Jorge Mateu and Álvaro Briz-Redón's contribution to the Discussion of ‘The Second Discussion Meeting on Statistical aspects of the Covid-19 Pandemic’

Discussion

It is difficult to outline good contributions to the statistical modelling of COVID-19 data, but these two papers are to be congratulated on their valuable and thought-provoking contributions within this rapidly developing field. We would like to focus our attention on several aspects that are arguably of interest to the community.

Bhatt et al. consider that $Rt$ can be estimated in each region in separate models. Clearly, this reproductive number can be better estimated if space–time structure is considered as, though a pragmatic assumption, it seems not very realistic that the reproductive number cannot vary in space and time, and can be affected by neighbouring structures. Another aspect is considering an alternative mechanistic form for the space–time intensity function (see Briz-Redón et al., 2022) bringing into play a deterministic part (where covariates of all types play a role) and an analytical form for the interaction. This seems a promising way to explore the behaviour of the spread of COVID-19.

Storviv et al. consider a stochastic susceptible-exposed-infectious-recovered (SEIR) model whose output is incorporated into their main statistical model to capture $Rt$ variation. This approach is novel and convenient to control for the time-varying size of the susceptible population. Besides, the proposal of Storviv et al. for the time-varying reproduction number can be easily extended to space–time by considering a low-rank approximation based on splines as in Martínez-Beneito et al. (2022). Proceeding this way, there is no need for such a large amount of involved parameters and Bayesian inference is easily adopted. We argue that heterogeneity across areas or counties should be considered and not avoided or simplified. Again, mobility patterns are crucial in the spread of COVID-19 and a natural way to outline its importance comes well explained in Slater et al. (2022). We wonder if a simpler while more model-based approach would be a good alternative to such mobility modelling.

Finally, we would like to highlight two common aspects of these contributions. First, they explicitly account for certain features of COVID-19 data, mainly the existence of underreporting and the temporal delay of the observations. This is something that has been overlooked in numerous related studies. Second, the Bayesian nature of both methodologies enables estimating $P(Rt>1)$ directly. Measuring the uncertainty around the reproduction number could better guide decision-making in future epidemic events.

References

Author notes