Background: More accurate and reliable stroke risk prediction tools are needed to optimize anticoagulation decision making in patients with atrial fibrillation (AF). We developed a new AF stroke prediction model using the original Anticoagulation and Risk Factors in Atrial Fibrillation (ATRIA) AF cohort and externally validated the score in a separate, contemporary, community-based inception AF cohort, ATRIA-Cardiovascular Research Network (CVRN) cohort.
Methods and results: The derivation ATRIA cohort consisted of 10 927 patients with nonvalvular AF contributing 32 609 person-years off warfarin and 685 thromboembolic events (TEs). The external validation ATRIA-CVRN cohort included 25 306 AF patients contributing 26 263 person-years off warfarin and 496 TEs. Cox models identified 8 variables, age, prior stroke, female sex, diabetes mellitus, heart failure, hypertension, proteinuria, and eGFR<45 mL/min per 1.73 m(2) or end-stage renal disease, plus an age×prior stroke interaction term for the final model. Point scores were assigned proportional to model coefficients. The c-index in the ATRIA cohort was 0.73 (95% CI, 0.71 to 0.75), increasing to 0.76 (95% CI, 0.74 to 0.79) when only severe events were considered. In the ATRIA-CVRN, c-indexes were 0.70 (95% CI, 0.67 to 0.72) and 0.75 (95% CI, 0.72 to 0.78) for all events and severe events, respectively. The C-index was greater and net reclassification improvement positive comparing the ATRIA score with the CHADS2 or CHA2DS2-VASc scores.
Conclusions: The ATRIA stroke risk score performed better than existing risk scores, was validated successfully, and showed improvement in predicting severe events, which is of greatest concern. The ATRIA score should improve the antithrombotic decision for patients with AF and should provide a secure foundation for the addition of biomarkers in future prognostic models.
Keywords: anticoagulants; atrial fibrillation; risk score; stroke.