Low performance of bleeding risk models in the very elderly with atrial fibrillation using vitamin K antagonists

Essentials Under-treatment of oral anticoagulation in the elderly with atrial fibrillation is common. As bleeding prediction is challenging, we compared HAS-BLED, ATRIA and HEMORR2 HAGES. All three were associated with major bleeding in the elderly, but with poor predictive abilities. Future studies with focus on elderly-specific risk factors for bleeding are warranted.

Background Anticipated bleeding complications contribute to underuse of oral anticoagulants, especially in elderly patients with atrial fibrillation (AF). Bleeding risk models could provide guidance; however, these were developed in the general AF population. Objective To study and compare the performance of the HAS-BLED, ATRIA and HEMORR2 HAGES for major bleeding in very elderly AF patients. Methods Subjects were a random sample (N = 1157) of AF patients ≥ 80 years using a vitamin-K antagonist with prospective clinical follow-up from 2011 to 2014. The primary outcome was major bleeding (International Society on Thrombosis and Haemostasis criteria). Results Patients aged 84 years (median; 25th-75th 82-87) were classified as low risk by HAS-BLED (25.2%), ATRIA (59.6%) and HEMORR2 HAGES (23.3%). Three-year rates of major, clinically relevant and any bleeding were 6.7%, 28.3% and 42.3%, respectively. We observed a statistically significant association for all models with major bleeding, but discriminatory abilities were rather poor (C-statistics < 0.60) without clear superiority for any of the three. Only two (anemia and antiplatelet therapy) of the various classical risk factors were associated with bleeding. An estimated risk-benefit profile indicated a favorable trade-off for oral anticoagulation in this specific cohort (number needed to treat, 22; number needed to harm, 91). Conclusions In this large prospective cohort of very elderly AF patients, the currently used bleeding risk scores were all associated with major bleeding, but with poor predictive abilities. Use of the ATRIA model may inadvertently result in less attention being paid to modifiable risk factors in this particular population. In light of the issues of under-treatment and the suggested favorable risk-benefit profile, future models with incorporation of elderly-specific risk factors may provide more guidance in this growing population of AF patients.