Utilizing Predictive Analysis to Aid Emergency Medical Services

Abstract

The study investigates the use of machine learning algorithms to predict patient outcomes in the Emergency Department. While triage systems help prioritize patients by rapidly evaluating patients’ acceptable waiting times, they cannot be directly employed for predicting patient clinical outcomes. Predicting clinical outcomes in resource constraint settings can be insightful and help make optimum resource management decisions, supporting overall better emergency medicine practices. Thus, we explored several machine learning techniques on a well-known dataset to classify patients into possible emergency outcomes for predicting the patient’s outcome. In contrast to the previous works, we include both the structured and free text variables to obtain the best Area Under the Receiver Operating Characteristic (AUC-ROC) of 0.76 with the Light Gradient Boost Machine model. We further demonstrate that using such models can help to better identify very critical patients in a clinical setting. To elaborate on these models’ usability in practical settings, we describe the user interface design of a prototype developed for emergency settings. Finally, we discuss the scope and limitations of the machine learning models used to predict clinical outcomes.