Human–AI Workflow Cuts Bilingual SNOMED CT Mapping Time in Single-Hospital Study
Published in Artificial Intelligence, Journals, Tools.
A study published on October 2, 2026 in the Journal of Medical Systems reports an AI-assisted workflow for mapping Korean and English clinical text to SNOMED CT terminology.
Researchers evaluated 2,261 de-identified text segments from a South Korean tertiary hospital. The system combined translation, abbreviation expansion and retrieval from a terminology database. Three health information managers compared manual mapping with AI-assisted mapping, retaining authority to accept, change or reject suggestions.
The assisted workflow increased the pooled first-choice hit rate from 83.7% to 86.8%. Total mapping time, including agent processing, fell from 1.57 to 0.72 minutes per segment—a 53.9% reduction.
The results concern terminology mapping, rather than diagnostic accuracy or patient outcomes. Evaluation used a single hospital and a fixed sequence: manual mapping first, assisted mapping later, separated by approximately four months. This design warrants caution when interpreting improvements and generalising them to other languages or institutions.
Source: Lee and colleagues, Journal of Medical Systems, October 2, 2026.