(PS5-64) Analyses of Protective and Risk Factors Predicting Youth Depression Trajectory Following Early Life Adversity: A Machine Learning Study

Background: Extensive research has shown that adverse childhood experiences (ACEs) pose a significant risk for psychopathologies, including major depressive disorder. However, not everyone who experiences adversity develops these psychopathologies. When investigating the relationship between ACEs and psychopathologies, there are many factors to consider, making it challenging for researchers to identify how each risk and protective factor impacts mental health outcomes. To reduce the barriers that traditional statistical techniques may pose in quantifying and analyzing an extensive number of interrelated factors, we used machine learning techniques to identify both risk and protective factors associated with the trajectory of depression in youth who experienced at least one traumatic event.

Method: A total of 4,153 participants from the Adolescent Brain Cognitive Development study were selected based on experiencing at least one trauma event on the Kiddie Schedule for Affective Disorders and Schizophrenia for DSM-5 (KSADS-5) subsection of traumatic events at baseline. Depression symptoms were measured with the Depression module of the Child Behavior Checklist (CBCL). Risk and protective factors included 39 influential features predicting adolescent depression trajectory from baseline to 3-year follow-up. Depression symptom trajectories were identified using Latent Class Growth Analyses (LCGA). These trajectories were used to train fourteen machine learning models. The most predictive model, defined by area under the curve (AUC), was selected for parameter tuning and applied to a new dataset to assess generalizability. Feature weights were extracted to estimate predictive value in discriminating between depression trajectories.

Results: LCGA analyses revealed four distinct trajectories of depression: 1) Persistently low depression (PLD) (N= 2,354); 2) Rising depression (RD) (N= 800); 3) Mitigating depression (MD) (N= 631); 4) Persistently high depression (PHD) (N= 368). Extra Trees Classifier was identified as the most predictive model and attained an AUC of 71.3%. The top six features that most contributed to the discrimination of these four depression trajectories included: sleep disturbances, prosocial behaviors, parent’s history of mental health problems, the overall sum of experienced trauma events, parent’s history of substance use problems, and neighborhood crime. Post-hoc analyses included one-way analysis of variance (ANOVA) and Tukey tests to identify differences between the four depression trajectory groups on the six features.ANOVA results indicated that all six of the top features revealed significant differences between at least two groups, with the PHD and PLD groups exhibiting the largest differences from each other (all p’s < 0.001).

Conclusions: These results show that each risk and protective factor identified as the most predictive feature of depression trajectories may discriminate between both higher and lower levels of depression. Ultimately, by focusing on youth who encountered trauma, this study delineates the interaction among various domains within a child’s life in the context specific to adversity and how each domain contributed to the course of depressive symptoms.