Predicting suicide attempts among Norwegian adolescents without using suicide-related items: a machine learning...

Area of Science:

• Psychiatry
• Machine Learning
• Adolescent Health

Background:

• Classifying adolescent suicide attempts often relies on sensitive data, posing challenges for population-level studies.
• Existing models struggle with data collection, particularly for adolescents.

Purpose of the Study:

• To assess the feasibility of classifying high-risk adolescent suicide attempters without sensitive suicide-related survey items.
• To identify key predictors of suicide attempts among adolescents.

Main Methods:

• Utilized nationwide survey data from 173,664 Norwegian adolescents (ages 13-18).
• Employed the Extreme Gradient Boosting (XGBoost) algorithm for binary classification.
• Analyzed 169 questionnaire items to identify suicide attempt predictors.

Main Results:

• XGBoost model achieved 77% sensitivity and 90% specificity.
• The model demonstrated strong performance with an AUC of 92.1% and AUPRC of 47.1%.
• Identified internalizing problems, substance use, interpersonal relationships, and victimization as significant predictors.

Conclusions:

• Machine learning offers a viable approach for population-scale screening of adolescent suicide attempts without sensitive data.
• Future research should focus on internalizing problems, interpersonal relationships, victimization, and substance use in suicidal behavior etiology.