Natural Language Processing for Substance Use Disorder Information Extraction: A Systematic Literature Review

Area of Science:

• Medical Informatics
• Computational Linguistics
• Public Health

Background:

• Substance Use Disorders (SUDs) pose a significant public health challenge.
• Accurate information extraction from clinical text is crucial for research and decision-making.
• Natural Language Processing (NLP) offers potential for automating this extraction process.

Purpose of the Study:

• To systematically review the application of NLP techniques for extracting information related to substance use disorders.
• To identify common NLP approaches, methods, software, and evaluation metrics used in SUD research.

Main Methods:

• A systematic review of 623 studies was conducted, with 35 meeting the inclusion criteria.
• Studies were analyzed based on the type of SUD addressed (alcohol, opioids, tobacco, multiple).
• NLP approaches (Rule-Based, Machine Learning, Deep Learning), methods, software (Python, cTAKES), and evaluation metrics (F1, ROC AUC) were categorized.

Main Results:

• The majority of included studies (65.7%) employed Rule-Based NLP approaches, with concept extraction being the most common method (43%).
• Opioid-related SUDs (34.3%) and multiple SUDs (45.7%) were frequently studied.
• Python emerged as the most utilized software (10 papers), followed by cTAKES (9 papers).
• Evaluation metrics like ROC AUC and F1 scores demonstrated acceptable to outstanding performance.

Conclusions:

• Rule-Based NLP methods, concept extraction, and common software like Python are prevalent in current SUD information extraction research.
• Existing NLP applications achieve good performance in identifying SUD-related information.
• Future research should explore advanced Machine Learning or Deep Learning approaches and specialized NLP software to enhance public health research and clinical decision-making.