Psychopathology profiles and longitudinal correlates of nonsuicidal self-injury in youth: a machine-learning approach.

Abstract

Nonsuicidal self-injury (NSSI) in youth is clinically heterogeneous. We aimed to identify distinct psychopathology-based profiles among children and adolescents reporting NSSI and their longitudinal correlates. Participants (N = 1 345) were drawn from the Brazilian High-Risk Cohort Study, which conducted extensive phenotypic assessments at baseline (ages 6-14 years) and across two follow-up waves (ages 9-18 and 13-23 years). First, we applied unsupervised machine-learning algorithms (Self-Organizing Maps and k-means clustering) to identify distinct psychopathology-based profiles among youth reporting NSSI at the second follow-up. We then employed three models to identify longitudinal predictors of these profiles: logistic regression, elastic net, and random forest. Analyses revealed two distinct profiles of youth reporting NSSI, characterized by high and low psychopathology. The high psychopathology profile (n = 117) was associated with factors identifiable earlier in life and characterized by persistent psychiatric symptoms and significant social adversity throughout development (e.g., family problems and bullying). The low psychopathology profile (n = 127) was marked by lower overall psychopathology and experienced mental health problems only later in development, with less severe challenges over time, such as school suspension and milder depressive symptoms. While the logistic regression did not provide overall significant performance, the elastic net (AUC = 0.72 95% CI 0.65-0.77) and random forest (AUC = 0.73 95% CI 0.67-0.78) did. The present study identified two distinct psychopathology-based profiles among youth reporting NSSI and their longitudinal correlates, using machine learning approaches. Early identification of youth in higher-risk profiles can inform early intervention strategies.