Future-proofing our COVID estimates
October 7, 2021
This article has not been updated recently
- Why is ZOE updating its COVID calculations again?
- Reverse engineering our vaccine effectiveness data
- How does the update affect COVID case estimates?
- We still need your data
Back in July, we updated our data methodology to minimise the impact of diminishing numbers of unvaccinated contributors. With unvaccinated users disappearing from our study, we’ve found a solution to continue predicting case numbers for the whole country.
Why is ZOE updating its COVID calculations again?
COVID-19 has changed a lot in 18 months, but people have changed more. Very few people had been vaccinated in 2020, while today 66% of the population are double dosed. This significantly alters your risk of being infected.
We have a huge pool of contributors, but it’s not the whole population - so we have to estimate figures and adjust for things like age and vaccination status as we go.
Over time, our contributors have rolled up their sleeves and been vaccinated. Today, among some of our age cohorts, unvaccinated contributors have all but vanished.
This means that we now need to review the methodology to ensure that we continue to present as accurate a picture as possible including those areas where we are underrepresented, including those who have not yet been vaccinated.
Reverse engineering our vaccine effectiveness data
One thing we have observed is that our vaccinated contributors’ risk of COVID infection is reduced. We have a model to predict vaccine risk reduction which has been published in The Lancet Infectious Diseases and you can read more about it on our website.
The table below outlines how vaccines reduce the risk of infection.
Using this data in reverse, we have developed a new weighting function that takes into account the differences between vaccinated and unvaccinated. So even if we lost all unvaccinated users from our study, we’d still be able to look at symptoms and positive tests in our vaccinated contributors, calculate the relative risk in the unvaccinated and extrapolate this to the UK population.
How does the update affect COVID case estimates?
With tiny cohorts, even a small number of test results can swing the data in one direction or another, so where studies have areas of underrepresentation it can impact the data results. Our new methodology effectively applies a weighted average to a cohort and smooths out the trendline.
We can see this when we place our previous and new incidence estimates alongside each other. The blue line moves more erratically, while the orange line is more stable. This is the effect of applying the average.
Because of these variations in the previous model, our confidence intervals, indicated by the faded blue areas on either side of the blue line, were wider than they are now. Our updated figures show the orange line passing within these confidence intervals. This reflects that our changes have now reduced this level of uncertainty and allows us to be even more precise in our calculations. We’ve also seen a slight level of the trajectory which reflects the levelling of rates which has also been seen in other studies.
All this means that while numbers might change a little on account of the methodology update, the trend is largely the same, with high numbers since July and and some variations occurring around the midpoint value.
We still need your data
It’s plain to see just how much we rely on the data of our contributors to make these predictions about the number of new COVID cases every day. We use this information to generate scientific research findings on COVID, inform and challenge government policy, and keep our contributors safe and aware of the situation and how it’s changing.
To do this, we need as many people as possible to download the ZOE COVID Study app and start logging daily health reports. It only takes a minute, but you’ll be contributing to life-saving research. If and when you’ve had a COVID-19 vaccine, or any other tests, be sure to log it in the ZOE COVID Study app and share the app with your friends and family as well.
Stay safe and keep logging.
As of the 20th of September 2021, the ZOE Infection figures in the UK are based on 730,738 (93%) fully vaccinated users, 11,531 (1.5%) partially vaccinated and 42,851 (5.5%) unvaccinated users. The vaccination rates in the study cohort are higher than those in the UK population: 66.7%, 6.2% and 27.1% respectively.
The ZOE COVID Study has been collecting key information on COVID-19 vaccines and infections since the start of the UK’s vaccination programme. Thanks to this data, ZOE and King's College London (KCL) have been able to assess the relative risk reduction from vaccines (Menni 2021) and track changes over time in the real world, a.k.a vaccine effectiveness. The relative risk reduction measured in the study can be used to predict an incidence rate in unvaccinated and partially vaccinated users from the incidence in the fully vaccinated group.
Menni, Cristina. “Vaccine side-effects and SARS-CoV-2 infection after vaccination in users of the COVID Symptom Study app in the UK: a prospective observational study.” The Lancet Infectious Diseases, vol. 21, no. 7, 2021, pp. 939 - 949.