Predicting ambulance offload delay using a hybrid decision tree model

Abstract

Ambulance offload delay (AOD) is a growing health care concern in Canada. It refers to the delay in transferring an ambulance patient to a hospital emergency department (ED) due to ED congestion. It can negatively affect the ability of the ambulance service to respond to future calls and reduce the efficiency of the system when the delay is significant. Using integrated historical data from a partnering hospital and an Emergency Medical Services (EMS) provider, we developed a decision-support tool using a hybrid decision tree model to predict the severity of AOD occurring within 1–5 h in an EMS system. The primary objective of this study is to provide an AOD prediction model based on the current system status, hour of the day, and day of the week. With this information, decision-makers can be proactive with efforts to mitigate AOD. Various prediction models are developed with different focuses and forecast periods. This research demonstrates a novel hybrid decision tree method applied with administrative data in a health care setting. A naïve Bayes classifier is first used to remove noisy training observations before decision tree induction. This hybrid decision tree algorithm was tested against the basic classification and regression tree (CART) algorithm, using classification accuracy, precision, sensitivity, and specificity analysis. The results indicate that the hybrid algorithm shows improvements in performance in the classification of the real-world problem. It is anticipated that the prediction model for AOD produced from this study will be directly transferable. It can be generalized to other EMS systems, where predicting AOD is important for efficient operations.