Feed

Internalizing disorders are among the most common psychiatric conditions in adolescence, often associated with long-term adverse outcomes. Early identification of at-risk youth is important for effective intervention, though it remains challenging due to the multifactorial nature of risk. Machine learning (ML) offers opportunities to integrate multiple data sources and improve risk prediction for internalizing disorders.

We used data from 13,743 adolescents (mean age 14.45 years; 52.7% female) participating in the Norwegian Mother, Father and Child Cohort Study (MoBa), linked to national health registries. Logistic regression with elastic net regularization was applied to predict the risk of an internalizing disorder (mood, anxiety or stress-related) occurring within one to five years after assessment. Nested models of increasing complexity incorporated sociodemographic, clinical, lifestyle, mental health, psychosocial, and genetic predictors. Model performance was evaluated in a hold-out test set. Simplified models combining three questionnaire scales were also evaluated.

Test-set performance increased with model complexity, reaching area under the receiver operating characteristic curve (AUC) of 0.731 for the full model. Mental health self-reported symptoms and psychosocial predictors contributed most to the discrimination. Simplified models using three questionnaire scales, alongside age and sex, achieved AUCs up to 0.718 and effectively stratified adolescents into high- and low-risk groups (OR80/20 ranged 6.11-9.35).

Multimodal ML models integrating registry information, mental health symptoms, psychosocial factors, and genetic data demonstrated moderate predictive performance. Simplified models with three questionnaire items reached comparable performance, highlighting their potential utility in the early identification of adolescents at elevated internalizing disorder risk.