Comparing Risk Prediction for Suicide Attempts and Deaths After Emergency Department Visits for Individuals With...

Area of Science:

• Public Health
• Clinical Psychology
• Data Science in Healthcare

Background:

• Emergency department visits for mental health disorders are common.
• Identifying individuals at high risk for suicide is a critical public health challenge.
• Predicting both nonfatal and fatal suicidal events requires nuanced approaches.

Purpose of the Study:

• To identify and compare predictors of nonfatal and fatal suicidal events within 180 days of emergency department (ED) visits.
• To assess the performance of machine learning models in predicting suicidal events.
• To evaluate the generalizability of models trained on nonfatal outcomes to fatal outcomes.

Main Methods:

• Longitudinal cohort analysis of 511,559 patients presenting to EDs between 2015-2022 with mental health disorders.
• Machine learning models used to predict nonfatal suicide attempts (n=4,525) and fatal suicide events (n=434).
• Analysis incorporated demographic, clinical characteristics, and feature importance rankings.

Main Results:

• Machine learning models showed stronger performance for nonfatal events (AUC=0.874) than for suicide deaths (AUC=0.787).
• High correlation (r=0.82) in feature importance between models, with 73% overlap in top predictors.
• Models trained on nonfatal outcomes demonstrated limited generalizability to fatal outcomes (AUC=0.724).

Conclusions:

• Nonfatal and fatal suicidal events share common predictors but differ in predictive strength and function.
• Challenges exist in using nonfatal outcome models to predict suicide death risk.
• Findings contribute to the debate on whether suicidal behaviors represent a continuum or distinct pathways.