Influenza forecasting for French regions combining EHR, web and climatic data sources with a machine learning...

Area of Science:

• Public Health
• Epidemiology
• Machine Learning

Background:

• Current French healthcare-based disease monitoring for influenza has a 1-3 week lag.
• This delay hinders timely public health interventions during outbreaks.
• A gap exists in real-time population-level disease activity data.

Purpose of the Study:

• To develop a machine-learning model for real-time influenza surveillance in France.
• To provide accurate, short-term influenza activity forecasts for all continental regions.
• To address the limitations of traditional, delayed surveillance systems.

Main Methods:

• Utilized a machine-learning ensemble approach.
• Integrated multiple data sources: Google search activity, weather data, Twitter data, electronic health records, and historical regional data.
• Developed models for real-time estimation and short-term forecasting of influenza activity.

Main Results:

• All individual data sources contributed to enhancing influenza surveillance accuracy.
• Machine-learning ensembles combining all data sources yielded the most accurate and timely predictions.
• The model provides real-time estimates and short-term forecasts for France's twelve continental regions.

Conclusions:

• Machine learning offers a powerful tool for real-time disease surveillance.
• Integrating diverse data streams significantly improves the accuracy and timeliness of influenza predictions.
• This approach can empower public health officials with actionable, up-to-date information for outbreak management.