Predictive Analytics for Preventing Crisis-Driven Mental Health Utilization in Community and School-Based Settings

Abstract

This study examined whether predictive analytics can support the early identification and prevention of crisis-driven mental health service utilization among adolescents and young adults in community and school-based settings, using publicly available school- and community-based youth survey data. A quantitative predictive analytics design was applied, employing LASSO-penalized logistic regression to identify independent predictors, estimate individual risk probabilities, and evaluate model performance using holdout validation, with assessment of discrimination, calibration, and risk stratification. Persistent sadness emerged as the strongest independent predictor of crisis-driven utilization (adjusted odds ratio = 3.21), alongside bullying victimization, substance use, chronic absenteeism, and low parental monitoring. The predictive model demonstrated acceptable discrimination (AUC ≈ 0.79) and strong calibration, with individuals in the highest predicted risk tertile experiencing nearly tenfold higher crisis-driven mental health service utilization compared with those at low risk. These findings indicate that crisis-driven mental health service utilization follows identifiable risk patterns and support the integration of calibrated risk stratification into school and community systems to guide targeted preventive outreach, strengthen cross-sector data coordination, and reduce reliance on emergency-based mental health care.