JMIR Publications

ABSTRACT

Background:

As the mental health issue becomes increasingly severe, adolescent suicide has become a critical global public health concern. However, early assessing the risk of suicide in adolescents remains challenging in clinical practice. In recent years, machine learning (ML)-based predictive models have gradually been applied to research on adolescent suicide; yet, systematic evidence evaluating their predictive performance is lacking.

Objective:

This study aims to review the performance of ML techniques in predicting the risk of suicide in adolescents.

Methods:

PubMed, Embase, Cochrane, and Web of Science databases were thoroughly retrieved until April 20, 2024, and the multivariate prediction model was employed for assessing the risk of bias. A meta-analysis was performed on non-suicidal self-injury (NSSI), suicidal ideation, suicide attempt, suicide attempt combined with suicidal ideation, and suicide attempt combined with NSSI, and their accuracy in the validation sets was evaluated.

Results:

42 studies involving 104 models and 1,408,375 adolescents were encompassed. The combined area under the receiver operating characteristic curve (AUC) values for ML models in predicting NSSI, suicidal ideation, suicide attempt, suicide attempt combined with suicidal ideation, and suicide attempt combined with NSSI were 0.79 (95% CI: 0.72–0.86), 0.77 (95% CI: 0.71–0.83), 0.84 (95% CI:0.83–0.86), 0.82 (95% CI:0.79–0.84), and 0.75 (95% CI: 0.73–0.76).

Conclusions:

The findings indicate that ML techniques exhibit a promising predictive performance for forecasting the risk of suicide in adolescents. Future research should develop suicide prediction tools based on ML methods, and offer scientific evidence and guidance for the formulation of intervention strategies for adolescent mental health. Clinical Trial: PROSPERO number: CRD42024566433.