Researchers have developed and validated an algorithm that can help healthcare professionals identify who is most at risk of dying from COVID-19 when they are admitted to a hospital, reports a study published today in eLife.
The tool, which uses artificial intelligence (AI), can help physicians direct critical care resources to those who need them most, and will be particularly valuable to resource-constrained countries.
“The presence of new SARS-CoV-2 variants, declining immune protection and mitigation measures means we are likely to continue to see increases in infections and hospitalizations,” explains the leader of this international project and senior author David Gómez-Varela , fhv. Max Planck group leader and current senior researcher at the Department of Pharmacology and Toxicology, University of Vienna, Austria. “Clinically valuable and generalizable triage tools are needed to help allocate hospital resources to COVID-19, especially in places where resources are scarce. But these tools need to be able to cope with the ever-changing scenario in a global pandemic and must be easy to implement. “
To develop such a tool, the team used biochemical data from routine blood tests performed on nearly 30,000 patients admitted to over 150 hospitals in Spain, USA, Honduras, Bolivia and Argentina between March 2020 and February 2022. This means that they were able to capture data from people with different immune statuses – vaccinated, unvaccinated and those with natural immunity – and from people infected with each SARS-CoV-2 variant, from the virus that originated in Wuhan, China, to the latest Omicron variant. “The inherent variability of such a diverse dataset is a major challenge for AI-based prediction models,” says lead author Riku Klén, associate professor at the University of Turku, Finland.
The resulting algorithm – called COVID-19 Disease Outcome Predictor (CODOP) – uses measurements of 12 blood molecules that are normally collected during hospitalization. This means that the predictable tool can be easily integrated into the clinical care of any hospital.
CODOP was developed in a multi-step process that initially used data from patients admitted to more than 120 hospitals in Spain to ‘train’ the AI system to predict the characteristics of a poor prognosis.
The next step was to ensure that the tool worked regardless of the patients’ immune status or COVID-19 variant, so they tested the algorithm in several subgroups of geographically dispersed patients. The tool still performed well in predicting the risk of death in hospital during this fluctuating scenario of the pandemic, suggesting that the measurements CODOP is based on are really meaningful biomarkers for whether a patient with COVID-19 is likely to worsen.
To test whether the timing of blood tests affects the performance of the tool, the team compared data from different blood sampling times before patients either recovered or died. They found that the algorithm can predict survival or death of hospitalized patients with high accuracy until nine days before both outcomes occur.
Finally, they created two different versions of the tool for use in scenarios where health resources are either functioning normally or are under severe pressure. Under normal operational burden, physicians may choose to use an “overtriage” version, which is very sensitive to picking up people at increased risk of death, at the expense of detecting some people who did not need critical care. The alternative ‘undertriage’ model minimizes the possibility of incorrectly selecting people at lower risk of dying, and gives doctors greater assurance that they are providing care to those at highest risk when resources are severely limited.
“CODOP’s performance in diverse and geographically dispersed patient groups and its ease of use suggest that it could be a valuable tool in the clinic, especially in resource-constrained countries,” notes Gómez-Varela. We are now working on a follow-up dual model tailored to the current pandemic scenario of increasing infections and cumulative immune protection, which will predict the need for hospitalization within 24 hours for patients in primary care and intensive hospitalization within 48 hours for those who “We hope to help healthcare systems restore previous standards of routine care before the pandemic takes hold.”
The CODOP predictor is freely available at: https://gomezvarelalab.em.mpg.de/codop/