Head-to-head comparison of CHADS2, CHA2DS2-VASc, CHA2DS2-VA, HAS-BLED, and HELT-E2S2 scores in edoxaban treated atrial fibrillation patients: a 2-year analysis from the ETNA-AF Global program

Introduction Risk stratification for thromboembolism and haemorrhage in atrial fibrillation (AF) remains a clinical priority. Although CHA2DS2-VASc is widely used, CHA2DS2-VA is recommended in European guidelines, whereas the CHADS2 score remains common in Japan. HAS-BLED and the Japan-specific HELT-E2S2 scores are also used for bleeding and stroke assessment. The relative prognostic utility of these risk scores in patients treated specifically with edoxaban remains unclear. Purpose This study compared the predictive value and clinical utility of multiple risk scores (CHADS2, CHA2DS2-VASc, CHA2DS2-VA, HELT-E2S2, and modified HAS-BLED) for effectiveness and safety outcomes in AF patients enrolled in the Global ETNA-AF program. Methods This post hoc analysis utilized data from the Global ETNA-AF registry, a multinational, prospective observational study of edoxaban users across Europe and Asia (N = 26,805). Primary effectiveness outcomes included stroke/systemic embolic events (SEE) and all-cause mortality; the primary safety outcome was major bleeding. Predictive performance was evaluated using two-year annualized event rates and the concordance index (C-index) derived from Cox regression models. Results Most patients were classified as "moderate risk" by effectiveness scores, while 61.2% were "high risk" per the modified HAS-BLED (Figure 1A). Two-year annualized event rates for all-cause mortality, stroke/SEE, and major bleeding increased progressively across higher risk categories for all scores (Figure 1B–D). Globally, for any stroke/SEE, the CHADS2, CHA2DS2-VASc, CHA2DS2-VA, and HELT-E2S2 scores demonstrated moderate discriminatory performance (C-index: 0.64–0.66). For all-cause mortality, the modified HAS-BLED score had a lower C- index (0.59) than other risk scores (0.63–0.65). For major bleeding, modified HAS-BLED performed similarly to other risk scores (0.61–0.63). Globally, differences in predictive performance were not observed between CHA2DS2-VASc, CHADS2, or CHA2DS2-VA for predicting any stroke/SEE (Table 1). In contrast, CHA2DS2-VA showed small but statistically significant improvements in predicting all-cause mortality and major bleeding compared with CHA2DS2-VASc and CHADS2 (C- index differences 0.01–0.02; p < 0.001), with regional variation. The modified HAS-BLED score showed significantly lower discrimination than CHA2DS2-VA for all-cause mortality (C- index difference 0.06; p < 0.0001), while showing no significant differences for stroke/SEE or major bleeding. HELT-E2S2 and CHA2DS2-VA demonstrated similar predictive performance across all outcomes. These findings were directionally consistent across countries. Conclusions HELT-E2S2 and CHA2DS2-VA scores provided the best predictive performance for all-cause mortality, any stroke/SEE, and major bleeding. These data provide real-world evidence that help contextualize the application of risk stratification tools in diverse clinical settings.