A team of researchers led by NYU has developed an AI tool and used it to identify which patients diagnosed with COVID-19 would go on to develop severe respiratory disease.

About 80 percent of those who get the virus only experience mild symptoms. But others can go on to develop dangerous, life-threatening symptoms and need intensive care, such as oxygen and ventilation, to survive. 

As COVID-19 is still new to the medical community, it can be hard for doctors to know straightway which cases will escalate to critical illness. In that regard, the researchers hope a version of this AI tool could be widely used, in the near future, to help doctors make critical decisions regarding which patients will need their care and which they can send home. 

“While work remains to further validate our model, it holds promise as another tool to predict the patients most vulnerable to the virus,” Megan Coffee, corresponding author of the study and clinical assistant professor in the Division of Infectious Disease and Immunology within the Department of Medicine at NYU Grossman School of Medicine, said in a news release. 

Through their study, the researchers determined their AI tool was able to predict an individual’s chances of developing acute respiratory distress syndrome (ARDS), a medical term for fluid build-up in the lungs, with 80 percent accuracy. That’s significant, since ARDS is a common killer, particularly for older COVID-19 patients and those with underlying health issues.

Demographic, radiological and laboratory findings were collected from 53 COVID-19 patients who tested positive for the virus in January at two Chinese hospitals, Wenzhou Central Hospital and Cangnan People’s Hospital. The majority of the patients experienced mild symptoms to begin with, including cough, fever and stomach ache. A few patients, however, developed more severe symptoms, including pneumonia, within a week. 

The researchers then used that data to train computer models designed to get smarter as more information is inputted. 

“Our goal was to design and deploy a decision-support tool using AI capabilities — mostly predictive analytics — to flag future clinical coronavirus severity,” co-author Anasse Bari, a clinical assistant professor in computer science at NYU’s Courant institute of Mathematics, said in the release. 

“We hope that the tool, when fully developed, will be useful to physicians as they assess which moderately ill patients really need beds and who can safely go home, with hospital resources stretched thin,” she added. 

Notably, the researchers were surprised to find that some characteristics widely considered to be “hallmarks” of COVID-19, like certain patterns in lung images, fever and strong immune responses, were not all that useful in determining which COVID-19 patients with mild symptoms would go on to develop severe lung disease. Neither age nor gender were helpful in predicting who would develop serious disease, either. Previous studies, however, have found that men over 60 are at highest risk. 

Instead, the AI tool found three factors —  levels of the liver enzyme alanine aminotransferase (ALT), reported myalgia (deep muscle aches) and high hemoglobin levels (the iron-containing protein that helps blood cells carry oxygen to bodily tissues) — to be the best indicators of future severe symptoms. 

The researchers determined that ALT levels, which increase when diseases like hepatitis harm the liver, were only a bit higher in COVID-19 patients. But they still played a big role in predicting disease severity.

Myalgia was also more common in COVID-19 patients, and past research has linked myalgia to inflammation in the body. 

And lastly, high levels of hemoglobin were also determined as predictors of respiratory distress. But, according to the release, this could be “explained by other factors, like unreported smoking of tobacco, which has long been linked to increased hemoglobin levels.”

The study had some limitations, according to the researchers, including the relatively small data set and limited clinical severity of the disease in the patients studied. The latter may in part be due to the lack of elderly patients included in the study, they explained. The average patient age was 43. 

“I will be paying more attention in my clinical practice to our data points, watching patients closer if they, for instance, complain of severe myalgia,” Coffee said in the release. “It’s exciting to be able to share data with the field in real time when it can be useful. In all past epidemics, journal papers only published well after the infections had waned.”

The study, although perhaps the most substantial, is just one example of medical experts’ interest in using AI to detect which COVID-19 patients are likely to experience the worst symptoms. 

Stanford researchers recently launched an initiative of their own to test whether an older AI tool could identify which COVID-19 patients need intensive care before their health fully deteriorates. 

The tool, which was built by the electronic health records vendor Epic, looks back on patients’ data and gives them a score from 0 to 100 based on how sick they are and how in need of care they may be. Currently, the researchers are working to validate the model in terms of how effective it is in regards to evaluating the needs of COVID-19 patients. And if they determine it to be effective, it will be administered throughout Stanford’s network of hospitals. 

In Israel, the nation’s largest hospital has administered an AI model geared towards predicting which COVID-19 patients will experience respiratory failure in the next 6-8 hours. 

And in China, researchers have reportedly built AI models capable of predicting which patients have a high chance of malignant progression. Their study has yet to have been peer-reviewed, though.