CUILESS2016: a clinical corpus applying compositional normalization of text mentions.

Area of Science:

• Natural Language Processing
• Medical Informatics
• Computational Linguistics

Background:

• Traditional clinical text normalization uses single identifiers (pre-coordinated concepts) or restricted multi-identifier concepts (post-coordinated concepts).
• Existing methods limit the semantic expressiveness of clinical text normalization.
• Compositional concepts, using multiple identifiers without defined relationships, offer a novel approach.

Purpose of the Study:

• To evaluate the utility of compositional concepts for clinical text normalization.
• To generate a freely available corpus of compositional concept annotations.
• To assess the impact on semantic coverage and annotator agreement.

Main Methods:

• Annotated 5397 disorder mentions from the ShARe corpus to SNOMED CT using compositional concepts.
• Allowed normalization to multiple Unified Medical Language System (UMLS) Concept Unique Identifiers (CUIs) without restricting semantic types.
• Computed annotator agreement using exact and hierarchical matching metrics.

Main Results:

• Generated the largest freely available clinical text normalization corpus to date.
• Successfully normalized 5389 out of 5397 disorder mentions using compositional concepts.
• Achieved annotator agreement ranging from 52.4% (exact match) to 78.2% (hierarchical agreement).

Conclusions:

• Compositional concepts significantly increase semantic coverage in clinical text normalization.
• This work presents the first freely available corpus of compositional concept annotation for clinical text.
• The developed methodology offers a more comprehensive approach to representing clinical concepts.