Identifying Suicidal Adolescents from Mental Health Records Using Natural Language Processing

Area of Science:

• Computational linguistics
• Adolescent mental health research
• Health informatics

Background:

• Suicidal ideation is a significant risk factor for self-harm and suicide, particularly in adolescents.
• Large-scale studies on adolescent suicidal behavior prevalence are limited.
• Electronic Health Records (EHRs) contain valuable data but often in unstructured free text.

Purpose of the Study:

• To adapt and evaluate a lexicon- and rule-based Natural Language Processing (NLP) approach for identifying suicidal adolescents.
• To assess the performance of the NLP approach on a large EHR database.
• To determine the feasibility of using NLP for suicide risk behavior surveillance in large cohorts.

Main Methods:

• Development of a comprehensive, manually annotated EHR reference standard for suicide risk.
• Adaptation and application of a simple lexicon- and rule-based NLP method.
• Evaluation of NLP performance at both document and patient levels using data from 200 patients (5000 documents).

Main Results:

• The NLP approach achieved promising performance, with an F1 score exceeding 80% at both document and patient levels.
• The study demonstrated the effectiveness of a straightforward NLP method for identifying suicidal adolescents.
• The developed approach showed potential for broader application in different populations and healthcare settings.

Conclusions:

• Simple NLP techniques are effective for identifying suicidal risk behavior in adolescents within EHRs.
• The proposed NLP approach offers a scalable solution for suicide risk surveillance.
• This methodology can be adapted for identifying at-risk individuals in other demographic groups and clinical environments.