Not for the first time, people could be excused for feeling a bit confused by conflicting data on Covid-19. This morning, many news outlets reported claims from the latest React study that levels of Covid-19 infections are no longer falling and may, in fact, be rising. This is somewhat at odds with government data that shows cases have peaked since the beginning of the third English lockdown. For days, the figure for new infections, as detected by NHS Test and Trace, and published on the government’s Covid-19 dashboard, have been falling.
In the past seven days, the number of new infections recorded across the UK was 294,172 — 21.5 per cent down on the previous seven day period.
What's going on? The Covid-19 dashboard figures are cases of infection picked up through the test and trace system — they are a count of people who have developed symptoms, been tested and returned a positive result. They suffer from the weakness, however, that they won’t be picking up all cases — they can’t count asymptomatic cases where people had no reason to get tested. This way of measuring prevalence is also a prisoner to the number of tests being conducted. These figures show, for example, that infections are running at multiples of what they were during the first peak — yet at that time few tests were being performed, with symptomatic people told not to bother ringing the NHS unless their symptoms were serious. However, in recent weeks the policy on testing has remained constant, with capacity for around 800,000 tests a day.
The React study, on the other hand, is based on testing a randomised sample of the population. Between 6 January and 15 January, it tested 142,909 people across the country, of which 1,962 were positive. That figure was then used to estimate that 1.58 per cent of the population had the virus between those dates.
But React has a major weakness in that it is not a continuous survey; rather it is carried out in rounds, the last of which was between 25 November and 3 December. It can tell us that prevalence grew between that period (when 0.91 per cent of the population was estimated to be infected) but there is no data for the intervening period.
How, then, does it come to the conclusion that cases may have risen since the beginning of the third lockdown, which began on 6 January? Interestingly, the Office for National Statistics’ summary of the React study does not make this claim. It merely states there is ‘no strong evidence for either growth or decay in prevalence across the period 6 January and 15 January’. It is only when you go to the paper itself, published by Imperial, that it makes the assertion that infection levels ‘may have started to rise’ at the end of the period — a claim it says it makes on the strength of analyses using ‘P-splines’ — an analytical technique which seeks to smooth out lines generated from limited amounts of data.
It also mentions data of people’s movements obtained from mobile Facebook apps — which might be a proxy for how well lockdown is being observed but hardly directly relates to infections. The paper admits that it might have missed a peak between early December and early January. There is another randomised study of infection prevalence which is continuous — the ONS weekly infection survey. However, this sheds little light on the matter because it wasn’t published last week — with the ONS blaming a delay in data.
All in all, React does not provide strong evidence for a possible uptick in cases between 6 and 15 January.