select search filters
briefings
roundups & rapid reactions
before the headlines
Fiona fox's blog
The Office for National Statistics (ONS) have updated estimates of the prevalence of post-acute symptoms among people with coronavirus (COVID-19) in the UK, 26 April 2020 to 1 August 2021.
Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:
“This ONS release is labelled as a ‘Technical article’, and it is pretty complicated. Actually what this mainly tells me is that what counts as ‘long Covid’ or, to be a bit more precise, ‘post-acute symptoms among people with coronavirus (COVID-19)’, is far from clear, despite the guideline definitions published by NICE last December1. The ONS article, which is based on results from their Covid-19 Infection Survey (CIS), provides estimates of the prevalence of long-term symptoms in people who previously had an infection with the virus that can cause Covid-19. There are three different ways of coming up with estimates, that give rather different results from one another. They are all plausible ways of analysing the data, and the main results reported on in the bulletin are for people experiencing symptoms at least 12 weeks after they were infected. The 12 weeks matches the length of time that NICE use to define “Post-COVID-19 syndrome”. (The ONS release also presents results for some shorter timescales, of 4 weeks or so, which NICE would call “Ongoing symptomatic COVID-19”.) However, the details of the way the cases are defined in this ONS article do not, and could not, match the NICE definitions exactly. The ONS results come from a survey that was already running before the NICE definitions were produced, so that the data collection does not match the NICE definitions because some of it predates them, and in any case information from a survey can’t possibly match all the subtlety in the NICE guideline.
“The ONS article also summarises the results from other work that seems to be estimating the prevalence and comes up with different results, including previous work by ONS itself, and gives possible explanations as to why the results differ so much. I think those summaries and explanations are important and interesting. It doesn’t concern me that, at this stage in the pandemic, numerical estimates about something as new and complicated as long Covid differ quite a lot – what’s important is that those involved should discuss openly what needs to be done to clarify the position. But the need to do that shouldn’t hold up the establishment and improvement of services to help people with these conditions, however they are defined and counted.
“It’s important to understand that this set of ONS estimates can’t match the estimates in the monthly series of ONS bulletins on “Prevalence of ongoing symptoms following coronavirus (COVID-19) infection in the UK.” The most recent of those came out on 2 September2. Those bulletins give estimates of the number and percentage of people in the UK community population who have ongoing symptoms. So, for instance, the most recent one was based on data from July and estimated that 1.26% of the whole community population, that is, about 1 in 80 of that population, had self-reported long Covid at least 12 weeks after they had had (or suspected they had had) Covid-19. In numbers, that’s about 817,000 people, a rather alarmingly high number in my view. But this new ONS release doesn’t aim to provide comparable figures at all. It does estimate percentages, but they are not percentages of the whole community population. Instead they are percentages of a smaller group – of the people who actually tested positive for the virus in a PCR test (and, indeed, in a PCR test carried out as part of the ONS CIS). The three different ways of doing the estimation produce estimates that 5%, 3% and 12% of people, who are very likely to have been infected because they tested positive, have symptoms that went on for at least 12 weeks. The three percentages are quite a lot different, which is very largely due to different definitions of symptoms being used for each method of estimation.
“The ONS bulletin doesn’t attempt to scale any of this up to estimate how many ‘long Covid’ cases there are in the population, and that can’t easily be done with their data – that wasn’t the aim. However, some interesting, if rather crude, calculations are possible. The ONS CIS produces estimates of the incidence of new infection. The latest set were for the week 14-20 August, and the total estimate for the whole of the UK community population is that there were 526,000 new infections that week – though that figure does have a wide margin of statistical error. If 3% of those new infections result in long Covid, that’s getting on for 16,000 new long Covid cases just from infection that week – and if the highest figure from the new ONS bulletin, 12% of the infections, result in long Covid, that would be over 60,000 new long Covid cases just from a week’s worth of infections across the UK. It’s reassuring, of course, that most people who are infected don’t go on to develop long Covid, but in absolute numbers, really a lot of people will be developing a quite long-term illness. Whether the prevalence estimates from today’s ONS bulletin will still apply now that so many people have been vaccinated is not yet clear, so one can at least hope that things won’t be as bad as that now – but there’s nothing in this article to say that that will be the case. Because this is an observational study, we can’t be entirely sure of cause and effect in any case.
“Another way of looking at the possible rate of new long Covid cases, assuming that the cases are indeed caused by the previous infection, is to use another estimate from today’s report. ONS estimate that about 18% of people who had symptomatic Covid-19 went on to have (self-reported) long Covid at least 12 weeks after their initial disease. The figures I’ve been discussing so far are of people who had an infection regardless of whether they had symptoms – that’s known from the CIS because people are tested regardless of whether they have symptoms. For the week of 4-10 September, there were slightly over 240,000 new confirmed cases of Covid-19 on the dashboard at coronavirus.data.gov.uk, and they would mostly have had symptoms. That ONS figure would imply that this would result in about 43,000 new long Covid cases just from that week’s worth of new confirmed cases. This assumes that the long Covid cases are indeed caused by the initial infection, and that the rates estimated in today’s ONS bulletin still apply at today’s level of vaccination, and the figure is a bit less than the 60,000 I estimated above – my estimates are really quite crude, I’m afraid – but if they are anything near the true level of future long Covid- I think they are pretty concerning.
