Explainable AI for Infection Prevention and Control: Modeling CPE Acquisition and Patient Outcomes in an Irish Hospital with Transformers

AI can help hospitals predict serious infection risks: The study used advanced machine learning (Transformer models) to predict which patients might acquire CPE infections and experience poor outcomes like readmission or extended hospital stays, achieving strong predictive performance (AUROC) that outperformed traditional methods.

Explainability is essential for clinical adoption: The researchers applied explainable AI (XAI) techniques to show why the model made specific predictions, identifying key risk factors like patient residence area, admission ward, and prior hospital exposure—making the AI's reasoning transparent enough for hospital staff to understand and trust.

Patient history and hospital networks matter more than expected: The analysis revealed that infection-related features (previous hospital exposure, admission context) and network measures (how connected a ward is) were highly influential in predicting CPE acquisition, suggesting that understanding patient movement patterns and ward connections is critical for infection prevention.

Modern AI approaches can handle complex medical data: Transformer-based models successfully processed multiple data types from electronic medical records—diagnostic codes, ward movements, demographics, and contact networks—demonstrating that sophisticated AI can integrate diverse clinical information to improve predictions.

Responsible AI requires both accuracy and interpretability: Rather than using only a "black box" model, the researchers prioritized explainability alongside performance, ensuring that hospital infection control teams could understand which factors drive predictions and act on that knowledge confidently.