Researchers at King’s College London, Massachusetts General Hospital, and ZOE Medical Company have developed artificial intelligence diagnostics that can predict if someone can have COVID-19 based on their symptoms. It is reported by Nature Medicine.
The AI uses data from the COVID Symptom Study application to predict COVID-19 infection by comparing human symptoms and the results of traditional coronavirus tests. Researchers say this will help populations where access to testing is limited. Two clinical trials in the UK and US are due to begin shortly.
More than 3.3 million people all over the world have downloaded the application and use it for daily reports on their health status regardless of whether they feel good or have any symptoms, such as persistent cough, fever, fatigue, and loss of taste and/or odor (anosmia).
In this study, scientists analyzed data collected from 2.5 million people in the UK and the USA, who regularly recorded their health status in the application, about a third recorded symptoms associated with COVID-19. Of these, 18,374 people reported that they did a coronavirus test, and 7,178 people had a positive result.
The research team examined which symptoms associated with COVID-19 were most often associated with a positive test. They found a wide range of symptoms compared to the common cold and therefore cautioned against assessing the possibility of infection only by the symptom of a fever or cough. Indeed, they found that loss of taste and smell (anosmia) was especially important: 2/3 of those with a positive result for coronavirus infection reported this symptom compared to more than 1/5 of those who tested negative. The findings suggest that anosmia is a stronger prognostic factor of COVID-19 than fever, confirming that loss of smell and taste is a common symptom of the disease.
The researchers then created a mathematical model that predicted with almost 80 percent accuracy the likelihood that a person would have COVID-19 based on his age, gender, and a combination of four main symptoms: loss of smell or taste, strong or persistent cough, fatigue, and absence appetite. Applying this model to an entire group of more than 800,000 symptomatic app users predicts that at that time, just under 1/5 of patients (17.42%) could have COVID-19.
Researchers suggest that combining this AI prediction with application distribution may help identify those who may be contagious as soon as the earliest symptoms appear, focusing efforts to track and test where it is most needed.