Development and International Validation of a Novel Imaging-Based Risk Score (IsCHEMiA) for the Prediction of Poststroke Epilepsy.

Abstract

Stroke is one of the most common causes of adult-onset epilepsy. We aimed to develop a model to predict poststroke epilepsy (PSE) after a first-ever ischemic stroke, incorporating neuroimaging features of incident stroke. We analyzed clinical and neuroimaging features of patients with first-ever acute ischemic stroke consecutively admitted to Massachusetts General Hospital, United States. We performed competing risk regression with all-cause mortality as a competing event and derived the final multivariable model using backward stepwise elimination by the Akaike Information Criterion. We externally validated the model in 3 international cohorts in Hong Kong (Queen Mary Hospital [HK-QMH], Ruttonjee Hospital [HK-RH]) and Japan (National Cerebral and Cardiovascular Center) by discrimination and calibration and compared its performance with the SeLECT and SeLECT2.0 scores. We included a final derivative cohort of 1,436 patients with a mean age of 67.4 years and a slight male predominance (54.7%), along with a total of 2,534 patients in the validation cohorts. PSE, defined as the occurrence or recurrence of unprovoked seizure >7 days after stroke, occurred in 5.5% of the overall study population. Six variables (infarct size [Is], cortical involvement [C], hemorrhagic transformation [H], early seizures [E], MCA involvement [Mi], and age younger than 65 [A]) were independent predictors included in the final model and formed the IsCHEMiA score. Model discrimination was consistent across all cohorts, with <i>c</i>-statistics of 0.870 (United States), 0.852 (HK-QMH), 0.857 (HK-RH), and 0.826 (Japan). The model was well calibrated at 1 and 3 years after stroke in the overall validation cohort. The IsCHEMiA score improved the prediction of PSE compared with SeLECT in all cohorts and the overall study population (<i>c</i>-statistic 0.848 vs 0.782, <i>z</i> = 5.170, <i>p</i> < 0.0001). For example, an IsCHEMiA score of 3 predicts a low risk of PSE at 1 year (2%) and 5 years (6%) while an IsCHEMiA score ≥8 predicts a high risk at 1 year (67%) and 5 years (78%). The IsCHEMiA score is an improved and readily applicable predictive model developed and validated using international stroke cohorts in the modern era of reperfusion therapies. It serves as a foundation for personalized management and may guide future clinical trials on antiepileptogenic therapies in acute ischemic stroke.