New AI Model Predicts Hospital Readmission Using Daily Data

Predicting whether a patient will be readmitted to the hospital within 30 days is crucial for patient care and resource management. Existing prediction models often simplify a patient's complex hospital stay into static representations, losing valuable day-to-day clinical changes that reflect their evolving health. This new research introduces Mr.Dec (Multimodal Readmission-risk prediction Decoder) to address this limitation. Mr.Dec treats each hospital admission as a natural, chronological sequence of daily multimodal events. It employs a Transformer Decoder to integrate daily updates from Electronic Health Records (EHR) and intermittent Chest X-ray (CXR) results, aligning with actual clinical workflows. To enhance its reliability, the model incorporates Disease-Specific Supervised Contrastive Learning, which helps structure the latent space with diagnosis-aware information. Evaluations on the MIMIC-IV and MIMIC-CXR datasets demonstrate that Mr.Dec achieves state-of-the-art performance by accurately preserving the integrity of the clinical sequence. Furthermore, the model can identify "Critical Days" during an admission, offering actionable and clinically interpretable insights for real-time risk stratification, which can guide interventions.