A PRISMA-ScR scoping review searched four databases and found only eight original studies from 2016-2026 applying AI or machine learning to adolescent suicidal ideation, plans, or attempts. Tree and ensemble models showed promising internal performance, but evidence was heterogeneous and insufficient for routine clinical use.
Key findings
- Sources included school surveys, national youth-risk datasets, family reports, structured assessments, and psychiatric records. Random Forest and XGBoost were common, but outcome definitions, validation methods, and metrics varied substantially, preventing direct performance comparison; most evidence was internally validated.
Why this matters globally
AI may help prioritize follow-up, but false negatives can miss risk and false positives can cause stigma or unnecessary intervention. Systems must remain accountable decision support with human oversight and response pathways, not replacements for professional assessment.
Thai researcher contribution
A Prince of Songkla University nursing researcher contributed to an interdisciplinary synthesis spanning mental health, nursing, and data science, emphasizing fairness, explainability, and ethics before deployment.
Limitations to consider
Only eight English-language studies were included. A scoping review maps evidence but does not establish pooled effectiveness. Ideation, plans, attempts, and composite outcomes are not interchangeable, and there is no evidence here that using a model reduces suicide outcomes.