Machine learning – or artificial intelligence-based computer algorithms that improve automatically through experience by using the collected data – was developed during the study and was found to be able to predict the risks at the different stages of illness.
The researchers studied 3,944 positive cases in Denmark and used positive cases taken by UK Biobank for "external validation" and took common risk factors such as age, BMI and hypertension into account to formulate the algorithm.The AI model predicted risk of death at different stages: at diagnosis, at hospital admission, and at Intensive Care Unit (ICU) admission.
Out of the 3,944 patients who were tracked for the study, 324 died of COVID-19. The men who died were all between 73 and 87 years old with clear signs of high blood pressure and BMI impacting the results.This group of men proved to be the one with the highest risk of mortality as a result, and so the AI program would predict that men in that age range with high blood pressure and BMI are at higher risk.
Surprisingly, some of the top risk features "shifted towards markers of shock and organ dysfunction in ICU patients" rather than the aforementioned common risk factors.
The study developed an algorithm which managed to predict the risk of death and the findings were further backed by the results in the external validation cohort.
Such technology could help hospitals and medical care facilities throughout the world take extra preventative measures and may help prioritizing some patients over others and therein preventing unnecessarily high mortality rates.
This is not the first study to present the potential use of machine learning in taking preventative measures amid the coronavirus pandemic. The Copenhagen study, however, points out that these studies focused on patients already admitted to the hospital while it is unclear "whether the classification ability transfers to other healthcare systems." Another concern was that they were not entirely accurate machine learning algorithms because they did not take milder cases into account.
In addition, the previous studies, according to the researchers, were based on Chinese models which are vulnerable to bias.