Researchers from King’s College London, Massachusetts General Hospital and the company ZOE have created an artificial intelligence (AI) system that can predict if someone is likely to have the coronavirus based on their symptoms.
Source: Unsplash
The new AI model uses data from a coronavirus symptom study app that can predict COVID-19 infection. The app compares symptoms and with the results of traditional tests. It could help people with limited access to testing figure out if they have or have been exposed to COVID-19. Around 3.3 million people have downloaded the app and are using it daily to track and report their health status and symptoms.
The team analyzed data gathered via the app from 2.5 million people in the U.K. and the U.S. who are regularly logging symptoms associated with COVID-19. Of the 18,374 people who reported being tested, 7,178 of those people were tested positive. They found that the symptoms associated with COVID were more likely to be associated with a positive COVID-19 test. The range of symptoms compared to cold and flu symptoms.
Based on this research, the team warns against focusing on fever and cough symptoms as the main indicators for COVID-19. They found that loss of taste and smell was found in one-fifth of the positive users reporting. Loss of smell was found to be a stronger predictor of COVID-19 than fever. This supports the early reports that predicted that loss of smell and taste were common COVID-19 symptoms.
The team created a math model that can predict with nearly 80% accuracy if a person is likely to have COVID-19. The predictions are based on age, sex and a combination of four key symptoms. The key symptoms are loss of smell, taste, severe or persistent cough, fatigue and skipping meals. The model was applied to a group of 800,000 app users who were experiencing symptoms.
Combining AI prediction with the widespread adoption of the app could help identify those who were likely to be infectious as soon as the earliest symptoms appear. This allows the focus of tracking and testing efforts where they are most needed.
A paper on this technology was published in Nature Medicine.
