Stroke risk scores for prediction of mortality and hemorrhages in atrial fibrillation patients.

Atrial fibrillation (AF) is an emerging epidemic worldwide, responsible for a twofold increase in mortality, independent of other risk factors. Stroke prevention is the cornerstone of AF management. However, oral anticoagulation imposes an increased risk of bleeding. Several risk scores have been developed for estimating both the thromboembolic and the bleeding risks. The aim of the study was to determine the usefulness of different stroke risk scores as predictors of mortality and hemorrhagic events in AF patients. We retrospectively enrolled 211 AF patients hospitalized in the Cardiology Ward of our tertiary hospital. The primary endpoints were mortality and non-minor bleeding events. The mean follow-up period was 378 days for bleeding events and 5 years and 1 month for mortality. For each patient, we evaluated the following stroke risk scores: CHADS2, CHA2DS2-VASc, R2CHADS2, ABC, ATRIA, GARFIELD. The mean age in our cohort is 66, with a slight predominance of women (52.2%). For a CHA2DS2-VASc ≥ 4 as well as for a score of 2-3, 5-year survival was worse than for patients with a score of 0-1(chi-squared=8.13; p=0.01). Similarly, all subgroups of patients with an ABC <2%, had a worse 5-year survival when compared with an ABC score of ≥2% (chi-squared=12.85; p=0.005). C-statistics show a modest predictive value for mortality, for all stroke scores except Garfield, with similar AUCs, the highest being for CHA2DS2-VASc (AUC 0.656; p=0.0001). CHA2DS2-VASc also correlates with bleeding events, having a good predictive ability (AUC 0.723; 95%CI 0.658-0.782, p=0.001), mildly superior to HAS-BLED (AUC 0.674; 95% CI 0.523-0.825; p = 0.04) and very close to Garfield-bleeding (0.765; 95%CI 0.702-0.80; p=0.0001). CHA2DS2-VASc is comparable to HAS-BLED and Garfield-bleeding in predicting bleeding events in AF patients. CHA2DS2-VASc and ABC correlate directly and consistently with mortality rate. For CHA2DS2-VASc, the AUCs for our endpoints are similar to the ones for stroke prediction, highlighting the potential of extending its applicability to various outcomes.