A GIS-Based Demand Forecast Using Machine Learning for Emergency Medical Services

Abstract

The objective for pre-hospital Emergency Medical Service (EMS) is to reach to, pick up, and deliver patients efficiently. By increasing the operational efficiency, the survival rate of major trauma patients could potentially be improved. In this research, the authors applied Moving Average, Artificial Neural Network, Linear Regression, and Support Vector Machine for the forecast of pre-hospital emergency medical demand. The results from these approaches, as a reference, could be used for pre-allocation of ambulances. The training of the models as well as the validation is conducted with data collected from the New Taipei City EMS. In order to represent the performance of forecast, the authors introduced Geographic Information System (GIS) to manage and visualize the spatial distribution of forecast demand and error. Moreover, the authors built a flexible model, which organizes demand data with the user given size of area and size of time step. With the easy use of model and acceptable prediction performance, the research outcome has its potential to be applied to the current practice.

Deep multitask ensemble classification of emergency medical call incidents combining multimodal data improves emergency medical dispatch

“…The potential capability of deep learning to enhance EMCI classification through the provision of decision support to non-clinical dispatchers, was spotted by the Health Services Department of the Valencian region, aware of the potential of these models: deep learning is at the state of the art of machine learning in tasks involving complex types of data, [23] e.g., high dimensional, unstructured, sequential, multimodal, [24][25][26][27] such as those found in EMCI databases. Likewise, this and other machine learning tools have already been applied to tackle EMD challenges such as ambulance allocation, [28][29][30]…”

“…The potential capability of deep learning to enhance EMCI classification through the provision of decision support to non-clinical dispatchers, was spotted by the Health Services Department of the Valencian region, aware of the potential of these models: deep learning is at the state of the art of machine learning in tasks involving complex types of data, [23] e.g., high dimensional, unstructured, sequential, multimodal, [24][25][26][27] such as those found in EMCI databases. Likewise, this and other machine learning tools have already been applied to tackle EMD challenges such as ambulance allocation, [28][29][30] prediction of emergency calls volume, [31] automatic stress detection of the caller, [32] interpretable knowledge extraction, [33] performance monitoring, [34] cardiac arrest calls assistance [35] or triaging unconscious and fainting patients. [36] Therefore, we can argue that deep learning models are a feasible and promising technology to improve EMD through EMCI classification.…”

Spatiotemporal Prediction of Ambulance Demand using Gaussian Process Regression

“…The weight of each Gaussian component is modulated by the phase of a prediction time within the weekly cycle. Multiple studies have implemented Artificial Neural Networks (ANNs) for callout prediction, most notably in Taipei [6,21] and Mecklenberg County, North Carolina [32]. While the studies for Taipei show promising accuracy at coarse test resolution, they struggle due to sparsity at the scale of resolution required here.…”

Deep ensemble multitask classification of emergency medical call incidents combining multimodal data improves emergency medical dispatch

“…The potential capability of deep learning to enhance EMCI classification through the provision of decision support to non-clinical dispatchers, was spotted by the Health Services Department of the Valencian region, aware of the potential of these models: deep learning is at the state of the art of machine learning in tasks involving complex types of data [23], e.g., high dimensional, unstructured, sequential, multimodal [24][25][26][27], such as those found in EMCI databases. Likewise, this and other machine learning tools have already been applied to tackle EMD challenges such as ambulance allocation [28][29][30], prediction of emergency calls volume [31], automatic stress detection of the caller [32], interpretable knowledge extraction [33], performance monitoring [34], cardiac arrest calls assistance [35] or triaging unconscious and fainting patients [36]. Therefore, we can argue that deep learning models are a feasible and promising technology to improve EMD through EMCI classification.…”