Predicting long-term bleeding after percutaneous coronary intervention

Objectives: To construct a model to predict long-term bleeding events following percutaneous coronary intervention (PCI). Background: Treatment with dual antiplatelet therapy following PCI involves balancing the benefits of preventing ischemic events with the risks of bleeding. There are no models to predict long-term bleeding events after PCI. Methods: We analyzed 1-year bleeding outcomes from 3,128 PCI procedures in the Patient Risk Information Services Manager (PRISM) observational study. Patient-reported bleeding events were categorized according to Bleeding Academic Research Consortium (BARC) definitions. Logistic regression analysis was used to develop a model predicting BARC ≥ 1 bleeding. Results: BARC 0, 1, 2 or 3 bleeding was observed in 574 (18.4%); 2382 (76.2%); 114 (3.6%); and 58 (1.8%) patients, respectively. Compared to patients who had no bleeding, patients with BARC ≥ 1 bleeding were more often female (30 vs. 23%), Caucasian (94 vs. 83%), had a higher incidence of drug eluting stent (DES) implantation (83 vs. 76%) and warfarin therapy (7.4 vs. 3.9%), and a lower incidence of diabetes (31 vs. 45%; P-value <0.01 for all comparisons). A 27-variable model had moderate discrimination (c-statistic of 0.674), and good calibration, as did a parsimonious model with 10 variables (c-statistic = 0.667). This model performed well in predicting BARC ≥ 2 bleeding events as well (c-statistic = 0.653). Conclusions: Bleeding is common in the first year after PCI, and can be predicted by pre-procedural patient characteristics and use of DES. Objective estimates of bleeding risk may help support shared decision-making with respect to stent selection and duration of antiplatelet therapy following PCI.