select search filters
briefings
roundups & rapid reactions
before the headlines
Fiona fox's blog
The Office for National Statistics (ONS) have released the latest data from their COVID-19 Infection Survey, and the government have released the latest estimates for the COVID-19 growth rate and R value.
Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:
“This week’s bulletin on the results of the ONS Infection Survey is particularly timely, for several reasons. There was no bulletin last week, because of technical issues that are now resolved. Because of the current state of the pandemic, as reflected in other data sources, and because of the very severe pressures on the NHS, it’s particularly important at this time to have good data on trends in the pandemic. Looking back slightly longer, data were not always available from all the usual sources over the Christmas period, so there are gaps that need to be filled. Also, the latest results from the REACT-1 survey, published yesterday1 in preprint form, showed some features that did not entirely match the recent trends in the data on confirmed cases of Covid-19 as published on the dashboard at coronavirus.data.gov.uk, so we need as much data as possible to try to see a reasonably clear picture.
“The value of the ONS Infection Survey, compared to the daily data on confirmed cases, is that it gives results on swab tests for the SARS-CoV-2 on a representative sample of people drawn from the community populations of each of the four UK countries. The confirmed cases data are based on people tested for specific reasons – maybe that they have symptoms that might be caused by Covid-19, or they are in hospital (for Covid-19 or something else), or that they live in a place where mass testing of people without symptoms is going on, or they work or study in a particular place or context. If the pattern of people being tested, or the numbers being tested, change over time, then that could lead to changes in the numbers of confirmed cases that do not relate to true changes in the rates of infection. The introduction of mass testing in some places and some contexts, and the wider use of lateral flow devices for testing, may have changed the pattern of people being tested in relevant ways. The ONS survey estimates should not be affected by these things, because the people in the survey are tested only to follow infection patterns across the country. (The same goes for REACT-1.) So it will not be affected by any biases that could arise from changes in the patterns of who is being routinely tested.
“I’ll concentrate in my comment on the patterns in England. That’s because the information from the ONS survey in England is more precise and more detailed. That is simply because the population of England is much bigger than the populations of the other three UK countries, so that the number of people sampled for the survey is bigger in England than in the other countries. So there are more features to explain and comment on in England than in the other countries. Briefly, the rates of infection, as estimated in the ONS survey, are lower in Wales, Northern Ireland and Scotland than they are in England overall, but the trends are in different directions in different countries. In Wales, rates of infection fell from mid-December but now appear to be levelling off; there appears to have been a very marked increase in infections in Northern Ireland since the end of December; and infection rates in Scotland seem to be levelling off.
“It’s important to understand that what’s estimated isn’t the number or percentage of people who are infected. It’s the number or percentage of people who would test positive, if they had a swab test. Every test can produce a results that do not match the true state of affairs; there will be false positives (people who test positive even though they are not infected) and false negatives (people who test negative even though the really are infected). Given what’s known about the chances of these incorrect results, and given the current levels of infection across the whole UK, false positives are likely to be very rare, but false negatives, while still much less common than true negatives, will be more frequent than false positives. The upshot of this is that the numbers and percentages of infected people will be slightly higher than the numbers and percentages of people who would test positive, though it’s not at all easy to say how much higher. But the general pattern of trends in infections over time will not be affected by this complication.
“The latest ONS estimates cover the week from 10 to 16 January 2021. For that week, ONS estimate that 1.88% of the English community population would test positive – that’s over a million people in all (1,023,700 in fact). That amounts to about 1 in 55 of the people in the community (aged 2 and over). Like all estimates from surveys, this is subject to some statistical uncertainty, and the data are compatible with the percentage being somewhere between 1.80% and 1.96% – that’s between 1 in every 50 people and 1 in every 55 (rounded to the nearest 5). That rate is rather lower, however, than the last weekly rate that ONS estimated, for the week 27 December to 2 January. Then, the estimate was that somewhere between 1 in 45 and 1 in 50 were infected. Both of those rates are higher than any other rate of infections estimated for England in the whole infection survey, back to when it began at the end of April last year. Infection rates had risen pretty rapidly for early September to a peak in early November, then fell back a certain amount during the second lockdown in England, but then rose again very rapidly after that lockdown ended at the start of December. So since the peak of infection, which was, according to the survey, at the turn of the year, infections have fallen. The pattern is not as clear as it might be, because there is no official ONS estimate for the week ending 9 January because of the technical issues, and there was also no estimate for a few days over Christmas because swabs were not collected. Another complication is that ONS actually publish three different sets of estimates for the rate of positive tests – their ‘official’ estimates for weekly periods, a set of daily ‘modelled’ estimates (whose accuracy will depend in a complicated way on the details of the statistical model used), and a set of estimates for non-overlapping 14 day periods that use rather simpler statistical methods. The two most recent fortnights for this third set of estimates are for the fortnights from 3 to 16 January, and from 20 December to 2 January, and those estimates do not differ significantly, so they do not show the same decline as in the ‘official’ estimates. However, if the true pattern is of a rapid increase in infection up to about 1 January, and a slower decline after that, these averages across whole fortnights would be expected to be similar to one another, so they are also compatible with that pattern.
