By Jessica Kent
December 21, 2020 – In the US, the dawn of winter has meant a surge of COVID-19 infections. Daily rates of new infections have jumped significantly since March, and the country recently surpassed 17 million confirmed coronavirus cases – a grim milestone that will only place more strain on an already-burdened healthcare system.
To get ahead of poor COVID-19 outcomes – and in turn, effectively allocate necessary resources – researchers are turning to artificial intelligence and machine learning technologies.
In a study published in Radiology: Artificial Intelligence, a team at Mount Sinai described an artificial intelligence tool that can rapidly predict outcomes of COVID-19 patients in the emergency room.
Using EHRs of patients between 21 and 50 years old, combined with their lab tests and chest x-rays, researchers trained the algorithm to determine who is at highest risk of intubation or death within 30 days of arriving at the hospital.
Because many patients with COVID-19 show non-specific symptoms when they come to the emergency room – such as cough, fever, and respiratory issues – clinicians can’t easily identify patients who will deteriorate quickly. The algorithm can calculate the likelihood that patients will need intubation before they get worse, helping providers make more informed care decisions.
READ MORE: Artificial Intelligence Identifies Asymptomatic COVID-19 Infections
The researchers found that the model could predict intubation and death within 30 days of arriving at the hospital with a sensitivity of 82 percent.
While other algorithms that predict COVID-19 outcomes do exist, these tools are used in admitted patients who have already developed more severe symptoms and have additional imaging and laboratory data taken after hospital admission.
With this new algorithm, providers can predict outcomes in COVID-19 patients while they’re in the emergency room, even in those with mild symptoms. The tool only uses data from the initial patient encounter in the emergency department.
“Our algorithm demonstrates that initial imaging and laboratory tests contain sufficient information to predict outcomes of patients with COVID-19. 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,” said Fred Kwon, PhD, Biomedical Sciences at the Icahn School of Medicine at Mount Sinai.
“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.”
READ MORE: How Artificial Intelligence, Big Data Can Determine COVID-19 Severity
Researchers at the University of California, Irvine (UCI) are also leveraging analytics tools to predict patient outcomes from the virus. A team of health sciences investigators recently developed a machine learning model to predict the likelihood that a COVID-19 patient will need a ventilator or ICU care.
Using a patient’s medical history, the tool can determine whether a patient’s condition will worsen within 72 hours, and is freely available online for any healthcare organization to use.
Researchers found that at UCI Health, the tool’s predictions were accurate about 95 percent of the time.
“The goal is to give an earlier alert to clinicians to identify patients who may be vulnerable at the onset,” said Daniel S. Chow, an assistant professor in residence in radiological sciences and first author of the study.
Researchers began collecting COVID-19 patient data at UCI Health in January 2020, enabling them to create a prototype of the tool by March. The algorithm uses pre-existing conditions, like asthma, hypertension, and obesity; hospital test results; and demographic data to estimate the likelihood that a patient will need a ventilator or ICU care.
READ MORE: Artificial Intelligence Could Speed COVID-19 Detection, Treatment
Although the study was based on UCI Health patients, researchers also tested the algorithm with 40 patients at Emory University in Atlanta to see whether it worked with a different patient population, with successful results.
The algorithm could help hospitals and health systems plan for COVID-19 surges, researchers noted.
“We might think about this tool in terms of predicting the number of ICU beds that we might need,” said Alpesh N. Amin, the Thomas & Mary Cesario Chair of Medicine and a study author.
While the calculator can predict the general severity score of COVID-19 patients at any hospital, it’s up to clinicians to make decisions on how to proceed based on local practices, number of beds, number of patients, and likely spread of the disease locally.
At UCI Health, the tool has guided patient care based on feedback from emergency, hospital medicine, critical care, and infectious disease physicians.
“You have to talk to your specialists, your doctors; you have to assess how many beds you have available and come together as a group to figure out how you want to use the tool,” said Peter Chang, the assistant professor in residence in radiological sciences who designed the machine-learning model.
Going forward, the team plans to expand the tool to other institutions and use it for further research. Their next study will aim to predict which patients are most likely to benefit from COVID-19 drug trials.