MedTime - Early Medical Emergency Prediction

The prediction of medical emergencies is critical to ensuring timely interventions and improving patient outcomes in Intensive Care Units (ICUs). This paper introduces MedTime, an advanced medical emergency prediction model that uses streaming data of vital signs—such as ECG, blood pressure, oxygen saturation, respiratory rate, and temperature—to predict possible deteriorations in a patient’s health condition. The model not only identifies imminent emergencies but also projects the possible states of these emergencies in the subsequent time step, enabling proactive decision-making. Our approach is based on real-time processing of ICU patient data using the MIMIC-III dataset and leverages machine learning models capable of handling streaming data with high accuracy. A system is designed to alert medical staff through an intuitive user interface, facilitating rapid response. MedTime was validated on large-scale ICU datasets, demonstrating its effectiveness in predicting emergencies and forecasting their potential progression.