2025-08-28
Can AI save teen lives?
Psychiatry
By Ana Espino | Published on August 28, 2025 | 3 min read
#AdolescentSuicidality #AI #MachineLearning #Prevention
Suicide ranks among the leading causes of death in adolescents, making it a particularly pressing public health issue. Traditional methods of risk assessment rely mainly on clinical interviews and standardized questionnaires. However, these approaches face major limitations: underreporting of symptoms, difficulty in detecting subtle and early warning signs, subjectivity in interpreting responses, and limited predictive power for anticipating suicidal behaviors.
A critical challenge lies in developing tools that can reliably, proactively, and individually detect at-risk youth, while accounting for temporal dynamics and diverse cultural contexts. In this respect, the rise of artificial intelligence and machine learning models opens new possibilities to analyze large datasets and uncover complex predictive patterns that traditional methods cannot detect.
This study was conducted to evaluate the capacity of machine learning models to predict suicidal risk among adolescents, identify the most relevant predictive factors, and determine the most effective algorithms for improving prevention and targeted intervention.
What if an algorithm saw what we
cannot?
Twenty-four studies published between 2018 and 2024, involving more than 14,000 adolescent and young adult participants, were included.
Data sources included psychometric questionnaires (PHQ-9, GAD-7), electronic medical records, and social media content analysis. Models tested encompassed both supervised and unsupervised algorithms such as Random Forest, Convolutional Neural Networks (CNNs), Long Short-Term Memory networks (LSTM), as well as more traditional approaches like logistic regression. The most powerful predictive factors were depression, anxiety, history of self-harm, social isolation, and family conflict.
The highest predictive performance was observed with Random Forest and CNN models, closely followed by LSTM. In comparison, simpler models performed significantly worse, confirming the value of advanced learning approaches.
Moreover, integrating multimodal data—combining clinical, behavioral, and social information—substantially improved predictive accuracy, underscoring the importance of a holistic and contextual approach to anticipating suicidal risk.
Adolescent suicide remains a major public health concern, requiring more effective detection tools than traditional clinical approaches. The challenge is to integrate AI into early, reliable, and personalized prevention strategies, capable of detecting weak signals before a crisis occurs. This study aimed to assess whether machine learning models could effectively predict suicidal risk from diverse data sources and to identify the most promising approaches.
Results confirm that certain algorithms—particularly Random Forest and Convolutional Neural Networks—offer high predictive power, especially when based on multimodal datasets combining clinical, behavioral, and social information. Future directions include the development of large-scale longitudinal cohorts, the integration of interpretable models for clinical application, and the evaluation of their effectiveness within prevention programs implemented in schools and healthcare facilities. Such steps are essential to strengthen proactive action against this critical issue.
About the author – Ana Espino
As a scientific writer, Ana is passionate about bridging the gap between research and real-world impact. With expertise in immunology, virology, oncology, and clinical studies, she makes complex science clear and accessible. Her mission: to accelerate knowledge sharing and empower evidence-based decisions through impactful communication.
#AdolescentSuicidality #AI #MachineLearning #Prevention
Suicide ranks among the leading causes of death in adolescents, making it a particularly pressing public health issue. Traditional methods of risk assessment rely mainly on clinical interviews and standardized questionnaires. However, these approaches face major limitations: underreporting of symptoms, difficulty in detecting subtle and early warning signs, subjectivity in interpreting responses, and limited predictive power for anticipating suicidal behaviors.
A critical challenge lies in developing tools that can reliably, proactively, and individually detect at-risk youth, while accounting for temporal dynamics and diverse cultural contexts. In this respect, the rise of artificial intelligence and machine learning models opens new possibilities to analyze large datasets and uncover complex predictive patterns that traditional methods cannot detect.
This study was conducted to evaluate the capacity of machine learning models to predict suicidal risk among adolescents, identify the most relevant predictive factors, and determine the most effective algorithms for improving prevention and targeted intervention.
What if an algorithm saw what we
cannot?
Twenty-four studies published between 2018 and 2024, involving more than 14,000 adolescent and young adult participants, were included.
Data sources included psychometric questionnaires (PHQ-9, GAD-7), electronic medical records, and social media content analysis. Models tested encompassed both supervised and unsupervised algorithms such as Random Forest, Convolutional Neural Networks (CNNs), Long Short-Term Memory networks (LSTM), as well as more traditional approaches like logistic regression. The most powerful predictive factors were depression, anxiety, history of self-harm, social isolation, and family conflict.
The highest predictive performance was observed with Random Forest and CNN models, closely followed by LSTM. In comparison, simpler models performed significantly worse, confirming the value of advanced learning approaches.
Moreover, integrating multimodal data—combining clinical, behavioral, and social information—substantially improved predictive accuracy, underscoring the importance of a holistic and contextual approach to anticipating suicidal risk.
Predicting to prevent: AI on the front line
Adolescent suicide remains a major public health concern, requiring more effective detection tools than traditional clinical approaches. The challenge is to integrate AI into early, reliable, and personalized prevention strategies, capable of detecting weak signals before a crisis occurs. This study aimed to assess whether machine learning models could effectively predict suicidal risk from diverse data sources and to identify the most promising approaches.
Results confirm that certain algorithms—particularly Random Forest and Convolutional Neural Networks—offer high predictive power, especially when based on multimodal datasets combining clinical, behavioral, and social information. Future directions include the development of large-scale longitudinal cohorts, the integration of interpretable models for clinical application, and the evaluation of their effectiveness within prevention programs implemented in schools and healthcare facilities. Such steps are essential to strengthen proactive action against this critical issue.
Read next:
About the author – Ana Espino
PhD in Immunology, specialized in Virology
Last press reviews
Single-cell epigenomic rewiring in alzheimer’s disease progression and cognitive resilience
By Lila Rouland | Published on August 21, 2025 | 2 min read