Prediction of first attempt of suicide in early adolescence using machine learning

Abstract

Background

Suicide is the second leading cause of death among early adolescents, yet the first onset of suicide attempts during this critical developmental period remains poorly understood. This study aimed to identify key characteristics associated with the first suicide attempt in early adolescence and to develop a predictive model for assessing individual risk.

Methods

We used data from the Adolescent Brain Cognitive Development Study, a longitudinal, population-based study in the US. The analysis focused on a cohort of 4,238 early adolescents (aged 11–12 years) who had no prior history of suicide attempts. To predict the onset of a first suicide attempt over the subsequent two years (2020–2022), we developed an extremely randomized tree model, incorporating 87 potential predictors from diverse bio-psycho-social domains pertinent to adolescent development.

Results

Among the 4,238 adolescents, 163 (3.8%) reported their first suicide attempt within the subsequent two years. Our predictive model demonstrated good discriminative ability, achieving an AUC of 0.82 (95% CI [0.79, 0.85]), with a sensitivity of 0.82 and a specificity of 0.69 at the optimized threshold. Key predictors included sex assigned at birth, sexual orientation, negative affect, internalizing and attention problems, and lifetime suicidal ideation, along with other significant factors from multiple domains.

Conclusions

These findings highlight the utility of machine learning algorithms in identifying predictors of suicide attempts among early adolescents. The insights gained from this study may contribute to the development of tailored screening tools and preventive interventions aimed at mitigating suicide risk in this vulnerable population.