The pandemic has put immense strain on ICUs, resulting in shortages of staff, beds, personal protective equipment and ventilators. It has also exposed the limitations of traditional predictive algorithms used to predict patient outcomes, manage capacity, and inform triage decisions.

Use of artificial intelligence can help refine the raw data and present more useful knowledge, especially in the ICU setting, by separating the clinically-relevant information from the noise in a data rich environment.

"This results in earlier recognition of changes in patient conditions and their evolving risks," said Dr. John Frownfelter, chief medical information officer at Jvion. "It also allows you to see patients holistically by bringing in data on behavioral risk factors that you wouldn't be able to see from the clinical data in the EHR."

Frownfelter, who will discuss the impact of AI on predicting ventilator utilization next week at HIMSS21, said given the tremendous amount of data available per patient, AI has the potential to organize that data and provide meaningful knowledge that enables the clinician to act, as opposed to reams of "raw data" that can often distract from what's important.

"From an operational standpoint, you can achieve better outcomes and prevent complications, while enabling staff to be more efficient and effective in their care by focusing the right resources on the patients at greatest risk of deteriorating," he said.

Specifically, in the context of COVID-19, AI can help triage patients by predicting which patients are at the greatest risk of needing ventilator support or intensive respiratory support in the next 24 hours, and which patients are at greatest risk of dying.

Frownfelter explained this allows care teams to better anticipate which patients can safely be discharged, which patients should be ventilated, and which patients are at such a high risk of dying that they would be best served by hospice care.

In a surge, this intelligence allows care teams to better allocate resources and ensure there are enough ventilators and ICU beds for the patients whose lives depend on them.

"AI has the potential to be the mechanism by which we achieve rapid and accurate understanding of new problems and challenges in the clinical environment," he said. "Unfortunately, we have also learned that AI can be done well and it can be done poorly, resulting in errors of omission, bias magnification, and erosion of trust with the medical community."

With that in mind, Frownfelter highlighted the need for a rigorous methodology to develop and deploy AI-powered insights in a way that engenders trust.

Dr. John Frownfelter will discuss the role of AI in predicting ventilator usage at HIMSS21 in a session titled "The Implementation of AI to Predict Ventilator Utilization." It's scheduled for Wednesday, August 11, from 11:45 a.m.-12:45 p.m. in Venetian Murano 3201A.

Nathan Eddy is a healthcare and technology freelancer based in Berlin.
Email the writer: nathaneddy@gmail.com
Twitter: @dropdeaded209