{"paper":{"title":"CUICurate: A GraphRAG-based Framework for Automated Clinical Concept Curation for NLP applications","license":"http://creativecommons.org/licenses/by/4.0/","headline":"CUICurate automates UMLS concept set curation with GraphRAG to yield larger and more complete sets than manual methods.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Blanca Gallego, Jamie Novak, Mathew Miller, Sze-yuan Ooi, Victoria Blake","submitted_at":"2026-02-20T03:00:13Z","abstract_excerpt":"Background: Clinical named entity recognition tools commonly map free text to Unified Medical Language System (UMLS) Concept Unique Identifiers (CUIs). For many downstream tasks, however, the clinically meaningful unit is not a single CUI but a concept set comprising related synonyms, subtypes, and associated concepts. Constructing these sets is labour-intensive, inconsistently performed, and poorly supported by existing tools. Methods We present CUICurate, a graph-based retrieval-augmented generation (GraphRAG) framework for automated UMLS concept set curation. A UMLS knowledge graph (KG) was"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"CUICurate produced substantially larger and more complete concept sets than the manual benchmarks. GPT-5 outperformed manual curation for all concepts and retained at least 95% of definitive gold-standard CUIs.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that LLM-based filtering (GPT-5 and Qwen3-32B) accurately distinguishes clinically meaningful relations from noise without systematic bias or hallucination, especially for concepts not observed in the 10,000 MIMIC-III notes used for validation.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"CUICurate uses GraphRAG on a UMLS knowledge graph plus LLMs to generate larger, higher-recall concept sets than manual curation for five clinical concepts.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"CUICurate automates UMLS concept set curation with GraphRAG to yield larger and more complete sets than manual methods.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"81d263ae8780ad419cd6a62d337611d7dc02f39e783b0f8b5343d34e80db4d68"},"source":{"id":"2602.17949","kind":"arxiv","version":2},"verdict":{"id":"055cea91-ef59-4104-a3e8-3975723644b9","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T21:14:14.730908Z","strongest_claim":"CUICurate produced substantially larger and more complete concept sets than the manual benchmarks. GPT-5 outperformed manual curation for all concepts and retained at least 95% of definitive gold-standard CUIs.","one_line_summary":"CUICurate uses GraphRAG on a UMLS knowledge graph plus LLMs to generate larger, higher-recall concept sets than manual curation for five clinical concepts.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that LLM-based filtering (GPT-5 and Qwen3-32B) accurately distinguishes clinically meaningful relations from noise without systematic bias or hallucination, especially for concepts not observed in the 10,000 MIMIC-III notes used for validation.","pith_extraction_headline":"CUICurate automates UMLS concept set curation with GraphRAG to yield larger and more complete sets than manual methods."},"references":{"count":17,"sample":[{"doi":"","year":2024,"title":"Medical Concept Normalization","work_id":"8065502f-2898-4a9d-8036-ce87f8160c7e","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"The Unified Medical Language System at 30 Years and How It Is Used and Published: Systematic Review and Content Analysis","work_id":"6d7b30ca-ef61-4ffd-b7ce-bc7a21ca8e26","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Clinical Concept Value Sets and Interoperability in Health Data Analytics","work_id":"35fc48cc-3ea4-4937-b51a-2987d73c9fd8","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"Quickumls: a fast, unsupervised approach for medical concept extraction","work_id":"724c5e38-0c54-4125-8d8e-aabee04d8206","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"MedCAT -- Medical Concept Annotation Tool","work_id":"303c64c2-5670-426c-806b-4061c378154d","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":17,"snapshot_sha256":"8f70c3d8424d905b529fd2a279e6de87e0dbea55bec5dca40a8fd9da7a32f9d8","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}