Improving risk prediction for death, stroke and bleeding in Asian patients with atrial fibrillation

Abstract

Background: The objectives of this study were to compare the GARFIELD Refitted model and CHA2DS2-VASc/HAS-BLED risk scores with the new model from the COOL-AF registry for all-cause death, ischaemic stroke/systemic embolism (SSE) and major bleeding in Asian patients with atrial fibrillation (AF). Methods: Patients with non-valvular AF in the nationwide COOL-AF registry were studied. Patients were enrolled from 27 hospitals in Thailand during 2014–2017. Main outcomes were all-cause mortality, SSE and major bleeding. Predictive models of the three outcomes were developed from the variables in the multivariable Cox-proportional Hazard model. Predictive values of the models were evaluated by C-statistics, calibration plots and decision curve analysis (DCA). The new COOL-AF models were compared with the GARFIELD Refitted models and CHA2DS2-VASc model for all-cause mortality, SSE/HAS-BLED model for major bleeding. Results: A total of 3405 patients were enrolled. The C-statistics for the COOL-AF models were 0.727 (0.712–0.742), 0.708 (0.693–0.724) and 0.706 (0.690–0.721) for all-cause mortality, SSE and major bleeding, respectively. Calibration plots showed good agreement between predicted probability the observed outcomes for the COOL-AF models with a calibration slope of 0.94–0.99. The predictive ability remains preserved after the internal validation with bootstraps and optimism (bias) correction. The COOL-AF predictive models tended to be superior to the GARFIELD Refitted, CHA2DS2-VASc and HAS-BLED models. Conclusion: The COOL-AF predictive models for all-cause mortality, SSE and major bleeding in Asian patients with AF had a good predictive ability. The COOL-AF model for all-cause mortality was superior to the GARFIELD Refitted and CHA2DS2-VASc model.