A-S.B.44: LOINC Laboratory Diversifier: A Benchmark Dataset for Interoperability
Author
Keywords
Biomedical Natural Language Processing, Interoperability, Large Language Models, Entity Mapping, Biomedical Ontologies
[sponser-meet-now-chat][/sponser-meet-now-chat]
Healthcare data interoperability enables different systems and digital health tools to exchange and use information without loss of meaning. Although standards such as HL7 and SNOMED-CT have improved this process, full interoperability remains challenging due to legacy infrastructures and persistent semantic inconsistencies.
In this work, we present an augmented dataset of LOINC laboratory terms designed to better capture the lexical variation and noise found in real-world clinical data. The aim is to improve term mapping robustness and support the development of more generalisable interoperability tools.
As a proof of concept, we show that providing a general-purpose LLM with examples of high-quality mappings derived from this dataset improves mapping performance across all evaluation metrics, with Micro Accuracy (Top-1) increasing from 0.696 to 0.836 and Macro Accuracy (Top-1) from 0.694 to 0.833. Our approach also outperforms several baselines, achieving a composite score of 0.5997 ± 0.0392 $.
Overall, the results suggest that targeted data augmentation for laboratory terminology can improve both synonym expansion and downstream mapping, highlighting the value of combining LLMs with biomedical ontology resources for healthcare data interoperability.
Co-authors: Ole Eigenbrod, Honghan Wu, T. Ian Simpson
Contact Attendee
Warning: Attempt to read property "user_email" on string in /home/1276969.cloudwaysapps.com/ydbgzhdjeq/public_html/wp-content/plugins/my-conference-now/functions.php on line 2826