Multimodal artificial intelligence using entire electronic health record and complete pathogen genome data for patient outcome prediction from life-threatening infection: the SuperbugAI Platform

A research team developed a multimodal artificial intelligence system combining electronic health record data and bacterial genome data to predict patient outcomes from life-threatening infections. The system was tested on 2,656 hospitalizations with bloodstream infections involving 2,535 patients. The deep learning model achieved 0.93 accuracy in predicting in-hospital mortality, significantly outperforming the traditional APACHE II score (0.77). The model also demonstrated strong performance in predicting the need for ICU admission (0.978), prolonged hospital stay (0.803), and unplanned readmission (0.696). Incorporating bacterial genomic features further improved prediction accuracy and enabled identification of key bacterial virulence pathways relevant to human disease. This approach is scalable to other medical specialties.