“Why do the three ONS estimation approaches produce results that differ so much? The first one, 5% of people who were infected having ongoing symptoms after 12 weeks, is based on the number of people who reported having any one (or more) of a list of 12 symptoms, at a time 12 to 16 weeks after their original infection. ONS point out, rightly, that the symptoms on the list can all be caused by other things. (The list is fever, headache, muscle ache, weakness/tiredness, nausea/vomiting, abdominal pain, diarrhoea, sore throat, cough, shortness of breath, loss of taste, and loss of smell.) They carried out a comparison with a control group of people, also amongst those included in the CIS sample, who were matched individually to the infected people in terms of age, sex, health status, and some aspects of when they took part in the CIS. However, unlike the infected people, those controls had never tested positive either for a current infection with the virus or for antibodies to the virus – so they very probably had not been infected. In that control group, 3.4% also reported having one of more of the 12 symptoms, at 12 to 16 weeks after the date when the infected person to whom they were matched had been infected. That’s not all that much less than the 5.0% for the infected people, which does show that having one or more of these symptoms isn’t uncommon regardless of Covid-19. But there were certainly more cases in the infected group.
“The second ONS approach used a tighter definition of what’s meant by long Covid, and included only people who had had one of the 12 symptoms continuously since their initial infection. That produced the lowest of the estimates, just 3.0% of infected people having long Covid after 12 weeks or more. In this case, the control group had a very considerably lower rate of having continuing symptoms, just 0.5%. ONS had produced an estimate using similar methods in April year, and got a much higher figure, about 14%. ONS put the difference down to improved statistical methodology, but also to people being followed up for longer, which allows a clearer measurement of when a previous period of continuous symptoms ends.
“The third ONS approach produced the highest estimate, 11.7% of infected people classifying themselves as having long Covid at least 12 weeks after their first infection. That’s obviously a much higher figure than the other two – but I suspect it is more realistic. The list of 12 symptoms used for the other approaches is quite limited, and does not include several symptoms (such as “brain fog” or certain types of pain) that are commonly reported by people who report that they have long Covid. Indeed, the ONS CIS survey include a longer list of symptoms, 21 of them, that are asked only of people who say that they have long Covid. The original 12-symptom list hasn’t been changed since the early days of the CIS, and is still used, I assume, to provide continuity, but it really is a bit too limited to provide good estimates of long Covid prevalence. Because the wider self-report question about long Covid, which includes the question about the 21-symptom list but also simply allows people to say they have symptoms that they would count as long Covid, isn’t asked of people who don’t believe they were infected, it wasn’t possible for ONS to analyse a comparable control group for this approach.”
Dr Michael Absoud, Honorary Reader, Department of Women and Children’s Health, King’s College London, said:
“The ONS are to be congratulated for engaging with clinicians and scientists to review their methodology and provide updated estimates on Post-Covid symptoms. The authors describe three very sensible approaches to ascertain estimates.
“We had previously raised questions on earlier estimates relating to children, particularly with needing a representative control group and a definition to an end of symptoms. The ONS results are based on data provided by a sample of over 20 thousand Coronavirus Infection Survey (CIS) participants (with matched controls), and using 12 key symptoms tracked over time.
“The ONS first published the approach in April 2021, and reported a 12-week prevalence of ‘Long-Covid’ in 14%. This has now been revised down to 3% in the latest estimate. The significant fall in the prevalence estimate is due to different interpretations of when symptoms come to an end, and also better long-term follow up for symptoms beyond the 12 week point. Continuous symptoms is taken to mean reporting any of the 12 symptoms at consecutive visits. Symptoms were tracked when reported by participants on the day of their first positive test, or that started within the next five weeks. Symptom discontinuation was defined as the first occurrence of at least two successive follow-up visits without reporting any symptoms. A consequence of these changes is that some participants would be re-classified as having experienced the end of their symptoms earlier, which would reduce the prevalence estimate.
“A striking finding is the overlapping confidence intervals in children. In 2-11 year olds, 4.1% (3.0-5.5) of controls had symptoms 12-16 weeks after symptom tracking began, compared to 3.2% (2.3-4.5) with Covid. These findings are in line with recent publications in children, and highlights the need for recovery and proper funding for all children experiencing symptoms in the pandemic. Children need support for education, physical and emotional wellbeing recovery.”
References:
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.