Named Entity Recognition in Electronic Health Records: A Methodological Review

Area of Science:

• Natural Language Processing (NLP)
• Health Informatics
• Clinical Data Mining

Background:

• Electronic Health Records (EHRs) contain substantial unstructured free text, limiting clinical decision-making utility.
• Named Entity Recognition (NER) is crucial for extracting valuable information from this unstructured data.

Purpose of the Study:

• To review current Named Entity Recognition (NER) methods.
• To trace the evolution of NER techniques in Electronic Health Records (EHRs) from 2011 to 2022.

Main Methods:

• Methodological literature review focusing on NER.
• Analysis of classification models, tagging systems (e.g., BIO), and languages used in corpora.
• Examination of Natural Language Processing (NLP) techniques like NER and Relation Extraction (RE).

Main Results:

• NER and RE methods can extract concepts, events, attributes, and relationships from EHRs.
• Most studies utilize English or Chinese corpora; bidirectional encoder representation from transformers with BIO tagging is common.
• Limited research exists on domain-specific NER/RE implementation in EHRs.

Conclusions:

• EHRs are vital for clinical information and automated decision support.
• Developing new, domain-specific EHR corpora is essential for advancing NER and RE models in clinical practice.