Background Effective communication in challenging patient interactions is crucial in medicine. While simulation labs are ideal for training, they lack universal accessibility. AI-based patient encounters provide an innovative tool for medical education that allow students to test new communication methods in a risk-free environment. We present an integrated educational module with best practice videos and an AI-based simulation app for honing skills in delivering difficult news to patients.

Methods The educational component includes videos on best practices for delivering bad news, followed by interactive training using an AI-powered app based on a Large-Language Model, programmed in Python. To build the AI simulation modules, a pilot case prompt was first created by specifying the clinical scenario and patient personality in ChatGPT-4o. This prompt was further refined to an appropriate length and personality specifications using an iterative prompting process within ChatGPT-4o. After testing this case in the app environment, multiple cases were designed using standardized prompts based on the pilot case specifying diagnoses, communication styles, and personality traits based on the Big Five model. The AI app incorporates diverse clinical scenarios presenting challenging situations and difficult conversations and provides personalized feedback based on the SPIKES protocol.

App Development The AI-powered app was developed to simulate realistic patient interactions. The OpenAI API was employed to generate patient responses and emotional cues using ChatGPT-4o from predefined cases. Each case includes a multidimensional prompt architecture, enabling the AI to portray various clinical scenarios and patient personalities accurately. A separate AI agent evaluates student performance using the SPIKES model criteria and provides real-time formative feedback on communication techniques.

Results The AI patient actors demonstrated accurate personality embodiments and significant emotional responses. In addition to giving appropriate responses, the app also provided body language and contextualizing statements in line with the encoded personality before beginning dialogue such as “fidgets nervously” or “takes a deep breath, trying to calm down.” The app assessed student performance against the SPIKES protocol and generated constructive feedback. The AI Patient Actor app for difficult conversations is openly available at https://patient-actor-dc.streamlit.app/.

Conclusion This fully integrated educational module, combining best practice videos and an AI simulation app, is an innovative tool for communication skills training that can provide formative skills training at scale. The module will be incorporated into Dartmouth's Geisel School of Medicine curriculum. Validation through cohort testing will follow.