AI identifies severe and fatal Covid courses
If a blood sample could be used to predict whether a Covid patient WILL take a severe course or whether the disease could even lead to death, that would help doctors with therapy. A research team from Germany and Austria is now presenting an AI-supported method that can do this in the journal “PLOS Digital Health”. That brings more information for doctors – ultimately also in dreaded triage situations, according to the South Tyrolean biochemist Markus Ralser.
It is important to recognize critical developments early on
Last year, Ralser, who works at the Berlin Charité and at the Francis Crick Institute (Great Britain), developed a method with colleagues that allows the remarkable protein structure in blood samples (proteome) to be determined using rapid and inexpensive mass spectrometry. Depending on how the body reacts to pathogens, death also changes the colorful bouquet of metabolic products.
New “Scanning SWATH” technology
In the work published last year in the journals Nature Biotechnology and Cell Systems, the team indicated that the new Scanning SWATH technology is also suitable for identifying proteins that indicate the severity of a Covid-19 infection Show . Since the method generates very complex data, the analysis requires the support of computer algorithms based on machine learning. A total of 54 proteins were discovered using artificial intelligence (AI), which serve as an indicator of the severity of the disease.
From this approach, Ralser and colleagues wanted to “pull out something clinically useful,” the researcher told the APA. “Then the question at the table was: Can you not only map the course of the disease as it looks at the moment, but can you also look into the future?”
In the new study, which was supported by numerous researchers from Innsbruck, including the head of the internal intensive care unit at Innsbruck University Hospital, Michael Joannidis, the prognosis was based on the proteomic data. The scientists analyzed 349 samples taken at different times from 50 patients with very severe Covid 19 courses who were treated at the Charité and the Innsbruck University Clinic. Once again, the new method and the AI approach are used to search among 321 quasi-suspicious proteins for those that indicate that a patient is more likely to survive.
AI in action
It turned out that even in this group with the most severe courses imaginable, 14 specific proteins strongly indicate whether someone was one of the 15 patients who did not survive as the disease progressed. The team then implemented a system, again using machine learning, that draws conclusions about the outcome of the disease from just one blood sample. In another group of very seriously ill patients, the prognoses largely matched the actual outcome: 18 of 19 patients who survived were correctly identified, as were the five who died.
Such “molecular signatures” also enable the course to be assessed even in a situation in which intensive care physicians can no longer predict how the clinical picture will develop, explained Ralser. Of course, this assessment is not 100% clear, “but it’s much better than it was before.”
One application for the method is to find out quickly and reliably, for example in smaller clinical studies, whether a drug has the desired effect. “The second situation that is of course in the room is the triage situation,” says Ralser. Especially in such a state of emergency, the doctor needs as much information as possible that he can get. However, a decision based solely on such a forecast cannot and should not be made, emphasized the researcher: “What we achieve is that there is as much information as possible on the table when such a decision is due.”
Prematurely watch out for difficult processes
Overall, the proteome data shows how much the metabolism changes in Covid-19. “It’s simply a matter of not expecting severe courses in selected stages,” emphasized Ralser, because the disease still offers few clues for clinicians to reliably assess a patient’s condition in the coming days and to carry out treatments. The particularly uncertain prognosis with sudden damage is considered a major problem in Covid 19 treatment. “This is very much dependent on the activation of one’s own immune system, and you can see that in the protein curves,” explained Ralser.
Now the focus is clearly on Covid-19. However, the approach using molecular markers can also be transferred to other infectious diseases. “This is a modern art of medicine that we are developing,” said Ralser with conviction.
Service: The “PLOS Digital Health” paper: https://doi.org/10.1371/journal.pdig.0000007; The work in “Nature Biotechnology” and “Cell Systems”: https://doi.org/10.1038/s41587-021-00860-4 and https://doi.org/10.1016/j.cels.2021.05.005
(APA/red, Photo: APA/HANS PUNZ)