New Scientist Default Image

Exclusive: Concerns raised about vital UK covid-19 infection survey

The Office for National Statistics’ survey is designed to test a representative sample of the UK population

Danny Lawson – WPA Pool/Getty Images

The UK’s largest scheme for tracking the spread of the coronavirus is at risk of providing a misleading picture of the epidemic, as a growing share of people invited to take part fail to respond or complete a test.

The UK’s Office for National Statistics (ONS) launched its survey in April to estimate how many people are infected with the virus each week. At first, it randomly sampled thousands of homes in England, later adding those in Wales and Northern Ireland. The UK government’s top scientific advisers consider it the gold standard for measuring the state of the epidemic because other methods such as testing can miss many cases.

Households who respond to the invitation to take part are visited by a survey worker, who provides the tests for people to complete themselves. When the survey began, 51 per cent of English households invited to take part completed at least one test. However, that figure has now dropped to just 5 per cent.

Advertisement

The survey’s most recent update, on 9 October, suggested that 1 in 240 people in England are currently infected and found a big increase in the incidence rate over the past six weeks, leading to calls for stronger interventions by the government.


The drop in the response rate has sparked concerns that the survey could become biased. If the tests are only being completed by a certain group of people, it may become less reflective of the wider population and the true state of infections.

The ONS says it weights the results to avoid this, but this may not work, says Sheila Bird at the University of Cambridge. “Reweighting may not deal sufficiently with systematic bias.”

One example of that bias would be if the small percentage of people completing tests are also those who are more observant of guidance, and therefore less likely to be exposed to covid-19 – raising the prospect of an underestimate of infections.

“The big question is whether or not this drop out could bias any results and that is a difficult one to answer,” says Paul Hunter at the University of East Anglia, UK. “In a large study, it is certainly possible to adjust the analyses for changing demographics in the remaining cohort by standardising for differences in age, sex and social group.” He says it is unsurprising the response rate is dropping, as he found the same during a 2016 infectious disease study.

One reason for the fall may be that the scheme has expanded hugely, from inviting 20,000 households initially to nearly 800,000 since 13 July, meaning it is simply reaching more people who are unwilling to take part.

The first batch of households had also taken part in previous ONS surveys, meaning they would be familiar with the organisation. Since 13 July, households have been chosen from a list of 26 million UK addresses, reaching people who haven’t engaged with the ONS before.

While the response rate in England is currently 5 per cent, the ONS suggests that shouldn’t be considered a final figure, as there is no time limit for how long invited households can take to respond.

A spokesperson for the ONS says: “By randomly testing a large representative sample of the UK population, regardless of whether or not they have symptoms, (the infection survey) continues to offer reliable estimates about the spread of this virus.”

Sign up to our free Health Check newsletter for a round-up of all the health and fitness news you need to know, every Saturday

Article amended on
16 October 2020

We have clarified the process by which participants are tested

More on these topics: