Using Machine Learning to Predict Patients Who Leave Without Being Seen in a Pediatric Emergency Department

Abstract

Background: Patients and their caregivers who seek care in an Emergency Department (ED) may ultimately choose to leave without being seen by a physician. This occurrence is labeled “left without being seen” (LWBS) and can account for up to 15% of all patients who come to an ED. Patients who LWBS do not receive the care they sought in the ED and may experience clinical deterioration related to delayed diagnosis or treatment. Objective: To describe a LWBS cohort and identify key LWBS attributes in a Canadian pediatric emergency department through thorough machine learning analysis. This prediction is intended to be used in practice to prevent adverse outcomes related to LWBS. This study focuses on the Pediatric Emergency Department at IWK Health in Halifax, Nova Scotia, Canada. Methods: This was a single-centre, retrospective analysis of administrative ED data from April 1, 2017, to March 31, 2020, from IWK Health Emergency Department in Halifax, Nova Scotia. Triage record data including 101,266 observations of children aged 16 and younger who visited the IWK Emergency Department during a three-year period were used. Several classification machine learning algorithms including Logistic Regression, Decision Tree Classifier, Random Forest Classifier, K-Nearest Neighbors, and Extreme Gradient Boosting were used to predict patients at high-risk for LWBS. SMOTE was used to handle the class imbalance and improve the performance of the machine learning algorithms. Feature importance was used on the best performing model to identify the features that are associated with LWBS. Results: The highest-performing model utilized SMOTE balancing and the XGBoost classification algorithm. Using this model, and data from our partner hospital, an easy-to-follow set of rules were developed for identifying patients at risk of LWBS in real time. Conclusions: Results show the feasibility of predictive analytics in identifying LWBS patients. This can support proactive decision-making about those patients who are at risk of LWBS.