What's Holding AI Back? Imagination

CommonSpirit Health CIO Daniel Barchi says AI development over the next five years won't focus on better technology, but on healthcare leaders finding the right way to use the tools to make healthcare more efficient.

The CIO of the nation's second largest nonprofit health system says AI governance isn't a revolutionary concept. It's a strategy built on how the industry has embraced new ideas and technologies in the past.

"It's important to remember that clinicians, hospitals and health systems have been caring for patients for many, many years through many advances in technology and many changes in clinical care," says Daniel Barchi, EVP and CIO of CommonSpirit Health, the Chicago-based network of 142 hospitals and more than 700 care sites spread across 21 states. "And if we hew to the guiding principles that clinicians have and that we as caregivers should aspire to have, we can apply those same ground rules, guiding principles and vision to AI in the same way that we've used other advanced tools safely."

Daniel Barchi, EVP and CIO of CommonSpirit Health. Photo courtesy CommonSpirit Health.

That's not to say that AI isn't causing problems with its rapid adoption, but Barchi says healthcare executives need to temper their concerns with a little common sense. They've been down this road before.

"Our goal is to make sure that we use it in the most efficient way for how fully it's developed at this point," he says. "And [we] use it as broadly as possible, with the proviso that we always have a clinician between the AI and the patient."

Reviewing AI Use Cases

CommonSpirit Health has a three-tiered approach to AI governance. Barchi says the health system runs any AI projects first through data management and patient advisory councils, then through the enterprise data and AI governance committee (EDAG), before finally going before the IT executive steering committee. Some 200 tools have made it through review and are now being used with the health system.

But there have been some that didn't make the cut.

"We've rejected 15 use cases for different reasons," he says. "Whether we didn't feel that they were clinically efficacious, [or] we were concerned about the ways that a third-party company might be using data, [or] whether or they were concerns about algorithmic bias."

Most, if not all, of those rejections come out of the EDAG committee, which meets every two weeks and is comprised of roughly 30 members, including ethicists, medical informaticists and representatives from legal, innovation, finance, clinical (including nursing), IT and cybersecurity departments.

"These people come together and evaluate every AI initiative that we have and determine if there are risks, if we're using data appropriately, if there are risks of algorithmic bias, what the upside is and whether we should approve it for use clinically and operationally in our health system," Barchi says.

Analyzing Agentic AI

He says he's particularly interested in how agentic AI evolves.

"We're creating the capability for tools to surface information, analyze it, make decisions and interact with others in ways that are very similar to what many of our colleagues do with data today," he says. "And management of these AI tools in agentic AI is almost akin to managing a team of workers. Thinking of this not as a technical process, but as a management challenge and a way to use operational efficiency safely is the next frontier for us as health system leaders."

"I anticipate over the next five years many of the advances and adoption of AI are not contingent on AI getting better," he adds. "It's health leaders thinking more intuitively about how AI can make our processes more efficient."

Which ties back to the idea that AI will replace doctors and nurses. Barchi says that isn't about to happen, as healthcare is still based on human interactions. But he does believe that healthcare organizations using AI will replace those who aren't on the bandwagon, and those using AI will become better at delivering healthcare.

"Physicians, nurses and other clinicians are more likely to be more thoughtful caregivers, because they can focus on the patient in front of them and allow AI to do more of the work behind them," he says. "And I've seen clinicians be very open-minded [about] what AI can provide them, whether it's data or insight, because they know at the end of the day, their overarching objective is improving the health of the patient in front of them."

That concept will also apply to patients using AI.

"We are better patients when we're better informed about our own health conditions," Barchi says. "We'll never be as educated as the neurosurgeon who's caring for us, but we have better insights to what he or she shares with us by being more educated and using AI and other tools to gain insight about the ways we might help with our own caregiving."

Charting a Future for AI

Looking ahead, Barchi sees AI evolving in two directions. He expects new tools and programs to be integrated directly into existing technology platforms like the EHR, so that workflows aren't negatively affected. He also believes that healthcare organizations will develop the capabilities to create their own AI tools.

"We'll simply adopt them in a way that nobody would go off and buy aftermarket parts for a car if you can buy them with the car itself," he points out. "And so yes, there will always be point solutions, but I think those are going to be fewer and more far between and developed internally. And then much of what we get from AI is simply going to be embedded in the tools that we buy and it's going to be harder for standalone companies to try to sell point solutions that are not embedded in our core platforms."

And he sees AI helping healthcare move out of the hospital, clinic and doctor's office and into the home.

"AI will begin to assimilate data and make inferences about our healthcare long before we as patients think about a clinical condition," he says. "It might monitor the number of times we open the refrigerator, the number of steps that we take, the way that we sleep, our online patterns, and look for patterns in a way that might inform an emerging condition long before you even begin to feel it physically."