Clinical Report: Predicting Adolescent Suicide Attempts Using Machine Learning
Overview
This study developed machine learning models to predict adolescent suicide attempts using data from the Korea Youth Risk Behavior Survey. The models demonstrated moderate predictive performance, with psychological and social variables being the strongest predictors.
Background
Adolescent suicide attempts are a significant public health concern, representing a leading cause of death in this age group. Early identification of at-risk adolescents is crucial for effective prevention and intervention strategies. Machine learning approaches may enhance prediction accuracy by capturing complex relationships among various risk factors.
XGBoost achieved the highest F1 score among the models evaluated.
Random forest demonstrated the highest sensitivity under screening conditions.
Hopelessness was identified as the strongest predictor of suicide attempts.
Other significant predictors included school violence, perceived stress, self-rated health, and household economic status.
Positive predictive values remained low despite improved sensitivity under screening conditions.
Clinical Implications
The findings suggest that machine learning models can aid in identifying adolescents at risk for suicide attempts, emphasizing the importance of psychological and social factors in risk assessment. Clinicians may consider integrating these models into routine screening practices.
Conclusion
Machine learning models provide a promising approach to predicting adolescent suicide attempts, highlighting the role of various psychological and social predictors. Further research is needed to enhance their predictive validity and clinical utility.
This Neuroscience Grand Rounds session, led by Yasaman Movahedi and Deanna Aghbashian, explores psychosis in adolescence through both clinical and neurocognitive lenses, emphasizing early recognition and multidisciplinary management.
