select search filters
briefings
roundups & rapid reactions
Fiona fox's blog
The Office for National Statistics (ONS) have published the latest update of the initial data from the COVID-19 Infection Survey.
Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:
“The ONS COVID-19 infection survey will soon move out of its pilot phase, and the weekly reports from it do always add a little more to the picture of infection across the whole community population of England (though not for people in hospitals, care homes, or other institutions). I know of no other current data source that provides evidence across the population in a representative way, and I continue to be impressed that ONS and its partners are doing this important work so rapidly and that they are being transparent about their findings and methods. But of course there are limitations – the data come from a survey and all survey results have an inevitable margin of uncertainty (because not everyone in the country was tested), and the number of people that it has so far been possible to test is still too small for the results to be very informative on certain aspects. But that will improve as more time passes and more people can be included in the survey sample. Also, the report makes it clear that not all the swabs sent for testing in the most recent week have yet been included in the results, so some details may change as results on these swabs are received.
“Most of the data analyses in this release are the same as in last week’s release, and the extra week’s data has not made a huge difference to several of the conclusions. But there are some interesting changes. Perhaps the most interesting change is that there’s some evidence of a decrease over time in the percentage of the population that actually has a SARS-CoV-2 infection. This comes from two different statistical analyses. One of them compares the percentage of positive swab tests in two non-overlapping periods of time (3 to 16 May and 17 to 30 May). In the first of these two-week periods, 0.25% were infected, that’s 25 people in 10,000. In the second, this had fallen to 10 in 10,000. In terms of the number of infected people in the whole community population in England, that fell from 139,000 to 53,000 – a decrease of more than half. We’ve got to be careful in interpreting this though – there’s quite a lot of statistical uncertainty in both these numbers, and the most recent one is provisional because not all of the relevant swab test results are included yet. It would appear that this decline in infection rate across the two fortnights is not statistically significant – that is, it might possibly be explainable by the statistical uncertainty that arises because the data come from a sample. However, there’s another new analysis in the report that shows something similar, and is reported to be statistically significant. This new analysis is experimental, developed by ONS’s partners at the Universities of Oxford and Manchester, and all the details have not yet been published. Also, it may also be affected by the provisional nature of the most recent data. But the methods used do look statistically appropriate to me, on the basis of the limited amount I know about them so far. The results do show a reasonably clear trend of a decline in the number of people infected each week during May.
“That’s very encouraging. The end of the period for which ONS now have data, 30 May, does include a time when some loosening of the lockdown had occurred. But there has been more loosening since, with (for example) the re-opening of some schools on 1 June. Will this survey be able to pick up any important changes in infection rates that happen after these more recent changes in lockdown? The survey results do have an advantage compared to, say, the weekly ONS figures on death registrations, because they look at current infection, and people who unfortunately die from the infection will not show up in death statistics until a certain length of time after they were infected. But the infection survey results are based on a sample of people, and though the sample is getting larger and larger over time, the statistical margin of error may initially make it difficult to pick up small changes. We’ll see in future weeks.
“The analysis of whether people who had a positive swab test showed any symptoms is repeated from last week. There are some relatively small increases in, for instance, the percentage of the people who have a positive test that report symptoms on the day of their test, but these do not really change the overall picture. Of people that tested positive at any time over the whole study period (now 26 April to 30 May), 29% reported one or more symptom at some point in the whole study period. That’s subject to a lot of statistical uncertainty – the 95% confidence bounds run from 19% to 40% – but it still indicates that the majority of people who tested positive did not report any symptoms. These numbers are changed very little from last week’s, when 30% of those who had tested positive had reported one or more symptom, and the change is well within the statistical margin of error. That’s hardly surprising, given that the numbers are based on only 88 individuals that tested positive – which is only one more individual than in last week’s report, indicating that there was only one new positive test in the most recent week.
“The other analyses looking at subgroups of the population (by age, by gender, by the type of work) again look much the same as last week’s. As before, because the numbers of people in the survey in these subgroups are mostly a lot smaller than the total number surveyed, there is more statistical uncertainty than in the overall figures on numbers infected, and differences between subgroups in the survey have to be quite large before they provide adequate evidence that there is a genuine difference in the community. So, for instance, there’s inadequate evidence of any differences in infection rates between men and women, or between different age groups. This doesn’t mean that no such differences exist – maybe they do, but this survey hasn’t been able to pick them up. There is, as last week, statistically significant evidence that health and care workers in patient-facing roles have a higher rate of infection than other working-age people.
