Using Natural Language Processing of Clinical Notes to Supplement Structured Electronic Health Record Data for...

Area of Science:

• Clinical Informatics
• Biomedical Data Science
• Natural Language Processing

Background:

• Electronic health records (EHR) often contain valuable clinical information in unstructured notes.
• Structured EHR data alone may incompletely represent patient phenotypes, impacting clinical research and care.
• Natural Language Processing (NLP) offers a method to extract information from unstructured text.

Purpose of the Study:

• To evaluate two NLP tools for extracting smoking and obesity phenotypes from unstructured EHR notes.
• To assess the added value of combining NLP-derived phenotypes with structured EHR data.
• To determine the scalability of NLP integration within a large health system.

Main Methods:

• Retrospective analysis of inpatient and outpatient EHR data from Mass General Brigham (2019-2020).
• Application of two rule-based NLP tools to extract smoking and obesity data from over 19 million clinical notes.
• Performance evaluation of NLP tools via manual review and comparison of phenotype prevalence using structured data alone versus combined data.

Main Results:

• Both NLP tools demonstrated high accuracy (0.99 for smoking, 0.91-0.92 for obesity).
• Combining NLP with structured data increased smoking patient identification by 29.5% (43.88% vs. 33.87%).
• NLP integration yielded a 19.3% increase in obesity case identification.

Conclusions:

• NLP extraction from unstructured EHR notes significantly improves patient identification for smoking and obesity.
• Integrating NLP-derived phenotypes with structured EHR data enhances case ascertainment at a health system scale.
• NLP is a valuable tool for unlocking comprehensive clinical insights from EHR data.