A School–Family Framework for Initial Identification of Suicide-Related Vulnerability in Children and Adolescents: Development and Geographic External Validation

Abstract Background Suicide-related thoughts and behaviors (STBs) are important indicators of suicide risk among children and adolescents and represent a major public health concern. Although validated suicide-specific instruments are essential for formal risk assessment, large-scale implementation may be challenging in school, community, and primary care settings where specialized mental health resources are limited. This study aimed to develop and externally validate a school–family-based framework using routinely available, non-sensitive information to support the initial identification of suicide-related vulnerability. Methods This multicenter school-based cross-sectional study included 29,662 children and adolescents from Chongqing, Shaanxi, and Yunnan provinces in China. STBs were operationalized as a broad composite indicator encompassing self-harm behaviors, suicidal ideation, or suicidal behaviors. The Chongqing cohort was used for model development and internal validation, while the Shaanxi and Yunnan cohorts served as independent external validation datasets. Predictors were selected using least absolute shrinkage and selection operator regression. Logistic regression, random forest, extreme gradient boosting, and light gradient boosting machine models were evaluated for discrimination, calibration, and potential utility. Shapley additive explanations were used for model interpretation. Results Twenty-two predictors were retained. Models demonstrated stable performance across validation cohorts, with AUROC values ranging from 0.783 to 0.817 in internal validation, 0.795 to 0.849 in Shaanxi external validation, and 0.785 to 0.840 in Yunnan external validation. Logistic regression showed comparable discrimination with favorable calibration and interpretability and was selected as the final model. Major contributors included child anxiety, family functioning, sleep characteristics, emotional and behavioral difficulties, and developmental factors. Conclusions Routinely available, non-sensitive school–family information supported the identification of suicide-related vulnerability across two geographically independent external populations. The proposed framework may provide a low-barrier approach to initial risk stratification in settings with limited specialized mental health resources, helping prioritize potentially vulnerable children for subsequent suicide-specific assessment and professional evaluation. Prospective studies are needed to evaluate its real-world implementation and impact.