emsReACT: A Real-Time Interactive Cognitive Assistant for Cardiac Arrest Training in Emergency Medical Services

EMS (emergency medical services) deals with cardiac arrest cases more frequently than any other fatal health conditions all over the world. In this situation, a sequence of complex and time sensitive interventions is performed to ensure the safe recovery of the patient. We have developed emsReACT, a real-time interactive cognitive assistant, to train EMS providers for cardiac arrest cases in an emergency situation. This customized tool interacts in real-time with the first-responder and collects critical information. Using the conversational audio data available at EMS training sessions, emsReACT provides responder-specific decision support during the training based on domain specific information extraction, context-aware tracking of cardiac arrest protocols, and the dynamically changing condition of the patient. emsReACT leverages a dynamic behavioral model and a task-graph of two frequently used cardiac arrest EMS protocols. We have developed an intelligent abstraction mechanism with a critical risk-rating that drives an anytime algorithm to meet time requirements for regular and critical situations. Our thorough experimentation reveals an average end-to-end time of 2.7 seconds and 1.8 seconds for regular and critical interventions, thereby meeting the time requirements of 7 and 3 seconds, respectively. A qualitative study also reflects that over 70% of the 31 surveyed EMS providers rate the system as helpful to properly train the first-responders for executing cardiac arrest protocols.