LAUSR

BACKGROUND AND AIMS Early detection of atrial fibrillation after stroke is important for secondary prevention in stroke patients without known atrial fibrillation (AF). We aimed to compare the performance of CHADS2, CHA2DS2-VASc and HATCH scores in predicting AF detected after stroke (AFDAS) and to test whether adding stroke severity to the risk scores improves predictive performance. METHODS Adult patients with first ischemic stroke event but without a prior history of AF were retrieved from a nationwide population-based database. We compared C-statistics of CHADS2, CHA2DS2-VASc and HATCH scores for predicting the occurrence of AFDAS during stroke admission (cohort I) and during follow-up after hospital discharge (cohort II). The added value of stroke severity to prediction models was evaluated using C-statistics, net reclassification improvement, and integrated discrimination improvement. RESULTS Cohort I comprised 13,878 patients and cohort II comprised 12,567 patients. Among them, 806 (5.8%) and 657 (5.2%) were diagnosed with AF, respectively. The CHADS2 score had the lowest C-statistics (0.558 in cohort I and 0.597 in cohort II), whereas the CHA2DS2-VASc score had comparable C-statistics (0.603 and 0.644) to the HATCH score (0.612 and 0.653) in predicting AFDAS. Adding stroke severity to each of the three risk scores significantly increased the model performance. CONCLUSIONS In stroke patients without known AF, all three risk scores predicted AFDAS during admission and follow-up, but with suboptimal discrimination. Adding stroke severity improved their predictive abilities. These risk scores, when combined with stroke severity, may help prioritize patients for continuous cardiac monitoring in daily practice.