A Study of Classification Methods for Structural Changes in Japanese Medical Institutions Using Generative AI.

This study examines the use of generative AI to classify structural changes in Japanese medical institutions, where institution code changes hinder longitudinal analysis. Using Ministry of Health data (2020-2024), cases of code change were assessed. The gpt-4o mini achieved an accuracy of 0.307, while implementing Google Search API-based RAG improved accuracy to 0.573, incorporating Chain-of-Thought (COT) prompting yielded a slight additional improvement to 0.601. While "Merged" and "Relocate" were classified relatively well, "New," "Organizational Change," and "Closed" remained challenging. The results suggest that integrating generative AI with web information can classify the reason of institutional change, though further refinement is needed for reliable automation.