Development of a spatially regularized detector for emergent/re-emergent disease outbreaks

Abstract

We propose a Bayesian framework to detect outbreaks of emerging diseases. Our method triggers when the observed case-count data disagree significantly from forecasted levels. Existing detectors compare data to baseline levels, which are difficult to set for emerging diseases because of noisy/missing data. In contrast, forecasting allows us to use socioeconomic parameters and spatiotemporal data on disease prevalence to compensate for low-quality epidemiological information. The proposed framework incorporates Poisson statistics to accommodate low-counts and Markov random fields to account for spatio-temporal correlations in the disease spread rate. Posterior distributions would be approximated with both via Markov-Chain Monte Carlo (at regional level) and variational inference (at the national level). Results computed with COVID-19 data from Nee Mexico, US, will be used to demonstrate the method. A figure-of-merit will be the time/date at which the new method detects the start of the outbreak.