Federated Learning for ICU Mortality Prediction: Balancing Accuracy and Privacy in a Multi-Hospital Setting

This study by the NStarX team demonstrates how multiple hospitals can build better AI models to predict ICU patient deaths without sharing any patient data. Using federated learning, each hospital trains AI models on their own data, then only shares model updates (not patient information) to create a superior combined model. The key innovation is proving this privacy-preserving approach works as well as traditional methods that require pooling sensitive medical data. Testing with 5,000 ICU patients across three simulated hospitals, their federated model significantly outperformed individual hospital models while maintaining strict data privacy. This enables secure collaboration between healthcare institutions for better patient care.