Developed at UCI, New Tool Helps Predict Patients’ Risk of Coronavirus Complications
Monday, May 18, 2020
While most people who develop complications or require intensive care due to COVID-19 tend to be older or have other conditions such as diabetes or heart disease, that’s not the case for every patient.
Now UC Irvine scientists have created a tool that uses artificial intelligence to help predict which patients may need that higher level of care.
Already in use at UCI Medical Center and soon to be launched at five other UC campuses with affiliated hospitals, the risk prediction tool was created by a collaboration of 10 doctors and researchers at the Irvine campus.
A single patient can generate reams of medical data over time, but when they come to the hospital it may be a doctor’s first time seeing them. Instead of sifting through voluminous files, health workers can enter specific information – including results of a COVID-specific blood panel – into the AI tool, which can also grab data from the hospital’s records system.
“Sometimes even if you have the risk factors that you know you should be looking for it can be very difficult to find them,” said Dr. Peter Chang. Chang’s medical specialty is neuroradiology, or brain imaging, but when he’s off duty he develops algorithms to improve patient care at UCI’s Center for Artificial Intelligence in Diagnostic Medicine.
Once it’s got the data, the tool creates a personalized risk prediction score reflecting the probability that the patient will need ICU care or a ventilator in the next three days, Chang said – and it can continually adjust when new data is added.
The risk scoring tool doesn’t make recommendations for patient care, and it doesn’t override or replace doctors’ experience or clinical judgment, Chang said.
But it can give a less experienced physician confidence in their assessment, or it can indicate which patients may need to be watched closely, even if they come in with few obvious risk factors and “on paper they should be fine,” he said.
UCI Medical Center has seen a number of what would seem to be lower-risk patients in their 40s who ended up in intensive care with serious complications.
While researchers were building the tool, “there was a patient that seemed otherwise OK, was sent home but came (up) with a very discrepant score on the calculator,” Chang said. “Within a few days, the patient came back to the hospital and needed to be in the ICU.”
That’s the type of situation Chang hopes the AI tool can prevent. If the risk prediction conflicts with the doctor’s observations, the medical team may want to keep a closer eye on that patient or discuss prescribing drugs or other proactive treatments.
Researchers have used the tool with more than 100 COVID-19 patients who visited UCI Medical Center at some point and stayed at least one night to receive care, and other UC hospitals have contributed data to help the tool learn.
One small silver lining of the pandemic is that it’s engendered a new willingness to cooperate on projects that typically would take far longer, Chang said. He built the beta version of the risk prediction tool in about two weeks, working with specialists in pathology, neurobiology, data science and other fields.
Chang hopes the scoring system tool could be used to address other diseases or to spot trends in global health data that could alert people to the next pandemic.
“The amount of things you can do with this same model are virtually unlimited,” he said.