Epic is going all in on generative AI in healthcare. Here's why health systems are eager to test-drive it

UNC Health has been tapped to test-drive Epic's generative AI tools to help overworked clinicians respond to a deluge of patient messages.

Large language models like ChatGPT and Microsoft's Azure OpenAI service have only been available to the public for a few months, but tech leaders at a handful of health systems are eager to be early adopters.

"We think this technology is going to be and is already becoming transformative and we just see it as a tremendous opportunity to provide assistance for our frontline care teams," Brent Lamm, UNC Health’s senior vice president and chief information officer, said in an interview.

The initial rollout will begin with five to 10 physicians at UNC Health. The program will use generative AI to auto-draft responses to some of the most common and time-intensive patient messages. The health system joins other early adopters from UC San Diego Health, the University of Wisconsin Health and Stanford Health Care.

Epic announced in April during the Healthcare Information and Management Systems Society Conference in Chicago that it was working with Microsoft to integrate large language model tools and AI into its electronic health record software. The health IT vendor and the tech giant are collaborating to combine the Azure OpenAI Service with Epic’s EHR software with an initial focus on drafting message responses.

More health systems will join the project in the next phase of rollouts, Epic said.

During the COVID-19 pandemic, as the use of virtual care boomed, doctors also saw a spike in digital messages from patients ranging from requests for medication refills to more complicated medical questions.

"Our physicians, through the pandemic and after the pandemic, have continued to really just be front-line heroes working with our patients, and they are as busy as ever. I think all of us in the industry are looking for ways to continue to drive assistance and implement technology to enable ease of use, productivity and efficiency for our front-line care team members and give them ideally more time to spend with our patients and more time at home at night with their families," Lamm said.

He added, "For us, the goal is to find ways we can thoughtfully and safely use AI to improve our teammates’ experience and help them focus on patients."

Healthcare has historically been slow to adopt technology and, as generative AI captures the zeitgeist, one option for health systems is to sit back and watch while others test it out.

Chapel Hill, North Carolina-based UNC Health wanted to be among the early adopters so physicians can develop use cases and ideas, Lamm noted.

"First and foremost, I think this is something that we want to take advantage of as soon as possible. Clearly, large language models and generative AI have been proven in the last six months to be very effective," he noted. "At UNC Health, we want to be an innovation-oriented, leading organization in healthcare. The academic and research mission of our organization drives us to be very forward-thinking."

The health system, comprised of 15 hospitals, 19 hospital campuses and more than 900 clinics, has built a strong foundational IT system and has been exploring other AI tools, Lamm noted. "We're able to be out front helping to shape and define how this new technology will be used to help our physicians and also help physicians far beyond just UNC Health. We want to take advantage of it and we also want to make sure that we can demonstrate leadership in the industry, as we have done in so many other ways."

The health system's partnership with Epic, a legacy EHR and health IT player, also boosts confidence that the pilot project will ease physicians' workload rather than just adding another app.

"We have a tremendous amount of competence in our relationship with Epic. They have become a trusted partner for us, and that is a key component of this," Lamm said. "For our physicians and our care team users, we want to make sure that the technology solutions that we are offering them do not require them to have to work in multiple systems. So the beauty of what Epic is doing here is it's seamlessly integrating this technology into the existing screens that our physicians are already using. It will be far easier for them to adopt this and far easier for them to use it, not having to think about opening up another window on their computer."

The pilot project will begin with a small subset of “more administrative” messages, similar to how smartphones can suggest responses to texts. The technology won’t replace the clinician’s judgment, Lamm noted. There is always a person reading and sending the message. It simply saves keystrokes.

"The way that Epic is thinking about this technology and the way that they're rolling it out is based on the idea that it is tool and an assistant for the physician. It's not there in any way, shape or form to replace their job," he noted.

UNC Health has tapped a team of physician informatics leaders spanning different specialties to work with the health system's chief medical informatics officer, David McSwain, M.D., to evaluate the generative AI tools.

"They're going to be using this in the early-stage adoption to look for things like, 'What is the accuracy of the text that the system is drafting?' and, 'What are the potential use cases in terms of how physicians can make it more effective for them?' We also think that having this multispecialty approach we may help Epic understand where there may be opportunities for specialty-specific tweaking of the models to make it more effective for a given specialty," Lamm said.

Physicians and informatics leaders also will be evaluating the tech's user interface to ensure it is easy for clinicians to use.

The health system expects to eventually scale up the generative AI-based messaging tool to a broader group of physicians. "At this point, we're pushing this as fast as Epic will allow," Lamm said.

Along with the draft message responses, Epic and Microsoft also are working to use generative AI to analyze medical records while looking for trends. The two companies working to bring natural language queries and interactive data analysis to SlicerDicer, Epic's self-service reporting tool. That functionality enables clinical leaders to "explore data in a conversational and intuitive way" to find trends, according to Phil Lindemann, Epic's vice president of data and analytics.

"The area where we focused first was, 'How can we make physicians more efficient?'" Lindemann said during an interview on the HIMSS23 exhibit hall floor.

Bringing generative AI and ambient voice technology into the exam room

Lamm sees promising opportunities to incorporate AI into EHR software to ease the burden of note taking for doctors and clinicians.

"I'm also very excited about what I'm seeing across the industry in terms of technologies that ambiently listen to a patient-clinician conversation," Lamm said. "I also think this will have a tremendous opportunity for virtual care as our physicians and care teams continue to see our patients through virtual care encounters. The artificial intelligence capabilities of being able to summarize key notes and documentation from those conversations as well; I really think the sky's the limit."

In fact, in a separate project, Epic recently worked with AI-powered voice solutions company Suki to integrate its generative AI assistant into its EHR software through its ambient application programming interface.

Suki Assistant helps clinicians complete time-consuming administrative tasks and helps ease workflow burdens for docs, according to the company. Punit Soni, CEO of Suki, says the company uses generative AI and large language models to listen to the patient-doctor conversation, identify the clinically relevant portions and then summarize that information as suggestions for the note, without human intervention. Clinicians can then review and accept, reject or edit content suggestions to ensure the accuracy of final notes. Once notes are complete, they sync back to Epic.

Suki has reported that its ambient note-generation tech is capable of reducing documentation time per note by as much as 72% for family medicine physicians. The company's voice-enabled digital assistant has been integrated into other EHR platforms including Athenahealth, Cerner and Elation Health.

"We believe you should be able to build an application where all modes of speech—command, dictation or ambient—just naturally work. Suki has been able to do that in a single app," Soni said. "We also believe that the physicians should be in complete control of the output of AI so you can put Suki in ambient mode and have it listen to a doctor-patient interaction and when you go to the note in Suki, it will make suggestions for you. Or you can just use voice [technology] and edit the suggestion."

Soni added, "We also believe that the most important thing in building experiences that reduce the administrative burden for docs is not AI. It's actually plumbing. So EHR integration, like what we have done with Epic, is a part of it. AI is tech that you can use but without plumbing, you can't build a product."

Suki differentiates itself in the ambient voice technology market as it's purely software-based without the use of hardware or humans, he noted.

"We believe it is actually going to be priced at a fraction of the cost of other ambient solutions,which means this product has a shot at being more mainstream. Today, ambient solutions cost thousands of dollars per doc per month. For a product that costs only a few hundred dollars a doctor a month, you would expect that it'll be available to family practices and all the other specialties that typically have no access to such expensive high-priced products," Soni said.