Comparison of the ATRIA, CHADS2, and CHA2DS2-VASc stroke risk scores in predicting ischaemic stroke in a large Swedish cohort of patients with atrial fibrillation

Aims Better stroke risk prediction is needed to optimize the anticoagulation decision in atrial fibrillation (AF). The ATRIA stroke risk score (ATRIA) was developed and validated in two large California community AF cohorts. We compared the performance of the ATRIA, CHADS(2), and CHA(2)DS(2)-VASc scores in a national Swedish AF (SAF) cohort. Methods and results We examined all Swedish patients hospitalized, or visiting a hospital-based outpatient clinic, with a diagnosis of AF from July 2005 through December 2010. Variables were determined from comprehensive national databases. Risk scores were assessed via C-index (C) and net reclassification improvement (NRI). The cohort included 152 153 AF patients not receiving warfarin. Overall, 11 053 acute ischaemic strokes were observed with mean rate 3.2%/year, higher than the 2%/year in the California cohorts. Using entire point scores, ATRIA had a good C of 0.708 (0.704-0.713), significantly better than CHADS2 0.690 (0.685-0.695) or CHA(2)DS(2)-VASc 0.694 (0.690-0.700). Using published cut-points for low/moderate/high risk, C deteriorated but ATRIA remained superior. Net reclassification improvement favoured ATRIA 0.16 (0.14-0.17) vs. CHADS(2) and 0.21 (0.20-0.23) vs. CHA(2)DS(2)-VASc. Net reclassification improvement decreased when cut-points were altered to better fit the cohort's stroke rates. Conclusion In this SAF cohort, the ATRIA score predicted ischaemic stroke risk better than CHADS(2) or CHA(2)DS(2)-VASc. However, relative performance of the categorical scores varied by population stroke rates. Score cut-points may need to be optimized to better fit local population stroke rates.