Research, published in the International Journal of Clinical Practice, reports on the number of people on the UK people who are likely to have already been infected by the COVID-19 virus.
Dr Konstantin Blyuss, Reader in Mathematics at the University of Sussex, said:
“There are several problems with this study. The first concerns the data that have been used, namely, the data analysed in this study only covers the period up to 23rd April. On the basis of those limited data, the study suggests the peak of the number of daily new confirmed cases occurred on the 8th April 2020, and the numbers of new cases has been below 4,000 since that data.
“Both of those conclusions directly contradict what the current data show, with the numbers of new cases exceeding 4,000 for most of April and the first week of May.
“A related problem is that this study based their estimates on 73,000 reported cases – this number now stands at a staggering 233,000, i.e. three times that earlier estimate, which can have a huge impact on the estimates of disease prevalence.
“The second issue concerns the methodology. The value of R is notoriously difficult to estimate, and as a result, the estimates always have a wide margin of error, which means that it is almost impossible to rely on accurate estimates of R for any significant population-wide conclusions.
“There have also been significant criticisms of using stepwise regression in statistical literature due to various biases and inherent shortcoming this methodology has, so relying solely on it as a main tool for drawing meaningful conclusions may not be justified.”
Dr Adam Kucharski, Associate Professor in Infectious Disease Epidemiology, London School of Hygiene & Tropical Medicine, said:
“In the absence of direct measurements, models can be useful for estimating the extent of infection from wider data sources. However, given how difficult it is to estimate the extent of unreported cases in a population from reported cases alone, it is likely that there is huge uncertainty in the estimates produced by the model used in this paper, and unfortunately this uncertainty is not reflected in the single value quoted in the paper and the press release.
“In addition, we now have direct measurements of infection from antibody testing in several countries, and the values found are generally much lower than the one suggested by this modelling analysis: one recent study found 5% had antibodies in Spain overall, another estimated 2% in Luxembourg, another 10% in Geneva. Even in areas that have been severely affected by COVID-19, the proportion of the population with evidence of past infection is so far relatively low: 10% in Wuhan, 10% in London, 11% in Madrid, 14% in Gangelt, Germany; 21% in New York.
“The only serological study the authors cite is a study from Santa Clara, California, which has received substantial criticism for likely overestimating the actual extent of infection in the population.
“Given how much antibody data is now emerging, it is increasingly important to focus on measurements rather than just modelling estimates.”
‘A phased approach to unlocking during the COVID-19 pandemic – Lessons from trend analysis’ by Mike Stedman et al. was published in the International Journal of Clinical Practice.
All our previous output on this subject can be seen at this weblink: