Country / Region
APAC
Tags
Artificial intelligence, Clinical Practice, Innovation, Mapping, Research
This is our experience in standardizing nursing records within the Korean Multi-Institutional Multi-modal Intensive Care (K-MIMIC) dataset, a Korean intensive care EMR database, using SNOMED CT. By comparing AI-based clustering results with classifications by clinical nursing experts, we explored the feasibility of AI-human collaboration in structuring nursing documentation and discussed practical implications for implementation.
Description
This study aims to structure and standardize free-text nursing records from the K-MIMIC dataset, which contains electronic medical records from Korean intensive care units. A total of 19,511 nursing phrases were collected. To perform clustering, we used sentence-level embeddings generated by three pretrained language models: Sentence-BERT, PubMedBERT, and S-BioBERT. Each phrase was embedded using these models, and semantically similar terms were grouped using the K-Means algorithm. We also examined the effect of translating Korean nursing phrases into English prior to embedding. The resulting AI-generated clusters were compared with classifications created independently by five experienced clinical nurses with 4, 5, 9, and two with 11 years of experience. This comparison was conducted to evaluate the clinical alignment and semantic accuracy of AI-derived clusters, assess the potential for semi-automated terminology mapping, and explore their suitability for integration with SNOMED CT.
Scope
SNOMED CT was chosen due to its comprehensive clinical coverage, including nursing domains, and its ability to support international interoperability and multilingual usage. Its alignment with global standards makes it well-suited for structuring Korean EMR data for both local implementation and international collaboration.
How SNOMED CT will be used
Nursing concepts identified through both AI clustering and expert review were mapped to SNOMED CT concept IDs. We utilized SNOMED CT's hierarchical structure and concept definitions to ensure semantic consistency and clinical relevance. The mapping process aimed to establish a foundation for structured nursing documentation and enhanced interoperability across systems.
Why SNOMED CT will be used
Contact
