Hybrid lexical-semantic retrieval over SNOMED CT: combining two retrieval paradigms to facilitate clinical data entry
- 1SNOMED International
Abstract
ObjectivesSearching SNOMED CT pits two retrieval technologies of opposite nature against each other. Deterministic lexical matching is precise and handles partial, out-of-order, as-typed input, but cannot cross a wording or language gap; learned semantic matching bridges para-phrase and language, but represents abbreviations and half-typed fragments poorly. Combining them naively lets the channel that is wrong for a given query degrade the one that is right, and the usual alternative, curating an additional local interface vocabulary, is labour-intensive to maintain. We ask whether the two paradigms can instead coexist in one retrieval flow over SNOMED CTs own curated descriptions, and study the query interpretation, rank fusion and re-ranking that make this coexistence safe.
Materials and MethodsWe built one open-source, reproducible implementation over the SNOMED CT International edition, combining order-independent multi-prefix lexical search, BioLORD-2023-M semantic embeddings, an optional cross-encoder re-ranker, and a local LLM for query normalization and entity extraction, with the two rankings fused by Reciprocal Rank Fusion and an optional hierarchy (descendant) filter. We measured search-only linking accuracy on 985 disease mentions from DisTEMIST (Spanish, zero-shot), ablating each optional component and each retrieval channel, and on 12,897 mentions from the SNOMED CT Entity Linking Challenge (English real-EHR discharge notes) under a field-scoped typeahead.
ResultsOn DisTEMIST the best and cheapest configuration, semantic search with re-ranking and no LLM pre-processing, placed the exact concept first for 57% of mentions (accuracy@1 = 0.57) and within the top-10 for 80% (recall@10 = 0.80). Channel ablation showed the two paradigms coexist without a trade-off: rank fusion followed by re-ranking keeps the on-task channels result and suppresses the off-task one. On the English real-EHR corpus a field-scoped typeahead reached the top-10 for 73% of mentions across all clinical domains (accuracy@1 0.52). Adding a contextual LLM re-ranker -- a second implementation of the re-ranking step that reorders the retrieved candidates using the querys context -- raised accuracy@1 further, to 0.61 on DisTEMIST and 0.56 on the challenge. LLM query normalization, by contrast, slightly hurt clean, terminology-like input but rescued the messy shorthand of real EHR text, most of all when given the surrounding note context, so it is best applied selectively rather than to every query. Results are specific to the evaluated configuration, not an architecture-independent estimate.
ConclusionLexical precision and semantic flexibility can coexist over SNOMED CTs maintained descriptions without trading one off against the other, secured by rank fusion and re-ranking rather than by choosing between the paradigms. This shifts maintenance from local phrase authoring toward governing and validating retrieval services rather than removing it, and may reduce additional local lexical curation where direct search is appropriate, while curated interface vocabularies retain a role in guided workflows.
Related
- Hybrid retrieval–verification framework for hallucination reduction in medical LLMs
- A Systematic Review of Spaced Repetition, Retrieval Practice, and Academic Performance in Undergraduate Medical Students
- A modified cryobiopsy specimen retrieval method using a thin bronchoscope and small guide sheath for the diagnosis of peripheral pulmonary lesions
- Modified interventional retrieval of a balloon with transverse rupture during percutaneous transluminal angioplasty of a vascular access： a case report
- Successful Bailout Retrieval of an Active‐Fixation Atrial Leadless Pacemaker Following Suspected Tissue Entrapment During Repositioning
BibTeX
@misc{osornio2026hybrid,
title = {{Hybrid lexical-semantic retrieval over SNOMED CT: combining two retrieval paradigms to facilitate clinical data entry}},
author = {A. Lopez Osornio and A. Randorff Hoejen and K. Kewley},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.10.26362796},
url = {https://www.medrxiv.org/content/10.64898/2026.09.10.26362796v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Hybrid lexical-semantic retrieval over SNOMED CT: combining two retrieval paradigms to facilitate clinical data entry
AU - A. Lopez Osornio
AU - A. Randorff Hoejen
AU - K. Kewley
PY - 2026
DA - 2026/09/10
PB - medRxiv
DO - 10.64898/2026.09.10.26362796
UR - https://www.medrxiv.org/content/10.64898/2026.09.10.26362796v1
AB - ObjectivesSearching SNOMED CT pits two retrieval technologies of opposite nature against each other. Deterministic lexical matching is precise and handles partial, out-of-order, as-typed input, but cannot cross a wording or language gap; learned semantic matching bridges para-phrase and language, but represents abbreviations and half-typed fragments poorly. Combining them naively lets the channel that is wrong for a given query degrade the one that is right, and the usual alternative, curating an additional local interface vocabulary, is labour-intensive to maintain. We ask whether the two paradigms can instead coexist in one retrieval flow over SNOMED CTs own curated descriptions, and study the query interpretation, rank fusion and re-ranking that make this coexistence safe. Materials and MethodsWe built one open-source, reproducible implementation over the SNOMED CT International edition, combining order-independent multi-prefix lexical search, BioLORD-2023-M semantic embeddings, an optional cross-encoder re-ranker, and a local LLM for query normalization and entity extraction, with the two rankings fused by Reciprocal Rank Fusion and an optional hierarchy (descendant) filter. We measured search-only linking accuracy on 985 disease mentions from DisTEMIST (Spanish, zero-shot), ablating each optional component and each retrieval channel, and on 12,897 mentions from the SNOMED CT Entity Linking Challenge (English real-EHR discharge notes) under a field-scoped typeahead. ResultsOn DisTEMIST the best and cheapest configuration, semantic search with re-ranking and no LLM pre-processing, placed the exact concept first for 57% of mentions (accuracy@1 = 0.57) and within the top-10 for 80% (recall@10 = 0.80). Channel ablation showed the two paradigms coexist without a trade-off: rank fusion followed by re-ranking keeps the on-task channels result and suppresses the off-task one. On the English real-EHR corpus a field-scoped typeahead reached the top-10 for 73% of mentions across all clinical domains (accuracy@1 0.52). Adding a contextual LLM re-ranker -- a second implementation of the re-ranking step that reorders the retrieved candidates using the querys context -- raised accuracy@1 further, to 0.61 on DisTEMIST and 0.56 on the challenge. LLM query normalization, by contrast, slightly hurt clean, terminology-like input but rescued the messy shorthand of real EHR text, most of all when given the surrounding note context, so it is best applied selectively rather than to every query. Results are specific to the evaluated configuration, not an architecture-independent estimate. ConclusionLexical precision and semantic flexibility can coexist over SNOMED CTs maintained descriptions without trading one off against the other, secured by rank fusion and re-ranking rather than by choosing between the paradigms. This shifts maintenance from local phrase authoring toward governing and validating retrieval services rather than removing it, and may reduce additional local lexical curation where direct search is appropriate, while curated interface vocabularies retain a role in guided workflows.
ER -