January 21, 2021

expert reaction to preprint giving the interim results from round 8 of the REACT-1 study on COVID-19 spread across England

A preprint, an unpublished non-peer reviewed study, reports on the latest data from the REACT-1 study on COVID-19 spread across England.

This Roundup accompanied an SMC Briefing.

Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:

“It’s good to see another set of results from the REACT-1 infection survey, run from Imperial College with collaborators from several other academic institutions and Ipsos MORI.  The survey had a bit of a break – the last round collected data up to 3 December, and the latest results, from round 8a (the first part of Round 8) started in earnest on 6 January.  That’s a considerably longer break than there has been between other REACT-1 rounds, and it happened when some other data sources on the progress of the pandemic also had gaps in their series of results over the Christmas and New Year break.  You might think that this doesn’t matter much, because we get daily data on confirmed cases of Covid-19 (on coronavirus.data.gov.uk) and that shows how the pandemic is changing.  But there are sometimes issues with the numbers of confirmed cases, that make them hard to interpret and potentially misleading.  Nobody can be counted as a confirmed case unless they have had a positive test for the virus, and so the number of confirmed cases depends on how many people are tested, which in turn depends on why people are asking for tests.  That will depend on whether they have symptoms, where they live, what job they do, and so on.  However, for REACT-1 a representative sample of the community population in England is tested for current infection, regardless of whether they have symptoms or where they work, but only to check the progress of the pandemic.  So the REACT-1 results, like those from the ONS Infection Survey, can keep a check on whether the confirmed cases numbers are truly showing what’s happening.  What is more, because the survey researchers ask more questions about the people included in their sample, REACT-1 (and the ONS survey) can tell us more about what type of people are being infected than we can learn from the confirmed case numbers.

“Another reason why I’m pleased to have these new results is that they cover swabs taken in a period (6-15 January) for which we don’t yet have results from the ONS survey.  Its latest published results go up to 2 January.  Last week’s bulletin on the ONS survey did not appear for technical reasons (but another one is promised on Friday this week).

“The most striking result from the new REACT-1 report is one that was clearly known from the confirmed case numbers – that the number of infected people in England has risen a great deal since the last REACT-1 report (which covered 25 November to 3 December).  Overall, the estimated number of infected people went up by 74% between the two rounds.  In fact the infection rate increased in nearly all of the subgroups of the population that the researchers looked at.  This time they looked at 46 different subgroups, based on gender, age, the region where people lived, household size, type of employment, level of deprivation of the area where they live, and more.  The infection rate went up over December in all but three of those subgroups, and often by a large amount.  For people aged 65 and over, the rate of positive tests went up by a huge 130% – that is, it doubled and then had almost another third of the rate from the start of December added on again.  The rate almost trebled in the East of England, and went up by very large amounts in the other regions that we know, from the confirmed case numbers, were badly affected during December – London and the South East.  But the rate of increase in the West Midlands was also very high, according to the REACT-1 researchers, indeed higher than the increase in London and the South East at almost 150%.  The three exceptions, where the rate in a subgroup fell, were in two regions that had been very badly affected in the last round (North East, where the fall was small, 6%, and Yorkshire & the Humber, where it was considerably larger at 40%), and also in people in the ‘Other’ ethnic group, where numbers tested are small and the rates cannot be estimated accurately.  But these results are based on much smaller numbers of people tested than the results for the whole country, so there’s quite a lot of statistical uncertainty.

“Many of those findings do match what was seen in the confirmed case numbers – but that does not apply to everything in these REACT-1 results.  The researchers estimated how numbers of infected people were changing during round 8a, that is, over the very short period between 6 and 15 January.  They found (overall) no strong evidence that numbers were either rising or falling.  That’s an average across the whole of England, and the picture did differ a bit from one region to another, though most of the regional differences don’t have clear statistical evidence.  This finding contrasts with what has been seen in confirmed cases, which fell reasonably consistently in England over the period in question, 6-15 January, from a peak around the turn of the year.  The REACT-1 researchers suggest various reasons for this difference.  First, they point out that the most recent published results from the ONS survey, covering 27 December to 2 January, showed a considerably higher prevalence of positive tests than was found in this REACT-1 round.  Perhaps there could have been a very sharp fall in numbers infected right in the first days of the new year, so that they (roughly speaking) reached a lower level, and levelled off, by the time the REACT-1 swabs were taken.

“Also the REACT-1 report presents some very interesting data from Facebook, on measures of people’s mobility across the whole period.  This shows a marked drop in mobility for the last two weeks of 2020, followed by a rise in the first week of 2021 (though not right back to the levels of mobility in the first part of December).  Perhaps people really were staying at home more over the Christmas and New Year break, despite the fears about too much travel and family meetings that could increase infection, and perhaps mobility increased again at the start of the year as people went back to work after the holidays.  The REACT-1 researchers speculate that higher transmission of the virus might lead to increased new infections, which will show up in the confirmed cases data very soon as a plateau or even an increase.  We’ll see soon enough whether that’s the case.

“Of course, another possible reason for the difference between the REACT-1 findings and the confirmed case trends is that one (or indeed both) might not be right.  On these trends, there’s certainly quite a strong possibility that the REACT-1 estimates are not very accurate.  I’ve previously expressed some concern about how accurate some of these short-term REACT-1 estimates of trend can be, when they are based on changes in the number of positive tests over a short period of time, and when they compare successive days on which the number of tests taken is not large and may not be very representative of the pattern across the country.  They are pretty sure to be less reliable than the estimated from the whole of a round (or even half a round like 8a) on overall infection rates, and infection rates in most subgroups.  This time, the REACT-1 researchers do draw attention to these caveats about their short-term trend estimates, and (rightly) urge caution in their interpretation.  Anyway, we’ll have a better idea soon enough as we get more data on confirmed cases and the next set of ONS infection survey estimates are published.

“The other major use of a survey like REACT-1, apart from checking and triangulating with other data sources on numbers of infected people, is to investigate how people’s different characteristics relate to their chances of testing positive.  In fact, some of those findings are clearer than they were in some previous rounds.  That’s partly because infection rates are now higher, overall, than they were in any previous round, and the higher the number of positive swabs, the more statistically precise the findings can be.  In some cases the differences seem to have become smaller than they were – for instance, though infection rates are still higher in younger age groups (13 to 34) than in older ones, the differences are less than they were in the previous round, showing how infection has spread rather more to the older age groups.  The increase in statistical precision of the estimates makes it clearer that infection rates are higher in health and care workers (particularly), and in other key workers, than in other groups of workers; in people of Asian and Black ethnicity compared to people of White ethnicity; and in people who live in larger households.  However, the differences between people living in areas at different levels of deprivation seem to have decreased since the last couple of rounds.  It’s still true that the more deprived the place you live in, the more likely you are to be infected, but the differences are less than they used to be in November.”

Preprint (not a paper): ‘REACT-1 round 8 interim report: SARS-CoV-2 prevalence during the initial stages of the third national lockdown in England’ by Steven Riley et al. was posted online at 00:01 UK time on Thursday 21 January 2021. This work is not peer-reviewed.

Declared interests

Prof Kevin McConway: “I am a Trustee of the SMC and a member of the Advisory Committee, but my quote above is in my capacity as a professional statistician.”