Developing algorithms to predict adult onset internalizing disorders: An ensemble learning approach.

Area of Science:

• Psychiatry and Mental Health
• Computational Neuroscience
• Epidemiology

Background:

• Machine learning (ML) is increasingly used for psychopathology risk prediction.
• Development of ML algorithms for internalizing disorder onset has been limited.
• Internalizing disorders (e.g., anxiety, depression) represent a significant public health burden.

Purpose of the Study:

• To develop and validate ensemble ML algorithms for predicting adult onset internalizing disorders.
• To identify individuals at high risk for developing conditions such as generalized anxiety, panic, social phobia, depression, and mania.

Main Methods:

• Utilized prospective survey data from the National Epidemiological Survey on Alcohol and Related Conditions (NESARC) Waves 1-2 (n=34,653).
• Operationalized 213 risk factors (features) from Wave 1 data.
• Employed super learning, an ensemble method, to combine multiple classifiers (e.g., random forests, k-nearest neighbors) for predicting incident internalizing disorders.

Main Results:

• Achieved cross-validated area under the curve (AUC) values ranging from 0.76 (depression) to 0.83 (mania), outperforming individual algorithms.
• The top 10% of individuals predicted to be at highest risk accounted for 37.97% (depression) to 53.39% (social anxiety) of all incident cases.
• Demonstrated acceptable-to-excellent prediction accuracy with high case concentration in high-risk groups.

Conclusions:

• Ensemble ML algorithms can effectively predict adult onset internalizing disorders.
• These algorithms show promise for identifying individuals who may benefit from early preventive interventions.
• Further validation and dissemination of these predictive algorithms are recommended to support mental healthcare.