Using Natural Language Processing to Predict Risk in Electronic Health Records.

Area of Science:

• Forensic Psychiatry
• Clinical Psychology
• Natural Language Processing

Background:

• Electronic health records (EHRs) contain valuable clinical narratives on patient behaviors and emotions.
• This rich data from forensic psychiatric centers has been underutilized for risk prediction.
• Previous studies used natural language processing (NLP) to identify symptoms from free-text records for risk prediction, but with limitations.

Purpose of the Study:

• To introduce an improved NLP approach for enhanced risk prediction in forensic psychiatric settings.
• To extract additional behavioral and emotional information from clinical narratives by incorporating a social interaction component.
• To improve the performance of machine-learning models for risk prediction using this enhanced data.

Main Methods:

• Creation of four dictionaries using the Diagnostic and Statistical Manual of Mental Disorders, Unified Medical Language System, sentiment analysis, and high-frequency word corpora.
• Development of an improved NLP approach incorporating a social interaction component to analyze patient behavioral and emotional states.
• Application of a machine-learning model utilizing the extracted social interaction features to enhance risk prediction.

Main Results:

• Keyword-based models alone showed limited predictive power.
• The improved NLP approach, incorporating social interactions, extracted additional relevant information from clinical narratives.
• The enhanced data significantly improved the performance of the machine-learning model in risk prediction.

Conclusions:

• An improved NLP approach with a social interaction component can effectively enhance risk prediction in forensic psychiatry.
• Analyzing social interactions within clinical narratives provides valuable insights beyond simple keyword identification.
• This methodology offers a promising avenue for more accurate risk assessment in psychiatric care.