Three months on, the data science team at ZOE and King’s College London behind the COVID Symptom Study app have taken the decision to update the way prevalence is calculated.
We are committed to sharing all our data and findings with our users, so we wanted to take the time to explain how and why these changes have been made and what that means in terms of the data.
Why change the model now?
As with any scientific study, over time things change and develop, so it’s important to constantly be reviewing and updating how we are doing things to make sure we are giving our users the most accurate information.
For the last two months, we have been running the testing programme with the Department of Health and Social Care in England, which has seen more than 350,000 users being invited to take swab tests when they begin to feel unwell. Users then log results of the test into the app, helping us to build a better picture of what symptoms are related to a positive test. We have also been inviting users who previously tested positive to be retested so we could understand for how long they are COVID positive and exhibiting symptoms.
This new data, combined with all the other data our users have kindly shared with us (including 100million symptoms assessments) means we have a greater insight into COVID symptoms. As a result we have been able to improve our predictions on who is sick, and most importantly we know how long people are likely to still be infectious and therefore test positive to a swab test.
How do we calculate prevalence rates?
Now that we have a better picture of how long it takes to recover from COVID, 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, prevalence is:
Yesterday’s active cases + today’s new cases - today’s recoveries = Number of active cases
We have run this formula from the start of our first testing programme (12th June) as we have large numbers of swab tests from this date. As a result we have a much more reliable estimate of daily active cases than we did before.
With this information, we have also recalibrated the prediction model of symptomatic cases to better understand the current levels of COVID in your area and updated the interactive map in data page. We will be separately estimating the numbers of people with long duration symptoms and updating our website with these figures. We want to emphasise that there are lots of people who continue to have symptoms long after they are no longer infectious - this is an area of huge importance, and one that our researchers are very keen to understand better with your help.
The introduction of this new model means that our prediction of prevalence has fallen significantly, because in the last few weeks our previous prevalence estimates included those people who are long term sufferers. These individuals are no longer in the prevalence figures and instead we will be reporting on them separately. It is also important to recognise that we do not identify people who show no symptoms whatsoever (asymptomatic).
What do the latest figures show us?
Our new prevalence figures show that 23,459 people in the UK currently have symptomatic COVID and highlight the big regional differences across the UK. While nations like Northern Ireland have almost no active cases, the rates for other English regions, like the Midlands, have been decreasing less steeply.
This estimate is in line with the latest ONS Infection survey in which 25,000 people in England were estimated to be infected with COVID-19 during the two week period that goes from the 14th to the 27th of June.
We will be continuing to update all the data on our data page: https://covid.joinzoe.com/data
We hope you found this blog informative and as always, we want to thank all our amazing users who make all this research and data possible!
[*] This analysis requires swab testing, which was kindly provided by the Department of Health and Social Care for England. As Scotland and Wales are not yet offering tests to app users, we provided indirect estimates using countrywide averages and wide confidence limits. Testing is happening in Northern Ireland, but the number of participants is too few to generate an accurate estimate. These figures exclude care homes as there is not enough data from the app to estimate this population.