Performance of Natural Language Processing for Information Extraction From Electronic Health Records Within Cancer:...

Area of Science:

• Biomedical Informatics
• Computational Linguistics

Background:

• Natural Language Processing (NLP) offers solutions for information extraction (IE) from clinical data.
• NLP applications in cancer research are growing, focusing on extracting cancer-related entities from clinical texts.

Purpose of the Study:

• To summarize and compare the performance of various NLP models for IE in cancer research.
• To provide an overview of the relative performance of existing NLP models for cancer-related entity extraction.

Main Methods:

• Systematic literature review of 3 databases (PubMed, Scopus, Web of Science).
• Inclusion of 33 articles on extracting cancer-related entities from clinical texts.
• Categorization of NLP models into rule-based, traditional machine learning, conditional random field-based, neural network, and bidirectional transformer (BT).
• Extraction and calculation of average F1-scores for performance comparison.

Main Results:

• Article performance varied, with F1-scores ranging from 0.355 to 0.985.
• Bidirectional transformer (BT) models demonstrated superior performance compared to all other categories, with average F1-scores between 0.2335 and 0.0439.
• An increasing trend in the implementation of BT models was observed over the years.

Conclusions:

• NLP is effective for extracting cancer-related entities from unstructured clinical text.
• Advanced NLP models generally outperform simpler ones.
• Bidirectional transformer (BT) models represent the top-performing category for this task.