A comparative analysis of HAS-BLED, ORBIT, DOAC and AF-BLEED bleeding risk scores in anticoagulated patients with atrial fibrillation: A report from the prospective Murcia AF Project III (MAFP-III) cohort.

Introduction: Atrial fibrillation (AF) patients often require oral anticoagulation (OAC), but predicting bleeding risk remains challenging. Classic risk scores like HAS-BLED and ORBIT, along with newer models such as DOAC and AF-BLEED, exhibit only modest predictive performance. Herein, we compared these four scores in AF patients on OAC to evaluate their predictive accuracy and clinical utility.

Methods: We included AF patients starting OAC between January 2016 and November 2021. Endpoints assessed were major bleeding, major bleeding/clinically relevant non-major bleeding (CRNMB), and intracranial hemorrhage (ICH). Score performance was assessed through discrimination, calibration, reclassification (NRI), decision curve analysis (DCA), and accuracy over a two-year follow-up period.

Results: 3,259 AF patients were included (median age 77 [IQR 70-83] years, 52.8% female). Over 2-years follow-up, 196 patients (6.35%/year) experienced major bleeding, 413 (10.3%/year) major bleeding/CRNMB, and 32 (0.5%/year) suffered ICH. All risk scores performed modestly for major bleeding (c-indexes <0.7) with ORBIT (c-index 0.664) and HAS-BLED (c-index 0.651) performing best. All four scores demonstrated good discriminatory ability (log-rank p<0.001). The ORBIT score showed a slight improvement in NRI for major bleeding compared to HAS-BLED, with a modest increase in discrimination, while no other score outperformed HAS-BLED. Calibration showed that HAS-BLED and ORBIT outperformed the DOAC and AF-BLEED scores for major bleeding and CRNMB. The Brier index indicated high accuracy across all scores.

Conclusion: All bleeding risk scores only had modest predictive ability. None demonstrated clear superiority in predicting major bleeding, major bleeding/CRNMB, or ICH.