariable was 0% to 1%.

Table 4 shows in-hospital bleeding events, odds ratios, the estimated number needed to treat to prevent 1 bleeding event, and the estimated reduction in bleeding events per 1000 patients treated with each strategy relative to manual compression in the matched, site-adjusted cohort. Data are displayed for the overall population and by preprocedural bleeding risk group. Independent of preprocedural risk, the use of vascular closure devices, bivalirudin, and vascular closure devices plus bivalirudin were associated with fewer bleeding events per 1000 patients treated (6.7 [95% confidence interval {CI}, 5.7-7.7], 8.5 [95% CI, 7.6-9.3], and 14.2 [95% CI, 13.5-14.8] events, respectively). In patients receiving both strategies, the high-risk group had fewer bleeding events per 1000 patients treated compared with the intermediate- and low-risk groups (30.5 [95% CI, 27.9-32.8] vs 12.5 [95% CI, 11.6-13.3] and 5.3 [95% CI, 4.5-6.0] events, respectively).

The frequency of use of bleeding avoidance strategies according to estimated pre-PCI risk of bleeding is shown in Table 5. Manual compression was used most often in the highest-risk patients and least often in the intermediate- and lowest-risk cohorts (40.3% vs 35.0% and 30.8%, respectively; P < .001), while the use of vascular closure devices plus bivalirudin was highest in low-risk patients (21.0%) and lower in intermediate-risk (17.8%) and high-risk (14.4%) patients (P < .001).

Using data from more than 1.5 million patients in the NCDR CathPCI Registry, we compared bleeding rates among patients undergoing PCI and receiving strategies to mitigate bleeding across a spectrum of preprocedural risk for bleeding. Among high-risk patients, the use of vascular closure devices plus bivalirudin was associated with an absolute 3.8% lower bleeding rate, which translates into an estimated number needed to treat of 33 to prevent 1 bleeding event as compared with manual compression. Lower rates of bleeding associated with this treatment strategy were proportionately less in intermediate- and low-risk patients. Despite the association between the use of vascular closure devices plus bivalirudin and lower bleeding rates among the highest-risk patients, these patients were the least likely to receive both strategies and most likely to receive manual compression. This apparent risk-treatment paradox highlights an opportunity for routine preprocedural risk stratification as a means to identify patients ideally suited for individualized bleeding avoidance strategies with the goal of increasing the safety of PCI.

Targeting bleeding complications as a quality-improvement goal holds great potential for improving the safety and cost-effectiveness of PCI. Among the more than 1 million PCIs performed annually in US hospitals, bleeding is a morbid and costly complication, occurring in 2% to 6% of patients,^23- 27 with wide variability across institutions. Major bleeding events result in an average 4- to 6-day increase in length of stay^28- 29 and, on average, increase hospital costs by $6000 to $8000.⁵ In addition to its association with nonfatal MI^28,30 and stroke,³⁰ periprocedural bleeding is also strongly associated with early^3- 4 and late³ mortality. Bleeding also exposes many patients to the added risk of blood transfusions.²³ According the Healthcare Cost and Utilization Project, 50% of patients experiencing bleeding following PCI receive a blood transfusion, which is associated with greater median length of stay (2 vs 6 days) and greater mean hospital charges ($48 000 vs $85 000).³¹

The current study found that higher-risk patients—those with the greatest potential to benefit from bleeding avoidance strategies—were the least likely to be treated with such strategies in contemporary clinical practice. Conversely, the lowest-risk patients were the most likely to receive bleeding avoidance strategies, supporting the presence of a risk-treatment paradox.^32- 35 This phenomenon has been demonstrated in a variety of medical treatment scenarios, including lower rates of angiotensin-converting enzyme inhibitor or β-blocker use in patients with severe heart failure,³² lower use of angiography after acute MI,^36- 38 and lower use of statins in higher-risk patients with angiographically confirmed coronary disease.³⁹

Several factors may contribute to the higher use of bleeding avoidance strategies in lower-risk patients found in this study. First, assessing the risk for bleeding in clinical practice is neither inherently intuitive nor commonly used. Second, physicians have more experience using bivalirudin in lower-risk patients, since it was first studied in patients undergoing elective PCI and only recently in higher-risk patients (ie, those with ST-segment elevation MI).⁴⁰ Lastly, the prior published data for bleeding mitigation with vascular closure devices has been limited.^41- 43

The results of this study suggest the need for additional research to better understand why higher-risk patients are least likely to receive bleeding avoidance strategies but also suggest the need to test interventions to overcome the risk-treatment paradox, such as enabling physicians to purposefully direct bleeding avoidance strategies to patients by providing preprocedural estimates of post-PCI bleeding. Translation of the findings in this study into clinical practice to optimally guide the use of such strategies will be challenging. Incorporating risk models into everyday care, which requires capturing risk factors, estimating individual risk, and presenting information in a clear and understandable manner at the point of care for physicians, is currently not feasible for many clinical environments. However, with the current national investment in health information technology, increasing literature on clinical decision support, increasing focus by regulatory and other agencies on the documentation of patient risk to guide care, and the development of quality-of-care tools by entities such as the ACC and the Society for Cardiovascular Angiography and Interventions, the health care environment is increasingly capable of supporting the implementation of point-of-care interventions to help guide effective and safe care.

It should be noted that there are more supporting data on the efficacy of bivalirudin than on that of vascular closure devices with regard to bleeding avoidance. In randomized trials including patients with broad indications for PCI (elective or acute coronary syndromes), bivalirudin has been demonstrated to reduce major bleeding following PCI to an extent comparable with the findings in this observational study.^24,44- 46 Randomized controlled trials demonstrating reductions in bleeding associated with the use of vascular closure devices were designed to assess ambulation time rather than bleeding. A meta-analysis of 30 clinical studies, including 4000 patients, showed a trend for reduced vascular complications associated with the Angio-Seal device (odds ratio, 0.46; 95% CI, 0.20-1.04; P = .06).⁴² A recent analysis from the Acute Catheterization and Urgent Intervention Triage Strategy (ACUITY) trial also suggests that a significant reduction in major bleeding is associated not only with the use of bivalirudin but also with vascular closure devices in patients undergoing PCI in the setting of unstable angina or non–ST-segment elevation MI.⁴³ Similar to our study, the greatest reduction in bleeding was seen in patients receiving both a vascular closure device and bivalirudin.

Given the findings of our observational study and that previous data linking vascular closure devices to bleeding are both limited and from nonrandomized studies, an adequately powered randomized trial of vascular closure devices evaluating post-PCI bleeding end points is needed. It is also important to note that data on vascular closure devices and bivalirudin were published during a period overlapping with data collection in the present study, and bivalirudin was not included in PCI guidelines⁴⁷ until 2006. Thus, there may be important temporal trends for rates of strategies not investigated in this study.

Several aspects of this work should be considered when interpreting our results and in translating these findings to clinical practice. First, this was not a randomized trial; thus, a causal relationship between bleeding avoidance and evaluated strategies cannot be concluded. Second, potential unmeasured confounding is a limitation of all observational studies. We sought to minimize confounding through the use of a center-adjusted, multinomial propensity-matched analysis, which successfully balanced the observable patient and treatment characteristics; nevertheless, some unmeasured confounding may have been present. The NCDR CathPCI Registry also does not currently collect data on complications related to vascular closure devices. However, the reported frequency of such complications is low.⁴⁸

Third, activated clotting time is not available in the NCDR CathPCI Registry, which limited our ability to assess the relationship between heparin dosing, level of anticoagulation, and bleeding, which could have further informed the observed variability in bleeding among the cohort receiving manual compression.

Fourth, some patients are not suited to receive bleeding avoidance strategies. For example, bivalirudin is not recommended in the setting of therapeutic anticoagulation with prior agents, including unfractionated or low-molecular-weight heparin with or without a glycoprotein IIb/IIIa agent. Bivalirudin is also not currently recommended in patients undergoing PCI for a chronic total occlusion.⁴⁹ Furthermore, mitigation of bleeding risk may be neutralized with the concomitant use of glycoprotein IIb/IIIa inhibitors and in the setting of renal failure. Vascular closure devices are also not recommended for use given several anatomical limitations, such as puncture at the site of an anatomical arterial bifurcation, presence of severe calcification, or presence of severe obstructive peripheral artery disease; the reasons for not using a vascular closure device for a given patient are not collected in the CathPCI Registry. To the extent that some of these examples cluster more frequently in patients at intermediate or high risk for bleeding, there will be a reduction in potential to offset the apparent risk-treatment paradox observed in this study.

In a large, national PCI registry, the use of vascular closure devices and bivalirudin was associated with significantly lower rates of periprocedural bleeding. However, there was an apparent risk-treatment paradox, whereby patients at greatest risk of bleeding were least likely to receive a bleeding avoidance strategy. These findings emphasize the opportunity to improve the safety of PCI and to further explore cost efficacy by directing such strategies to those patients most likely to benefit from them.