We developed a Deep Heterogeneous Multimodal Learning-Based Technique for Predicting Hospital Readmission and Mortality Risk on heart failure patients. This technique integrates heterogeneous modalities, including clinical data, electrocardiograms, and chest X-rays, to predict short-term and long-term mortality and readmission risks. Our technique, recognized by the 20th National Innovation Award, is under patent application, multi-hospital validation, and planning for SaMD certification.

Our Multimodal Learning-Based Techniques for Predicting Hospital Readmission and Mortality Risk in Patients with Heart Failure extract features from diverse clinical modalities, integrate them via heterogeneous feature fusion technologies and weighting strategies, and construct a stacked ensemble classifier to predict short- and long-term readmission and mortality risks. Our approach outperforms traditional methods, achieving the highest AUCs above 0.8 for mortality and readmission prediction.

With the global population aging, heart failure has become a leading cause of hospitalization, affecting over 20% of people aged 65+, highlighting the need for accurate risk tools. Our technique integrates heterogeneous data to predict readmission and mortality risks at 1, 3, 12, and 36 months. It supports personalized care and clinical decisions and can connect with health databases, wearables, and PHRs via secure cloud infrastructure across hospitals, outpatient, and telemedicine platforms.