Data source and study population
Among the National Cardiovascular Data Registries, the CathPCI Registry is the largest quality improvement program for PCI in the world. Cosponsored by the American College of Cardiology and the Society for Cardiovascular Angiography and Interventions, it captures detailed clinical data on baseline patient and hospital characteristics, clinical presentation, procedural complications, and in-hospital outcomes among patients undergoing PCI at >1,500 sites across the United States. Details of the design and conduct of this registry have been previously described. Data are systematically collected using third-party software platforms certified by the American College of Cardiology or through a secure, Web-based platform and are regularly audited for data completeness and accuracy (14). The Duke University Medical Center Institutional Review Board granted a waiver of informed consent and authorization for this study, as data are collected without individual patient identifiers.
Between July 2009 and June 2013, we identified 2,516,937 patients undergoing PCI at 1,453 hospitals. We excluded patients with missing variables necessary to define bleeding (n = 10,210) and those who underwent coronary artery bypass grafting (n = 30,238). We further excluded patients undergoing PCI at sites that reported no bleeding events (n = 13,752); patients who had contraindications to, were blinded to, or had missing information for bivalirudin administration (n = 2,573); and patients who presented at hospitals performing <50 PCIs annually (n = 478).
Data definitions and outcomes
Patients were treated using any BAS if: 1) they underwent PCI via radial artery access; 2) bivalirudin was used for periprocedural anticoagulation regardless of arterial site of access, or, in case of femoral access; 3) they received vascular closure device to assist with hemostasis at the conclusion of the procedure.
The outcome of interest was the rate of CathPCI bleeding, defined as site-reported arterial access-site bleeding (either external or a hematoma >10 cm for femoral access, >5 cm for brachial access, or >2 cm for radial access); retroperitoneal, gastrointestinal, or genitourinary bleeding; intracranial hemorrhage; cardiac tamponade; post-procedural hemoglobin decrease of ≥3 g/dl in patients with pre-procedural hemoglobin levels ≥16 g/dl; or post-procedural non-bypass surgery-related blood transfusion for patients with pre-procedural hemoglobin levels ≥8 g/dl.
Statistical analysis
We compared baseline demographic, clinical, presentation, and hospital characteristics for patients by tertile of hospital-level use of BAS following PCI. Continuous variables are expressed as median values with interquartile ranges (IQRs), and categorical values are presented as percentages. We used Pearson chi-square tests for categorical variables and Wilcoxon rank sum tests for continuous variables.
The observed rate of bleeding for each hospital was calculated as the observed number of bleeding events divided by the total number of admissions. To estimate adjusted hospital bleeding rates by patient risk, we used logistic regression with random intercepts for hospital. The log odds for random hospital were assumed to be normally distributed, with mean equal to the intercept and variance equal to the random-effect variance or variation in log odds attributable to between-hospital differences. We estimated these parameters and transformed from the log-odds scale to the probability scale. The hospital-specific intercepts were used to estimate hospital-specific bleeding rates.
To assess whether the use of BAS attenuates the variation in adjusted hospital-level bleeding rates, we fit a series of 8 models and estimated random-effect variance. We used the percentage of proportional change in variance (PCV) to assess the incremental effect of adding variables to the model. The PCV is calculated as follows: PCV = (Va − Vb)/Va × 100, where Va is the variance of the initial model and Vb is the variance of the model with more terms. The 8 models were: 1) unadjusted; 2) patient risk adjusted; 3) patient risk adjusted plus radial; 4) patient risk adjusted plus bivalirudin; 5) patient risk adjusted plus vascular closure device; 6) patient risk adjusted plus radial and bivalirudin; 7) patient risk adjusted plus bivalirudin and vascular closure device; and 8) patient risk adjusted plus radial, bivalirudin, and vascular closure device. We calculated the PCV for model 2 versus 1 and for all other models versus model 2. The patient risk-adjusted models included variables from the previously validated CathPCI bleeding model (15). Specifically, included covariates were sex, age, body mass index, cerebrovascular disease, peripheral vascular disease, chronic lung disease, prior PCI, diabetes, left ventricular ejection fraction, chronic kidney disease (no disease, dialysis, mild [glomerular filtration rate 45 to 59 ml/min], and moderate [glomerular filtration rate 30 to 44 ml/min]), cardiogenic shock, PCI status (emergency or salvage vs. urgent), New York Heart Association functional class (I to IV), cardiac arrest, pre-procedural hemoglobin (as a continuous linear spline with 1 knot at 13 g/dl), ST-segment elevation myocardial infarction, pre-procedure TIMI (Thrombolysis In Myocardial Infarction) flow grade, number of diseased vessels, in-stent thrombosis on some lesion previously treated within 1 month, Society for Cardiovascular Angiography and Interventions class, and lesion segment category.
The rate of any BAS use for each hospital was calculated as the number of patients who received some BAS divided by the total number of admissions. To account for the possibility of the risk-treatment paradox for BAS (10), we examined patient-level use of BAS by predicted bleeding risk. To describe whether hospitals with higher use of any BAS had lower adjusted bleeding rates, we used a scatterplot to display the association between hospital adjusted bleeding rates and rates of any BAS use. We also grouped hospitals by deciles of rate of any BAS use and describe mean risk-adjusted hospital bleeding rate. To quantify the association between hospital's rate of any BAS use and patient-level bleeding events, we fit a mixed-effects logistic regression model adjusting for the CathPCI Registry bleeding model covariates and rate of any BAS. We checked for linearity of hospital rate of any BAS using splines.
All analyses were performed at the Duke Clinical Research Institute, which served as the data analytic center, using SAS version 9.4 (SAS Institute, Cary, North Carolina).