Over the past few weeks the COVID Symptom Study app has been releasing the number of daily new cases of COVID in the UK, but we aren’t the only source of this information in the UK.
We have been receiving lots of questions about the different sources of data so we wanted to share more details that may help clear things up.
There are currently two main sources of information: the COVID Symptom Study app (us) and the Government’s Office of National Statistics (ONS). Both sources provide estimated prevalence and incidence rates. Prevalence is the total number of estimated cases at any given time, whereas incidence is the estimated number of daily new cases. The Government also provides the total number of lab confirmed cases everyday.
In this post we will explain how each of the surveys are calculated and some of the important things to consider.
What do the different figures say right now?
*The figures on this blog are up to date on 9th July 2020. If you are reading after 9th July please follow the links to get the latest figures.
The number of lab confirmed cases
This is the rawest form of data that includes records of all the swab positive tests on any given day in the UK, including hospital cases.
Things to consider
- Many cases of COVID are not recorded. So lab confirmed cases are just a fraction of the total story.
- This data is the most immediate and only includes a small lag, which makes it suitable for spotting outbreaks, especially in hospitals.
- The recorded cases come from a variety of sources, predominantly constitute tests from hospitals, care homes, care workers, healthcare professionals and other communities at high risk of infection. The reported numbers can offer crude estimates but are sensitive to testing patterns, test availability and other changes in government policies.
The COVID Symptom Study figures
How do we calculate incidence rates?
We have a two step process. First we identify app users who log new symptoms after being healthy for at least 9 days, and estimate the proportion of newly sick app users. These individual users are then invited to have a COVID swab test provided by the Department of Health taken within 24 hours and are asked to continue logging their symptoms in the app.
The results of these tests allow us to identify the proportion of newly sick individuals who test positive for COVID-19. This two step process enables us to do more targeted swab tests focussed on sick users, who are more likely to be positive. We scale the proportion of sick users with the proportion of those testing positive to get the incidence rate and then generalize it to the wider population to give the estimated number of daily new cases across the UK.
How do we calculate prevalence rates?
We now have a better picture of how long it takes to recover from COVID and as a result we have built a recovery model that tells us how many people recover within a specific number of days from symptom onset. For example, we observe that only 52.2% of people recover within 13 days. We have combined this recovery model with our daily new cases model to produce our prediction of the number of daily active cases.
In a nutshell, our prevalence is:
Yesterday’s active cases + today’s new cases - today’s recoveries = Number of active / symptomatic cases.
Things to consider:
- Our targeted testing strategy enables us to record more positive tests. For example, last week ONS published estimates based on 10 positive tests, whereas we published estimates using 56 positive tests. The higher number of positive tests makes it possible to estimate incidences at a regional level, which permits us to reliably compare and comment about the different regions in the UK.
- The app uses all the test results that we receive and estimates incidence for the entire UK. ONS surveys are for England only at present.
- We are only able to invite those who are using the app for COVID testing, and our user base is not fully representative of the entire country. So, our estimates are slightly biased towards the demographic and behavioural characteristics of our app users, although we adjust for age and deprivation when extrapolation to UK population This means that although users can log for relatives, our estimates don't include many cases in hospitals, care homes and other communities where the app usage is subpar or non-existent.
- The app is not able to monitor asymptomatic users (i.e. those without any symptoms) and thus our estimates don’t account for them. So, our estimates will be lower than ONS estimates, which include asymptomatic cases, although they don't screen for the same number of symptoms as we do.
- The incidence estimates are updated everyday, but we have a four day lag to provide ample time for our users to get tested and log their test results. The prevalence estimates also use the symptoms logged by 3 million users in the UK to predict how likely someone is to test positive on each day.
The ONS figures
The ONS randomly invites around 8,000-9,000 households in England to participate in its testing program and uses these test results to estimate incidence and prevalence. The ONS publishes an estimated prevalence and incidence every two weeks based on their infection survey test results.
Things to consider:
- As it is a random invitation survey, ONS can test people across all ages with enough representation from various regions and communities in England. So, its estimates are more representative of the population, although will still have some biases. For example, it will only measure people who are willing to engage in a repeated survey for the government.
- The randomness of inviting people ensures that both symptomatic users and asymptomatic users are tested, unlike our app.
- It is a small sample - only 8-9,000 households vs the millions in the UK. It cannot see hotspots in particular areas therefore.
- The estimates are not updated daily. It is once a week with an extra four day lag. This means that the figures are not as sensitive to sudden changes/outbreaks
- The ONS publishes numbers only for the whole of England (and not for regions) because of small numbers, though numbers may increase.
- It only accounts for private households, it can not account for the number of cases in hospitals and other care situations.
The lab confirmed cases is a standalone figure, which is not comparable to the prevalence and incidence rates from the other two survey sources. The ONS and COVID Symptom Study app provide roughly the same estimates over the last month which is reassuring.
The results over the past few weeks of surveys have suggested that the rates have been falling over time, though our app has shown regional differences.
We will be continuing to release weekly updates on the data on Thursday each week and we hope that this blog has helped outline the differences between the data and what each figure is telling us about the current COVID situation in the UK.