Prediction of mental health risk in adolescents

Abstract

Prospective prediction of mental health risk in adolescence can facilitate early preventive interventions. Here, using psychosocial questionnaires and neuroimaging measures from over 11,000 children in the Adolescent Brain and Cognitive Development Study, we trained neural network models to stratify general psychopathology risk. The model trained on current symptoms accurately predicted which participants would convert into the highest psychiatric illness risk group in the following year (area under the receiver operating characteristic curve = 0.84). The model trained solely on potential etiologies or disease mechanisms achieved an area under the receiver operating characteristic curve of 0.75 without relying on the child's current symptom burden. Sleep disturbances emerged as the most influential predictor of high-risk status, surpassing adverse childhood experiences and family mental health history. Including neuroimaging measures did not enhance predictive performance. These findings suggest that artificial intelligence models trained on readily available psychosocial questionnaires can effectively predict future psychiatric risk while highlighting potential targets for intervention. This is a promising step toward artificial intelligence-based mental health screening for clinical decision support systems.

Figure 1.. Model Performance Stratified by Metric and High-Risk Groups.

Receiver Operating Characteristic (ROC) and Precision-Recall (PR) curves comparing predictive model performance for mental health risk assessment for the test set (n = 1,142 participants). Each panel illustrates the performance of symptom-driven (CBCL scales) and mechanism-driven (Questionnaire) models: (a) Conversion ROC curve: prediction of participants transitioning from no-, low-, or medium-risk groups to the high-risk group. (b) Persistence ROC curve: prediction of participants that remain in the high-risk group. (c) Agnostic ROC curve: prediction of participants entering the high-risk group from any prior group. (d) Conversion PR curve: positive predictive value for participants converting to the high-risk group. (e) Persistence PR curve: positive predictive value for participants persisting in the high-risk group. (f) Agnostic PR curve: positive predictive value for predicting high-risk status regardless of prior group. The curves highlight that while symptom-driven models (CBCL scales) outperform mechanism-driven models (Questionnaire), the sensitivity and precision of mechanism-driven models are strong.

Figure 2.. Predictor Category Shapley Additive Explanations (SHAP) Analysis.

SHAP values represent the influence of each predictor category on the model’s predicted probability of a participant being classified as high-risk. Predictor categories are ranked by their absolute SHAP values sums, with higher values indicating a larger impact on model predictions. Absolute SHAP value sums are displayed as mean ± 95% confidence interval (n = 1,142 participants). The Sleep Disturbances category emerged as the most influential predictor, followed by Prosocial Behaviors, Adverse Childhood Experiences (ACEs), Family Mental Health History, and Family Conflict. Each predictor category is further stratified by respondent type (youth vs. parent), with parent-reported data generally showing more substantial influence on model predictions than youth-reported data. The questionnaire predictor responses can be found in Supplementary Table 4.