Artificial intelligence and machine learning for detecting suicidal ideation and related suicide outcomes among adolescents: A scoping review

AI Summary

AI and machine learning show potential as exploratory decision support for adolescent suicidal ideation detection but current evidence is heterogeneous and insufficient for clinical implementation.
Random Forest and XGBoost were common, showing promising internal performance; key predictors included depression, anxiety, loneliness, non suicidal self injury and prior attempts.
Future research must emphasise adolescent specific samples, external validation, calibration, explainability, fairness assessment, ethical safeguards, and clear outcome distinctions.

AI Summary

OBJECTIVE: This scoping review aimed to map the existing evidence on the application of artificial intelligence and machine learning in detecting suicidal ideation and related suicide outcomes, including suicide plans and suicide attempts, among adolescents.

Eligible studies involved adolescent populations or reported adolescent-specific findings, used real human-subject data, and applied artificial intelligence or machine learning methods to detect suicidal ideation, suicide plans, suicide attempts, or composite suicide-risk outcomes.

Future research should prioritize adolescent-specific samples, external validation, calibration, explainability, fairness assessment, ethical safeguards, and clear distinction between suicidal ideation, suicide plans, suicide attempts, and composite suicide-risk outcomes.

Basic summary

Acta Psychol (Amst). 2026 Jul 9;268:107418. doi: 10.1016/j.actpsy.2026.107418. Online ahead of print.

ABSTRACT

METHODS: A scoping review was conducted following the Arksey and O’Malley framework, refined by Levac et al., and reported according to PRISMA-ScR guidance. Searches were conducted in Scopus, PubMed, PsycINFO, and Web of Science for English-language original empirical studies published between 2016 and 2026. Eligible studies involved adolescent populations or reported adolescent-specific findings, used real human-subject data, and applied artificial intelligence or machine learning methods to detect suicidal ideation, suicide plans, suicide attempts, or composite suicide-risk outcomes. Data were charted descriptively and synthesized narratively.

RESULTS: Eight original studies met the eligibility criteria and were included in the final synthesis. The studies used school-based surveys, national youth risk behavior data, family-reported data, structured clinical assessments, and psychiatric clinical data. Tree-based and ensemble models, particularly Random Forest and XGBoost, were commonly used and generally reported promising internal performance. Key predictors included depression, anxiety, loneliness, emotional pain, non-suicidal self-injury, prior suicide attempts, family functioning, parental support, and clinical severity. However, the studies varied substantially in outcome definitions, data sources, validation approaches, and performance metrics.

CONCLUSION: Artificial intelligence and machine learning show potential as exploratory decision-support tools for detecting suicidal ideation and related suicide outcomes among adolescents. However, current evidence remains methodologically heterogeneous and insufficient for routine clinical implementation. Future research should prioritize adolescent-specific samples, external validation, calibration, explainability, fairness assessment, ethical safeguards, and clear distinction between suicidal ideation, suicide plans, suicide attempts, and composite suicide-risk outcomes.

PMID:42424915 | DOI:10.1016/j.actpsy.2026.107418

Document this CPD