Jason Wyse, Dhorasso Temfack, Eishita Yadav and James Sweeney's contribution to the Discussion of ‘Some statistical aspects of the Covid-19 response’ by Wood et al.

We congratulate the authors on a thought-provoking and timely paper evaluating the effectiveness and impact of decisions made during the COVID-19 pandemic on society at large. The authors produce compelling arguments as to why some of the approaches taken to modelling, model checking, and prediction may not have been optimal. They suggest that narratives around disease risks and the benefits of disease mitigation strategies, communicated to the public, did not meet statistical best practice. We fully agree with their sentiments on the inappropriateness of data being made available only to siloed groups as opposed to the research community at large. However, while robust, we do wonder if the authors’ criticisms of the models that were used fully appreciate the context in which they were built and deployed. We have applied the Section 5 code to Irish death case data where it performs well on a long time series of 1.5 years, matching the timeframe of the analysis in the paper. However, at pandemic outset where data is sparse and the pathogen poorly understood, how well would the approach work from an advisory perspective? The first lockdown was announced in the UK in March 2020; however, we note that the paper is reliant on a metanalysis (McAloon et al., 2020) that was published in July 2020. How would the authors’ approaches have performed using contemporaneous estimates of infection to death duration and data available at time of first lockdown in March 2020? We also wonder how the authors’ framework would perform for diseases characterized by abrupt changes in the disease dynamics, especially to reconstruct the change in Rt (Section 5 of the paper). Can the smoothness assumption of the model lead to underestimating the sharpness of real changes, causing the estimated Rt change to appear as though it starts earlier and ends later than may actually be the case? There is, however, considerable food for thought in the paper. It is undisputed that countries worldwide were not as prepared as they may have been. Given the societal upheaval of COVID-19, preparation for the next pandemic is vital towards ensuring economic and social stability. Despite the paper offering many interesting critiques and discussion points, we felt there was some missed opportunity in providing concrete guidance on what might be done differently when such a crisis scenario arises again. In such circumstances, could retrospective analysis like that reported in Section 5 provide useful insights? For example, if a novel pathogen emerges for which a vaccine is not available, what is statistical best practice to inform the public and policy makers on foot of the authors’ analysis? How should the population be sampled to estimate case incidence? How should key transmission settings be identified in real-time? None declared. The author replied later in writing as follows:

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.