Information Extraction from Patient Care Reports for Intelligent Emergency Medical Services

Every year, more than 30 million emergency medical incidents are responded to in the U.S. Upon arrival at an incident scene, responders assess the situation and provide emergency medical care to the patients before transporting them to hospitals. In this process, the responders collect substantial amounts of data with different levels of importance and confidence, including the patient's present medical conditions, past medical history, and interventions performed. Although there are several standards and tools for collecting, storing, and sharing EMS data, less attention has been given to reliably translating this wealth of information into actionable knowledge for assessing the performance of emergency operations and evaluating response protocols. This paper presents the analysis of over 35,900 EMS pre-hospital electronic Patient Care Reports (ePCR) from an urban ambulance agency. We used both the structured and unstructured information in the dataset to develop a domain-specific EMS ontology with a standardized lexicon for medications, procedures, responders’ impressions, call types, chief complaints, and signs and symptoms. The EMS ontology was used to develop methods for automated segmentation of narratives, detection and correction of incorrect/incomplete information in the reports, and generation of time-series data to represent the progression of incidents and the most common sequences of response actions (models of EMS protocols). Finally, we performed an analysis on the relationships among different aspects of incidents to provide insights for the design of future EMS assistive technologies.