Translating Natural Language Questions into SNOMED Expression Constraint Language

Abstract

The SNOMED Expression Constraint Language (ECL) is a powerful but complex tool for querying clinical concepts within SNOMED CT, playing a critical role in clinical decision support, data analysis, and healthcare interoperability. However, its steep learning curve-requiring both syntactic expertise and in-depth knowledge of SNOMED CT's coding system-creates significant challenges for non-specialists. To address this issue, a novel approach is proposed that leverages state-of-The-Art Large Language Models (LLMs) to translate natural language questions into ECL queries. The model is designed to perform bidirectional tasks: generating ECL queries from user questions and providing natural language explanations for ECL queries, making the language more accessible and easier to understand. This work represents the first research effort to tackle this specific translation challenge, supported by the development of custom datasets and a novel pipeline that integrates multiple AI agents. Evaluation results demonstrate that the proposed model achieves 83.78% accuracy, highlighting the significant potential of LLMs for translating natural language questions into SNOMED ECL.