“Overall, then, I’d say that there’s a reasonably clear picture of a turnround in the fast increase of infections that was occurring during December. That turnround seems to have started at some time between the very start of January and 10 January, though it’s not clear exactly when. That is, of course, covers when the most recent lockdown in England started. It’s impossible to say what exactly caused the turnround, but at least there’s an indication that the lockdown contributed. This seems to be an indication that the lockdown measures can in fact contain the highly infectious new variant.
“There are some discrepancies, though, between these ONS estimates and data from other sources. The daily number of confirmed cases of Covid-19, in England, fell by almost a third between its peak on January 1 and January 14, the latest date for which a smoothed 7-day average is currently available. (This is based on the averages of 7-day periods ending on those dates, and uses the results based on the dates when someone’s specimen was tested, not the data when the result was reported.) That’s a much greater rate of decrease than is shown in the estimates of who would test positive in the ONS survey. They fell by only about 9% between the week ending 2 January and the week ending 16 January. (Because of the statistical uncertainty of survey-based estimates, the fall could be a bit less than that, or a bit larger.) There are some obvious reasons for the difference, though. Most importantly, the confirmed cases data are for new cases (‘incidence’, in the jargon), while the ONS estimates are for anyone who would test positive, even if they might have been initially infected some time beforehand (so ‘prevalence’, in the jargon). Because in the ONS survey people are tested more than once, ONS can in principle provide estimates of new infections (incidence) as well, and indeed they did provide these estimates until late November. But they have paused that series to improve the statistical methods used. That means that ONS trends in the rate of testing positive aren’t really comparable with trends in the confirmed case numbers. Generally they would go up and down roughly in step, though changes in the direction of the trend would not necessarily occur at the same time – and generally, I’d expect the rate of new cases to change more quickly than the rate of positive tests. So maybe there’s no real underlying difference in the ONS infection survey and confirmed cases trends. I do have a suspicion, though it’s no more than a suspicion, that the trend in the ONS data provides a more reliable picture than the trend in confirmed cases, because there have been some changes in the number and, probably, in the pattern of the routine tests being carried out under Test and Trace and similar programmes. But to check that would require more data than I have.
“One reason for my feeling that the trends shown in the ONS data are probably more reliable than those in the confirmed cases numbers for England is that, in a very general sense, the ONS survey estimates are compatible with the REACT-1 latest results published yesterday. REACT-1 does not publish results so frequently as does the ONS survey. Its latest results are for the period 6-15 January, which is similar (but not identical) to the latest week in the ONS results (10-16 January). But the previous REACT-1 results were for late November and the first few days of December. The rate estimated by REACT-1 was much higher in its latest period than for its previous round in November, which isn’t at all surprising. REACT-1’s results are compatible, as they say in their report, with there having been a peak of infection at the very end of 2020, followed by a fall until their current round started sampling in large numbers on 6 January. That general pattern would match the ONS results.
