Classifying Conduct Disorder Using a Biopsychosocial Model and Machine Learning Method.

Area of Science:

• Neuroscience
• Psychiatry
• Data Science

Background:

• Conduct disorder (CD) is a prevalent syndrome with significant societal impact.
• Existing predictive models for CD are limited by focusing on single risk factors or domains.
• Machine learning (ML) offers a novel approach to analyze multidomain data for predicting CD development.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting conduct disorder (CD) using a comprehensive set of risk factors.
• To determine if a multidomain approach outperforms single-domain models in predicting CD.
• To identify key social, psychological, and biological predictors of CD.

Main Methods:

• Utilized data from the Adolescent Brain Cognitive Development Study (N=2368) including social, psychological, and biological risk factors.
• Applied a feed-forward neural network machine learning model to predict CD diagnoses.
• Risk factors were assessed at ages 9-10, with CD diagnoses predicted two years later.

Main Results:

• A multidomain ML model achieved 91.18% accuracy in predicting CD diagnoses, outperforming single-domain models.
• Identified significant predictors within each domain: social (low parental monitoring, household aggression, low income), psychological (ADHD/ODD symptoms, cognitive deficits), and biological (network disruptions).
• Highlighted specific social, psychological, and biological factors associated with CD development.

Conclusions:

• An accurate, sensitive, and specific predictive model for CD can significantly aid prevention and intervention efforts.
• Key risk factors for CD encompass unpredictable, impulsive, deprived, and emotional contexts.
• Multidomain analysis using machine learning provides a powerful tool for understanding and predicting complex conditions like CD.