Presentation (CTI)

OBJECTIVES: Cross-country inconsistencies reduce comparability, and manual creation limits codelist scalability and reproducibility. Japan presents a particularly acute challenge: no Japanese OMOP vocabulary exists, and six distinct terminology systems (disease receipt codes mapped to ICD-10, YJ drug codes, medical and dental procedure Kubun codes, and JLAC laboratory codes) are maintained by separate authorities with different update frequencies. We aimed to develop an AI-driven code list generation tool supported by a unified Japanese medical dictionary to enable governed, reproducible, and cross-country code list development.
METHODS: The AI Code List Generation Tool was developed across four stages: (1) harvesting and structuring existing validated study code lists into a searchable, metadata-tagged repository; (2) generating reproducible code lists using hierarchical dictionary expansion, rule-based inclusion/exclusion logic, and version-controlled audit trails; (3) building an AI agent architecture combining natural language understanding with a knowledge layer spanning internal code list repositories, medical dictionaries, and published literature; and (4) aligning with OHDSI Athena vocabularies for cross-market semantic translation. To address Japan-specific requirements, JADE (Japanese Analytical Dictionary for Evidence) was developed as an upstream contributor, systematically harmonising Japanese disease, drug and medical/dental procedures with lab codes, enriched with AI-assisted Japanese-to-English translation and longitudinal version tracking.
RESULTS: The disease dictionary component of JADE was delivered, unifying 27,687 unique disease codes across 47 versions with 24 unique columns and 10 documented schema changes. Drug codes and medical/dental procedure dictionaries are currently in development. The code list tool prototype, integrating JADE, UK, and US dictionaries, is targeting completion by mid-2026.
CONCLUSIONS: Combining an AI-driven code list generation tool with a harmonised Japanese medical dictionary addresses critical gaps in reproducibility, scalability, and cross-country comparability of RWD code lists, while establishing governed infrastructure for future AI-enabled real-world evidence generation.