The research team trained an AI algorithm using clinical data to better triage patients in low-testing environments and identify the infection at the point of care, as standard screening methods often take up to 48 hours to produce a diagnosis.
Routine blood tests conducted during the patient’s arrival, blood gas results and vital signs were used to train the algorithm.
The AI model used for diagnosis among COVID-19-presenting patients in the emergency department resulted in 77 percent sensitivity and 96 percent specificity, while the model used for diagnosis among patients admitted with COVID-19 resulted in 77 percent sensitivity and 95 percent specificity. More than 99 percent of patients the algorithm diagnosed as negative were in fact free of the disease.
Ecditor’s note: This article was updated July 14 at 9:00 a.m. CDT.
More articles on artificial intelligence:
Harvard, Stanford researchers lead COVID-19 AI data initiative: 4 things to know
4 AI systems outperforming medical experts
How UC San Diego Health, AWS implemented an AI imaging algorithm to detect COVID-19 in 10 days
