2020/11/07 by Peter Congdon, Congdon, Peter
Economics, Econometrics and Finance · Mathematics · Medicine · #Applications (stat.AP) #COVID-19 Pandemic Impacts #COVID-19 epidemiological studies #FOS: Computer and information sciences #Zoonotic diseases and public health
paper · pdf · doi:10.48550/arxiv.2011.03818
openalex publication_date 2020/11/07 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
The evolution of the COVID-19 epidemic has been accompanied by accumulating\nevidence on the underlying epidemiological parameters. Hence there is potential\nfor models providing mid-term forecasts of the epidemic trajectory using such\ninformation. The effectiveness of lockdown interventions can also be assessed\nby modelling later epidemic stages, possibly using a multiphase epidemic model.\nCommonly applied methods to analyze epidemic trajectories include\nphenomenological growth models (e.g. the Richards), and variants of the\nsusceptible-infected-recovered (SIR) compartment model. Here we focus on a\npractical forecasting approach, applied to interim UK COVID data, using a\nbivariate Reynolds model (cases and deaths). We show the utility of informative\npriors in developing and estimating the model, and compare error densities\n(Poisson-gamma, Poisson-lognormal, Poisson-logStudent) for overdispersed data\non new cases and deaths. We use cross-validation to assess medium term\nforecasts. We also consider the longer term post-lockdown epidemic profile to\nassess epidemic containment, using a two phase model. Fit to mid-epidemic data\nshows better fit to training data and better cross validation performance for a\nPoisson-logStudent model. Estimation of longer term epidemic data after\nlockdown relaxation, characterised by protracted slow downturn and then upturn\nin cases, casts doubt on effective containment. Many applications of\nphenomenological models have been to complete epidemics. However, evaluation of\nsuch models based simply on their fit to observed data may give only a partial\npicture, and cross-validation against actual trends is also useful. Similarly,\nit may be preferable to model incidence rather than cumulative data, though\nthis raises questions about suitable error densities for modelling often\nerratic fluctuations. Hence there may be utility in evaluating alternative\nerror assumptions.\n