Using AI to Predict COVID-19 Outcomes

Shibani S. Mukerji, MD, PhD, associate director of the Neuro-Infectious Diseases Unit of the Department of Neurology at Massachusetts General Hospital, talks about her work on developing the COVID-19 Acuity Score (CoVA), which can be incorporated into an electronic health record system. In this video, Dr. Mukerji elaborates on the use of the CoVA score to identify the key characteristics of patients that may be hospitalized within a week.

Subscribe to the latest updates from Neuroscience Advances in Motion

Thank you for subscribing!

Error: Please enter a valid email address.

Email Address Submit

Transcript

So we developed a CoVA score, which was one of the most exciting studies that we've been part of and has been a really lovely collaboration. So the COVID acuity score that providers can use to determine whether or not the patient in front of them may have an increased risk of hospitalization, ICU admission or death within the next seven days. And so this collaboration, I think, is fun because it was done with Brandon Westover here at Mass General as well as his team of engineers and we'd already had this established collaboration looking at machine learning for HIV to try and identify our brain age. We translated that, sort of, machine learning and sort of, this group collective understanding to try to see if we can identify using the electronic health medical record—really try to identify what are the key characteristics of patients who are going to be seen at a respiratory outpatient clinic that may eventually be hospitalized again within the next seven days. And so the CoVA score is looking at multiple factors that include vital signs, that included past medical history, and in general, once we took a look at all potential factors. We identified about 30 risk predictors that could identify with quite high sensitivity, individuals that would potentially be at risk for a negative outcome.

The reason why it is extremely exciting and important is that it was built off of a development set of over 10,000 patients and then we were able to use over 2,000 patients to validate that score, to make sure what we had developed could be applied to another data set. So using that, what we'll be using this CoVA score for is to put them in respiratory clinics potentially is our goal, building it hopefully into the EPIC system where we can then show providers, "Here's the score, here is some of these risks that this patient in front of you may have to eventually have a negative outcome in the next seven days."

Learn more about the Department of Neurology

View all COVID-19 updates