Detecting outbreaks using a spatial latent field.

Area of Science:

• Epidemiology
• Spatial Statistics
• Computational Statistics

Background:

• Estimating disease infection rates across multiple regions presents challenges due to data variability.
• Existing epidemiological models often focus on single areal units, limiting spatial analysis.
• Accurate spatial-temporal disease surveillance is crucial for public health interventions.

Purpose of the Study:

• To develop a robust method for estimating the spatial-temporal infection rate of a disease.
• To extend single-unit epidemiological models to a multi-unit framework using Bayesian approaches.
• To create an anomaly detection system for identifying new epidemic waves.

Main Methods:

• Utilized time-series case-counts from multiple areal units.
• Extended a single-unit epidemiological model to a multi-unit context.
• Employed Bayesian inference with a Gaussian random field prior.
• Applied adaptive Markov chain Monte Carlo for parameter sampling.
• Validated the model using COVID-19 case data from New Mexico counties.

Main Results:

• Spatial correlation between neighboring regions regularizes estimations, especially with high-variance data.
• The calibrated model accurately forecasts infection rates across areal units.
• An anomaly detector based on estimated infection rates significantly outperforms methods relying solely on case-counts.
• Demonstrated improved epidemic wave detection capabilities.

Conclusions:

• Integrating spatial dependencies enhances the reliability of infection rate estimations.
• The proposed Bayesian framework provides a powerful tool for spatial-temporal disease surveillance.
• The developed anomaly detection method offers a more sensitive approach to identifying emerging epidemic waves.