Predicting Mental Health Trajectories After Potentially Traumatic Events: A Machine Learning Approach

Objective: This study aimed to investigate the trajectories of internalizing and externalizing problems following childhood potentially traumatic events (PTEs) and analyse a comprehensive set of baseline variables (PTEs, individual, environmental) to elucidate their predictive role as contributors to different mental health trajectories. Method: The sample consisted of 4141 participants ( M = 9.48, SD = 0.51 years at baseline; 48.7% girls; 72.1% White) from the Adolescent Brain Cognitive Development study who had experienced at least one PTE. Participants’ mental health problems were assessed using the Brief Problem Monitor self-report form. Latent Growth Mixture Modelling was used to identify trajectories of youth´s internalizing and externalizing problems across the six assessments. Extreme Gradient Boosting, a machine learning approach, was utilized to investigate 37 predictors of different trajectories. Results: Three distinct trajectories were identified: “Resilient”, “Mild stable” and “Moderate chronic increasing”, for internalizing and “Resilient”, “Mild increasing” and “Moderate chronic decreasing” for externalizing problems. Predictors of the “Moderate chronic” versus “Resilient” trajectories were identified using machine learning. The three most important predictors of the internalizing problems trajectory were: behavioural inhibition, female gender, and less parental monitoring, whereas predictors of the externalizing problems trajectory were family conflicts, screentime and behavioural inhibition. Conclusion: The findings can help characterize individual variation in mental health trajectories following childhood PTEs and provide potential targets for intervention to foster mental health.