A Model for Highly Fluctuating Spatio-Temporal Infection Data, with Applications to the COVID Epidemic.

Area of Science:

• Epidemiology
• Statistical modeling
• Infectious disease dynamics

Background:

• Spatio-temporal infection data exhibit complex dynamics, including endemic stability, epidemic outbreaks, and spatial spread.
• Existing models often struggle to capture the non-stationarity and regime-switching inherent in infection patterns.

Purpose of the Study:

• To develop a flexible spatio-temporal mixture-link model for infection counts that accounts for regime-switching between epidemic and endemic states.
• To incorporate adaptive regime-switching for modeling neighborhood spillover effects in infection spread.
• To improve the in-sample fit and short-term predictive accuracy of infection models.

Main Methods:

• A generalized Poisson regression model is employed for infection counts, avoiding data transformations.
• A mixture-link model with regime-switching is proposed to capture distinct epidemic and endemic phases.
• The model incorporates spatial autocorrelation through neighborhood spillover effects, also governed by adaptive regime-switching.

Main Results:

• The proposed model demonstrates a better in-sample fit compared to existing methods for COVID-19 area-time data.
• Out-of-sample, short-term predictions were improved, particularly during epidemic phases.
• Case studies involving COVID-19 in London and Southeast England highlight the model's ability to capture spatial dynamics.

Conclusions:

• The developed spatio-temporal mixture-link model effectively captures the balance between epidemic and endemic tendencies in infection data.
• The model offers enhanced accuracy for short-term forecasting of infectious disease spread.
• The findings underscore the importance of regime-switching and spatial dependencies in understanding and predicting epidemic dynamics.