Using a Trauma Knowledge Graph to Develop an Educational Tool for Identifying Injury Associations

Abstract

Network analysis offers a very powerful way of visualizing and deriving relationship information from data. Network graphs allow unique probability computation by identifying the multiple relationships that exist, like a web, between data. We created a network graph of all trauma injuries registered in the National Trauma Data Bank (NTDB) in 2016 to help visualize the predicted association between any given injury and another potential injury. The study is a retrospective computational network analysis of all trauma injuries listed in the NTDB in 2016. Nodes represented unique ICD-10-CM codes mapped to their description while edges—the connections between nodes—represented the rate of co-occurrence during an injury encounter. Bayesian posterior probabilities were determined across the entire trauma graph and probabilities for injury sequelae were generated with likelihood ratios. Mechanism of injury was simplified to categorical descriptors and included as a stratification element. R (version 3.5.1) with the iGraph package was used for all analyses. Shiny was used for web development and deployment. A total number of 3,400,946 injury entries from 968,665 encounters from the NTDB were digested to create a network graph composed of 177 unique injury classifications that served as nodes. An online tool was created to display a visual representation of the network subgraph along with probable injury sequelae and likelihood ratios. In summary, we created an educational tool that allows users to identify constellations of injury associations from large scale registry data. As an educational tool, our visualization and model are an innovative platform to teach and recognize injury patterns. It can serve as a tool for triage or protocol development. Tertiary trauma care centers often receive patients with a limited number of diagnosed injuries. This tool could educate surgeon and non-surgeon physicians to identify undiagnosed injury thereby aiding in triage management.