Infection Surveys and Symptom Panels

The ONS survey and the ZOE app use different methods.

How many people have a SARS-CoV-2 infection? During this pandemic, there is a vital interest in tracking infections over time.

This article looks at differences between infection surveys and the ZOE symptom study.

Tracking the pandemic

In the news, we often hear the daily number of new cases.

A case is a new person with a positive test result. That result can be via a laboratory or a lateral flow device.

Confirmed cases are not all cases. Testing regimes and accuracy influences these statistics. If testing availability changes, then recorded case numbers can increase or decrease. No country tests everyone: public health agencies will miss some cases.

There are also reporting lags. The daily reported number can reflect time taken to process results in labs. The types of tests are changing: with figures including lateral flow device tests.

This is the number of confirmed cases by date of report. (Image: Public Health England)

The Office for National Statistics infection survey

We need to understand how many people would test positive, if given a test. To that end, the Office for National Statistics conducts an infection survey.

The ONS survey is of people living in private households. This is a random sample. It does not include people in hospitals, care homes, or other institutional settings.

The weekly reported estimates continue to fall. (Image: ONS)

People aged two and over have a throat and nose swab. For children aged 11 and under, parents and carers do the swab. Laboratories then process these swabs, conducting polymerase chain reaction (PCR) tests. University of Oxford researchers then model headline estimates of positive test results.

Surveys provide estimates, subject to many sources of potential error.

ZOE COVID Symptom study

Another estimation method comes via the ZOE COVID Symptom study. The health company ZOE made the app. Researchers at King’s College London conduct the analysis.

One main purpose is to understand which symptoms predict whether someone has SARS-CoV-2. This is not a representative survey. Volunteers log into the app each day, noting any symptoms they may be experiencing.

People aged 18 or over can download the app — participants can report the symptoms of children. With consent, users can log symptoms in other adults. Users can also record any tests for SARS-CoV-2 they have, and vaccinations.

The measurement is of symptomatic cases. It does not include anyone who have SARS-CoV-2 but not experience symptoms. When users feel unwell, they can request a swab test.

Their prevalence model uses active cases, adding new cases and subtracting recoveries. Their model adjusts for age, gender, and region.

This study is estimating “Symptomatic COVID”. (Image: ZOE)

There is also the Real-time Assessment of Community Transmission surveys. Imperial College London and Ipsos MORI run this survey. The main difference from the ONS survey is that REACT does not need follow-up visits.

In short, the main differences are:

  • Random sample versus self-selecting panel: The random probability survey can have sampling errors. The ZOE app estimates are via volunteers, and so suffers from self-selection bias.
  • Definition of a case: The ZOE estimates measure symptomatic cases. The ONS survey includes cases that do not show symptoms.
  • Prevalence models: The ZOE estimates prevalence by adding new cases and subtracting recoveries. The ONS estimates prevalence by using survey results in Bayesian modelling.

It is important to understand differences in methods when comparing two estimates.

This blog looks at the use of statistics in Britain and beyond. It is written by RSS Statistical Ambassador and Chartered Statistician @anthonybmasters.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store