Large language model-based simulated patient training for heart failure palliative care communication: a pilot study

Area of Science:

• Medical Education
• Artificial Intelligence in Healthcare
• Palliative Care

Background:

• Effective communication in heart failure (HF) palliative care is crucial but challenging to train at scale.
• Traditional training methods like role-playing require significant resources (facilitators, standardized patients).
• Large language models (LLMs) offer a potential solution for scalable communication skills development.

Purpose of the Study:

• To evaluate the feasibility of a web-based LLM-driven communication training application for HF palliative care.
• To assess the early educational impact of the LLM on physicians' self-efficacy in conducting these sensitive conversations.

Main Methods:

• A single-arm pilot study involving physicians using a Japanese-language LLM application simulating advanced HF patients.
• Automated, framework-based feedback was provided by the LLM.
• Physician self-efficacy was measured using pre- and post-session questionnaires.

Main Results:

• Ten sessions were analyzed, with physicians engaging in an average of 7.6 dialogue turns.
• Significant improvements were noted in knowledge of palliative care communication (adjusted P < 0.01) and confidence in HF palliative care communication (adjusted P = 0.03).
• LLM response and feedback generation times were rapid (approx. 3s and 17s, respectively).

Conclusions:

• A web-based LLM-simulated patient system is feasible for training HF palliative care communication.
• The LLM demonstrated an early positive educational signal on physician self-efficacy.
• LLM-driven training may supplement traditional methods, warranting further investigation in larger controlled studies.