“However, there are still differences between the two surveys to explain. The ONS estimate for the number testing positive for 10-16 January is 1.88%, or 1 in 55 of the population. The REACT-1 estimate for 6-15 January is 1.58%, or 1 in 65 of the population (to the nearest 5). Both estimates are based on surveys, so there is statistical uncertainty, but the intervals given in the two reports to describe that uncertainty are not enough to explain a difference this large between the estimates. The results from the two surveys, for periods when both are running, are never quite identical, but ONS have helpfully produced a diagram (Figure 2 in their weekly bulletin bringing together data from several sources2) comparing the two sets of estimates, and that indicates that the difference this time is larger than it usually is. I am not entirely sure of the reason for this larger-than-usual discrepancy. It may be because the REACT-1 results are based, so far, on only the first part of their latest round of testing, so that their statistical uncertainty is greater than it sometimes is. Also it’s perhaps possible (though I have no evidence for this) that its sampling was less representative for just part of the round than it will eventually be for the whole round. Another possibility might arise because the way the swabs are taken is rather different in the two surveys, and in previous reports the REACT-1 researchers have indicated that their results would be rather more affected by false negatives than the ONS researchers believe to be the case. But in any case, none of this really affects the picture on overall trends.
“Another discrepancy is that the ONS daily modelled results indicate a decline, albeit a very slow one, in infection rates over the latest week in their data, while the REACT-1 researchers did mention that they had seen a suggestion of an increase, rather than a decrease, in the last few days of their latest period. (This suggestion was more prominent in the press releases than in the actual REACT-1 preprint, and in my view more prominent still in some of the reporting.) My own opinion on this is that neither survey really has enough appropriate data to look definitively at trends over such a short period. The REACT-1 ‘suggestion’ is based on data for just a few days, when the number of swabs tested was relatively small, and may (or may not) have been untypical. Overall their data on changes within their latest part-round is compatible with a considerable decline over that time, or a considerable increase, or anything in between really, and we will have to wait until the report on the full round, and probably even longer than that, before the pattern becomes clear. The ONS modelled estimates shows a clear decline partly because the model is looking at smooth trends over a considerably longer period than just the latest week. So these very short-term differences in trend are, in my view, simply not well enough supported by the data to be worth taking much account of.
“As usual, the ONS bulletin also presents results on subgroups of the English population, by age and by region. A problem with all these estimates is that they are subject to considerably greater statistical uncertainty than the overall figures for England, because the number of people tested within each subgroup is obviously smaller than the number for the whole country. On age, there is clear evidence of decreasing infection in most age groups (secondary school, and all the groups over 34), and levelling off in the two age groups between about 17 and 34. The pattern in children of primary school age and younger isn’t so clear.
“For the English regions, despite the statistical uncertainty, there is pretty clear evidence that the rate of infection has declined in the regions in the south and east where rates were increasing very fast, and were already very high, before Christmas. The ONS data also show clear evidence of falling rates in Yorkshire & the Humber, and in the East Midlands. The picture isn’t so clear in the rest of the country, with some indication that the rate may still be increasing in the South West (though it is still low there). As is often the case, the picture is least clear in the North East region, where there is a great deal of statistical uncertainty because the population size, and hence the sample size for the survey, is considerably smaller than all the other regions.
“But the regional data that interest me most are those that compare the positive tests compatible with the new variant of the virus, and those compatible with other variants. In the regions where, in earlier bulletins at the end of last year, the new variant was not yet dominant, it has continued to increase. That is the case in the North East (though there is a lot of statistical uncertainty), the North West, Yorkshire and the Humber, the West Midlands, and the South West. In all these regions, the ONS modelled estimates show increases in infections with the new variant between 31 December and 16 January, and a decrease in infections with other variants (except in the South West, where infections with other variants increased as well, though much more slowly). In some of them (South West, West Midlands) infections with the new variant now outnumber those with other variants, and in others (North East and North West) the new variant is close to overtaking other variants. To that extent, the patterns in those regions match what was happening in the south-eastern regions (including London) before Christmas – with the very important exception that total infections are not showing signs of the overall rapid increases we saw in London and the surrounding areas. That indicates that the current lockdown, and/or people sticking even more carefully to the rules, may indeed be enough to keep the new variant in check.
