UC San Diego

Undifferentiated chest pain is a major driver of emergency department (ED) overcrowding and contributes to morbidity and mortality. Early risk stratification for major adverse cardiac events (MACE) allows for early discharge and ED decongestion. We developed a multimodal machine learning model integrating EHR data, laboratory results, demographic and comorbidity features and chest x-ray impressions to predict 30-day MACE in patients presenting to the ED with chest pain. Using 49,348 ED encounters from two healthcare systems, our model achieved a high area under the curve (AUC) of 82.0% at the development site and 74.6% at the external validation site. Predictions occurred within 6 h of ED arrival and demonstrated consistently high negative predictive values across time, supporting safe early-rule out. This model demonstrates the potential of integrating AI modeling to enhance early MACE risk stratification and improve ED overcrowding and efficiency.