select search filters
briefings
roundups & rapid reactions
factsheets & briefing notes
before the headlines
Fiona fox's blog

expert reaction to unpublished paper modelling what percentage of the UK population may have been exposed to COVID-19

An unpublished modelling study, from the University of Oxford, reports that up to half of the UK population may have already been exposed to COVID-19.

 

Dr Simon Gubbins, Group Leader – Transmission Biology, The Pirbright Institute, said:

Is this good quality research?

“Yes.

Does it show that over half the UK population has been infected with COVID-19?

“No.  What it shows is there are scenarios consistent with the available data in which a high proportion of the population (68%) could have been infected with SARS-CoV-2 by 19 March.  However, the estimates for proportion of the population infected depends on the assumptions made about proportion of the population at risk of severe disease, which is unknown.  The high level of infection is predicted only if this proportion at risk is small (0.1%).  If the proportion at risk is 1% (another scenario consistent with the data), the proportion infected by 19 March would be much lower (36-40%).

Are there any limitations or caveats to be aware of when reporting this work?

“The authors use the reported deaths from COVID-19 in the UK and Italy to back-calculate the number of people infected with SARS-CoV-2.  This uses the fact the number of deaths is (approximately) proportional to the number of infected individuals at some earlier time (in this case around 17 days previously).  Back calculation requires an estimate of the proportion of the population at risk of severe disease (and so death), but this cannot be estimated as part of the analysis, so it has to be assumed.

“The model treats the UK and Italy as a single well-mixed populations.  This means the model will overestimate the rate of spread and, hence, the proportion of the population infected.

What does this paper tell us about asymptomatic infection and mortality rate?

“Nothing much about either.

Why do these result appear to be so different to other studies?

“Most other studies analyse the reported cases and use these to estimate things such as the reproduction number, the serial interval (i.e. the time between a person developing symptoms and the people they infect developing symptoms), the generation time (i.e. the time between a person becoming infected and them infecting others) or the incubation period.  In this study, the focus is on analysing reported deaths and extrapolating from these to the underlying level of infection in the population.

Does this highlight the need for large scale serological surveys?

“Yes.  Serological surveys are required to assess the proportion of a population infected.  They are also essential to accurately assess the mortality rate.

Is there any over speculation?

“Yes and no.  It’s more that the study presents scenarios consistent with the available data than making predictions about how many people have actually been infected.”

 

Prof James Naismith, Director of the Rosalind Franklin Institute, University of Oxford, said:

“This theoretical simulation rests on a key assumption which may be or may not be correct.  The work is a contribution to the scientific debate and science often advances by challenges to what seems perceived wisdom.  The paper calls for widespread serological testing and this will be necessary to test the paper’s hypothesis.  The need for, the science behind and plans to implement such serological testing are accepted and moving forward across the globe.  This will take time.  At this moment, nothing in paper calls for or could be used justify any change in current policy that is unless we all follow the current government advice on social distancing, the UK will see many thousands of deaths that could have been avoided.”

 

Prof James Wood, Head of Department of Veterinary Medicine, and researcher in Infection dynamics and control of diseases, said:

“This work is some simple modelling that tries to infer infection rates by fitting models to observed mortality.  The work models one of the most important questions – how far has the infection really spread – in the total absence of any direct data.  The authors acknowledge that their results are very sensitive to the assumptions that they have made, but still draw conclusions from the model fit.  The work merely makes assumptions about asymptomatic infection and mortality rates, but cannot measure them.

“The results simply should be used to emphasise the need to conduct serological studies in areas where epidemic spread has occurred to determine what is needed in the way of ongoing controls, rather than to infer that large proportions of the populations are already infected.  The current version of the paper does substantially over-speculate and is open to gross over interpretation by others.”

 

Prof Paul Hunter, Professor in Medicine, University of East Anglia, said:

“The paper by Lourenço and colleagues present an SIR (Susceptible-Infected-Recovered) model that they calibrate to the epidemic trajectory in both Italy and the UK using Bayesian approaches. Using a range of assumptions, they conclude that already a large proportion of the UK population, possible up to 68% may already have been exposed to infection. If the authors are correct this would have potentially large implications for the utility of our current control strategies.

“A key point, admitted by the authors, is that we actually do not have any hard data on the proportion of the public who have been infected but not severely enough to be diagnosed. In that we are in agreement. We do desperately need to know what proportion of the population have been infected and have not become ill, and whether those people develop immunity and if they do not become ill whether they pose a risk of infection to others.

“However, in my view the model presented by Lourenço and colleagues suffers from a number of key failings which make me doubt its utility.

“My main criticism is that a very simple SIR model that assumes complete mixing of the population which is almost always wrong at a country level. We do not all have an equal random chance of meeting every other person in the UK, infected or otherwise.  Ro is also a very clumsy post-hoc measure of disease multiplication and is likely to vary with time and with the complexity of the social networks through which the disease is spreading. In essence this is a simple model that is being back-fitted, albeit in a sophisticated way, to one measure of disease outcome that is itself dependent on different processes that do not appear to be captured in the model.

