Speaker
Description
AI simulated patients (AISPs) powered by large language models (LLMs) offer a scalable alternative to conventional simulated‑patient training. However, building a clinical training platform with customizable AISPs that supports interactional competence (not only medical content recall) remains challenging. I present a clinical training platform designed around a dual‑layer framework for grounded theory‑informed clinical communication training. The first layer represents clinical content structure (e.g., chief complaint, HPI attributes, relevant ROS, contextual details), so that patient information is delivered by the AISP in response to appropriate questioning. The second layer represents interactional structure (openings, agenda setting, empathy sequences, clarification/repair, transitions, closing), linking conversational moves to the AISP persona (e.g., anxious, skeptical, and reserved) and scaffolded difficulty parameters. This interactional layer aligns with established clinical interview frameworks (e.g., Calgary‑Cambridge), while my grounded theory findings informed the design requirements and AISP behaviors. Within the platform, the two layers are combined so that learner questions trigger both content retrieval and appropriate interactional responses, enabling practice of patient‑centered interviewing strategies such as reflective listening, summarizing, and managing resistance while maintaining clinical relevance. I will describe the workflow, scenario authoring approach, and instructor/learner controls for adjusting difficulty to match desired educational outcomes. Applying this framework to an AI‑powered training platform aims to make AISPs more reliable for clinical interview education by explicitly modeling interaction, not only content recall.