Methods:

The authors evaluated five bleeding risk prediction models: HEMORR₂HAGES, Shireman, HAS-BLED, ATRIA, and ORBIT. To validate these models, the authors used a subset of atrial fibrillation subjects (n = 3,623) from the RE-LY (Randomized Evaluation of Long-Term Anticoagulation Therapy) trial who had had a previous TIA or stroke and were randomized to either dabigatran or warfarin. During 6,922 person-years of follow-up, there were 266 major bleeding events. Model performance was assessed with C-statistics. (A C-statistic is equal to the area under a receiver operating characteristic [ROC] curve and ranges from 0 to 1.0. A C-statistic of 0.5 suggests that a model is no better than random chance at predicting risk. A C-statistic of ≥0.8 is generally considered a strong model.) In addition, the authors looked at the 1-year net benefit of warfarin for patients in the low, moderate, and high ORBIT bleeding risk categories by comparing the risk reduction in recurrent ischemic stroke to the increased risk of major bleeding.