New! Sign up for our free email newsletter.
Science News
from research organizations

Development of an ensemble model to anticipate short-term COVID-19 hospital demand

Date:
May 9, 2022
Source:
Institut Pasteur
Summary:
Scientists identified the most relevant predictive variables for anticipating hospital demand and proposed using an ensemble model based on the average of the predictions of several individual models.
Share:
FULL STORY

For the past two years, the COVID-19 pandemic has exerted pressure on the hospital system, with consequences for patients' care pathways. To support hospital planning strategies, it is important to anticipate COVID-19 health care demand and to continue to improve predictive models.

In this study published in the Proceedings of the National Academy of Sciences, scientists from the Mathematical Modeling of Infectious Diseases Unit at the Institut Pasteur identified the most relevant predictive variables for anticipating hospital demand and proposed using an ensemble model based on the average of the predictions of several individual models.

The scientists began by evaluating the performance of 12 individual models and 19 predictive variables, or "predictors," such as epidemiological data (for example the number of cases) and meteorological or mobility data (for example the use of public transport). The scientists showed that the models incorporating these early predictive variables performed better. The average prediction error was halved for 14-day-ahead predictions. "These early variables detect changes in epidemic dynamics more quickly," explains Simon Cauchemez, Head of the Mathematical Modeling of Infectious Diseases Unit at the Institut Pasteur and last author of the study. "The models that performed best used at least one epidemiological predictor and one mobility predictor," he continues. The addition of a meteorological variable also improved forecasts but with a more limited impact.

The scientists then built an ensemble model, taking the average of several individual models, and tested the model retrospectively using epidemiological data from March to July 2021. This approach is already used in climate forecasting. "Our study shows that it is preferable to develop an ensemble model, as this reduces the risk of the predicted trajectory being overly influenced by the assumptions of a specific model," explains Juliette Paireau, a research engineer in the Mathematical Modeling of Infectious Diseases Unit at the Institut Pasteur and joint first author of the study.

This ensemble model has been used to monitor the epidemic in France since January 15, 2021.

The study demonstrates an approach that can be used to better anticipate hospital demand for COVID-19 patients by combining different prediction models based on early predictors.

The full results of the study can be found on the Modeling page : https://modelisation-covid19.pasteur.fr/realtime-analysis/hospital/


Story Source:

Materials provided by Institut Pasteur. Note: Content may be edited for style and length.


Journal Reference:

  1. Juliette Paireau, Alessio Andronico, Nathanaël Hozé, Maylis Layan, Pascal Crépey, Alix Roumagnac, Marc Lavielle, Pierre-Yves Boëlle, Simon Cauchemez. An ensemble model based on early predictors to forecast COVID-19 health care demand in France. Proceedings of the National Academy of Sciences, 2022; 119 (18) DOI: 10.1073/pnas.2103302119

Cite This Page:

Institut Pasteur. "Development of an ensemble model to anticipate short-term COVID-19 hospital demand." ScienceDaily. ScienceDaily, 9 May 2022. <www.sciencedaily.com/releases/2022/05/220509112038.htm>.
Institut Pasteur. (2022, May 9). Development of an ensemble model to anticipate short-term COVID-19 hospital demand. ScienceDaily. Retrieved October 31, 2024 from www.sciencedaily.com/releases/2022/05/220509112038.htm
Institut Pasteur. "Development of an ensemble model to anticipate short-term COVID-19 hospital demand." ScienceDaily. www.sciencedaily.com/releases/2022/05/220509112038.htm (accessed October 31, 2024).

Explore More

from ScienceDaily

RELATED STORIES