Predicting Major Bleeding in Ischemic Stroke Patients With Atrial Fibrillation

BACKGROUND AND PURPOSE: Performance of risk scores for major bleeding in patients with atrial fibrillation and a previous transient ischemic attack or ischemic stroke is not well established. We aimed to validate risk scores for major bleeding in patients with atrial fibrillation treated with oral anticoagulants after cerebral ischemia and explore the net benefit of oral anticoagulants among bleeding risk categories. METHODS: HAGES (hepatic or renal disease, ethanol abuse, malignancy, older age, reduced platelet count or function, hypertension [uncontrolled], anemia, genetic factors, excessive fall risk, and stroke), Shireman, HAS-BLED (hypertension, abnormal renal/liver function, stroke, bleeding history or predisposition, labile international normalized ratio, elderly, drugs/alcohol concomitantly), ATRIA (Anticoagulation and Risk Factors in Atrial Fibrillation), and ORBIT scores (older age, reduced haemoglobin/haematocrit/history of anaemia, bleeding history, insufficient kidney function, and treatment with antiplatelet) with C statistics and calibration plots. Net benefit of oral anticoagulants was explored by comparing risk reduction in ischemic stroke with risk increase in major bleedings on warfarin. RESULTS: During 6922 person-years of follow-up, 266 patients experienced a major bleed (3.8 per 100 person-years). C statistics ranged from 0.62 (Shireman) to 0.67 (ATRIA). Calibration was poor for ATRIA and moderate for other models. The reduction in recurrent ischemic strokes on warfarin was larger than the increase in major bleeding risk, irrespective of bleeding risk category. CONCLUSIONS: Performance of prediction models for major bleeding in patients with cerebral ischemia and atrial fibrillation is modest but comparable with performance in patients with only atrial fibrillation. Bleeding risk scores cannot guide treatment decisions for oral anticoagulants but may still be useful to identify modifiable risk factors for bleeding. Clinical usefulness may be best for ORBIT, which is based on a limited number of easily obtainable variables and showed reasonable performance.