October 9, 2020

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A preprint, an unpublished non-peer reviewed  study, reports on the latest data from the REACT-1 study on COVID-19 spread across England.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

“This preprint covers the full results of Round 5 of the REACT-1 survey, unlike the interim report that came out last week, which covered only about half the results of that round.  In that interim report, several of the estimates had wide margins of error, in statistical terms, because they were based on smaller amounts of data from Round 5 than this week’s report.  But, allowing for that, most of the key results from the full round don’t differ all that much from those in the interim report, except that the statistical uncertainties are less.

“These results apply to England only. But unlike data from testing of people within the Test and Trace programme and other testing schemes in hospitals and similar, REACT-1 tests a large representative sample of people from the English population.  Therefore its results don’t depend on the availability of tests to the general public, and aren’t affected by changes in availability or in the types of people in the general public who are being tested.  The results are therefore generally reliable, in my view.  Since they are based on a survey, and since not everyone in the country is tested, there is inevitable statistical uncertainty in the estimates.  But another important strength of the REACT-1 survey, and other reports based on properly conducted representative surveys (such as the ONS Infection Survey), is that they can give good estimates of how statistically uncertain the main estimates are.  In REACT-1, this is done by giving so-called credible intervals that show the range of numbers that are consistent with the survey results.

“The most positive news from the interim report, arguably, was that the researchers estimates that the R number seemed to have fallen slightly from the high value based on their previous data. In very broad terms, that’s still the position from the full data.  In the previous round of REACT-1, for which the swab tests were done between 20 August and 8 September, the R number was estimated at 1.7 (with a credible interval, showing the statistical uncertainty, from 1.4 to 2.0).  In Round 5 for which the swabs were taken between 18 September and 5 October, the R estimate is 1.16 (with an interval from 1.05 to 1.27).  The main estimate is not so much different from the estimate, 1.06, in the interim report, taking into account that the interval showing the uncertainty in that interim figure was very wide, from 0.74 to 1.46.  The reduction in uncertainty reflects the facts that the number of samples for the full round is much larger (175,000, compared to about 85,000 in the interim report), and that data are available for a longer run of days, allowing a more precise estimate of how fast things are changing.  The new estimate is smaller than the Round 4 estimate from late August and early September, and the intervals showing the uncertainty around them do not overlap. It is also smaller than the estimate from putting Rounds 4 and 5 together (showing the change over a longer period of time), which is 1.39 (with interval 1.34 to 1.43).

“However, the smaller statistical uncertainty in the estimate from the full round does make it very unlikely that R is now below 1.  (The interim data did still allow that as a possibility.)  So the epidemic is continuing to grow, though, it seems, a bit more slowly than it was growing in late August and the first part of September.  The REACT-1 team are now estimating that the number of infected people is doubling in 29 days, but there’s quite a lot of statistical uncertainty about that, with the interval that shows the uncertainty ranging from 17 days to 84 days.  The survey can’t directly give reasons as to why the R number has decreased, or why the doubling time may have increased, so it can’t directly tell us whether local lockdowns and other measures are effective – but the slowing in the rate of increase does indicate that something is having a positive effect.

“The researchers estimate the national prevalence of positive tests as 0.60%  – that is, 6 people in every 1,000, or about 1 in 170 people in England testing positive.  The credible interval for that rate is 0.55% to 0.71%.  They estimate that this means 450,000 people in England are infected on any one day (interval 410,000 to 530,000).  This estimate does depend on an assumption about the false negative rate of the test (that is, the percentage of infected people whose test would, wrongly, be negative), and the figure they use for that, 25%, is higher than some other studies have concluded.  But in any case, even if that estimate of the false negative rate is too high, this indicates a high number of infected people, and considerably higher than in previous rounds.  The new report presents data on the prevalence rate (the percentage of people who test positive) in a large number of different subgroups of the population (based on age, gender, region, type of employment, ethnicity, household size, and so on).  They give data on more than 50 such groups, and in all but one of them, the prevalence is higher in the latest round (Round 5) than it was in Round 4, and generally a lot higher.  (The single exception is a very small subgroup, based on whether someone’s child had had hospital contact, and there were no infections in that subgroup in either round.)  There is quite a lot of statistical uncertainty about these infection rates in the smaller subgroups, but the fact that they have pretty well all show increases in infections is telling.  One of the clearest differences between different parts of the population is the difference between regions, where the new data reflect what has been reported from other sources – much higher infection rates in the North of England (and to a lesser extent, the Midlands and London) than in the South of England (outside London).”

Dr Jonathan Pearson-Stuttard, School of Public Health, Imperial College London, said:

“The REACT study along with the ONS infection survey data provide the most comprehensive estimates as to the true level of COVID-19 across the country therefore today’s findings are  very concerning.  Particularly, the 8-fold increase in infections in those 65 years or older who are most risk of hospitalisation and poor outcomes from the virus is likely to result in rising hospitalisation and strain on the NHS over coming weeks.”

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

www.sciencemediacentre.org/tag/covid-19

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.”

None others received.