OUTCOMES IN MACHINE LEARNING MODELS FOR CHILD PSYCHIATRY: A SYSTEMATIC REVIEW OF THE LITERATURE.

Abstract

Machine learning (ML) offers powerful tools to address the complexity and data richness of mental health research. By detecting subtle patterns, integrating diverse datasets, and supporting precise decision-making, ML holds promise for enhancing diagnosis, prognosis, and personalized treatment. In child and adolescent psychiatry - characterized by marked clinical heterogeneity and developmental variability - ML may help disentangle complexity and guide clinical care. This systematic review examined studies applying ML to psychiatric disorders in individuals aged 0-18 years. Of 65 identified studies, 33 met inclusion criteria. Most focused on attention-deficit/hyperactivity disorder (ADHD) and autism spectrum disorder (ASD), with others addressing schizophrenia, bipolar disorder, eating disorders, suicidal behaviors, and depression. Overall, the emphasis was on diagnostic applications. Findings were heterogeneous due to variability in algorithms, datasets, and outcome measures, with performance ranging from modest to high. However, small sample sizes, lack of external validation, and overfitting remain major barriers. ML in child and adolescent psychiatry is at an early stage but shows considerable promise, requiring standardized methods, interpretability, and ethical safeguards for clinical translation.