A Natural Language Processing Tool to Extract Quantitative Smoking Status from Clinical Narratives

Area of Science:

• Medical Informatics
• Natural Language Processing
• Clinical Data Extraction

Background:

• Accurate extraction of quantitative smoking information from clinical notes is crucial for lung cancer risk assessment.
• Standardizing smoking data (e.g., Pack-Year, Quit Year) aids in large-scale data analysis and research.

Purpose of the Study:

• To develop and evaluate a natural language processing (NLP) tool for extracting quantitative smoking data from clinical notes.
• To standardize extracted smoking information into a uniform unit (Pack-Year).

Main Methods:

• Annotation of 200 clinical notes from patients undergoing low-dose CT for lung cancer screening.
• Development of a two-layer rule-engine based NLP system.
• System training and evaluation using distinct training and test data sets.

Main Results:

• The NLP system achieved high performance on a test set of clinical notes.
• Achieved F1 scores of 0.963 (lenient) and 0.946 (strict) for quantitative smoking information extraction.
• Demonstrated effectiveness in standardizing smoking data into the Pack-Year unit.

Conclusions:

• The developed NLP tool accurately extracts and standardizes quantitative smoking information from clinical notes.
• This tool can facilitate research in lung cancer screening by providing reliable smoking history data.
• The system's performance indicates its potential for integration into clinical data management systems.