Mid-Epidemic Forecasts of COVID-19 Cases and Deaths: A Bivariate Model applied to the UK

TL;DR

This paper develops a bivariate forecasting model for COVID-19 cases and deaths in the UK, demonstrating the importance of informative priors and error density choices for accurate mid-term epidemic predictions.

Contribution

It introduces a practical bivariate Reynolds model with informative priors and compares various error densities, enhancing epidemic forecasting accuracy and evaluation methods.

Findings

Poisson-logStudent model performs best in cross-validation.

Longer-term post-lockdown data suggests containment is unlikely.

Modeling incidence rather than cumulative data may improve fit.

Abstract

The evolution of the COVID-19 epidemic has been accompanied by accumulating evidence on the underlying epidemiological parameters. Hence there is potential for models providing mid-term forecasts of the epidemic trajectory using such information. The effectiveness of lockdown interventions can also be assessed by modelling later epidemic stages, possibly using a multiphase epidemic model. Commonly applied methods to analyze epidemic trajectories include phenomenological growth models (e.g. the Richards), and variants of the susceptible-infected-recovered (SIR) compartment model. Here we focus on a practical forecasting approach, applied to interim UK COVID data, using a bivariate Reynolds model (cases and deaths). We show the utility of informative priors in developing and estimating the model, and compare error densities (Poisson-gamma, Poisson-lognormal, Poisson-logStudent) for overdispersed data on new cases and deaths. We use cross-validation to assess medium term forecasts. We also consider the longer term post-lockdown epidemic profile to assess epidemic containment, using a two phase model. Fit to mid-epidemic data shows better fit to training data and better cross validation performance for a Poisson-logStudent model. Estimation of longer term epidemic data after lockdown relaxation, characterised by protracted slow downturn and then upturn in cases, casts doubt on effective containment. Many applications of phenomenological models have been to complete epidemics. However, evaluation of such models based simply on their fit to observed data may give only a partial picture, and cross-validation against actual trends is also useful. Similarly, it may be preferable to model incidence rather than cumulative data, though this raises questions about suitable error densities for modelling often erratic fluctuations. Hence there may be utility in evaluating alternative error assumptions.