“The same applies in the three south-eastern regions (London, East of England, South East) that were worst affected by the new variant – overall, infections are falling there too, and in the East Midlands too. The ONS data, then, do not support fears that the current lockdown isn’t enough to contain the new variant. But there’s something odd about the balance between the old and the new variant in those three south-eastern regions. In all of them, between 31 December and 16 January according to the ONS modelled estimates, infections with the new variant fell faster than infections with other variants. (In fact, in London infections with other variants were at the same level on those two dates. In the other two regions, infections with other variants did fall, but more slowly than infections with the new variant.) As a result, the percentage of positive tests that were with the new variant fell between those two dates. In London on 31 December, according to the ONS model, 78% of infections were with the new variant, but on 16 January it was only 70%. The corresponding figures in the East of England were 71% and 57%, and in the South East, 70% and 67% (so a smaller difference). There’s some evidence of the same kind of pattern in the East Midlands, but the various differences there are really too small to interpret.
“However, I’m really not at all sure why this is happening. I can speculate, but it’s not really more than speculation. It’s possible, perhaps, that there is an explanation to do with the way the statistical modelling works, or the way that positive tests whose result is compatible with the new variant are recorded, rather than this being a real effect. But its consistency over at least three regions may make that unlikely. I suppose one possibility is that the greater infectivity of the new variant changes its transmission more in some groups of people than in others, and the current lockdown restrictions may also change the transmission of the virus more in some groups than others. Maybe those differences between different groups of people interact in a way that means that, in the regions in the south and east where the new variant was running pretty rampant before Christmas, the lockdown has happened to suppress the new variant more than other variants. Another possibility might be that infections with other variants (that is, not compatible with the new variant that was first found in the UK) actually involve different variants in different parts of the country. But I’m no virologist so I will end my speculations there.”
1 https://spiral.imperial.ac.uk/bitstream/10044/1/85583/2/REACT1_r8a_final.pdf and https://www.sciencemediacentre.org/expert-reaction-to-preprint-giving-the-interim-results-from-round-8-of-the-react-1-study-on-covid-19-spread-across-england/
Prof Sheila Bird, Formerly Programme Leader, MRC Biostatistics Unit, University of Cambridge, said:
“The number of SARS-CoV-2 positives per positive household increased in England but decreased in Scotland between non-overlapping 2-week periods at the end of November 2020 and at the start of January 2021. Does this give insight about national adherence to self-isolation? I’m not sure!
“During the first fortnight (3 to 16, re-weighted)) of January 2021, ONS Infection Survey estimated SARS-CoV-2 prevalence in England as 2.13% (95% CI: 2.05% to 2.22%), substantially higher than Scotland’s 1.16% (95% CI: 0.98% to 1.36%).
“My Table contrasts the number of SARS-CoV-2 positives per positive household between England and Scotland for non-overlapping 6-week eras. The ONS Infection Survey is substantially more powerful in England than in Scotland due to the overall number of participant households in the most recent 2-weeks (England: 83,091 versus Scotland: 9,611) and also to England’s higher SARS-CoV-2 prevalence.
“Even so, Scotland’s ratio of positive participants per SARS-CoV-2 household has shifted down in Scotland but increased in England. Does the differential shift tell us anything about compliance with self-isolation within infected households? At best indirectly but may warrant keeping an eye on.”
Dr Yuliya Kyrychko, Reader in Mathematics, University of Sussex, said:
“It is extremely encouraging to note that the latest estimates of R number have reduced slightly from last week.
“We have to be careful, though, with interpreting this is as a firm evidence of the lockdown being successful, because R number estimates are by their very nature based on lagging data, therefore, they rather represent a situation a week or two ago.
“Also, there appears to be a significant variation between different regions of England, with some them actually exhibiting a growth in disease prevalence since last week. This provides some of the reasons for still very high numbers of new daily confirmed cases.”
Dr Konstantin Blyuss, Reader in Mathematics, University of Sussex, said:
“The latest data suggest that the rate of growth of new infections, as represented by the R number, is decreasing in England.
“Of course, this is great news, but we have to be cautious not to overlook the fact that in parts of the country, and more specifically, the North East and the Midlands, the prevalence has actually increased since last week, indicating that the infection is still growing in those regions.
“One particular aspect of the story is that the proportion of cases associated with the new more infectious strain is also changing in time, therefore, we still have to wait and see whether this overall reduction of the R number in England will continue over the coming weeks.”
