Oral anticoagulant therapy is associated with an increased risk of hemorrhage, which can be assessed by bleeding risk scores. We evaluated the performance of five validated scores for predicting major and clinically relevant non-major bleeding events in patients receiving warfarin.Methods and results
We conducted an ambispective, single-center cohort study of 321 consecutive patients enrolled in an academic anticoagulation clinic. The following scores were calculated: modified Outpatient Bleeding Risk Index, Contemporary Bleeding Risk Model, HEMORR2HAGES (Hepatic or Renal Disease, Ethanol Abuse, Malignancy, Older Age, Reduced Platelet Count or Function, Re-Bleeding, Hypertension, Anemia, Genetic Factors, Excessive Fall Risk and Stroke), ATRIA (Anticoagulation and Risk Factors in Atrial Fibrillation), and HAS-BLED (Hypertension, Abnormal Renal/Liver Function, Stroke, Bleeding History or Predisposition, Labile International Normalized Ratio, Elderly, Drugs/Alcohol). Main outcomes were major bleeding and a composite of major plus clinically relevant non-major bleeding. Incidence rates for all group were 3.8 (95% confidence interval [CI] 2.0–6.4) and 11.9 (95% CI 8.6–16.4) events per 100 patient-years for major bleeding and major plus clinically relevant non-major bleeding, respectively. Agreement among the five scores was low to moderate (Kendall's tau-b coefficients 0.22–0.54). For major bleeding, the c-statistics ranged from 0.606 to 0.735, whereas for major plus clinically relevant non-major bleeding, they ranged from 0.549 to 0.613. For all scores, the 95% CI for the c-statistics crossed 0.5 or was very close. Among high-risk patients, the hazard ratios for major bleeding ranged from 0.90 to 39.01, whereas for major plus clinically relevant non-major bleeding, they ranged from 1.52 to 8.71. For intermediate-risk patients, no score, except the Contemporary Bleeding Risk Model, produced statistically significant hazard ratios.Conclusion
The scores demonstrated poor agreement and low to moderate discriminatory ability. General clinical implementation of these scores cannot be recommended yet.