Using a natural language processing system to extract and code family history data from admission reports

Area of Science:

• Clinical Informatics
• Natural Language Processing
• Medical Data Mining

Background:

• Extracting structured clinical data from unstructured text is challenging.
• Family history is crucial for risk assessment but often buried in free text.
• Automated methods are needed to efficiently code family history data.

Purpose of the Study:

• To develop and evaluate a rule-based natural language processing (NLP) system.
• To assess the system's accuracy in extracting and coding family history from hospital admission notes.
• To determine the system's performance for disease and relative categorization.

Main Methods:

• Developed a rule-based NLP system for clinical data extraction.
• Applied the system to hospital admission notes focusing on family history sections.
• The system was designed to identify 12 specific diseases and the degree of relation.
• Evaluated performance using sensitivity and positive predictive value (PPV).

Main Results:

• The NLP system achieved a sensitivity of 0.96 for disease extraction.
• Positive predictive value (PPV) for disease extraction was 0.97.
• Sensitivity for relative categorization reached 0.96.
• PPV for relative categorization was 0.93.

Conclusions:

• The developed rule-based NLP system demonstrates high accuracy in extracting and coding family history data.
• This automated approach offers a reliable method for capturing crucial family history information from clinical text.
• The system's performance suggests its potential utility in clinical decision support and epidemiological studies.