Identification of hospital outliers in bleeding complications after percutaneous coronary intervention

Post-percutaneous coronary intervention (PCI) bleeding complications are an important quality metric. We sought to characterize site-level variation in post-PCI bleeding and explore the influence of patient and procedural factors on hospital bleeding performance. Hospital-level bleeding performance was compared pre- and postadjustment using the newly revised CathPCI Registry(®) bleeding risk model (c-index, 0.77) among 1292 National Cardiovascular Data Registry(®) hospitals performing >50 PCIs from 7/2009 to 9/2012 (n=1,984,998 procedures). Using random effects models, outlier sites were identified based on 95% confidence intervals around the hospital's random intercept. Bleeding 72 hours post-PCI was defined as: arterial access site, retroperitoneal, gastrointestinal, or genitourinary bleeding; intracranial hemorrhage; cardiac tamponade; nonbypass surgery-related blood transfusion with preprocedure hemoglobin ≥ 8 g/dL; or absolute decrease in hemoglobin value ≥ 3 g/dL with preprocedure hemoglobin ≤ 16 g/dL. Overall, the median unadjusted post-PCI bleeding rate was 5.2% and varied among hospitals from 2.6% to 10.4% (5th, 95th percentiles). Center-level bleeding variation persisted after case-mix adjustment (2.8%-9.5%; 5th, 95th percentiles). Although hospitals' observed and risk-adjusted bleeding ranks were correlated (Spearman ρ: 0.88), individual rankings shifted after risk-adjustment (median Δ rank order: ± 91.5; interquartile range: 37.0, 185.5). Outlier classification changed postadjustment for 29.3%, 16.1%, and 26.5% of low-, non-, and high-outlier sites, respectively. Hospital use of bleeding avoidance strategies (bivalirudin, radial access, or vascular closure device) was associated with risk-adjusted bleeding rates. Despite adjustment for patient case-mix, there is wide variation in rates of hospital PCI-related bleeding in the United States. Opportunities may exist for best performers to share practices with other sites.