How Artificial Intelligence, Big Data Can Determine COVID-19 Severity

When COVID-19 first started spreading across the country, hospital capacity and the allocation of resources were among the top concerns for healthcare leaders.

For more coronavirus updates, visit our resource page, updated twice daily by Xtelligent Healthcare Media.

In other countries, overcrowded ICUs, limited numbers of ventilators, and short-staffed facilities were the unfortunate reality, and a similar situation threatened to erupt in the US.

“In Northern Italy, doctors were put in awful, medically saturated situations where healthcare resources were over tapped, and that led doctors to have to make decisions about which patients get the ventilators or who gets the ICU beds, and that didn’t lead to good outcomes,” John McDevitt, PhD, professor of biomaterials at NYU College of Dentistry, told HealthITAnalytics.

“Our lab understood that we had to find a better way. The virus was soon going to be in America, and we knew we were going to be pushed in our resources here as well. The clinicians who have finite resources now need the ability to look into the future and prioritize their patients,” added McDevitt, who is also a professor of chemical and molecular engineering at NYU Tandon School of Engineering.

To help providers detect which patients are most likely to have severe cases of COVID-19, McDevitt and his team leveraged artificial intelligence and big data to produce COVID-19 severity scores.

Utilizing data from 160 hospitalized patients in Wuhan, China, the researchers identified four biomarkers measured in blood tests that were significantly elevated in patients who died versus those who recovered, including the C-reactive protein, myoglobin, procalcitonin, and cardiac troponin I.

These biomarkers can signal complications that are relevant to COVID-19, such as acute inflammation, lower respiratory tract infection, and poor cardiovascular health.

“From a data science perspective, we had the benefit of being able to look back at patterns and trends,” McDevitt said.  

The team then built a model using the biomarkers as well as age and sex – two established risk factors. Using machine learning algorithms, researchers trained the model to determine patterns of COVID-19 and forecast its severity. When a patient’s biomarkers and risk factors are entered into the model, it produces a numerical COVID-19 severity score ranging from zero to 100.

“The severity score basically tells the attending clinician the probability of mortality from the information that comes from these specific tasks,” McDevitt said.

“We were able to further validate the model in New York City, the epicenter of the pandemic. We also developed this clinical decision support model into a mobile app that is delivered to clinicians at the point of care.”

The overall aim of the validated model and app is to arm providers with the data needed to make informed care decisions, leading to better outcomes for patients.

“The real motivation behind this was to put actionable data into the hands of clinicians and help them make judgments on life-and-death decisions for their patients. We wanted to render that information in a way that’s much faster than following the more normal time course of enrolling patients, watching symptoms evolve over multiple days, and looking at chest x-rays, biomarker profiles, and oxygen saturation levels,” McDevitt explained.  

“It’s series of different clinical metrics that become part of the information flow to the doctor, and clinicians can learn in real time how to deal with this disease. Providers don’t have beautiful flow charts that say exactly what to do with the patients. A lot of healthcare providers have to figure this out, and we’re still figuring this out as a medical and scientific society.”

The COVID-19 severity score builds off of a model McDevitt and his team had previously developed to forecast outcomes for patients with cardiovascular disease.

“The study of the 160 patients from Wuhan was from a cardiac population, so they specialize in cardiovascular complications. It put us into our sweet spot, going back again to this cardiac scorecard situation,” McDevitt stated.

“We also see a lot of patients with cardiac complications in New York City. These patients tend to have the most severe complications and ultimately many of them pass away from the disease.”

While the researchers’ lab is located in Manhattan, the team retrospectively evaluated the app in the Family Health Centers at NYU Langone in Brooklyn, which serve more than 102,000 patients each year.

“We found that there’s actually a bigger need in the adjacent boroughs of New York City. That’s again in the epicenter, and represents the linkage off into the community clinics where people are often undertreated in more rural settings,” McDevitt said.

“It’s really important to be able to make the call on an individual before their health goes rapidly downhill. To have lead time and to be able to determine which patients are at elevated risk when they first approach the hospital is a really key gap that currently exists. This severity test has big health consequences that deal with this issue of the life-and-death decisions that clinicians need to make.”

Right now, the COVID-19 app can be used only with existing laboratory tests and requires oversight by an authorized clinician. Ultimately, the team hopes to develop and scale the ability to test a drop of blood for COVID-19 severity biomarkers – similar to how diabetic patients can determine their blood sugar levels, McDevitt noted.

“Our vision was that we would ultimately have a gadget that is as simple as a blood glucose measurement system – a glucometer that takes a drop of blood and shoots back a glucose value. With a tool like that, the patient themselves knows what to do if the value is above or below 100. There is an actionable thing to do based on that single number,” he concluded.