Predicting regional influenza epidemics with uncertainty estimation using commuting data in Japan

Area of Science:

• Epidemiology
• Network Science
• Public Health

Background:

• Accurate regional influenza prediction is vital for medical institutions and public health planning.
• Existing models often fail to account for population movement between areas, limiting their effectiveness.
• Influenza's person-to-person spread necessitates models that incorporate inter-regional connectivity.

Purpose of the Study:

• To develop an improved regional influenza prediction model.
• To incorporate human mobility patterns, specifically commuting data, into influenza forecasting.
• To enhance the accuracy and practical utility of epidemic spread predictions.

Main Methods:

• Proposed a novel method utilizing an extended Graph Convolutional Network (GCN).
• Integrated commuting data to represent the flow of people between geographical areas.
• Developed a prediction interval suitable for cyclic time series data, applied to weekly influenza patient data in Japan.

Main Results:

• The GCN model incorporating commuting data demonstrated significantly improved predictive accuracy compared to baseline models.
• The model showed enhanced performance both temporally and spatially in predicting influenza patient distribution.
• An appropriate prediction interval was successfully provided for medical institutions.

Conclusions:

• Graph Convolutional Networks (GCNs) are effective tools for predicting epidemic spread, particularly when incorporating mobility data.
• The proposed model offers practical benefits for public health authorities in decision-making, resource allocation, and managing vaccine demand.
• Accounting for population flow is crucial for accurate regional epidemic forecasting.