Predicting COVID-19 Outcomes in Emergency Department Patients | Emergency Medicine | JAMA | JAMA Network
[Skip to Navigation]
Sign In
Views 7,856
Citations 0
Biotech Innovations
February 9, 2021

Predicting COVID-19 Outcomes in Emergency Department Patients

JAMA. 2021;325(6):522. doi:10.1001/jama.2021.0227

A new artificial intelligence algorithm uses chest x-ray severity scores and clinical variables collected during emergency department (ED) visits to predict whether patients with coronavirus disease 2019 (COVID-19) will be intubated or will die. If further validated in larger studies with additional patient populations, the proof-of-concept model, described in Radiology: Artificial Intelligence, could be used to appropriately triage ED patients before they become seriously ill with COVID-19.

Researchers at the Icahn School of Medicine at Mount Sinai in New York City trained and validated the algorithm using deidentified electronic health record data from patients with COVID-19 who were treated at 3 Mount Sinai Health System EDs in the Brooklyn, Manhattan, and Queens boroughs over 2 weeks last March. The training and validation data comprised outcomes, x-rays with radiologist-assigned severity scores, blood pressure readings, and blood work like creatinine, D-dimer, and troponin levels from 338 patients aged 21 to 50 years.

The algorithm learned how to assign severity scores to x-rays. The researchers then tested the model’s ability to predict outcomes for 161 different patients with COVID-19 seen at the EDs the following week in March. Two-thirds of this group were older than 50 years. Overall, the approach predicted intubations with 86% accuracy and deaths with 82% accuracy, but it performed better for younger patients.

“The algorithm can help clinicians anticipate acute worsening (decompensation) of patients, even those who present without any symptoms, to make sure resources are appropriately allocated,” Fred Kwon, PhD, the study’s lead author, said in a statement. The chest x-ray severity scores alone could be useful. “We are working to incorporate this algorithm-generated severity score into the clinical workflow to inform treatment decisions and flag high-risk patients in the future,” Kwon said.