AI NOW: Paving the Way for True Transformation

A panel of healthcare executives at this week’s HealthLeaders virtual summit identifies key opportunities and challenges to deploying agentic AI

Healthcare organizations are embracing agentic AI as a care management and clinical decision support platform, but it requires some fancy footwork around data management and protection and a clear understanding of what you’re accomplishing.

“What defines safety?” asked Jim Blum, MD, FCCM, AME, Chief Health Information Officer at University of Iowa Health Care. “What are the human interventions that need to be [integrated], and what are the human brakes that need to be built into agentic AI. [That] is going to be a real significant concern going forward in the next three to five years.”

Blum was a participant in this week’s HealthLeaders AI NOW virtual summit, joining Byron Yount, PhD, Chief Data and AI Officer at Mercy, and Zafar Chaudry, MD, MS, MIS, MBA, SVP, Chief Digital Officer and Chief AI and Information Officer at Seattle Children’s, on a panel that focused on taking the next steps with agentic AI.

Using Agentic AI Now

The hour-long panel gave the audience an opportunity to look at how three prominent health system executives are embracing agentic AI to address very specific issues, a crucial caveat to using AI in healthcare. For Blum, the issue is patient handoffs, and the health system contracted with a company called Evidently to develop an AI tool that surfaces patient data across the enterprise.

The tool, he said, offers not only a summary, but “the missing elements that you need to complete this evaluation.” Taken further, he sees the tool offering recommendations on care pathways.

“That's an area in healthcare where there's a lot of handoffs, [and] there can be a lot of drops and errors,” noted Yount, whose organization is also developing agentic AI tools for handoffs. “it's also a space where a lot of information needs to be moved from one care team to the next because the care team doesn't move with the patient throughout the system, either going from the emergency department to the floor or the floor to the home.”

Byron Yount, PhD, Chief Data and AI Officer at Mercy. Photo courtesy Mercy.

“So we've built agentic systems that take a lot of the information that's being collected on the patient, both during that encounter as well as historically, and create the standard handoffs that are needed,” he added.

At Seattle Children’s, Chaudry says the health system developed its own AI tool in-house, called Pathway Assistant. The tool assesses thousands of PDFs on traditional pediatric care treatments and creates a platform to guide providers on care management.

“That agent now can interact with the clinician,” he said. “Clinicians can ask questions, can feed in information about that patient, [such as] the age, the weight, the symptoms. And then the agent will start interacting with the clinician by asking the clinician further questions, probing deeply, asking the clinician to … do certain tests, building a differential diagnosis stack, at which point it's great for trainees, but it's also great to get to the answers very, very quickly.”

Build or Buy?

As agentic AI becomes more popular, health systems and hospitals are tasked with deciding whether to build in-house or partner with vendors. Chaudry said building in-house gives Seattle Children’s complete control over its own data, a key element in ensuring that PHI isn’t compromised.

At UI Health, Blum says the health system has to think very carefully about partnerships.

“The big concern that we have with AI is how is the data being used to train models, in particular large language models and potentially world models, as it goes out into our vendors’ hands,” he said. “Our solution to that has just been, in addition to meeting all the security requirements, that you have to have a BAA with us.”

James Blum, MD, FCCM, AME, Chief Health Information Officer at University of Iowa Health Care. Photo courtesy UI Health.

“You just can't train on our data, period,” he added. “That is our solution. We do not believe that you can truly de-identify to the point that particularly with large language [models], well, you will not have PHI [or] identifiable information embedded.”

Yount said a health system has to check several boxes in developing AI tools, whether in-house or through a vendor partnership.

“Making sure that you've got the guardrails in place, how decision rights work, where the accountability boundaries are, how you escalate, how you report events, the oversight structures, all that needs to be thought out in advance,” he said.

Finding the Right Positives

On the other side, though, the potential benefits are intriguing.

“I think it is a tool on a tool belt,” Chaudry said. “It's going to help our clinical teams focus on what they wanted to do in the first place, which was take care of patients and not sit at the desk and type away forever so that they can meet an insurance claim.”

Zafar Chaudry, MD, MS, MIS, MBA, SVP, Chief Digital Officer and Chief AI and Information Officer at Seattle Children's. Photo courtesy Seattle Children's.

“We haven't seen financial savings where we could just clearly say we can eliminate people because of the tools,” he added. “But we have seen improvements in burnout rates. We have seen clinicians be much happier, especially with the ambient AI type tools. They get to spend more time with the patient. You know, amazingly, some people can go home on time, which is a huge win, and they're not working through the weekend trying to catch up on the clinical documentation.”

The panelists also noted that because agentic AI is under the microscope, it has become what Yount calls “a mirror back to the organization.” By closely analyzing how AI uses data, clinical leadership is able to spot other care gaps and problems in care coordination and management, and can take other, transformative actions that don’t even have anything to do with AI.

“You're trying to solve problems in the health system and it's usually not the AI solution alone that is there solving the problem,” Blum pointed out.

AI Replacing People?

Finally, the panelists addressed the idea that AI would take away jobs. And they all began by noting that the healthcare industry is experiencing a workforce shortage, so the potential jobs under threat aren’t even being filled now.

“Long term, we don't know the answer of exactly how much certain occupations are going to be impacted,” Blum said. “But I do think that the writing is on the wall for certain roles. And that those roles will dramatically evolve and we'll likely … need less people engaged in certain activities.”

“Probably the biggest example of that is coding,” he added. “Agentic coding is proven. It's in certain domains, so if you're doing radiology coding right now, you probably won't be doing radiology coding in the next five years. … You may be doing quality assurance, you may be doing other things.”

“I think there's definitely going to be a seismic shift, but the shift will be elimination of roles that nobody wanted to do in the first place,” noted Chaudry. “If we can use the AI tools to automate those functions or some of those very trivial back office admin roles, then those roles will absolutely in the future be gone.”

“You take five coders that are serving, I don't know, 500,000 outpatient appointments in terms of coding those episodes, and now the same number of people can do a million outpatient appointment coding episodes,” he added. “So you're not hiring more people. Therefore your costs are staying static. Your volumes are going up, you're getting better patient satisfaction because people can get seen. That's where the shift is going to happen. I don't think it's a straightforward model that you see from technology companies where they tend to come in and say, ‘Oh yeah, you can just get rid of 50 FTEs and it's just going to work.’ I think you have to be smart about what that shift is, where you're going to shift the workload between AI and human.”

Yount said this is all just a part of reinventing and transforming healthcare – and that is needed.

“We're just trying to keep up with demand [and] growth, looking for improvements, figuring out how to give the provider and the patient that relationship more effectively, and candidly looking at other opportunities, which is working with payers and the broader health healthcare ecosystem,” he said.