What factors can help identify teenagers at higher psychiatric risk? Researchers found that sleep problems, family history and social stress may provide important early warning signs.
A new artificial intelligence tool being tested by researchers at Duke University could help doctors identify young people at risk of serious mental health problems before their symptoms become severe. ()
The tool is designed to look beyond an existing diagnosis or a single symptom. It analyzes information about a young person’s behavior, health, family environment and other factors to estimate the likelihood of worsening mental health.
The approach could be particularly useful in primary care, where doctors often have limited time to conduct detailed mental health assessments and access to child and adolescent mental health specialists remains limited.
AI Model Looks for Early Warning Signs
The Duke team developed the model using data from more than 11,000 children who took part in the Adolescent Brain Cognitive Development Study.
The system examines factors including family conflict, social circumstances, health information and sleep patterns. Researchers found that sleep disturbances were among the strongest predictors of future psychiatric risk.
Rather than waiting for a child to develop severe symptoms, the aim is to identify warning signs earlier, when doctors and families may have more opportunities to intervene.
In the study, a model trained on current symptoms predicted which children would move into the highest psychiatric-risk group within the following year with an area under the receiver operating characteristic curve of 0.84. A second model that focused on possible underlying risk factors achieved an AUC of 0.75 without relying on current symptom levels.
Sleep and Family Stress Emerged as Important Factors
One finding that stood out to researchers was the role of sleep. The model identified sleep disturbances as the most influential predictor among the factors examined. Family mental health history, adverse experiences, problematic behavior and family conflict were also associated with increasing psychiatric risk.
Researchers caution that the model does not show that poor sleep or family conflict directly causes mental illness. Instead, these factors can help identify children who may need closer attention or support.
The Tool Is Not Designed to Replace Doctors
The purpose of the system is to support clinicians rather than make a diagnosis on its own. Researchers envision the tool being used during routine medical visits, where a relatively brief questionnaire could provide doctors with an indication of a child’s mental health risk. A clinician could then decide whether the child needs further assessment, monitoring or specialist support.
The Duke team is now working to test the technology in clinical settings. That step is important because a model that performs well using research data still needs to demonstrate that it works reliably with patients in everyday healthcare.
Research Supports Earlier Mental Health Risk Detection
The work builds on a 2025 study published in Nature Medicine, in which Duke researchers used psychosocial questionnaires and neuroimaging data from more than 11,000 children to develop neural-network models for predicting mental health risk.
The researchers found that the model could identify young people who were likely to experience escalating psychiatric illness over the following year. Importantly, the study also found that neuroimaging did not improve the model’s performance, suggesting that information collected through relatively simple questionnaires may be enough to provide useful risk estimates.
The findings offer a possible way to expand early screening without requiring every child to undergo an extensive psychiatric evaluation.
Why Early Mental Health Screening Matters for Teens
Mental health problems during adolescence can develop gradually, and warning signs may not always be obvious to parents, teachers or primary care doctors. The CDC data show the scale of the problem: more than 40% of U.S. high school students reported feeling persistently sad or hopeless, while nearly one in five reported seriously considering suicide.
That makes earlier identification an important goal, but using AI in this setting also requires careful attention to privacy, accuracy and the consequences of false alarms. A recent review published in npj Digital Medicine noted that evidence on the safety and effectiveness of AI chatbots for youth suicide risk remains limited and called for more clinical research and stronger safety standards.
The Duke tool takes a different approach from AI chatbots because it is intended as a clinical decision-support system rather than a source of mental health advice. Its next test will be whether it can reliably help doctors identify young people who need support before a mental health crisis develops. For now, the research points to a potential role for AI in making early mental health screening more practical, but it does not replace professional assessment or clinical care.
References:
- Prediction of mental health risk in adolescents – (https://www.nature.com/articles/s41591-025-03560-7)
Source-Medindia