A data-driven multivariate bleeding risk model incorporating modifiable and non-modifiable bleeding risk factors: the GLORIA-AF bleeding risk score
Y Xie, A Askarinejad, T Gu, W Li, G Boriani, T Chao, M V Huisman, Y Zheng, Y Meng, T Huang, G Y H LipAbstract
Background
Major bleeding remains a serious complication in patients with atrial fibrillation (AF) receiving oral anticoagulation (OAC), making accurate risk prediction essential to guide timely interventions and optimize treatment. Bleeding risk reflects the interplay of modifiable and non-modifiable factors, as captured by the simple HAS-BLED bleeding risk score. As with most scores based on clinical factors alone, HAS-BLED has limited predictive performance and may not generalize well across diverse patient populations.
Purpose
To identify key demographic variables, alongside comorbidities and treatments associated with the outcome of major bleeding among patients with AF. Second, to develop and externally validate a multivariable risk model to predict major bleeding, and to compare its performance with the HAS-BLED score.
Methods
Using data from the global, observational and prospective GLORIA-AF registry, a total of 18 variables, including demographic factors, clinical characteristics and antithrombotic treatment were examined as potential predictors of major bleeding outcome. A multivariable bleeding risk model was developed to predict the risk of major bleeding, and its predictive performance was evaluated and compared with the HAS-BLED score using the area under the receiver operating characteristic curve (AUC), sensitivity and calibration metrics. External validation was performed on the combined datasets of 2 large European and Asia-Pacific AF registries (EORP-AF and APHRS-AF registries).
Results
Among 20,478 eligible patients recruited in Phase II/III of the GLORIA-AF registry (mean age 69.9 ± 10.3 years; 45.1% female), major bleeding occurred in 679 patients (3.3%). Patients aged >65 years (odds ratio [OR]: 2.17, 95% confidence interval [CI]: 1.76-2.71), females (OR: 1.22, 95% CI: 1.03-1.44), white ethnicity (OR: 1.52, 95% CI: 1.21-1.92), smoking (OR: 1.47, 95% CI: 1.24-1.74), hyperlipidaemia (OR: 1.19, 95% CI: 1.01-1.41), diabetes (OR: 1.21, 95% CI: 1.01-1.45), prior bleeding history (OR: 1.67, 95% CI: 1.27-2.16), respiratory disease (OR: 1.33, 95% CI: 1.07-1.63), and abnormal kidney function (OR: 2.90, 95% CI: 1.95-4.18) were associated with increased risks of major bleeding. In the derivation cohort, the GLORIA-AF bleeding risk model exhibited superior discriminatory performance to the HAS-BLED score (AUC: 0.66 (95% CI: 0.64-0.68) vs. 0.53 (95% CI: 0.52-0.54); sensitivity: 0.66 vs. 0.15, respectively). The difference in ROC curves was significant on DeLong’s test (p < 0.001). On external validation in the EORP-APHRS dataset (n = 10084, mean age 69.0 ± 11.6, 38.2% female), the GLORIA-AF bleeding risk model achieved AUC of 0.62 (95% CI 0.58-0.67).
Conclusions
Several key clinical and demographic factors are associated with major bleeding in patients with AF which allows development of a novel, data-driven multivariate risk model with superior predictive performance to the simple HAS-BLED score.Forest Plot and ROC Comparison