Setting and Problem
Empathy and strong communication skills are vital in medical care, yet these skills are often underemphasized in medical training, particularly in the context of serious illness conversations. Moreover, serious illness conversations can be emotionally taxing for clinicians and may contribute to stress, compassion fatigue, and burnout. Residents often feel unprepared to initiate these conversations due to limited formal training.1
Traditional teaching methods, such as workshops with standardized patients, are resource-intensive, time-consuming, and difficult to scale. Simulation centers require trained actors, faculty facilitators, and protected time, creating logistical and financial barriers to implementation.1,2 As a result, many programs lack consistent opportunities for residents to practice these critical communication skills.
To address these barriers, we developed an innovative, AI-powered training tool that allows residents to practice serious illness conversations in a realistic, low-risk environment using artificial intelligence conversational agents (AI-CAs) and virtual reality (VR).
Intervention
We created an AI-CA, named “Lisa,” to simulate patient-clinician interactions about serious illness and end-of-life care. Lisa was constructed using Convai (Convai Technologies Inc), a platform for avatar-based dialogue generation. This system uses natural language processing to generate dynamic, unscripted dialogue, allowing for realistic conversations. The AI-CA was refined iteratively with a palliative care specialist to ensure clinical and emotional realism.
Lisa was integrated into a VR environment to create an immersive training experience that can be easily scaled. The curriculum included 3 sequential components:
Initial AI-CA Session: Residents engaged in a baseline conversation with Lisa to assess pre-training communication approaches.
Educational Modules: Residents participated in two 30-minute pre-recorded lectures led by a palliative care physician covering communication frameworks for serious illness discussions, including delivering serious news, responding to emotions, and eliciting patient goals and values.
Follow-up AI-CA Session: Residents completed a second VR interaction with Lisa, applying these skills in a simulated patient encounter.
This structure combines experiential practice with structured teaching while requiring minimal additional costs and no live actors.
Outcomes to Date
To date, one cohort of 4 incoming anesthesiology interns completed the AI-CA based serious illness communication training. All participants completed pre- and post-session surveys assessing feasibility, realism, and self-reported confidence.
Feasibility and Acceptability: All 4 participants completed the intervention. Three of the 4 reported the experience felt “low-pressure” and allowed them to “practice at their own pace.” Cognitive workload was assessed using the NASA Task Load Index (NASA-TLX), a validated multidimensional instrument that measures perceived workload across mental, physical, and temporal demand, effort, performance, and frustration. The median NASA-TLX mental workload score was 56 out of 100, suggesting moderate cognitive demand with minimal stress (mean irritation score 19 out of 100).
Perceived Realism and Engagement: Three out of 4 described the AI-patient Lisa as somewhat or very realistic, and all participants reported being at least “somewhat emotionally engaged.” Qualitative themes reflected emotional authenticity and psychological safety (“I felt calm and safe trying to have the conversation even if it included making mistakes.”)
Educational Impact: Three out of 4 reported feeling more prepared to lead goals-of-care discussions compared with lecture alone. Participants emphasized learning to “listen more,” “assess patient understanding,” and “deliver headlines clearly.”
These preliminary findings demonstrate strong feasibility, high learner engagement, and early educational benefit. Because the system requires minimal faculty oversight and can be deployed with existing VR hardware, it represents a scalable model for integrating communication skills training across residency programs.
References
- 1.Cegelka D,, Jordan TR,, Sheu JJ,, Dake JA,, Assaly R. End-of-life training in US internal medicine residency programs: a national study. Sci Open Access J. 2017;1:030. [Google Scholar]
- 2.Schmit JM,, Meyer LE,, Duff JM,, Dai Y,, Zou F,, Close JL. Perspectives on death and dying: a study of resident comfort with end-of-life care. BMC Med Educ. 2016;16(1):297. doi: 10.1186/s12909-016-0819-6. doi: [DOI] [PMC free article] [PubMed] [Google Scholar]