Chipsoft

Employing Large Language Models to clarify SNOMED CT concepts to patients

EMEA

Artificial intelligence, Collaboration, Innovation, Research, Translation

In the Netherlands, SNOMED CT concepts are provided with patient-friendly terms and clarifications by the National Release Center and a medical translation agency through an extensive, mostly manual translation project. However, large language models (LLMs) offer a promising alternative for manual translation by enabling the automatic clarification of SNOMED CT terms. In this presentation we present these alternative translation approaches and share the results of a comparative analysis between manually and automatically generated clarifications. Additionally, we will explore methods to employ LLMs to automate steps in the translation process, examine the limitations of different translation pipelines, and discuss the implications for terminology developers and clinical practice.

Description

Medical jargon is difficult to grasp for persons with low health literacy, but it is important for patients and their health outcomes to understand their condition and treatment. In the Netherlands, SNOMED CT concepts are provided with patient-friendly terms and clarifications by the National Release Center at Nictiz (The Hague, Netherlands) and a medical translation agency through an extensive, mostly manual translation project. However, large language models (LLMs) offer a promising alternative for manual translation by enabling the automatic clarification of SNOMED CT terms. We carried out a comparative analysis of the clarifications that are currently released in the SNOMED CT Netherlands Edition with clarifications generated by a state-of-the-art large language model (LLM). We limited the scope to diagnoses because this was also the current focus of the translation project from the Netherlands National Release Center.

Scope

Data in electronic health records (EHRs) are increasingly encoded with SNOMED CT: the comprehensive, international and multilingual healthcare terminology standard that represents health information from various healthcare domains. When clarifications are provided to SNOMED CT concepts, they can thus be used in a wide range of applications and settings. As a reference terminology SNOMED CT can also help to clarify data from other coding systems that are mapped to SNOMED CT. Additionally, tools are available to encode free-text EHR content with SNOMED CT concepts, which then can be used to provide the clarifications to lay end users.

How SNOMED CT will be used

We will describe the current translation method of Nictiz, that involves translation by a medical translation agency, and validation by terminologists, clinicians and patients. Additionally, we present the results from a study in which we used the large language model (LLM) Gemini 2.0 Flash (from Google, Mountain View, CA, USA), and a prompt based on the requirements from Nictiz about clarifications, to generate clarifications for diagnoses that are registered in hospitals in the Netherlands. A purposeful random sample of the clarifications was validated by two medical doctors. Next, an online survey was carried out with participants from the general population to evaluate the quality of the clarifications and to compare the quality of the clarifications from the sample and from SNOMED CT Netherlands Edition. Finally, we will discuss alternative methods to automate steps in the translation process, examine the limitations of different translation pipelines, and discuss the implications for terminology developers and clinical practice.