‘Introduction’ Open Access

The current COVID-19 pandemic has brought to wide attention how important the statistics discipline is to public health monitoring and decision-making, from the design of studies through the analysis of data to the reporting of results and their correct interpretation. In the United Kingdom, data visualisations, estimation of the current state of the epidemic and model-based predictions have featured prominently in daily briefings by the country’s most senior politicians and their scientific advisers.

As part of its response to the pandemic, The Royal Statistical Society established at the end of March 2020 a COVID-19 Task Force, whose aim is ‘to ensure that the RSS can contribute its collective expertise to UK national and devolved governments and public bodies, regarding statistical issues during the COVID-19 pandemic’. For more information on the Task Force, see: https://rss.org.uk/policy-campaigns/policy-groups/covid-19-task-force/.

Throughout the current epidemic there has been considerable focus on the use of the reproduction number or ‘R-number’ (roughly speaking the expected number of additional cases per primary case at time t) as a headline summary of the state of the epidemic. Opinions vary on the value of this summary measure, not least because it reduces the transmissibility of infection to a single number. The reality is much more complex, with transmissibility depending on a range of natural and social environmental factors that vary across different parts of a country and across identifiable subpopulations, as well as being influenced by earlier interventions and acquired immunity.

The goal of this Special Topic Meeting, held over three sessions on 9 and 11 June 2021, was to discuss statistical issues around the general topic of COVID-19 transmission, including but not restricted to local (to particular areas or subpopulations) versions of the R-number. In the tradition of RSS Discussion Meetings, the programme consisted of oral presentations based on peer-reviewed papers that were made available beforehand, followed by invited and open discussion contributions and responses from the authors. The complete proceedings are now published as a permanent record of some of the many significant contributions that statistical modellers have made to tracking the evolution of the COVID-19 pandemic in real time, with a view to providing sound evidence for policy makers.

The first session brought complementary views from Parag et al. (2022) and Jewell and Lewnard (2022) on the value and common misunderstandings of R as a summary measure, its sensitivity to model assumptions, inherent difficulties of estimation and its use or misuse for guiding public health interventions. These authors further discussed the usefulness of additionally quantifying time changes in R and of estimating growth rates. Coffeng and de Vlas’s paper (2022) focused on demonstrating the impact of heterogeneous settings on transmission as a warning that making a simplifying assumption of homogeneity can lead to overestimation of the transmission rate. The two papers in the second session squarely target the estimation of R at a local geographical scale, typically lower tier local authorities (LTLAs). They both adopt a renewal equation framework for the epidemic modelling and a Bayesian estimation framework, differing in how they treat the spatial dynamics of transmission. Mishra et al. (2022) do not explicitly model spatial dependence between areas, whereas Teh et al. (2022) adopt a spatiotemporal hierarchical formulation and resort to some approximations to address the resulting computational issues. The final session returned to the topic of the R-number with a detailed discussion by Pellis et al. (2022) of its properties and of the statistical issues faced for its estimation as the underlying model is progressively complexified. In contrast to widely used semi-mechanistic modelling approaches, Bekker-Nielsen, Dunbar and Held (2022) adopt a time series approach that incorporates endemic and epidemic components in an autoregressive manner, and demonstrate the utility of this approach in assessing the effect of a non-pharmaceutical intervention on the course of a local epidemic.

REFERENCES