Researchers at Northwell’s Feinstein Institutes for Medical Research have developed an AI-powered predictive tool they say can assess patients for their risk of respiratory failure within 48 hours.
WHY IT MATTERS
New research, led by assistant professors Theodoros Zanos and Dr. Douglas Barnaby of the Feinstein Institutes, and published in the Journal of Medical Internet Research, showed accuracy in identifying at-risk patients for earlier interventions such as critical care consultation and closer patient monitoring.
The project centered on electronic health record data from 11,525 patients who were admitted to 13 Northwell hospitals in spring of 2020, when the pandemic was peaking in the New York area. Of those patients, 933 (8%) were placed on ventilators within 48 hours of admission.
To develop those predictive clinical tools, Northwell researchers also used other EHR data, such as vitals and labs, along with demographic information.
They created and compared three machine learning models to help score patients upon admission. One of the tools – gradient boosted decision trees, or XGBoost – had a predictive accuracy of 92%, the highest of the three.
That data can help predict which patients may be most at risk of respiratory failure – helping triage those who need the most interventions to reduce mortality.
The model performance was compared against (and outperformed) the Modified Early Warning Score to calculate the chance of respiratory failure within 48 hours. Having shown that level of accuracy, Northwell will deploy the new tool at some of its hospitals.
Just this past week, we reported on a new model from Israeli researchers that can potentially predict the length of COVID-19 hospitalization using an array of clinical and demographic data and also assess the chance of in-hospital mortality.
ONC’s new Synthetic Health Data Challenge hopes to build on these advances, looking for other new approaches to AI-powered analytics, seeking innovative algorithms for disease modeling.
ON THE RECORD
“Respiratory failure is the leading cause of death among patients with COVID-19, and it is imperative that our clinicians on the front lines have the most unbiased, informative data to aid their decision-making in the emergency department,” said Zanos, assistant professor at The Feinstein Institutes’ Institute of Bioelectronic Medicine, in a statement.
“The clinical tool we propose could identify patients who are at greater risk of deteriorating shortly after being admitted to the hospital and helps show that machine learning and AI can be powerful clinical assets in treating this complex virus.”