What type of study is this?

This is a Quantitative Study study.

September 19, 2025

Development and Evaluation of SNOMED CT Automated Mapping Tool: Advancing Terminology Standardization and Semantic Interoperability (Preprint)

YPYoungsun ParkYonsei University HKHannah KangKorean Academy of Science and Technology JKJi-Won KimNational Institute of Animal Science

Key Points

Mapping accuracy for diagnostic terms reached over 98% across four hospitals, with significant workload reductions.
The tool decreased manual mapping rates by 30% and reduced validation times by 75-90% across various institutions.
Utilizing a large language model, the automated mapping process enhances both efficiency and scalability for healthcare data standardization.
Despite improvements, challenges like ambiguity and granularity gaps highlight the need for further advancements in autonomous mapping.

Abstract

BACKGROUND Kakao Healthcare has built the Healthcare data Research Suite (HRS) to enable multi‑institutional research, where semantic interoperability depends on clinical terminology standardization. Manual mapping of local terms to Systematized Nomenclature of Medicine Clinical Terms (SNOMED CT) is labor‑intensive, inconsistent across sites, and poorly suited to bilingual data and institution‑specific granularity. OBJECTIVE We designed a large language model (LLM)‑based tool to support multi‑site research by standardizing terminology, combining automated SNOMED CT mapping and new‑concept authoring via post‑coordination in one process. METHODS The automated mapping pipeline included preprocessing local terms, syntactic and vector similarity mapping leveraging LLM-based embeddings, and iterative enrichment based on validated results. Translation and semantic representation used GPT-4o and Gemini. New concepts were authored through a structured post-coordination process. Performance was evaluated using diagnostic and surgical procedural terms from four major hospital networks (nine university hospitals) in South Korea, with clinical terminologists providing usability feedback on system results. RESULTS Using reference terms, Top-5 accuracy for diagnostic mapping reached 98.67%, 89.69%, 98.52%, and 92.78% across the four institutions, and 99.19%, 82.62%, 98.73%, and 84.69% for surgical procedural mapping. Implementation of the tool reduced manual mapping rates by 30% and overall manual workload by up to 90%. Average mapping and validation times per item decreased by 75–90%. CONCLUSIONS The tool improves mapping accuracy, efficiency, and scalability in terminology standardization, but challenges remain with ambiguous terms, granularity gaps, and abbreviation resolution. Future work will focus on integrating external abbreviation databases, Retrieval-Augmented Generation techniques, and ontology-driven modeling to further advance autonomous mapping performance and quality assurance.

Perguntar à IA

Bookmark

View Full Paper