Alexandra Siegel (left), current student in the BSN Entry to Doctor of Nursing Practice pathway, and Xinxin Huang (right), senior educational technologist, discuss a new generative AI clinical simulation tool (photo credit: Elisabeth Fall).
How UCSF Experts are Using Generative AI to Revolutionize Nursing Education
A key part of nursing education is learning how to interact with patients and provide a diagnosis based on both the patient’s medical history and what they say during an appointment. But having nursing students practice with actors before working with patients can be costly and difficult to schedule.
Nursing schools are also hemmed in by who they can hire to play these roles. Not only are there limitations to hiring child and adolescent actors to pretend to be sick, but they often have trouble finding a diverse group of available and skilled actors who reflect the actual patient population nurses will serve.
“We’re limited by the population of actors we have,” said Bridget Gramkowski, MS, RN, CPNP-PC, associate professor and current student in UCSF School of Nursing’s Post-Master’s Doctor of Nursing Practice pathway. “We want to represent a range of humanity, but we may not have that range of humanity in our actor pool.”
The solution: a new generative artificial intelligence (GenAI) tool that allows students to interact, in a simulated telehealth visit, with an adolescent patient created with already existing case studies. A team including Gramkowski; Xinxin Huang, MS, senior educational technologist; and Mary Gallagher, DNP, MPH, CPNP-PC, assistant clinical professor at UC Davis, has created a training module where different AI “agents” portray adolescents complaining of headache. Through three different simulations, nursing students review the patient’s chart and talk with the patient to try to reach a diagnosis — an interaction that is recorded and can be reviewed.
The AI program “opens up opportunities because most simulations don’t have to be scheduled around time and space availability,” said Gallagher. “It costs less money and also allows us to be a little more creative in our case offerings.” For example, factors like socioeconomic status and food insecurity can be added into simulations to make sure that nurses are prepared to ask about the context of the whole patient before reaching a diagnosis, not just what’s being said in an appointment or what’s on a medical chart.
The tool came together rather quickly, said Huang. They received funding from the UCSF Innovation Fund in July of 2024. She used multiple tools including Versa, UCSF's secure GenAI platform, ChatGPT, Midjourney, Wellsaid and D-ID to create the text, image, video and audio of the agent.
Huang was also able to make the experience more complex than she had anticipated when drafting the project proposal. “Initially we only had the image, not an interactive agent,” she said, adding that using D-ID made it possible to talk to the agent, perhaps the most impressive part of the technology.
In October 2024, the team presented the agent at the ATXpo IdeaLab, a Bay Area conference, and were overwhelmed by the response. Most attendees asked if the agent was an actor and wanted to know how Huang had created such a realistic, interactive AI patient. “I thought it was really great, but didn’t fully appreciate how excellent it was. Other AI and technology professionals were stopping in their tracks to take notes,” said Gramkowski.
Huang said she has also been approached by other institutions to share her work. “This is a low-cost learning opportunity,” Huang said. “As for now, a few other schools have reached out to me because they want to do something similar.”
The team then ran a pilot program with UCSF nursing students in December 2024, to assess how the tool worked in an educational setting and to collect feedback with the goal of improving the agent.
The team now plans to continue perfecting their simulated headache cases and expand the tool. That would mean more types of diagnoses, and more types of patients at different ages that are created through the AI simulator. They are even exploring pairings where a mother and child would be on the same AI-generated telehealth call.
“We could really address health equity in terms of the pediatric population,” said Gramkowski. That means adding factors like if a patient is in foster care. Not only will a robust batch of potential simulated cases help prepare nurses to encounter a wider range of patients, but the ability to review and receive feedback on their interactions will also help those nurses assess if they are “treating patients differently, which can then be addressed in our course content.”
It will also help augment student nurses’ in-the-field studies to ensure student nurses are seeing a gamut of cases, Gallagher added, because not every nurse is going to have practical experience with every kind of patient before graduating. “Students might not necessarily work with a child with type 1 diabetes, but they may have run through a case and had the opportunity to practice taking a history through a simulation like this.”
