Tracking the national and regional COVID-19 epidemic status in the UK using weighted principal component analysis

Journal article

One of the difficulties in monitoring an ongoing pandemic is deciding on the metric that best describes its status when multiple intercorrelated measurements are available. Having a single measure, such as the effective reproduction number R, has been a simple and useful metric for tracking the epidemic and for imposing policy interventions to curb the increase when R>1. While R is easy to interpret in a fully susceptible population, it is more difficult to interpret for a population with heterogeneous prior immunity, e.g. from vaccination and prior infection. We propose an additional metric for tracking the UK epidemic that can capture the different spatial scales. These are the principal scores from a weighted principal component analysis. In this paper, we have used the methodology across the four UK nations and across the first two epidemic waves (January 2020–March 2021) to show that first principal score across nations and epidemic waves is a representative indicator of the state of the pandemic and is correlated with the trend in R. Hospitalizations are shown to be consistently representative; however, the precise dominant indicator, i.e. the principal loading(s) of the analysis, can vary geographically and across epidemic waves. This article is part of the theme issue ‘Technical challenges of modelling real-life epidemics and examples of overcoming these’.

Authors: Swallow, B; Xiang, W; Panovska-Griffiths, J

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: The Royal Society
• Journal: Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences
• Volume: 380
• Issue: 2233
• Pages: 20210302
• Article number: 20210302
• Publication date: 2022-08-15
• Acceptance date: 2022-03-10
• DOI: 10.1098/rsta.2021.0302
• EISSN: 1471-2962
• ISSN: 1364503X, 1364-503X
• Language: English
• Keywords: multivariate statistics; spatial epidemiology; COVID-19; dimension reduction; principal component analysis