UNC Health has been piloting Epic's artificial intelligence tool that drafts clinician responses to patients' questions and comments for over a month, and the health system's team traveled to Epic's headquarters to discuss strategies for future AI innovations within EHRs.
Epic's generative AI tool works by drafting in-basket message responses for clinicians to review, edit and send back to the inquiring patient. It is meant to save time physicians currently spend on the computer answering patient messages within the My UNC Chart portal.
Phase one has been the testing phase — with 10 clinicians using the tool in a test environment — and UNC Health will be adding additional waves of physicians to the pilot once it goes live in Epic on August 6.
The Chapel Hill, N.C.-based health system has gained two significant insights from the initial pilot. First, it has reinforced the necessity for clinicians to continue active engagement with the messages. Second, it has emphasized the importance of ensuring clinicians understand that AI is meant to assist with, not take over, clinician tasks.
"Most importantly, it's that a clinician needs to always remain in the loop when corresponding with patients," David McSwain, MD, chief medical informatics officer at UNC Health, told Becker's. "It's important for our clinicians to understand, and for our patients to understand, that this is not AI taking over medical decision-making. This is AI making the medical decision-making easier for our clinicians and more streamlined."
The health system has been working very closely with Epic by having weekly check-ins to provide feedback that is directly incorporated into the application's design to improve clinician experience and patient outcomes.
"The feedback from our clinicians who are piloting this effort is that the message drafts are generally helpful, especially for some of the more routine responses like work excuses and medication refills, but they can't be a substitute for our clinicians," Dr. McSwain said. "We are working specifically with Epic on how to ensure that our clinicians engage with the message and how to draw their attention to the parts of the draft where the AI isn't best equipped to answer."
UNC Health said, when used correctly, the technology can reinforce the value of their expertise, emphasizing that the AI tool is not a substitute for clinicians.
"Having just gotten back from our trip to Epic, we are really excited for the work that is ongoing, and working alongside them," Dr. McSwain said.
The health system discussed AI initiatives and strategies with Epic, leveraging the insight discovered during the initial pilot to shape their approach moving forward into subsequent phases.
Those phases will entail the addition of clinicians in waves every few weeks, expanding the pilot roster with dozens of physicians and members of the nursing staff who assist in responding to patient messages. As the pilot progresses, the health system will be closely evaluating the impact the technology has on clinician efficiency and burnout across different specialties.
"One of the things that we're excited about with this project is that we're engaging not only clinicians who have informatics and technological expertise, but we're also engaging clinicians who are directly involved with our wellness initiatives at UNC, as well as clinicians across multiple specialties," Dr. McSwain said.
As the tool continues to evolve, it will ultimately learn from clinician input and adapt its responses accordingly. Until the technology reaches that point of self-learning, it utilizes prompt engineering, which is the ability to modify the prompts that control the generation of message drafts.
Prompt engineering allows the customization of the input instructions that the chatbot follows to produce the message draft, acting as a manual method of incorporating necessary feedback to produce the best possible results.
The work in AI at UNC Health "spans clinical work, patient engagement, access to care, revenue cycle, analytics and several other areas," Dr. McSwain said. "There are so many opportunities to improve the efficiency of our healthcare processes and to support our teams using this technology."