Bayesian analysis of diffusion-driven multi-type epidemic models with application to COVID-19

Abstract We consider a flexible Bayesian evidence synthesis approach to model the age-specific transmission dynamics of COVID-19 based on daily mortality counts. The temporal evolution of transmission rates in populations containing multiple types of individuals is reconstructed via an appropriate dimension-reduction formulation driven by independent diffusion processes. A suitably tailored compartmental model is used to learn the latent counts of infection, accounting for fluctuations in transmission influenced by public health interventions and changes in human behaviour. The model is fitted to freely available COVID-19 data sources from the UK, Greece, and Austria and validated using a large-scale prevalence survey in England. In particular, we demonstrate how model expansion can facilitate evidence reconciliation at a latent level. The code implementing this work is made freely available via the Bernadette R package.