The machine learning tool may help understand how and why some people develop long COVID symptoms.
Research led by Justin Reese of Lawrence Berkeley National Laboratory, Berkeley, CA, and Peter Robinson of The Jackson Laboratory for Genomic Medicine, Farmington, CT, has resulted in the development of a machine learning tool to help define the different types of long COVID.
The AI tool, described in a research paper published in eBioMedicine, analyzed the electronic health records (EHRs) of people diagnosed with long COVID.
According to the paper, after analyzing the EHRs of 6,469 individuals with a confirmed COVID-19 and a subsequent long COVID diagnosis, the team found six sub-types or clusters of long COVID symptoms.
These include clusters with distinct pulmonary, cardiovascular, and neuropsychiatric abnormalities, and a cluster associated with severe disease and increased mortality.
They also analyzed relationships between long COVID symptoms and pre-existing diseases, age, and other demographic factors. In addition, the AI tool was able to show that identified clusters were generalizable across different hospital systems.
“We compare all symptoms for the pair of the patients in this way, and get a score of how similar the two long COVID patients are. We can then perform unsupervised machine learning on these scores to find different subtypes of long COVID,” said Justin Reese in a news release.
According to the study authors, this ability to determine clusters among those with chronic COVID provides a base for classifying subgroups for treatment or therapies. In addition, the machine learning approach self-adapts to different record systems. This means researchers can collect data from a wide range of medical facilities.