生成AIを活用した精神科教育の向上：対象関係論トレーニングにおけるAI支援パイロットスタディ

To assess whether object relations theory (ORT) formulations generated by large language models (LLMs) enhance resident education in psychodynamic psychiatry.

This institutional review board (IRB)-approved observational study was conducted at a single academic program from January to March 2025. Eleven postgraduate year (PGY) 3 and 4 residents submitted de-identified psychotherapy case narratives to a GPT-4 model using a standardized ORT prompt. Residents evaluated accuracy, clarity, and educational value using a 10-item, 5-point Likert scale and provided free-text feedback. Descriptive statistics were calculated. Thematic analysis was used to identify key perceptions.

Ten residents completed the rating process. Mean scores ranged from 3.5 to 4.2 across all items, with the highest ratings for identification of objects, affective tone, and educational value. Qualitative analysis identified five themes: (1) improved clarity and structure, (2) stronger conceptual anchoring to ORT, (3) risk of overgeneralization, (4) maternal emphasis bias, and (5) importance of faculty oversight. Residents characterized the outputs as concise and clinically applicable, although they noted occasional factual inaccuracies and misalignments with specific case details.

With appropriate supervision, AI-generated ORT formulations may serve as a valuable adjunct to psychodynamic education by standardizing terminology and supporting conceptual development. Future controlled studies are needed to evaluate effects on Milestone-aligned assessment and applicability to additional psychotherapy models.