Abstract Body: Background
Previous risk assessment tools for patients with atrial fibrillation (AF) have not primarily focused on those with acute ischemic stroke (AIS). Prediction models that rely solely on clinical parameters have shown limited predictive ability, underscoring the need for models that incorporate neuroimaging parameters to improve the prediction of stroke recurrence in AIS patients with AF.

Method
The East-Asian Ischemic Stroke Patients with Atrial Fibrillation (EAST-AF) study was a nationwide, multicenter prospective registry that enrolled AIS patients with AF from 15 stroke centers. Twenty-six potential variables were selected based on previous literature regarding stroke recurrence and AF. A subdistribution model, with non-stroke death as a competing risk, was employed. Variables for the primary model were chosen based on the Akaike Information Criterion (AIC), with ΔAIC < 0 indicating model simplification. Final variables were selected from the primary model based on statistical significance (P-value) and included in the secondary model. Scores were assigned to each variable using the beta coefficients divided by 0.1. The scoring model was validated through bootstrapping and compared to CHADS2, CHA2DS2-VASc, and ATRIA scores.

Result
Among 15,936 acute ischemic stroke patients from 2017 to 2020, 2,577 consented to participate. Of these, 2,165 patients who had lesion-positive imaging, underwent MRI and angiography, and survived at discharge were analyzed. The mean age was 75.0 (±9.99) years, and 78.8% were prescribed direct oral anticoagulants. The cumulative incidence of recurrent stroke was 4.83% (3.95-5.84%) at 1 year and 9.27% (7.90-10.77%) at 3 years. Ten variables with ΔAIC < 0 were chosen for the primary model. From these, four variables—concomitant large artery steno-occlusion (cLASO), acute infarction multiplicity, chronic embolic infarction, and pre-stroke anticoagulation—were included in the secondary model (Image 1), with 5, 4, 5, and 5 points assigned, respectively. The area under the curve of the scoring system was 67.5% at 1 year and 62.6% at 3 years (Image 2), with similar results in internal validation. The new model demonstrated better predictability than conventional scores (Image 3).

Conclusion
A novel risk stratification model that incorporates clinical and neuroimaging parameters showed improved predictability for secondary prevention in AF patients compared to conventional scoring systems.