Machine Learning to Predict the Severity Score of Trauma Pediatric Patients

Background Many scoring systems currently exist to predict severity and mortality among trauma patients. However, controversy remains over the validity of these scoring systems in pediatrics and their use in guiding treatment. Additionally, many of these tools utilize variables that can only be collected days after admission, prohibiting predictions from being determined in the immediate hours following the trauma. The goal of this study is to develop an appropriate scoring system utilizing variables collected in the pre-hospital setting and on admission of a trauma pediatric patient that could be used to help determine the minimal necessary level of care for survival. Methods A retrospective chart review was performed for 1277 pediatric patients presenting to an inner-city Level 1 trauma hospital via emergency medical services (EMS) status post motor vehicle collision. Many variables were collected from the EMS documentation, including the restraints in use, position in vehicle, location of the accident, and pre-hospital Glasgow coma scale (GCS). Additionally, variables were collected from the medical charts including admission vital signs, mechanism and speed of the collision, as well as injuries sustained and procedures performed in the operating room (OR). This data was then provided to the Human-Interactive Robotics (HIRo) Lab at Purdue University to analyze with their machine learning technology. Results and Conclusion Data is currently being analyzed by the HIRo machine learning lab at Purdue University. Analysis and conclusions will be completed in time to be presented at the symposium in March.