Automated Detection of Substance-Use Status and Related Information from Clinical Text.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Clinical Data Mining

Background:

• Extracting patient substance use information from unstructured clinical text is challenging.
• Accurate identification of substance use (smoking, alcohol, drugs) is crucial for patient care.
• Existing methods often lack efficiency and comprehensive attribute extraction.

Purpose of the Study:

• To develop and evaluate an automated system for extracting patient substance use status and attributes from clinical discharge records.
• To improve the accuracy and efficiency of substance use information retrieval from unstructured medical text.
• To demonstrate the system's generalisability on unseen datasets.

Main Methods:

• A four-stage system employing keyword search, enhanced NegEx negation detection, windowing/chunking for temporal status, and regular expressions/syntactic patterns for attribute extraction.
• Utilized Natural Language Processing (NLP) and rule-based techniques for robust information extraction.
• System evaluated for sentence-level and document-level data processing.

Main Results:

• Achieved high F1-scores: up to 0.99 for substance-use identification, 0.98 for negation detection, and 0.94 for temporal status.
• Excellent performance in attribute extraction: F1-scores up to 0.98 (amount, frequency, type, period) and 1.00 (type).
• Outperformed state-of-the-art systems on an unseen dataset, demonstrating strong generalisability.

Conclusions:

• The developed automated system effectively extracts comprehensive patient substance use information from clinical text.
• The NLP and rule-based approach offers high accuracy and generalisability for clinical data mining.
• This system has the potential to enhance clinical decision-making and patient record analysis.