Tobacco use status from clinical notes using Natural Language Processing and rule based algorithm.

Area of Science:

• Medical Informatics
• Clinical Data Mining
• Natural Language Processing

Background:

• Electronic Health Records (EHR) contain valuable clinical notes.
• Extracting specific patient data like tobacco-use status from unstructured text is challenging.

Purpose of the Study:

• To develop a rule-based algorithm using NLP for determining current tobacco-use status.
• To assess the accuracy of NLP in classifying patient tobacco-use from clinical notes.

Main Methods:

• Retrospective study analyzing 5,371 clinical notes from 363 patients.
• NLP software classified documents into "Current Smoker", "Past Smoker", "Nonsmoker", and "Unknown" categories.
• Manual classification by two coders served as the gold standard for comparison.

Main Results:

• High inter-rater reliability (Cohen's kappa = 0.9889) for manual classification.
• NLP achieved F-measures ranging from 0.700 to 0.988 for document-level classification.
• Patient-level classification F-measures ranged from 0.580 to 0.933.

Conclusions:

• NLP and rule-based algorithms are useful for determining patient tobacco-use status from EHRs.
• The developed tool demonstrates potential for improving clinical data extraction.
• Further precision enhancements are targeted to increase the tool's translational value.