This article is more than 1 year old

No, AI can't tell if you've got COVID-19 by listening to your coughs

UK study pours cold water on claims after government already spent money trying to build an app

Machine learning algorithms cannot accurately predict whether someone has COVID-19 by analyzing the sound of their coughs, according to a study led by the UK's Alan Turing Institute. 

Claims that AI could detect the difference in cough sounds between those with and without COVID-19 with up to 98.5 per cent accuracy were first reported in a paper from researchers led by the Massachusetts Institute of Technology. The result led to efforts to build an app powered by the algorithms to provide people with a cheap and easy method to test for the novel coronavirus.

The UK's Department of Health and Social Care even went so far as to award two contracts, collectively worth over £100,000, to Fujitsu to develop the government's so-called "Cough In A Box" initiative in 2021, Politico reported. The software would collect audio recordings of coughs from users to analyse on its COVID-19 app.

But tests performed by a team of researchers led by the Alan Turing Institute and the Royal Statistical Society found the technology doesn't quite work after all. They collected and examined a dataset of audio recordings from over 67,000 people recruited from the National Health Service's Test and Trace and REACT-1 programs, which asked a random portion of the population to perform and send back nose and throat swabs to test for COVID-19.

Participants were asked to record samples of them coughing, breathing and talking as well as the results of their swab tests. Over 23,000 of them had tested positive for the respiratory disease. The team trained a machine learning model on these sounds, comparing them with people's COVID-19 test results to see whether coughs could act as an accurate biomarker.

"But as we continued to analyse the results, it appeared that the accuracy was likely due to an effect in statistics called confounding – where models learn other variables which correlate with the true signal, as opposed to the true signal itself," explained Kieran Baker, a statistics PhD student at King’s College London and research assistant at the Alan Turing Institute.

The confounding was due to recruitment bias in the Test and Trace system, which required participants to have at least one symptom in order to take part. The researchers performed more tests grouping participants of the same ages and genders into pairs, with only one of them having COVID-19. 

"When we evaluated these models on the matched data, the models failed to perform well, and so we conclude that our models cannot detect a COVID-19 bio-acoustic marker from this data," Baker said.

Chris Holmes, lead author of the paper released last month, professor of biostatistics at the University of Oxford, and program director for health and medical Sciences at The Alan Turing Institute, said: "Finding new ways to quickly and easily diagnose viruses like COVID-19 is really important to stop its spread. While it's disappointing that this technology doesn't work for COVID-19, it may still work for other respiratory viruses in the future," the UK Authority reported. ®

More about


Send us news

Other stories you might like