June 06, 2025
Overview:
Accurate forecasting is essential for supporting timely public health action and ensuring that health systems are prepared to manage surges in illness. Despite extensive efforts to develop models for predicting COVID-19 trends, there were many gaps in the accuracy and reliability of these tools.
In this collaborative study, One Health Trust researchers developed a new machine learning model using an improved version of an existing forecasting method called N-BEATS. They added features that help the model consider outside factors such as changes in public behavior or new virus variants and estimate how likely different outcomes might be. They tested the model against other widely used COVID-19 forecasting tools and also applied it in a hospital setting, adapting it to local data to see how well it could work in practice.
The Question:
Can we build a better model to predict COVID-19 hospitalizations so that health systems can prepare more effectively for future waves of illness?
The Findings:
The new model was more accurate than existing forecasting tools, reducing prediction errors by 34 to 37 percent. It also delivered accurate, actionable forecasts when implemented at a large academic medical center, helping hospital leadership with resource planning and COVID-19 surge preparedness.
Read the article in Open Forum Infectious Disease here.
