Can auxiliary indicators improve COVID-19 forecasting and hotspot prediction?

Area of Science:

• Epidemiology
• Public Health
• Data Science

Background:

• Accurate short-term forecasting of pandemic indicators like cases, hospitalizations, and deaths is crucial for public health decision-making.
• Real-time COVID-19 indicators have been collected and made publicly available since early 2020.
• Traditional forecasting models often rely on public health reporting data.

Purpose of the Study:

• To evaluate the forecasting utility of five diverse COVID-19 indicators derived from medical insurance claims, online surveys, and Google search trends.
• To determine if incorporating these indicators into an autoregressive (AR) model improves predictive accuracy compared to an AR model without external features.

Main Methods:

• Utilized an autoregressive (AR) model as a baseline for COVID-19 forecasting.
• Assessed the impact of including five distinct data sources (medical claims, survey symptoms, Google searches) as external features in the AR model.
• Compared the predictive accuracy of the augmented AR model against the baseline AR model.

Main Results:

• Inclusion of each of the five studied indicators significantly improved the overall predictive accuracy of the AR model.
• Predictive gains were most substantial during periods when COVID-19 cases were trending flat or downward.
• Google search activity emerged as a particularly valuable indicator during periods of increasing COVID-19 trends.

Conclusions:

• Diverse data streams beyond traditional public health reporting can enhance the accuracy of short-term COVID-19 forecasting models.
• The optimal set of indicators for improving forecast accuracy may vary depending on the current pandemic trend (e.g., increasing, decreasing, or stable).
• Real-time indicators from non-traditional sources offer valuable insights for pandemic surveillance and response.