[Paper Review] Biomedical term normalization of EHRs with UMLS

Naiara Pérez, Montse Cuadros+1 more|arXiv (Cornell University)|Feb 8, 2018

Biomedical Text Mining and Ontologies15 references3 citations

TL;DR

This paper presents a multilingual, cross-lingual prototype for biomedical term normalization in Spanish electronic health records (EHRs) using the UMLS Metathesaurus. It leverages Apache Lucene for indexing, the IXA pipeline for NLP, and UKB for word sense disambiguation, achieving moderate agreement (kappa ~0.6) with MetaMap on two parallel corpora, and includes a web-based demo for interactive concept exploration.

ABSTRACT

This paper presents a novel prototype for biomedical term normalization of electronic health record excerpts with the Unified Medical Language System (UMLS) Metathesaurus. Despite being multilingual and cross-lingual by design, we first focus on processing clinical text in Spanish because there is no existing tool for this language and for this specific purpose. The tool is based on Apache Lucene to index the Metathesaurus and generate mapping candidates from input text. It uses the IXA pipeline for basic language processing and resolves ambiguities with the UKB toolkit. It has been evaluated by measuring its agreement with MetaMap in two English-Spanish parallel corpora. In addition, we present a web-based interface for the tool.

Motivation & Objective

• To address the lack of existing tools for biomedical term normalization in Spanish clinical texts using UMLS.
• To develop a multilingual, cross-lingual pipeline for mapping clinical terms to UMLS concepts.
• To evaluate the system’s performance against MetaMap on parallel English-Spanish EHR corpora.
• To provide an accessible web interface for clinicians and researchers to explore UMLS mappings interactively.
• To identify limitations in concept coverage due to lexical variability and missing abbreviations in the Metathesaurus.

Proposed method

• Indexing the UMLS Metathesaurus using Apache Lucene to enable fast retrieval of mapping candidates from EHR text spans.
• Using the IXA pipeline for tokenization, part-of-speech tagging, and named entity recognition in Spanish clinical text.
• Applying the UKB toolkit for word sense disambiguation to resolve lexical ambiguities in clinical terms.
• Generating candidate mappings by matching text spans against indexed UMLS concepts using fuzzy and exact string matching.
• Implementing a web-based interface using Angular2 that displays annotated terms with color-coding by semantic type and provides concept details via UMLS API.
• Enabling users to filter results by UMLS semantic types and navigate concept hierarchies (hypernyms/hyponyms).

Experimental results

Research questions

• RQ1 How well can a UMLS-based normalization pipeline perform on Spanish EHRs when no dedicated tool exists for this language?
• RQ2 To what extent does the system’s performance align with MetaMap, a mature English-term normalization tool?
• RQ3 How effective is the combination of Lucene indexing, IXA NLP, and UKB WSD in handling lexical ambiguity and term variation in clinical text?
• RQ4 What are the main sources of false negatives in the system’s concept mapping, and how do they affect overall recall?
• RQ5 Can a web-based interface effectively support interactive exploration of UMLS mappings in clinical texts?

Key findings

• The system achieved moderate agreement with MetaMap on two parallel English-Spanish EHR corpora, with a reported kappa value of approximately 0.6.
• The best-performing configuration of the pipeline reached a precision of 0.71 and recall of 0.65 on a subset of the evaluation data.
• False negatives were primarily due to lexical variability such as misspellings, non-standard abbreviations, and morphological variants not covered by the Metathesaurus.
• Phrase-based span detection errors contributed to missed term annotations, especially when lower-level NLP components failed to identify relevant noun phrases.
• The web-based demonstrator successfully enabled users to explore UMLS concepts with semantic type filtering, concept details, and hierarchical relationships (e.g., hyponyms of 'Aspergillus').
• The system demonstrated feasibility for Spanish EHR normalization but highlighted the need for enhanced handling of term variation and expansion of abbreviation coverage.