“The authors state that “Our overall approach rests on the assumption that only a very small proportion of the population is at risk of hospitalisable illness”. This is a big assumption and it is far too early in the epidemic to know what this value is. It is also likely to relate to differences in immunity (that vary with age) presence of comorbidities (which also tend to increase with age) and network connectivity (also age dependent) all of which will vary with the progression of the epidemic. 

“As far as I can tell, the model also assumes that all those infected, whether they are asymptomatic, mildly ill or severely ill are equally infectious to others. This is almost certainly false. Also, the model does not take any account of differences in risk, severity of illness, asymptomatic transmission and possibly infectivity with age or other comorbidities that are clearly impacting on the mortality rates we see.

“Personally, I accept that this model might well generate matches with the observed epidemic trajectories to date, but in my view it should not be given much credibility and should certainly not influence choice of strategies for mitigating the spread of Covid19 or predicting the ultimate size of the epidemic in the UK should it have been left to run its course.”

 

Prof Mark Woolhouse, Professor of Infectious Disease Epidemiology, University of Edinburgh, said:

“The paper by Laurenco et al. reports an mathematical exploration of one of the huge unknowns about the epidemiology of COVID-19: the frequency of undetected infections. It is a pre-print and has not yet been subjected to peer review.

“There are two very different possibilities. 

“First, as the National Health Commission of China and the World Health Organisation apparently believe, there is the ‘what-you-see-is-what-you-get’ hypothesis. This assumes that COVID-19 (similar to SARS) occurs mainly as symptomatic infections.

“Second, there is the ‘tip-of-the-iceberg’ hypothesis that suggests that there are many times more cases occurred than were detected. This has been proposed before based on epidemiological analyses. Laurenco et al. favour this hypothesis but take it further by estimating that the fraction of hidden cases could be very large indeed: more than 1000 for every death.

“This is a legitimate idea but is very hard to prove by fitting models to early-stage epidemic data, which is what Laurenco et al. attempt. It therefore remains a hypothesis rather than fact. As the authors say, a proper test will come from serological surveys – which will tell us how many people have been exposed. Alternatively, the idea could be tested by large-scale surveys of virus genome diversity – this would confirm the much earlier introduction date that is central to Laurenco et al.’s theory (because this allows a longer period of, mostly hidden, exponential growth). Both serological testing and virus genome analysis are planned in the UK.

“If Laurenco et al. are correct this would have huge implications. It would imply that the main reason why COVID-19 epidemics peak is the build-up of herd immunity. Though that would not change current policy in the UK – which is focussed reducing the short-term impact of the epidemic on the NHS, it would change enormously our long term expectations, making a second wave significantly less likely and raising the possibility that the public health threat of COVID-19 will diminish all around the world in the coming months.

“So this paper is useful in highlighting a major unknown about COVID-19: are the cases we see only a very small tip of a very large iceberg? Though this paper does not give a definitive answer to this question we should have a definitive answer in the coming weeks.”

 

Dr Simon Clarke, Associate Professor in Cellular Microbiology, University of Reading, said:

“We need to know much more about the immunology of this virus before we make such confident predictions. It was just over a week ago we were being told that infecting around 60% of the population would be enough to confer herd immunity which would protect everyone else so much more research needs to be done to carefully assess whether these predictions are correct and what level of immunity people have once they have been infected and recovered from the virus.”

 

Prof Rowland Kao, Sir Timothy O’Shea Professor of Veterinary Epidemiology and Data Science, University of Edinburgh, said:

“This model addresses a critically important issue – how many individuals might have been already exposed to Covid-19, when we know that only a small proportion will display clinical signs and an even smaller proportion hospitalised. What the modelling work shows is that a very good fit to the epidemic curve can be achieved when assuming that a large proportion of the population has been exposed. This model usefully highlights the urgent need for more extensive testing to be done.

“However, the exact figures must be viewed with caution. The analysis is based on an abstract model which does not account for spatial distribution of infection, and neither the countries in this analysis (Italy or Great Britain) has presented an even geographic distribution of infection. This unevenness of exposure must be taken into account before the figures being presented here can be considered to be broadly applicable. in areas which have already been exposed for a long time, that also have a higher density population with likely higher levels of contact, such as London, the suggestion of high levels of exposure should be taken seriously (though even then the exact proportion exposed remains highly conjectural). However much of the country is likely to be substantially behind London in the course of the epidemic. In those other areas, there is a good chance substantially fewer people are exposed.”

 

Prof Jonathan Ball, Professor of Molecular Virology, University of Nottingham, said:

“This is interesting work, but is hampered by the same issues that impacts on all epidemiological models – they rely on assumptions that at the moment are based on only a paucity of scientific fact about how thus virus transmits.

“The reliable way to answer the really important question about levels of exposure is to carry out serology-based studies – Detecting the presence of specific novel coronavirus antibodies in the wider population will give us the real answer.

“This is key data as it tells us about the rates or serious disease and death, and will also give an accurate idea of future waves of infection.”

 

* https://www.dropbox.com/s/oxmu2rwsnhi9j9c/Draft-COVID-19-Model%20%2813%29.pdf?dl=0

 

All our previous output on this subject can be seen at this weblink:

www.sciencemediacentre.org/tag/covid-19/

 

Declared interests

None received.

in this section

filter RoundUps by year

search by tag