Healthcare AI solutions tout their ability to identify more at-risk patients and irregularities imperceptible to physicians, all while keeping a human in the loop.
But are there enough humans to handle all this additional demand brought on by AI?
That’s a question healthcare leaders are grappling with as the technology expands across the industry.
“You don’t want to be, say, implementing something that’s going to scan every patient for a particular disease, which costs you a lot of money if you can’t do anything about it, because you don’t have the appointments downstream to actually manage that,” said Michael Pfeffer, MD, senior vice president and chief information and digital officer of Palo Alto, Calif.-based Stanford Health Care, at Becker’s 16th Annual Meeting in April. “So you have to look at the entire workflow and value chain to see: Is it the right tool to put in?”
As for keeping a human in the loop on AI, Dr. Pfeffer said that’s just not feasible — or even necessary — in every instance. Research is showing that physicians increasingly trust AI and are not going to check every summary and citation the technology makes. Where the human element becomes critical is if, say, AI detects a hospital patient is deteriorating from a lack of fluids — a human clinician then has to administer fluids.
“We’ve been thinking exactly about the same thing, and we hold ‘human in the loop’ as sort of a bulwark for safety,” said Sri Adusumalli, MD, vice president and chief health information officer of Philadelphia-based Penn Medicine, during the panel discussion. “But we know we humans are terrible at vigilance of algorithms and other technology tools. So banking on humans in the loop as that bulwark is not sustainable. Plus, there are not enough humans.”
Dr. Adusumalli is a cardiologist by trade, and in the cardiology space there are a number of AI products that scan CT scans or EKGs to surface risk or predict the presence of a disease.
“Well, what do you do based on that? You need to get an ultrasound, an echo. Well, how long is the wait? How long does it take to get to the echo then? How long does it take to get in for the appointment to then action that?” he said. “There are opportunities for AI to be able to help augment and enable those downstream processes too. But we need to think about it all together rather than just implementing that one tool that might identify the opportunistic finding.”
At Philadelphia-based Jefferson Health, neurorestoration staffers developed AI to review MRIs from the past year to flag potential neurological conditions and then bring the patients back in, said Luis Taveras, PhD, executive vice president and chief digital and information officer, at the panel.
“But are we ready to handle that? Because it’s coming back with a lot of great answers, a lot of great things that we need to look at,” he said. “But the team is not … big enough to handle the volume that is coming back out of this.”
Stanford’s AI framework includes utility, determining, for example, if the health system were to launch an application that analyzes echocardiograms, how many more cardiologists would it need to handle the increase in volume, Dr. Pfeffer said.
“Let’s do the calculation before we roll something out and create chaos,” he said. “You really need to look at the entire process. And if the tool is really that good and it’s going to be that beneficial to patients, then you need to fix the receiving parts of the workflow so that you then can deploy the tool.”
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
