Psychometric validation and predictive efficacy of a comprehensive depression risk model for undergraduates

Background Undergraduate depression is prevalent, yet traditional screening is unidimensional and inefficient. We developed a biopsychosocial risk classification model for the cross-sectional identification of current depressive symptoms. Methods A cross-sectional study enrolled 898 undergraduates from a medical university in Western China (March–June 2024). The participants were randomized to training ( n = 719, 80%) and test ( n  = 179, 20%) sets. Assessments included demographics, somatic symptoms (Somatic Symptom Scale), insomnia severity (Insomnia Severity Index, ISI), and depressive symptoms (Patient Health Questionnaire-9, PHQ-9). To avoid circularity, all PHQ-9 items were excluded from predictors; the total score defined the outcome (≥5). Independent risk factors were identified via univariate and multivariate logistic regression. Three models—random forest, XGBoost, and logistic regression—were developed. Performance was evaluated using discriminative metrics (AUC, accuracy, sensitivity, specificity, PPV, NPV, and F1 score), calibration plots, and decision curve analysis. Internal validation utilized fivefold cross-validation and bootstrap resampling (1,000 iterations). Subgroup analyses stratified the results by gender, grade, and somatic symptom severity. Results The point prevalence of depressive symptoms (PHQ-9 ≥5) was 45.21% (406/898), which was significantly higher in women (OR = 2.98). Multivariate analysis identified severe somatic symptoms (OR = 37.94), moderate somatic symptoms, and social isolation as key independent risk factors. Excluding PHQ-9 items to avoid circularity, the random forest model achieved an AUC of 0.872 (95% CI: 0.841–0.903), outperforming scale-only (ΔAUC = 0.110, p  p  = 0.042). Feature importance consistently highlighted somatic distress, insomnia severity, and lack of close friends over emotional items. Calibration was excellent (Hosmer–Lemeshow p > 0.05), and decision curve analysis supported net clinical benefit (thresholds 0.2–0.9). Conclusion A comprehensive model combining physiological, psychological, and social factors yields excellent cross-sectional discriminative capability and stability for identifying undergraduates currently at risk for depressive symptoms. The proposed three-step clinical pathway (universal screening, targeted re-evaluation, and precision intervention) can facilitate large-scale, early identification in university settings. Due to the cross-sectional design, the term “prediction” is not used in a temporal or causal sense; rather, the model estimates the probability of concurrent depressive symptoms.