Supporting the annotation of chronic obstructive pulmonary disease (COPD) phenotypes with text mining workflows

Area of Science:

• Medical Informatics
• Natural Language Processing
• Pulmonology

Background:

• Chronic obstructive pulmonary disease (COPD) poses a significant public health challenge.
• Personalized treatment for COPD relies on identifying patient phenotypes from electronic health records.
• Phenotypes are often embedded in unstructured clinical text, necessitating advanced text mining tools.

Purpose of the Study:

• To develop a semi-automatic methodology for creating a corpus to support text mining tools.
• To facilitate the identification of chronic obstructive pulmonary disease (COPD) patient groups based on phenotypes.
• To expedite the extraction of phenotypic information from clinical records.

Main Methods:

• A corpus of 30 full-text papers was curated based on COPD specialist input.
• An annotation scheme was designed for fine-grained, computable COPD phenotype annotations.
• A semi-automatic annotation workflow was implemented using the Argo platform and integrated text mining tools.

Main Results:

• The semi-automatic workflow achieved a micro-averaged F-score of 45.70% when evaluated against gold standard annotations.
• The developed corpus demonstrated potential in training improved COPD phenotype extraction models.
• The methodology shows promise for enhancing the accuracy of phenotype identification.

Conclusions:

• A semi-automatic workflow integrating text mining tools was established to support COPD phenotype curation.
• The ongoing development of the corpus shows encouraging results for future automatic phenotype extractors.
• This approach has the potential to significantly improve the identification of COPD patient phenotypes.