Anxiety onset in adolescents: a machine-learning prediction

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Adolescence is a critical period for anxiety disorder onset.
• Longitudinal studies suggest MRI correlates of anxiety symptoms, but predictive value is unestablished.
• Machine learning may enhance the clinical relevance of predictive markers for anxiety.

Purpose of the Study:

• To evaluate the predictive value of gray matter volumes and psychometric scores for prospective clinical anxiety in adolescents using machine learning.
• To identify key features contributing to the prediction of pooled anxiety disorders and generalized anxiety disorder (GAD).

Main Methods:

• A voting classifier combining Random Forest, Support Vector Machine, and Logistic Regression was employed.
• Data included gray matter volumes and psychometric scores from 14-year-old adolescents (N=424 healthy, N=156 with future clinical anxiety).
• Shapley values were used for feature importance interpretation.

Main Results:

• Prospective prediction of pooled anxiety disorders achieved moderate performance (AUC=0.68), primarily driven by psychometric features (neuroticism, hopelessness, emotional symptoms).
• MRI regional volumes did not enhance prediction for pooled anxiety disorders beyond psychometric features alone.
• MRI volumes, particularly caudate and pallidum, improved GAD prediction accuracy.

Conclusions:

• Clinical anxiety onset in adolescence can be individually predicted 4-8 years in advance.
• Psychometric features are crucial for predicting general anxiety disorders.
• Neuroanatomical data (caudate and pallidum volumes) are valuable additions for predicting GAD in adolescents.