OBJECTIVES: Hospitals are increasingly focused on reducing patient harm associated with anticoagulant therapy.  However, because treatment decisions may be based on prognosis, estimates of treatment effects obtained from observational data may suffer from "confounding by indication." To address this concern, we used a grouped-treatment approach to determine the impact of choice of anticoagulant on the risk of severe bleeding and in-hospital death in patients undergoing percutaneous coronary intervention (PCI). METHODS: We analyzed the Premier Perspective database on patients aged ≥18 years admitted to Premier hospitals with a diagnosis of ST-elevated myocardial infarction (STEMI) and ≥1 procedure code for PCI between Q12004 and Q12008 (N=71,296). We constructed individual-level models of severe bleeding and in-hospital death (all-cause) similar to those in a conventional multivariate analysis, except that each individual's actual treatment variables were replaced with grouped-treatment variables (the proportion of patients receiving each treatment at the hospital/year in which treatment occurred). We used logistic regression to assess the impact of the likelihood of treatment with bivalirudin or heparin ± a glycoprotein IIb/IIIa inhibitor on severe bleeding and in-hospital death, controlling for other treatments (including stent use, other drug use,  CABG); patient demographics, concomitant diagnoses, insurance status, physician specialty; and hospital region, size, teaching status. We calculated confidence intervals allowing for the clustering of errors at the hospital/year level. RESULTS: Bivalirudin treatment was associated with a significantly reduced risk of severe bleeding (OR=0.45, 95% CI=0.21-0.97) and a reduced risk of in-hospital mortality (OR=0.83, 95% CI 0.54-1.28) compared to heparin+GPI. CONCLUSIONS: Increasing the proportion of a hospital's patients treated with bivalirudin was associated with a significant reduction in severe bleeding. These results demonstrate the benefits of bivalirudin for clinical outcomes in a real-world setting.