Identifying Patients With Hypoglycemia Using Natural Language Processing: Systematic Literature Review.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Data Analysis

Background:

• Accurate identification of hypoglycemia is critical for patient safety and mortality reduction.
• Natural Language Processing (NLP) offers a scalable and efficient method for extracting clinical information from large electronic health record (EHR) datasets.
• NLP utilizes computational algorithms to process and analyze text data, making it suitable for identifying specific medical conditions like hypoglycemia within clinical notes.

Purpose of the Study:

• To systematically review and synthesize existing literature on the application of NLP techniques for extracting hypoglycemia data from EHR clinical notes.
• To evaluate the effectiveness of NLP in identifying hypoglycemic events compared to traditional methods.

Main Methods:

• Conducted comprehensive literature searches across multiple databases including PubMed, Web of Science, CINAHL, PsycINFO, IEEE Xplore, Google Scholar, and ACL Anthology.
• Included studies published in English that utilized NLP to identify hypoglycemia, reported relevant outcomes, and were available as full papers.
• Keywords for the search included "hypoglycemia", "low blood glucose", "NLP", and "machine learning".

Main Results:

• The systematic review included 8 studies, highlighting heterogeneity in reported hypoglycemia prevalence rates, ranging from 3.4% to 46.2%.
• NLP application in analyzing clinical notes demonstrated improved capture of undocumented or missed hypoglycemic events, outperforming International Classification of Diseases (ICD-9/ICD-10) codes and laboratory testing alone.
• Combining NLP with ICD-9 or ICD-10 codes significantly increased the identification rate of hypoglycemic events, with prevalence rates reaching 32.2% for combined methods compared to 12.4% for ICD codes and 25.1% for NLP alone. All reviewed studies employed rule-based NLP algorithms.

Conclusions:

• The application of NLP to clinical notes effectively improves the detection of hypoglycemic events within EHR data.
• Combining NLP with established coding systems like ICD-9/ICD-10 and laboratory test results offers a superior strategy for comprehensive hypoglycemia identification.
• NLP presents a valuable tool for enhancing clinical data analysis and improving patient care by accurately identifying at-risk individuals.