PsyMooc: Empowering Simulation-Based Educational Systems with LLM Agents to Train Clinical Interviewing Skills

Abstract

Psychiatric clinical interviewing is a core yet challenging skill for psychiatric residents, requiring clinicians to navigate open-ended dialogue, interpret emotional cues, and manage diagnostic uncertainty. However, existing simulation-based education (SBE) tools often fail to provide sufficient opportunities for realistic and autonomous interview practice, largely due to their reliance on scripted and rigid interactions. In this paper, we explore how large language model (LLM) agents can be leveraged to support SBE systems for psychiatric clinical interview training. We first conducted a formative study to identify the psychiatry-specific learning needs and interactional challenges that should shape system design. Based on these insights, we developed PsyMooc, an LLM-enhanced SBE system designed to support open-ended interview interactions and deliver context-aware, competency-oriented feedback through LLM-driven agents. We then evaluated PsyMooc through a small-scale between-subjects user study to examine usability and learning-related outcomes. The results provide preliminary evidence that PsyMooc was perceived as usable and engaging, and was associated with improvements in residents' clinical confidence, interview performance proxies, and patient-centered communication behaviors.