An infodemiological framework for tracking the spread of SARS-CoV-2 using integrated public data

Area of Science:

• Epidemiology
• Data Science
• Public Health

Background:

• The SARS-CoV-2 pandemic presented challenges in early detection and prediction due to scarce initial data.
• Heterogeneous public data sources offer potential for early warning signals of infection waves.

Purpose of the Study:

• To characterize temporal pandemic indicators using integrated public data.
• To apply these indicators to a Prophet model for predicting COVID-19 trends.

Main Methods:

• Developed a natural language processing pipeline to extract time-series signals from news articles.
• Utilized Kleinberg's burst detection algorithm to identify signal bursts.
• Correlated Google Trends, news volume, and wastewater SARS-CoV-2 data with weekly COVID-19 case numbers.
• Integrated identified predictors into a Prophet model for forecasting.

Main Results:

• High correlations (0-3 week lags) were observed between Google Trends, news volume, wastewater data, and COVID-19 case numbers across US states.
• Incorporating these predictors significantly enhanced the Prophet model's performance.
• The model achieved average mean absolute errors of 0.38 and 0.46 for one and two-week COVID-19 case number predictions, respectively.

Conclusions:

• Publicly available data streams serve as effective early predictors for COVID-19 trends.
• The integration of these diverse data sources improves the accuracy of epidemiological forecasting models.
• This approach offers valuable tools for managing and mitigating future pandemic waves.