Trends and inequities through real-time injury surveillance (TETRIS): novel use of physician-informed natural language processing in emergency department data for pediatric injury surveillance -…

Trends and inequities through real-time injury surveillance (TETRIS): novel use of physician-informed natural language processing in emergency department data for pediatric injury surveillance - although injuries are the leading cause of death and disability among U.S. Children, timely and accurate injury surveillance does not exist. The long-term goal is to decrease pediatric injuries by analyzing trends and changes over time of important injury mechanisms. The overall objectives in this application are to create a novel pediatric injury surveillance data network using natural language processing (NLP) in a multi-site emergency department (ED) electronic health record (EHR) data system, the pediatric emergency care applied research network (PECARN) registry. The central hypothesis is that a timely and accurate pediatric injury surveillance system can be created by applying an NLP graphical user interface (GUI) in ED EHR data as soon as IT is available (2 months after ED visit). The rationale for this project is that given current data gaps, creating a multi-site ED based injury surveillance system, as most critical trauma and critical illness are treated in EDS, is needed for timely identification of injury trends. The central hypothesis will be tested by pursuing three specific aims: 1) establish a near-real time pediatric ED injury surveillance system, which is more accurate and timely than current sources, using NLP in multi-site ED data; 2) determine near real-time trends of pediatric injuries using the PECARN registry; and 3) demonstrate feasibility of local pediatric injury surveillance with on-site application of NLP methodology to internal EHR data at 1 hospital, not using PECARN registry data. Under the first aim we will apply an NLP GUI to the PECARN registry to identify 3 specific injury mechanisms (suicide attempts, opioid overdose injuries, micromobility device/scooter injuries). We will compare test characteristics for identifying these injuries with NLP to only using diagnosis codes. For the second aim we will report trends and changes over time for these 3 injury mechanisms. For the third aim we will apply the NLP GUI only to internal EHR data at 1 hospital for local injury surveillance of the same 3 mechanisms with more timely data (1 week after ED visit). The research proposed in this application is innovative, in the applicant's opinion, because IT uses physician domain expertise on injuries applied to NLP methodology, which is brought to the extant data sources. The proposed research is significant because IT is expected to create a timely, accurate pediatric injury surveillance system to advance prevention of critical trauma and critical illness from injuries.