your123 - stock.adobe.com
Epic's take on agentic AI designed to boost patient experience
Epic Systems is dipping its toes into agentic AI, leveraging the technology to enable scalable, high-touch patient engagement for post-op patients.
Agentic AI didn't used to be a common phrase in Epic Systems exec Sean Bina's vernacular. But now, it's nearly all he hears about as Epic explores agentic AI for the patient experience.
"Three days ago, I'd never heard the word agentic, and now I think I've heard it like 500 times in the last three days," Bina, the VP of Patient Access and Experience at Epic, said during a pre-HIMSS phone interview. "We've been working on these kinds of agent concepts for a long time."
Poised to be unveiled at the 2025 HIMSS conference in Las Vegas, Epic's foray into agentic AI is part in parcel with the health IT company's overall efforts for promoting a better patient experience, Bina said alongside his colleague, Epic R&D Director Trevor Berceau.
Agentic AI is a system that can complete autonomous action and decision-making. These tools can complete certain tasks without human intervention, relying on reinforcement training and evolutionary algorithms to handle ambiguous or new situations and queries.
Managing patient access and the overall healthcare experience serves as a key use case for agentic AI, as healthcare organizations and the health IT companies that work with them seek to streamline patient interactions, according to Berceau. These tools are knowledgeable, he said, but they can also complete key tasks, delivering a value proposition that actually meets patient needs, unlike the chatbots of yore.
Agentic AI might succeed where chatbots fall short
Indeed, AI is already being deployed to support the patient experience, mainly in the form of chatbots. Many healthcare organizations have adopted AI-powered chatbots to answer patient queries and streamline the call center.
But these tools often fall short of patient and provider expectations. Patients using digitized appointment scheduling, for example, often find themselves reaching for the phone to book manually, anyway. And it's not uncommon for folks using a hospital or health system's customer service chatbot to ultimately try to connect with a human.
One 2024 study in the Journal of Medical Internet Research indicated that technical difficulties and user experience issues beleaguered chatbot performance.
Still, the data is mounting that chatbots can be effective tools for disseminating patient information. Systems like ChatGPT are good at answering basic patient queries, and some studies have indicated that generative AI can even respond to some patient portal messages, although these responses require human review.
Indeed, Berceau and Bina explained prior to HIMSS that Epic's technology uses Gen AI to parse through free-text provider notes to give patients self-management recommendations and patient education.
Up until now, gen AI chatbots haven't proven to be able to service patient access, but most experts agree they can handle simple tasks. This removes burden from human staff members who are then free to address more complex patient needs.
Agentic AI presents an opportunity to create a high-touch patient experience, one that the industry could never dream of scaling, in a sophisticated way, Bina explained.
"Pulling agent concepts all together into this intelligent infrastructure, which combines all these different elements of everything from search to appointment navigation to being able to identify my clinical history, to being able to help answer questions -- it's all coming together in this pretty amazing way," he remarked.
"You will no longer have this issue where people perceive that these bots are really [burdensome]," Bina continued. "You won't say 'talk to a human agent right away.' Instead, the bots will be able to service stuff up so much more quickly than a human will. People might be talking to a human and say, 'take me to the AI agent.'"
Agentic AI for healthcare relies on heavy datasets
The difference is in how sophisticated the AI programming is, according to Berceau.
"The difference between a poor experience with a bot and an amazing experience with a bot is the level of integration that it has with both knowledge and the ability to take action," he said.
Because agentic AI has more autonomy than generative AI, the chatbots it powers can complete more sophisticated patient-facing tasks.
Take, for example, a post-op patient: this is a significant area where many hospitals and health systems leverage patient engagement best practices to achieve good clinical outcomes.
"If we had infinite people and infinite money, organizations would have a person dropping by every patient's home and asking them a bunch of questions and making sure their next appointments were lined up and taking a look at their wounds, seeing how it's going," Berceau explained. "But that doesn't scale. We don't have the people for it. We don't have the resources for it."
Agentic AI can start to do many of those tasks, Berceau said. The technology could automate a phone call to the patient two days post-op and ask about knee pain, mobility, swelling and signs of infection. It could check in about changing wound dressings, provide education around that and even prompt the patient to book their next post-op visit. The patient hasn't had their flu shot yet? Leave it to the AI agent to get that done, too, Berceau said.
"Being able to do all of those things, that's what makes that a worthwhile check-in," he pointed out.
But a lot goes into creating an AI agent that can do all of that. The AI needs to know about the patient's care journey, including when they had surgery, on which body part and who was part of the clinical care team. The tool also needs to know how to navigate the local healthcare system, get a look into the patient's upcoming appointment schedule and assess where the patient might want to access their care.
"On the surface, that type of really slick interaction, it feels like it's just this conversational chatbot interface, but what actually is going to make the difference," Berceau asserted.
Does the agentic AI have integration with clinical and organizational knowledge? Can it take action in different systems and understand constraints on scheduling?
"With the agentic design paradigms, we can build those reusable components to say, this one knows how to look at the upcoming schedule for a given type of visit and figure out when it's going to be available," Berceau explained. "This one knows how to compare our local protocols for a knee replacement clinical check-in to what the patient's answering."
Epic sees a future where agentic AI can support a better patient experience across use cases. There are possibilities for streamlining the pre-visit patient intake process and supplementing the preventive care scheduling process, Berceau stated.
"We're really excited about the possibilities of being able to create all of these deeply integrated agents and then construct them into different workflows for patients at different points in time," he concluded.
Sara Heath has covered news related to patient engagement and health equity since 2015.
