Time Series Models for Predicting the Number of Patients Attending the Emergency Department in a Local Hospital

EasyChair Preprint 13713

Abstract

The daily influx of patients at  emergency departments (EDs) is highly unpredictable and a major cause of overcrowding in hospitals. This study aims to provide decision-making support and establish a shared situational awareness among medical and administrative personnel. By accurately forecasting daily attendances, this project attempts to effectively reduce overcrowding issues and improve overall patient care. To address this issue, this study focuses on studying different models to predict the number of visits to the emergency departments and investigating the factors affecting daily demand. Hospitals can benefit from accurately forecasting the number of patients arriving at the ED, allowing for early planning and mitigating overcrowding. As the subject of the study, a real database collected from Asunción Klinika from 2004 to 2022 was examined in Tolosa, Gipuzkoa. For this purpose, models such as ARIMA, LSTM and GRU are proposed. The study revealed that weekly patterns as well as calendar and meteorological information have an impact on the volume of  daily patient arrivals. Over the years, several forecasting models using time series analysis have been proposed to address this challenge. Results showed that hybrid models outperformed the others in terms of the Mean Absolut Error metric (MAE). Predictions  have yielded an average daily error of 5.2 individuals, which accounts for 13%.

Keyphrases: ARIMA, RNN-GRU, RNN-LSTM, daily patient arrivals, prediction models

@booklet{EasyChair:13713,
  author    = {Silvia Aguirre and Jon Kerexeta and Moises Espejo},
  title     = {Time Series Models for Predicting the Number of Patients Attending the Emergency Department in a Local Hospital},
  howpublished = {EasyChair Preprint 13713},
  year      = {EasyChair, 2024}}