The Office for National Statistics (ONS) have published more results from their COVID-19 Infection Survey, this time on the characteristics of people testing positive for the disease in England.
Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:
“This detailed set of results from statistical analyses of data from the ONS infection survey comes out nearly six weeks after the previous similar report. There’s a lot of interesting information, but some of it isn’t easy to interpret. There are several changes from the conclusions in the previous report. Some changes arise simply because the infection survey has been going on longer and more data have been collected. Some differences exist because the numbers in the two reports relate to different times, and there’s statistical evidence that things have changed. And some differences arise from different approaches by ONS and their collaborators in analysing the data.
“As in most of the ONS weekly reports on the infection survey, this report covers two different types of test carried out on a representative sample of people from the community population of England. (As always, they do not cover people living in communal establishments such as care homes or prisons – that’s a very small proportion of the whole population, but a larger proportion of very elderly people.) The two types of test are swab (antigen) tests, that test for a current infection with the new coronavirus, and antibody tests, that test whether someone was infected in the past and has developed antibodies against the virus.
“On swab tests, this report looks at how the chance of having a positive test is related to various individual characteristics – age, ethnicity, household size, and several more. These characteristics were also investigated in the previous report from early July, but this time the analysis covers a different period of time (8 June to 2 August – the previous report covered the time from when the infection survey started in late April until 27 June). Also, ONS have changed their method of analysing the data. In the previous report, they looked at associations between personal characteristic and the chances of a positive test one at a time, without allowing for other characteristics in their analysis of any one specific characteristic. This time, they use a statistical approach that allows them to adjust the estimates for one characteristic in the light of other characteristics. I think this is a better way of doing the analysis, generally. For instance, the previous analysis provided some (rather weak) evidence that the rate of positive tests was lower in smaller households (with one or two people) than in larger ones. But this arose from looking at household size without taking other characteristics into account. It would seem very likely that the ages of people living in smaller households tend to be quite a bit different from those of people in larger households, so that the observed difference in positive tests might be to do with age – or perhaps some other characteristic – rather than directly with household size. But the previous analysis could not take account of this possibility. The new analysis, this month, does take that into account, and it found statistical evidence that the rate of positive swab tests is higher in one-person households than in all the other household sizes, when all the other characteristics are taken into account. That is, if one could compare people living in one-person households with people living in larger households, while the other characteristics (age, gender, ethnicity and so on) of people in both sets of households were identical, one would find a higher rate of positive tests in the one-person households than the others. This makes sense for a lot of reasons, I think, and it’s a more informative comparison than the previous one that didn’t take any other characteristics into account. It could be that the difference between this report and the previous one, in relation to household size, also has something to do with a change in the dates that are covered, but I think this unlikely (and that possibility isn’t put forward by ONS as a reason for the difference between the two reports).
“Also on swab tests, the new analysis found statistically significant evidence of a higher rate of positive swab tests in people of Asian or Asian British ethnicity compared to people of White ethnicity, after taking the other personal characteristics into account. No other differences on swab test results between ethnicities had good statistical evidence, though that may (to some extent anyway) be because these results come from a survey, and there is considerably more statistical uncertainty about the rate of positive tests in some ethnic groups because the numbers of people surveyed in those groups are not high. And the survey also found a greater rate of positive tests in people who reported that they had had contact with known or suspected COVID-19 cases – that’s hardly surprising. But there wasn’t evidence of differences between males and females, or between different age groups.
“A separate analysis of the swab test results, just based on data from people of working age (16 to 74), found similar patterns as for the analysis of data from everyone (aged 2 and over) for the characteristics that I have just discussed. But it found no statistically significant evidence, after allowing for the other characteristics, between people who worked in different types of location (inside or outside the home), or between patient-facing healthcare workers and others. Previous analyses had found higher rates of positive tests in patient- or resident-facing health and care roles. ONS suggest, on this particular comparison, that the change in findings may well be due to a real change over time rather than arising because of the change in analysis method, with infection rates having decreased in patient-facing workers relative to others in recent weeks. Assuming that really is the position, I could speculate that it’s because there are many fewer people with COVID-19 in hospitals and care homes than was the position in May, and also because availability of PPE has improved – but this survey can’t give direct information about that.
