Prediction of late seizures after ischaemic stroke with a novel prognostic model (the SeLECT score): a multivariable prediction model development and validation study.

ABSTRACT

BACKGROUND: Stroke is one of the leading causes of acquired epilepsy in adults. An instrument to predict whether people are at high risk of developing post-stroke seizures is not available. We aimed to develop and validate a prognostic model of late (>7 days) seizures after ischaemic stroke. METHODS: In this multivariable prediction model development and validation study, we developed the SeLECT score based on five clinical predictors in 1200 participants who had an ischaemic stroke in Switzerland using backward elimination of a multivariable Cox proportional hazards model. We externally validated this score in 1169 participants from three independent international cohorts in Austria, Germany, and Italy, and assessed its performance with the concordance statistic and calibration plots. FINDINGS: Data were complete for 99·2% of the predictors (99·2% for Switzerland, 100% for Austria, 97% for Germany, and 99·7% for Italy) and 100% of the outcome parameters. Overall, the risk of late seizures was 4% (95% CI 4-5) 1 year after stroke and 8% (6-9) 5 years after stroke. The final model included five variables and was named SeLECT on the basis of the first letters of the included parameters (severity of stroke, large-artery atherosclerotic aetiology, early seizures, cortical involvement, and territory of middle cerebral artery involvement). The lowest SeLECT value (0 points) was associated with a 0·7% (95% CI 0·4-1·0) risk of late seizures within 1 year after stroke (1·3% [95% CI 0·7-1·8] within 5 years), whereas the highest value (9 points) predicted a 63% (42-77) risk of late seizures within 1 year (83% [62-93] within 5 years). The model had an overall concordance statistic of 0·77 (95% CI 0·71-0·82) in the validation cohorts. Calibration plots indicated high agreement of predicted and observed outcomes. INTERPRETATION: This easily applied instrument was shown to be a good predictor of the risk of late seizures after stroke in three external validation cohorts and is freely available as a smartphone app. The SeLECT score has the potential to identify individuals at high risk of seizures and is a step towards more personalised medicine. It can inform the selection of an enriched population for antiepileptogenic treatment trials and will guide the recruitment for biomarker studies of epileptogenesis. FUNDING: None.