During the ongoing COVID-19 pandemic, it is important to be quick in diagnosis, in order to prevent the spread of the pandemic. While major efforts were undertaken to raise the amount of tests conducted, oftentimes it was not enough; one can only make so much tests within a day. Add to that the fear factor – many people, who experience COVID-19 related symptoms panic and out of fear go en masse to testing facilities. Some of these people do end up being tested positive for having the coronavirus, but many results are (thankfully) negative, Still, this creates the issue of overcrowding of such facilities, making testing even slower.

Thankfully, with the help of Artificial Intelligence, people can now reach out to the help of a certain application to fasten the diagnosis process. While this does not eliminate the need for testings, it can facilitate a great supporting factor in the screening process, to determine which cases should be focused upon.

COVID Symptom Study App – how AI helps to improve diagnosis screenings.

Research that developed this app was conducted at the King’s College London, Massachusetts General Hospital and ZOE, a health science company. In turn, they developed a test, using AI, that can determine when anyone is likely to have COVID-19 based on their symptoms. The AI model utilizes data from the COVID Symptom Test app to compare COVID-19 symptoms by contrasting individual signs with the outcomes of standard COVID studies. To this end, 18,374 people used the application for the research to report what symptoms they were experiencing. This information was then matched with the data gathered through standard COVID-19 tests, and the results were matched to see who has a higher chance of being infected. Out of the 18,374 people, 7,178 people were tested as positive for having the coronavirus.

Since then, over 3 million people have downloaded the application. They use it to monitor their health condition on a regular basis, whether they sound good or have some signs, such as constant cough, nausea, exhaustion and lack of taste or scent.
The study team analyzed which of the signs believed to be correlated with COVID-19 were more likely to be linked with an affirmative result. They found a wide range of symptoms compared to cold and flu, and warned against focusing only on fever and cough. In fact, the loss of taste and smell was particularly striking, with two thirds of those experiencing the symptom being tested positive for coronavirus. Researchers used all of the obtained data to develop a mathematical model to diagnose whether someone is likely to have COVID-19 based on their age, sex, and a combination of four symptoms: anosmia (loss of scent and/or taste), severe or persistent cough, fatigue, and not eating. They found that the AI ‘s predictions were accurate almost 80 percent of the time. Among the more than 800,000 users of the app who claimed they were not well, only up to 20% were predicted to actually have COVID-19.

Why this matters

Combining this AI forecast with universal use of the software may help classify those that are likely to be contagious as soon as the early signs begin to emerge, concentrating on monitoring and evaluating when they are most important. As the pandemic continues to spread and take lives daily, it is imperative to recognize symptoms as early as possible to save the lives of countless people and to bring hope that this will come to an end.
Even if the precision of the test leaves a lot to be desired, it may help to resolve the limitations of coronavirus research, limitations that include mainly the amount of daily tests a country can do and to prevent overcrowding of research facilities.

Keep in mind that while this pre-diagnosis can be helpful for further testings, it does not substitute them. It is meant as a means to support the tests, and most importantly – to rule out any cases that have an improbability of being actually COVID-19.

Reference: Menni, C., Valdes, A.M., Freidin, M.B. et al. Real-time tracking of self-reported symptoms to predict potential COVID-19. Nat Med (2020). https://doi.org/10.1038/s41591-020-0916-2