An emergency department simulation and a neural network metamodel

Area of Science:

• Operations Research
• Healthcare Management
• Artificial Intelligence

Background:

• Hospital emergency departments (EDs) face challenges in managing patient flow and wait times.
• Discrete event stochastic simulation is a powerful tool for analyzing ED operations but can be computationally intensive.
• Metamodeling offers a computationally efficient approach to approximate complex simulation outputs.

Purpose of the Study:

• To develop and evaluate an artificial neural network (ANN) metamodel for a hospital emergency department simulation.
• To compare the performance of the ANN metamodel against the original simulation for estimating key performance metrics.
• To assess the utility of ANNs in creating surrogate models for complex healthcare system simulations.

Main Methods:

• A discrete event stochastic simulation model of a hospital emergency department was created.
• Artificial neural networks were employed as the metamodeling technique.
• The ANN metamodel was trained using output data generated from the simulation.
• The accuracy of the ANN metamodel in estimating the mean and variance of patient time in the ED was statistically compared to the simulation's results.

Main Results:

• The developed ANN metamodel demonstrated comparable performance to the full simulation in estimating the mean and variance of patient time.
• The metamodel provides a significantly faster estimation of these performance metrics compared to running the original simulation.
• This indicates the potential for ANNs to serve as effective surrogate models for ED simulations.

Conclusions:

• ANN metamodels can accurately and efficiently approximate the performance of complex hospital emergency department simulations.
• Metamodeling using ANNs offers a valuable approach for rapid analysis and prediction of patient flow dynamics in EDs.
• This methodology can aid healthcare administrators in optimizing ED operations and resource allocation.