AI algorithm identifies hospital patients most likely to die from COVID-19

AI algorithm identifies hospital patients most likely to die from COVID-19

Photo: Early Warning System predicts who needs critical care for COVID-19 (Image courtesy of Unsplash)

Scientists have developed and validated an algorithm that could help healthcare professionals identify people who are most at risk of dying from COVID-19 when they are hospitalized. The tool, which uses artificial intelligence (AI), could help clinicians direct critical care resources to those who need them most, and would be especially valuable for resource-limited countries.

To develop the tool, an international team led by the University of Vienna (Vienna, Austria) used biochemical data from routine blood draws performed on nearly 30,000 hospitalized patients in more than 150 hospitals in Spain, the United States, Honduras, Bolivia and Argentina between March 2020 and February 2022. That means they were able to capture data from people with different immune conditions — vaccinated, unvaccinated, and naturally immune — and from people infected with every type of SARS-CoV-2, from the virus that emerged in Wuhan, China, to the latest Omicron variant.

The resulting algorithm – called the COVID-19 Disease Outcome Prediction (CODOP) – uses measurements of 12 blood molecules that are naturally collected during admission. This means that the predictive tool can be easily integrated into the clinical care of any hospital. CODOP was developed in a multi-step process, initially using data from hospitalized patients in more than 120 hospitals in Spain, in order to “train” the AI ​​system to predict the hallmarks of a poor prognosis. The next step was to make sure the tool worked regardless of patients’ immune status or COVID-19 variant, so they tested the algorithm in several subgroups of geographically distributed patients. The tool still performs well in predicting the risk of in-hospital death during this volatile epidemic scenario, indicating that CODOP-based measurements are really useful biomarkers of whether or not a patient with COVID-19 is likely to deteriorate.

To test whether when the blood tests were taken affects the tool’s performance, the team compared data from different time points of blood drawn before patients recovered or died. They found that the algorithm could predict hospital patients’ survival or death with high accuracy up to nine days before either outcome occurred. Finally, they created two different versions of the tool for use in scenarios where healthcare resources are operating normally or are under severe stress. Under normal operating burden, clinicians may choose to use the “overload” version, which is very sensitive in capturing people at increased risk of death, at the expense of detecting some people who did not need critical care. The alternative “no responsibility” model reduces the possibility of wrongly selecting people at lower risk of death, providing clinicians with greater certainty that they are directing care to those most at risk when resources are most limited.

“CODOP’s performance in diverse and geographically dispersed patient groups and ease of use suggest that it can be a valuable tool in the clinic, particularly in countries with limited resources,” said this international project leader and first author David Gómez-Varela. , former leader of the Max Planck group and current chief scientist at the Department of Pharmacology and Toxicology at the University of Vienna. “We are now working on a dual follow-up model tailored to the current pandemic scenario of increased infection and cumulative immune protection, which will predict the need for hospitalization within 24 hours for patients within primary care, and admission to intensive care within 48 hours for those already hospitalized. Hopefully in helping health care systems restore previous standards of routine care before the outbreak of the pandemic.”

Related links:
University of Vienna

2022-05-18 12:45:15

Leave a Comment

Your email address will not be published. Required fields are marked *