Prof James Naismith, Director of the Rosalind Franklin Institute, and University of Oxford, said:
“Today’s release of data (ONS and R-values) support the view that we are past this peak of infections. This is true across all the UK’s nations. Scotland has less people infected than the others. London has been hit very hard again. The number of daily deaths in the UK is likely to increase and peak in the next 10 days or so. As we saw with the first wave, cases rocket up but only float back down. Each death will have broken the heats of those left behind. We cannot be numb to the scale of this tragedy.
“The ONS data has some areas of concern that bear watching – in some regions, the decrease in prevalence has slowed and perhaps even levelled off. Hopefully, next week we will see a resumption in decrease. Similarly, some age groups might also show levelling off in the decrease of prevalence. There was no data release last Friday and any slow down in decrease may be an illusion.
“The success of the lockdown can be seen most clearly in the decrease in the new strain. Equally, had we not locked down, the consequences are surely clear. These would have included overwhelming of the NHS in early January with horrific consequences.
“The NHS is under severe stress but has not buckled, a tribute to its staff. Due to the slow decline in cases the NHS will remain under high pressure for some time to come.”
Prof Liam Smeeth, Professor of Clinical Epidemiology, and Dean of the Faculty of Epidemiology and Population Health, London School of Hygiene and Tropical Medicine, said:
“Some cautious optimism is warranted. The data suggest we are beginning to control transmission in most areas, although there are exceptions where a levelling off is less certain, and indeed growth may be continuing, notably in the midlands, northern England, the South West and Northern Ireland. Circulating levels of virus and new infections remain high everywhere, and this does mean that hospital admissions and deaths caused by Covid will remain high for some weeks to come. But yes, the data do suggest that the current measures – while obviously very difficult for many – are beginning to control spread. If we all continue to follow the guidance, we can hope that over the coming weeks we will see continued falls, likely to be helped by the vaccination roll out.”
Dr Simon Clarke, Associate Professor in Cellular Microbiology at the University of Reading, said:
“As tempting as it is to find some cause for celebration in infection rates that are beginning to fall, as seen in both ONS data and the Government’s publication of the R number, we still see very high levels of Covid-19 in every region of the UK. The national lockdown in England, and equivalents in Wales and Scotland, are beginning to arrest the spread of Covid, including the new variant seen in this country. However, the lag between infections and then hospitalisations and deaths mean that healthcare services still face unprecedented demand, and we are still likely to see recorded deaths continue to increase for some time yet.
“The new publication of health indicators1 shows just how much of a Covid-19 impact the NHS is experiencing. More than twice as many people are being admitted to hospital in the latest publication, compared to early December. As there is an expected lag between case rates and hospitalisations, and deaths, that 35 people in every 100,000 are being hospitalised is a sobering and should serve as a clear reminder for people to stay at home.
“The data continues to be a reliable set for comparative purposes, and the new weekly overview combining multiple models and other data gives more useful data in one place. Notably, it provides information about preventative measures being taken, and shows that the vast majority of people are complying with Covid restrictions, although not a high enough proportion to drastically reduce transmission.”
Dr Jeffrey Barrett, Director of the SARS-CoV-2 Genomics Initiative at the Wellcome Sanger Institute, said:
“The shortcut of using “S gene target failure”, SGTF, to track proportions of the B.1.1.7 variant can classify some test results very accurately using the three channels for the S, N and ORF1ab genes as follows:
“But things become more complicated when only N or ORF1ab is positive, or if any of the positive channels are weak (also known as having “high Ct value”). The first item in the “Notes” on the ONS page explains their definition, which classifies these edge cases in the “Other positives” category, which may partially explain the apparent decrease in proportion of B.1.1.7. While the ONS Infection Survey is robust and the data is useful, these ‘edge’ classifications may make a difference to what patterns look like – and we do know from pillar 2 testing that the proportion of SGTF is now almost 90% nationally and has been increasing week on week.”
ONS Infection Survey: https://www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/conditionsanddiseases/bulletins/coronaviruscovid19infectionsurveypilot/22january2021
R number and growth rates: https://www.gov.uk/guidance/the-r-number-in-the-uk
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 its Advisory Committee. I am also a member of the Public Data Advisory Group, which provides expert advice to the Cabinet Office on aspects of public understanding of data during the pandemic. My quote above is in my capacity as an independent professional statistician.”
None others received.