Bayesian inference for high-dimensional discrete-time epidemic models: spatial dynamics of the UK COVID-19 outbreak

TL;DR

This paper introduces a Bayesian MCMC method with novel samplers for fitting high-dimensional spatial epidemic models, demonstrated on UK COVID-19 data to inform policy decisions.

Contribution

It develops a new Bayesian inference approach for complex spatial epidemic models, enabling real-time analysis of COVID-19 spread at the national level.

Findings

Real-time spatial epidemic insights for UK COVID-19

Effective Bayesian inference for high-dimensional models

Policy-relevant outbreak control recommendations

Abstract

Stochastic epidemic models which incorporate interactions between space and human mobility are a key tool to inform prioritisation of outbreak control to appropriate locations. However, methods for fitting such models to national-level population data are currently unfit for purpose due to the difficulty of marginalising over high-dimensional, highly-correlated censored epidemiological event data. Here we propose a new Bayesian MCMC approach to inference on a spatially-explicit stochastic SEIR meta-population model, using a suite of novel model-informed Metropolis-Hastings samplers. We apply this method to UK COVID-19 case data, showing real-time spatial results that were used to inform UK policy during the pandemic.