Empirical, automated vocabulary discovery using large text corpora and advanced natural language processing tools

Area of Science:

• Medical Informatics
• Natural Language Processing
• Clinical Terminology

Background:

• Electronic medical records (EMRs) are hindered by incomplete clinical vocabularies.
• Existing systems lack expressiveness for clinical diagnoses, findings, severity, acuity, and temporal factors.
• Manual vocabulary construction fails to capture diverse clinical language.

Purpose of the Study:

• To address the limitations of current clinical vocabularies in EMRs.
• To utilize natural language processing (NLP) to identify unrepresented clinical terminology.
• To compare NLP-identified terms against the Unified Medical Language System (UMLS) Metathesaurus and quantify discovery effort.

Main Methods:

• Application of advanced NLP tools to a clinical findings corpus.
• Comparison of identified terminology with the UMLS Metathesaurus.
• Quantification of human and computational resources for terminology discovery.

Main Results:

• A substantial number of clinical phrases and modifiers were absent from the UMLS Metathesaurus.
• NLP effectively identified diverse and previously unrepresented clinical language.
• The effort required for discovering additional terminology was modest.

Conclusions:

• Advanced NLP methods can significantly enhance clinical vocabularies for EMRs.
• The UMLS Metathesaurus coverage of clinical findings is incomplete.
• NLP offers an efficient approach to expand clinical terminologies, improving EMR expressiveness.