This abstract was part of the 2026 UCSF AI and Education Symposium
Serious illness communication (SIC)—which includes ‘breaking bad news,’ sharing prognostic information, and exploring goals of care to inform treatment recommendations—is an essential component of high-value, patient-centered care. Training of these skills at the undergraduate medical level is inconsistent between institutions and occurs mostly in the elective curriculum.
Artificial intelligence (AI)-simulated patients (AI-SPs) may provide advantages for SIC education. AI-SPs can facilitate a psychologically safe learning environment for trainees. They are far more accessible and scalable than human standardized patients. They can be customized to a variety of clinical situations, adapt their outputs in real-time to learners’ skill level, and maintain conversational memory across encounters. These features are especially relevant for SIC education given the longitudinal nature of prognostic disclosure, evolving preferences over time, and recursive goals-of-care conversations.
We have partnered with MedSimAI—a multi-institutional AI simulation platform and have applied for an UCSF AME Innovation grant and are pending review. As part of the grant process, we have designed a standardized, adaptable, iterative AI-SP cases focused on developing core SIC competencies that will be an adjunct to the existing pre-clerkship medical school curriculum.
Our aim is to investigate: the authenticity of AI-SP simulations and their impact on pre-clerkship medical students’ SIC skill development; the quality and safety of real-time AI-generated SIC feedback; and students’ and educators’ perceptions regarding utility and feasibility of using AI-SPs to support students’ evolving SIC learning needs through clerkship years.
In addition to providing background on the need for further SIC educational tools for students, we would like to share one of our case simulations of SIC with the daughter of Ms Ramirez, Sophia. An audience participant will be able to participate in a 5-minute simulation with Sophia around the gravity of her mother’s recent admission, diagnosis of aspiration pneumonia, and review of goals of care.
After a five-minute simulation, the participant will be provided real time feedback by the Agent based on our SPIKES protocol and Serious Illness Conversation Guide grading rubric.
We will close by touching on our planned effectiveness and implementation evaluation plan as well as further directions to start assessing non-verbal communication during SIC through AI-SPs and customization of cases using UCSF OpenAI’s ChatGPT Enterprise HIPAA compliant LLM once integrated into APEX.
Contacts
Suresh Rangarajan, [email protected]
Daniel Bui, [email protected]
Teva Brender, [email protected]
Yulin Hswen, [email protected]