Using machine learning-based Natural Language Processing to quantify emergency department presentations related to...

Area of Science:

• Public Health
• Health Informatics
• Computational Linguistics

Background:

• Suicide and self-harm represent a global health challenge requiring accurate data for effective interventions and policy development.
• Current methods for identifying emergency department (ED) presentations for self-harm are often inefficient or underestimate the true prevalence.
• Timely and comprehensive data are crucial for understanding and addressing suicide and self-harm behaviors.

Purpose of the Study:

• To evaluate a novel machine learning-based Natural Language Processing (NLP) program for quantifying ED presentations related to suicidal ideation or behavior.
• To compare the NLP program's performance against traditional methods like ICD-10-AM coding and keyword searching.
• To assess the accuracy, efficiency, and comprehensiveness of the NLP approach in identifying self-harm related ED visits.

Main Methods:

• Developed and implemented a machine learning-based NLP program to analyze ED triage notes for suicide and self-harm indicators.
• Compared the NLP program's identification of presentations with those identified using International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification (ICD-10-AM) codes.
• Utilized keyword searching as an additional comparative method for identifying relevant ED presentations.

Main Results:

• The NLP program identified 27,298 ED presentations related to suicide or self-harm between July 2015 and June 2022, significantly more than the 10,399 identified by ICD-10-AM codes.
• The NLP program demonstrated high performance with a precision of 0.89 and a recall of 0.94 in identifying relevant presentations.
• Manual review confirmed the NLP identification as the most accurate, comprehensive, and efficient method.

Conclusions:

• Existing methods identify less than 40% of ED presentations related to suicide or self-harm in the Australian Capital Territory.
• The developed NLP program offers a superior method for accurately quantifying suicide and self-harm presentations in emergency departments.
• Improved data identification through this NLP tool will enhance the understanding and management of suicide and self-harm issues.