MS Disability Progression Prediction

This study presents the first systematic evaluation of personalized federated learning (PFL) for predicting 2-year disability progression in multiple sclerosis using multi-center real-world data from over 26,000 patients extracted from the international MSBase neuro-immunology registry. Unlike standard federated learning, which struggles with institutional data heterogeneity, PFL adapts shared models to local data distributions while preserving privacy and significantly improves predictive performance. Personalized methods, particularly FedProx and FedAvg, achieved strong results with ROC-AUC scores around 0.84, underscoring the importance of personalization for scalable, privacy-aware clinical prediction.