“I won’t comment this week on the results of blood antibody tests (the tests that give an indication of whether a person was infected with SARS-CoV-2 at some previous time), because this new release incudes no new antibody testing data.”
Prof Sheila Bird, formerly Programme Leader, MRC Biostatistics Unit, University of Cambridge, said:
On survey size
“The 5th report from the ONS Infection Survey looks at how the percentage of people testing swab-positive for COVID-19 has changed from 3-16 May (weeks 1 and 2) to 17-30th May (weeks 3 and 4). During that time it has gone from 2.5 per 1000 people (95% CI: 1.2 to 4.6) to 1.0 per 1000 people (95% CI: 0.5 to 1.8), a reduction of 60%.
“The all-important 95% confidence intervals per fortnight do overlap but the difference approaches statistical significance and, when analysed for week-on-week trends, there was a steady and unambiguous decrease in positive tests, from 3.1 per 1000 people in week 1 (95% CI: 2.0 to 5.0) down to 1.1 per 1000 people in week 4 (95% CI: 0.6 to 1.9).
“A key consideration now is whether to increase the monthly survey-size if prevalent infections were to decrease further. Why? Because we need a 50% increase or 50% decrease in prevalent infections to be discerned powerfully, including by the yardstick of statistical significance.
On intelligence gleaned
“Using its data from 26 April to 30 May 2020, the ONS Infection Survey has nailed the increased 5-week infection-risk associated with working outside the home (7.2 per 1000 people, 95% CI: 4.7 to 10.5) as compared to working from home (2.2 per 1000 people, 95% CI: 1.0 to 4.1). Hence, even as lockdown eases, if you can work at home, please continue to do so!
“Also nailed, and contrary to the impression given in the recent Public Health England report on ethnicity and other risk factors, is the substantially higher 5-week infection-risk for individuals working in patient-facing healthcare or resident-facing social care roles (18.7 per 1000 people, 95% CI: 10.7 to 30.2) versus those of us not working in such roles (3.2 per 1000 people, 95% CI: 2.6 to 4.4).
“During 26th April to 30 May, of those individuals in residential households reporting a cough and/or fever and/or loss of taste/smell on the day of test, 77 per 1000 people (95% CI: 39 to 134) were swab-test positive, as were 3.6 per 1000 people who reported no such symptoms on the test-day (95% CI: 2.9 to 4.5). For Test & Trace to pick up early on such asymptomatic infections in quarantined close contacts, and to learn more efficiently about asymptomatic infections, Test & Trace needs to build-in the offer of a randomly-timed swab-test for those in quarantine.
On new infections per 1000 persons followed for 1 week
“Using its data from 26 April to 30 May 2020, the ONS Infection Survey has estimated the new infection/incidence rate as 0.7 per 1000 people (95% CI: 0.5 to 1.0) or 5,600 new infections per day in the community (95% CI: 3,700 to 7,900).”
Prof Keith Neal, Emeritus Professor of the Epidemiology of Infectious Diseases, University of Nottingham, said:
“This set of data shows a number of interesting a key findings.
“The rate of infection continues to decline and is half what it was 2 weeks ago. Changes over a longer period are now statistically significant. This is highly compatible with the fall in diagnosed cases. The contact tracing service has more than enough staff to cope with the current level of infection. The main problem is people not getting tested for COVID-19 when they have symptoms.
“Those working outside the home had higher risks but that includes those in patient-facing occupations.
“The most important finding is the risk to those in patient-facing occupations have a much higher rate of COVID-19 detection. This significantly higher figure suggests that a significant part of the current COVID-19 epidemic may be an infection control issue rather than a series of community outbreaks. The control strategies are different in hospitals and care homes which requires infection control whilst community transmission requires social distancing and contact tracing. We are given an overall R but what is important for the lifting of some restrictions is the community R.
“The main limitation of the study is the small numbers testing positive gives wide confidence intervals. No study is perfect but by following the same group of people it is a very powerful tool to identify the trend.”
‘Coronavirus (COVID-19) Infection Survey pilot: 5 June 2020’ and ‘Analysis of death registrations not involving coronavirus (COVID-19), England and Wales: 28 December 2019 to 1 May 2020’ were published on Friday 5th June 2020.
Declared interests
Prof McConway – “Prof McConway is a member of the SMC Advisory Committee, but his quote above is in his capacity as a professional statistician.”
Others – None received