Abstracts

Rationale: Stroke is the most common cause of acquired epilepsy in adults. The latent period between the initial infarct and late (> 7 days) seizures offers time for diagnostic and therapeutic procedures to prevent epileptogenesis. An instrument to predict those who will develop seizures and would benefit from these procedures is not available. We developed and externally validated a prognostic model of late seizures after ischemic stroke. Methods: Using backward elimination of a multivariate logistic regression model, we derived a score (SeLECT) in 1200 ischemic stroke patients from the Stroke and Epilepsy Registry of Eastern Switzerland (SERES) to predict late seizures (> 7 days after insult) as primary outcome measure. The secondary outcome were recurrent late seizures. We externally validated this score in 1427 patients from three independent international cohorts (Austria, Germany, Italy) and assessed its performance with the concordance (c) statistic, calibration plots, and the Hosmer-Lemeshow calibration test. Results: Overall, 138 of 2627 (5%) patients suffered late seizures and they were recurrent in 90 (3%). The lowest SeLECT value (0 points) was associated with a 5-fold decrease of late seizure risk, whereas the highest value (10 points) predicted a 12-fold increase of relative risk (10-fold decrease vs. 10-fold increase for recurrent seizures). The model had an overall c statistic of 0.76 in validation cohorts. Calibration plots and nonsignificant Hosmer-Lemeshow tests (p > 0.3) indicated good fit of predicted and observed outcomes. Conclusions: This easily applicable instrument was a good predictor of late seizures after stroke in triple external validation. The SeLECT score has the potential to individualise clinical care by identifying those who would benefit from optimal management and appropriate follow-up. It can inform the selection of an enriched population for antiepileptogenic treatment trials and guide the recruitment for biomarker studies of epileptogenesis. Funding: None.