Machine Learning Prediction of First Suicide Attempts in Early Adolescence Among Child Ideators

Area of Science:

• Child and Adolescent Psychiatry
• Developmental Neuroscience
• Public Health

Background:

• Adolescent suicide is a major public health concern.
• Early identification of individuals at high risk for suicide attempts is critical for prevention.
• Predicting the transition from suicidal ideation to attempt is a key challenge.

Purpose of the Study:

• To predict the first suicide attempt in adolescents with baseline suicidal ideation.
• To identify sociodemographic, clinical, neurocognitive, functional, and structural brain predictors.
• To develop a machine learning model for early suicide risk assessment.

Main Methods:

• Utilized data from the longitudinal Adolescent Brain Cognitive Development (ABCD) study (N=11,864).
• Trained machine learning models on 70% of children (aged 9-10) with suicidal ideation at baseline.
• Validated models using data from holdout sites, focusing on the primary outcome of first suicide attempt.

Main Results:

• The final model, excluding brain imaging, predicted first suicide attempts with an AUC of 0.75.
• Top predictors included female sex, self-harm, access to means, anxiety disorders, impulsivity, and suicidal ideation severity.
• The model demonstrated good generalization and was unbiased across subgroups.

Conclusions:

• A predictive model using clinically accessible features can identify children at risk for first suicide attempts.
• Modifiable risk factors like impulsivity and anxiety symptoms present intervention targets.
• Findings support current suicide theories by identifying key risk factors for the ideation-to-attempt transition.