Extracting Sexual Trauma Mentions from Electronic Medical Notes Using Natural Language Processing.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Public Health

Background:

• Patient history of sexual trauma is clinically significant due to adverse health outcomes for survivors.
• Accurate identification of sexual trauma history in electronic medical records is crucial for healthcare providers.
• Existing methods may not effectively capture nuanced mentions within free-text clinical notes.

Purpose of the Study:

• To develop and evaluate a natural language processing (NLP) pipeline for identifying sexual trauma mentions in electronic medical notes.
• To create a domain-specific lexicon to enhance the accuracy of sexual trauma detection.
• To assess the performance of the NLP tool at both mention and patient levels.

Main Methods:

• Development of a natural language processing pipeline for information extraction from free-text electronic medical notes.
• Creation of a domain-specific lexicon for identifying sexual trauma.
• Scaling the pipeline to a large corpus of notes from US Veterans Health Administration facilities.
• Evaluation of extracted snippets by trained human reviewers.

Main Results:

• The NLP pipeline achieved an overall positive predictive value (PPV) of 0.90 for identifying sexual trauma mentions.
• A patient-level PPV of 0.71 was reported for the identification of sexual trauma history.
• Performance metrics were notably superior for records pertaining to female patients.

Conclusions:

• The developed NLP tool demonstrates high accuracy in identifying sexual trauma mentions within electronic medical records.
• This method can assist healthcare providers in recognizing patient histories of sexual trauma, facilitating improved care.
• Further refinement may enhance patient-level accuracy and applicability across diverse patient populations.