The effect of non-pharmaceutical interventions on COVID-19 cases, deaths and demand for hospital services in the UK: a modelling study

Nicholas G Davies; Adam J Kucharski; Rosalind M Eggo; Amy Gimma; W. John Edmunds

The model presented here is subject to several limitations. Because the model does not explicitly structure individuals by household, we are unable to evaluate the impact of measures based on household contacts, e.g. household quarantine, i.e., where all members of a household with a suspected COVID-19 case remain in isolation. Such contact-targeted measures could increase the impact of a package of interventions by limiting spread in the community. However, the presence of asymptomatic infections [23] means that isolation based on symptomatic case identification would be unlikely to fully prevent ongoing transmission. We also do not include individual level variation in transmission (i.e. 'superspreading events', [24] ). There are several examples of such events for COVID-19 [25] , and individual-level variation is likely important in influencing the success of control measures in the very early stages of an outbreak [5] . However, as outbreaks of directly-transmitted infections become larger, the population-level dynamics will predominantly be driven by the average mixing pattern between key epidemiological groups, particularly between different ages [11, 26] . We therefore used a stochastic model implementation to capture variation in these population-level dynamics. We also assumed that subclinically-infected individuals were 50% as infectious as clinical cases. A study of 2,147 close contacts in Ningbo, China estimated that the mean onward transmission from asymptomatic infections was 65% (95% HDI: 20-120%) that of symptomatic cases [23] . However, symptomatic cases were found to be more likely to generate new symptomatic infections compared to asymptomatic infections. This suggests that the overall relative contribution of asymptomatic individuals to new infections may be lower than 65%, and hence 50% is a plausible assumption. We used mixing matrices for the UK measured in 2006 [12] , and changes in contact patterns since then may alter the potential effect of interventions. The fractions of hospitalisation, ICU use, and death are estimated using data from China, and any differences in UK populations could affect our estimates of health care demand.