QuantitativeSmokingStatusExtractionNLPTool

A rule-base NLP system to extract quantitative smoking information (e.g., pack-year) from clinical notes

Aim

The package is used for extraction of quantitative smoking information from clinical notes.

Smoking information type

• pack per day
• smoking year
• quit year (e.g., quit for 10 years)
• year at quit (e.g., quit at 2008)
• pack-year

Requirement

• java 1.8

Input format

For input, we expect a CSV file with encoding as UTF-8. The data table should have no header (only real data in table) and 5 columns as

1. note ID
2. patient ID
3. note Date
4. note Type
5. note text (You can use dummy text for 1-4)

Output format

The output is a TSV with encoding as UTF-8.

1. note ID
2. patient ID
3. note Date
4. note Type
5. extracted data type
6. extracted data value
7. a snippet of where the extracted value located in text (50 characters before and after the value) (1-4 is copied from input data)

How to run

1. change to the project directory
2. run java -jar RuleBaseSmokingInfoExtraction.jar or use the run.sh (modify arguments necessary)

Example

• we provide sample.csv for testing, see run.sh

Note

• this is a rule-based system
• we are keeping update rules to cover special cases

Release

• we released the RuleBaseSmokingInfoExtraction.jar
• we will release source code

Citation

Yang X, Yang H, Lyu T, Yang S, Guo Y, Bian J, Xu H, Wu Y. 
A Natural Language Processing Tool to Extract Quantitative Smoking Status from Clinical Narratives.
2020 IEEE International Conference on Healthcare Informatics (ICHI), 2020, pp. 1-2. 
doi: 10.1109/ICHI48887.2020.9374369.
PMID: 33173920; PMCID: PMC7654916.

https://ieeexplore.ieee.org/abstract/document/9374369