COVID-19 hospitalizations forecasts using internet search data.

Area of Science:

• Epidemiology
• Public Health
• Data Science

Background:

• The ongoing COVID-19 pandemic necessitates reliable real-time hospitalization forecasts.
• Accurate predictions are crucial for effective medical resource allocation and public health decision-making.
• Emerging variants underscore the need for adaptable forecasting models.

Purpose of the Study:

• To forecast national and state-level COVID-19 new hospital admissions in the United States for the upcoming two weeks.
• To develop an enhanced forecasting model leveraging public search behavior and existing time-series data.
• To evaluate the model's performance against established forecasting methods.

Main Methods:

• Utilized a modified AutoRegression with GOogle search data (ARGO) model.
• Employed LASSO-penalized linear regression to integrate Google search trends with COVID-19 time-series data.
• Implemented dynamic training and rolling window prediction for continuous model updating.
• Conducted retrospective out-of-sample evaluation from January 4, 2021, to December 27, 2021.

Main Results:

• The proposed model demonstrated substantial error reduction compared to other models in the COVID-19 Forecast Hub.
• The method proved flexible, self-correcting, robust, and accurate in retrospective evaluations.
• Google search data significantly contributed to improving hospitalization admission predictions.

Conclusions:

• The developed forecasting method offers a powerful, interpretable tool for public health officials.
• This approach can aid in managing medical resources during current and future infectious disease outbreaks.
• Integrating behavioral data, like search trends, enhances infectious disease forecasting accuracy.