Bayesian inference across multiple models suggests a strong increase in lethality of COVID-19 in late 2020 in the UK

Abstract

We apply Bayesian inference methods to a suite of distinct compartmental models of generalised SEIR type, in which diagnosis and quarantine are included via extra compartments. We investigate the evidence for a change in lethality of COVID-19 in late autumn 2020 in the UK, using age-structured, weekly national aggregate data for cases and mortalities. Models that allow a (step-like or graded) change in infection fatality rate (IFR) have consistently higher model evidence than those without. Moreover, they all infer a close to two-fold increase in IFR. This value lies well above most previously available estimates. However, the same models consistently infer that, most probably, the increase in IFR preceded the time window during which new variant B.1.1.7 became the dominant strain in the UK. Therefore, according to our models, the caseload and mortality data do not offer unequivocal evidence for higher lethality of a new variant. We compare these results for the UK with similar models for Germany and France, which also show increases in inferred IFR during the same period, despite the even later arrival of new variants in those countries. We argue that while the new variant(s) may be one contributing cause of a large increase in IFR in the UK in autumn 2020, other factors, such as seasonality, or pressure on health services, are likely to also have contributed.

Inferring the COVID-19 infection fatality rate in the community-dwelling population: a simple Bayesian evidence synthesis of seroprevalence study data and imprecise mortality data

“…Our estimates are based on data from 2020, some of which were obtained more than a year ago (see dates listed in Table 1). It is likely that, with continual viral mutation of SARS-CoV-2 and advances in treatment, the current IFR in many places is now markedly different than it was earlier, and our estimates are therefore likely to be outdated (Pei et al, 2021, Pietzonka et al, 2021, Walensky et al, 2021). In particular, at the present time, India is experiencing a rapid increase in COVID-19 fatalities which suggests that the current IFR in India may be much higher now than during earlier phases of the pandemic (Padma, 2021, Thiagarajan, 2021).…”

Efficient Bayesian inference of fully stochastic epidemiological models with applications to COVID-19

“…Given the modelling assumptions ( particularly the well-mixed assumption, without over-dispersion), this result shows that the method has promise. Further work on more detailed and accurate models [40] will provide new and stringent tests of its applicability. Data accessibility.…”

“…Other classes may be either infectious or non-infectious: the canonical example is an SIR model which corresponds to L = 3, in which case the recovered (R) class is non-infectious. The analysis presented here is straightforwardly generalized to more complex compartment models, as might be used (for example) to model different pathogen strains, or vaccinated individuals with reduced susceptibility, or testing and quarantining [ 40 ].…”

“…It is likely that fits of similar quality could be achieved by a model with significantly fewer parameters; the dependence of parameters on age is also not very strong, so the number of age cohorts might also be reduced without much loss of accuracy. However, one purpose of this example is to test the capacity of the approach to handle models of this complexity, with a view to future work with (for example) compartments for quarantined/vaccinated individuals [ 40 ] and/or multiple variants of the virus. In this example, the inference computations are within the capability of desktop workstations, although long runs were required for MCMC and evidence computations, see below for details.…”

“…The intended future applications of these methods are to similar (population-level) models with higher complexity, e.g. [ 40 ].…”