Generating a spatial coverage plan for the emergency medical service on a regional scale: Empirical versus random forest modelling approach

Abstract

Adequate spatial coverage by the emergency medical service and the ability to reach any location in the area of interest in the shortest possible time are crucial for the survival of patients with serious conditions. Knowledge of blind spots (i.e. sites that cannot be reached within the required time) represents key information for improving the service quality and may lead, e.g. to a relocation of bases or to other active interventions. Spatial coverage can be derived from experience based on historical data. Such an approach may be problematic if a larger area is being analysed, especially if data is not available for some parts of such areas or if no data is available. To eliminate such problems, we created a prediction model utilising the random forest ensemble learning method. The model is capable of predicting the travel time based on available historical data on ambulance movements (GPS) and the geometric and construction characteristics of individual road segments. We therefore collaborated with the regional public administration and emergency medical service authorities to deliver a time- and resource-efficient solution for emergency spatial planning practice. The outputs from the newly built model were subsequently validated against data from an empirical model currently used by the regional authorities. The results from both models were compared from the perspective of performance in various seasonal and time-of-day conditions. The prediction of travel times using the new model improved according to all the evaluated validation metrics. The importance and applicability of the foregoing model lies in the fact that it can be incorporated into the current emergency medical service management system in a simple manner in terms of data availability and the required computational resources. We conclude that the dynamic model presented in this paper represents an improvement relative to the reference data, and discuss the possibilities of further improving the proposed model.