Predicting COVID-19 community infection relative risk with a Dynamic Bayesian Network

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• COVID-19 continues to pose a global health challenge, necessitating accurate local-scale spread prediction.
• Emerging variants underscore the need for robust infectious disease surveillance and forecasting methods.
• Understanding spatial and temporal dynamics is crucial for effective community-level interventions.

Purpose of the Study:

• To develop and validate a Dynamic Bayesian Network (DBN) for predicting COVID-19 infection risk.
• To assess the model's performance in forecasting relative risk at the census tract scale in Indiana.
• To incorporate socio-environmental vulnerability factors into spatial-temporal epidemiological modeling.

Main Methods:

• Development of a Dynamic Bayesian Network (DBN) model.
• Integration of social and environmental vulnerability indicators, including environmental determinants of infection.
• Spatial-temporal prediction of COVID-19 relative risk one month into the future.
• Comparative analysis against five established spatial epidemiological modeling techniques.

Main Results:

• The DBN model demonstrated superior predictive performance compared to five other methods.
• The model successfully predicted community-level relative risk of COVID-19 infection.
• The DBN framework proved effective for spatial-temporal forecasting and "what-if" scenario analysis.

Conclusions:

• Dynamic Bayesian Networks offer a powerful approach for predicting infectious disease spread.
• Incorporating socio-environmental factors enhances the accuracy of spatial-temporal epidemiological models.
• Future research should explore AI integration with DBNs for dynamic process modeling in epidemiology.