Integrating information from historical data into mechanistic models for influenza forecasting

Area of Science:

• Epidemiology
• Mathematical Modeling
• Public Health

Background:

• Seasonal influenza significantly impacts global health, causing millions of consultations annually.
• Existing forecasting models are either mechanistic (lacking historical data integration) or statistical (lacking mechanistic assumptions).
• Mechanistic models often produce forecasts deviating from past epidemic trajectories.

Purpose of the Study:

• To develop an improved mechanistic model for influenza forecasting.
• To integrate historical epidemic data into mechanistic modeling for enhanced accuracy.
• To leverage France's extensive influenza surveillance data for model development and validation.

Main Methods:

• Developed a novel mechanistic model incorporating epidemic data from training seasons.
• Employed a particle filter for parameter estimation on observed data.
• Utilized a second particle filter to generate forecasts aligned with historical epidemic trajectories.
• Calibrated and tested the model on 35 years of French influenza-like-illness (ILI) surveillance data (1985-2019).

Main Results:

• The new model demonstrated improved accuracy over standard mechanistic approaches.
• Retrospective testing showed enhanced precision for short-term forecasts (1-4 weeks ahead).
• The model improved predictions of epidemic peak timing and intensity.

Conclusions:

• The developed model successfully integrates statistical strengths into a mechanistic framework.
• This approach offers more accurate influenza forecasting by maximizing the utility of long-term surveillance data.
• The findings contribute to better preparedness and management of seasonal influenza epidemics.