Machine learning-based prediction for self-harm and suicide attempts in adolescents

Area of Science:

• Adolescent mental health
• Machine learning applications in psychology
• Suicidology

Background:

• Adolescent self-harm and suicide attempts pose significant public health challenges.
• Accurate risk prediction is crucial for timely intervention and prevention strategies.
• Existing prediction models often lack comprehensive variable selection capabilities.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting adolescent self-harm and suicide attempts.
• To identify key predictors of self-harm and suicide attempts in adolescents using ML.
• To compare the predictive performance of ML models against traditional methods.

Main Methods:

• Secondary analysis of cross-sectional data from the Longitudinal Study of Australian Children.
• Utilized random forest classification to select optimal predictors and generate risk predictions.
• Included variables related to mental health, socio-demographics, and psychosocial factors at ages 14-15 to predict outcomes at 16-17.

Main Results:

• ML models demonstrated fair predictive accuracy for self-harm (AUC: 0.7397) and suicide attempts (AUC: 0.7220).
• These models significantly outperformed prediction based solely on prior self-harm or suicide attempts (AUC: 0.6).
• Key predictors included depressed feelings, Strengths and Difficulties Questionnaire scores, self-perception, and school/parental factors.

Conclusions:

• Machine learning, specifically random forest classification, is effective in identifying critical predictors for adolescent self-harm and suicide risk.
• ML models offer a promising approach to enhance the accuracy of risk prediction in adolescent mental health.
• Further research is warranted to validate and scale ML techniques in clinical mental health settings.