COVID-19 Impact on Forecasting Emergency Department Visits Performance
19 Pages Posted: 26 Jul 2022
Abstract
The COVID-19 pandemic has been affecting the normal functioning of every hospital and healthcare system. Government restrictions and the installed fear led to significant changes in the dynamic of the emergency department (ED) visits, establishing a new challenge for forecasting systems that aim to predict the number of visits in advance. This study provides an analysis on the performance of three distinct forecasting models during the full year of 2020.SARIMA, Prophet and Random Forest were tested for 5 years of data (2016-2020) and evaluated for each month with the forecasting horizon set to 1-day ahead. The performance of each model was also contrasted with the government restrictions to measure its influence on the models.This study evidences the underperformance of the models, with MAPE raising between 12%-34% from the moment that the first cases of infection were reported. Some of the relevant variations were confirmed to have been influenced by some specific government restrictions that highly contributed to the decrease in the number of ED visits and, consequently, model’s accuracy. SARIMA, a baseline model holds a great equilibrium by being both robust when predicting in different periods of the calendar and quick to adapt to the structural break, provoked by the pandemic. While Prophet can only achieve the best results when seasonal factors drive the data, Random Forest on the other hand, react better to the structural break, but is only marginally more accurate when compared to SARIMA.
Note:
Funding Information: None.
Declaration of Interests: None.
Consent Statement: All ED visits data were anonymous, and its collection was approved by the hospital’s ethics committee (CES).
Keywords: forecasting, Emergency Department, predictive models, resource management, hospital visits
Suggested Citation: Suggested Citation