Predicting the onset of internalizing disorders in early adolescence using deep learning optimized with AI.

Area of Science:

• Developmental Psychology
• Computational Psychiatry
• Machine Learning in Mental Health

Background:

• Internalizing disorders (depression, anxiety, somatic symptom disorder) are prevalent in early adolescence, significantly impacting daily functioning.
• Identifying predictors of these disorders is crucial for developing effective intervention and prevention strategies during this critical developmental period.

Purpose of the Study:

• To construct individual-level predictive models for the onset of depression, anxiety, and somatic symptom disorder in early adolescence.
• To analyze a wide range of candidate predictors across cognitive, psychosocial, neural, and biological domains using advanced machine learning techniques.

Main Methods:

• Utilized deep learning with artificial neural networks, guided by an evolutionary algorithm for hyperparameter optimization and automated feature selection.
• Analyzed approximately 6,000 candidate predictors from the ABCD cohort for children aged 9-10 years to predict onset at ages 11-12 years.
• Compared the predictive power of different predictor domains (psychosocial, cognitive, neural, biological).

Main Results:

• Achieved robust prediction of internalizing disorder onset in early adolescence with Area Under the Curve (AUROC) values ≥0.90 and accuracy ≥80%.
• Identified specific predictor sets for each disorder, with parent behavioral traits and sleep disturbances emerging as cross-cutting themes.
• Demonstrated that psychosocial predictors were more influential than cognitive, neural, or biological factors in predicting these disorders.

Conclusions:

• Deep learning models can accurately predict the onset of early adolescent internalizing disorders.
• Psychosocial factors play a significant role, highlighting potential targets for early intervention and prevention programs.
• Further research is needed to validate findings in diverse datasets and explore applications to other developmental stages and mental health conditions.