“(It’s perhaps worth mentioning that one potential technical statistical concern about the new analysis method is, in fact, not important for these results. The new analysis of swabs reports ratios of odds for the different characteristics, whereas the previous one looked directly at the chance (probability) of a positive test in different groups. You might be aware that comparing odds rather than probabilities can, under some circumstances, give different numerical answers from one another, so that a doubling in odds doesn’t correspond to a doubling in probabilities. But that isn’t actually an issue here, because the number of positive tests is low as a percentage of all the people in the sample.)
“One final point about the swab test results – as in previous analyses, this one shows that a considerable majority of people who had a positive swab test did not report any COVID-19 symptoms, either on the day of their positive test or on the days of their previous or subsequent tests (which would typically have been the week before and the week after the positive test). This time, that applied to 72% of those with a positive swab test, though statistical uncertainty means that the data are consistent with a percentage between 64% and 78%. ONS give reasons why, in fact, some of these people might not have been entirely asymptomatic, but the data do tend to indicate that being infected without having noticeable symptoms is likely to be common. For this particular analysis, the statistical method was not changed (so that other personal characteristics were not taken into account), and, as in previous analyses, this one covers the whole period from when the infection survey began in late April. (Compared to the previous analysis, this one does include slightly more than a month’s worth of extra swab tests, but the percentages are not hugely different from those in the previous report.)
“There’s more detail on the analysis of the antibody test results than there was in the previous report from early June. Here, though, the method of statistical analysis hasn’t changed – associations between each personal characteristic and the chance of a positive antibody test are considered separately, without any adjustment for other personal characteristics. Again, the data used over the period right through from when antibody testing began in late April, up to late July. The estimates about antibody testing are less statistically precise than for swab tests, because not everyone in the survey sample is tested for antibodies so that the numbers of tests are lower. But the extra month’s worth of antibody tests has allowed ONS to look at a wider range of characteristics than in the previous report – that only reported on differences in rates of positive antibody tests by ethnicity. The new report finds statistically significant differences in rates of positive antibody tests, by age (with lower rates of positive antibody tests in older people, indicating that a smaller percentage are likely to have been previously infected), by ethnicity (with lower rates in White people than in the other ethnic groups taken together), and by whether people work in patient- or resident-facing health and care roles (with people in those roles having higher antibody positivity rates, not surprisingly given their work during the height of the pandemic at a time when the number of infected people was high and PPE was not always adequately available). But there isn’t statistical evidence of a difference in positive antibody test rates between men and women. In general terms these patterns match those reported on antibody tests from the big REACT-2 study last week.”
Prof James Naismith, Director of the Rosalind Franklin Institute, and Professor of Structural Biology, University of Oxford, said:
“This release is very informative. It suggests around 70 % of infected people have had no symptoms. Whilst this is good news since for the vast majority COVID-19 is not a serious illness, it makes it much harder to spot its spread. For elderly, the medically vulnerable and the unlucky it is a very serious disease with a significant risk of death. It is of concern that some ethnic minorities are bearing a heavier disease burden, more needs done to reduce this inequality. Women are as likely to be infected as men, but we know women are less likely to fall seriously ill or die from COVID-19. I note that people aged 20 to 49 were as likely to be infected as people aged over 70. Of course, there are more people aged 20 to 49, but lowering the infection rate in those above 70 should be a priority. A person living on their own, is more likely to test positive.”
* Coronavirus (COVID-19) Infection Survey: characteristics of people testing positive for COVID-19 in England, August 2020
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 member of the SMC Advisory Committee, but my quote above is in my capacity as a professional statistician.”
None others received.