Development and validation of a machine learning model to predict prognostic outcomes in infantile epileptic spasms syndrome

Objective: To develop and validate a machine learning (ML) model for predicting seizure outcomes in infants with infantile epileptic spasms syndrome (IESS). Methods: This retrospective study enrolled pediatric patients diagnosed with infantile epileptic spasms syndrome (IESS) from Wuhan Children's Hospital. The cohort was randomly split into training and validation sets at a 7:3 ratio. Independent prognostic factors were identified using Cox regression analysis. Six machine learning algorithms were then applied to develop predictive models. Model performance was evaluated in terms of discrimination (e.g., AUROC), calibration (calibration curves), and clinical utility (decision curve analysis, DCA). The optimal model (XGBoost) was interpreted via decision tree visualization and SHAP analysis. Results: Poor seizure outcome was observed in 56% of the cohort. MRI findings of tuberous sclerosis complex or malformations of cortical development were independent risk factors. Among the models, XGBoost demonstrated the best overall performance, achieving an AUROC of 0.921 in the validation set, along with robust calibration and clinical utility. Conclusion: The developed ML model reliably and interpretably predicts poor seizure outcomes in IESS patients using routine clinical data, potentially aiding in clinical decision-making and follow-up planning.