Comparative validation of HAS-BLED, GARFIELD-AF and ORBIT bleeding risk scores in Asian people with atrial fibrillation treated with oral anticoagulant: A report from the COOL-AF registry

Abstract

Aims: Comparative data between the HAS-BLED, GARFIELD-AF and ORBIT score are limited in anticoagulated Asian patients with atrial fibrillation (AF). We compared the performance of the 3 scores in a nationwide registry. Methods: AF patients treated with oral anticoagulants in the COOL-AF registry were studied. We fitted the variables of the HAS-BLED, GARFIELD-AF and ORBIT score to major bleeding in Cox model. We explored a modified HAS-BLED by addition of sex and body weight. Discrimination, calibration, net reclassification index (NRI) and decision curve analysis were used to compare the performance of the 3 models. Results: Of 3402 patients in the registry, 2568 patients who received oral anticoagulant at baseline were studied. Majority of patients (91.1%) received warfarin. The rate of major bleeding was 2.11 per 100 person-years. The C-statistics of the GARFIELD-AF, HAS-BLED, modified HAS-BLED and ORBIT score were 0.65 (95% confidence interval [CI] 0.63–0.67), 0.66 (95%CI 0.64–0.68), 0.69 (95%CI 0.67–0.71) and 0.64 (95%CI 0.62–0.66) respectively. There was good agreement between predicted and observed bleeding in the deciles of HAS-BLED and GARFIELD-AF scores, while the modified HAS-BLED score and ORBIT score overestimated the risk in the last decile. The modified HAS-BLED score had superior NRI than the HAS-BLED score (26.9%, 95%CI 9.7%–42.2%) and the ORBIT score (31.9%, 95%CI 9.0–53.6%). The NRI between the modified HAS-BLED and GARFIELD-AF score was similar. The net benefit curve of the 4 models were overlapping among different thresholds. Conclusions: The clinical utility for bleeding prediction of GARFIELD-AF, HAS-BLED, modified HAS-BLED and ORBIT scores were similar in anticoagulated Asian patients with AF participating in the COOL-AF registry. We found no advantage of the ORBIT over HAS-BLED score for bleeding risk prediction, even in direct oral anticoagulant users.