Key factors and serial mediation analysis of anxiety and depression among university students: a cross-sectional study

Background: Anxiety and depressive symptoms are prevalent among university students; however, their risk factors are unlikely to operate in isolation. Instead, they may be organized in a multilevel structure that progressively approaches emotional outcomes, spanning from distal adversity exposure and the current family relational context, through resource pathways, to proximal functional status. This study aimed to identify the key risk factors for anxiety/depressive states among university students and to examine whether a sequential statistical association structure exists among childhood adversity exposure, family interaction, social support, and sleep status. Methods: A cross-sectional online survey was conducted in seven universities in Harbin (Sep-Nov 2025; N = 9, 796). Depressive and anxiety states were assessed by PHQ-9 (≥5) and GAD-7 (≥5). Family cohesion and adaptability (FACES II-CV), the Parent-Adolescent Communication Scale (PACS) and the Social Support Rating Scale (SSRS) were administered, and childhood harmful exposure, family conflict, recent negative life events, lifestyle, and sleep status were measured. Hierarchical logistic regression and LightGBM, MLP, SVM models were used for analysis, with SHAP applied to identify key correlates, and serial mediation models were used to examine the sequential statistical associations among variables. Results: The prevalence of depressive and anxiety states was 19.7% and 19.5%. For both outcomes, sleep status showed the strongest and most stable association (depression OR = 1.431; anxiety OR = 1.335). childhood harmful exposure, problem communication, family conflict, and recent major negative life events were consistent risk correlates for both outcomes, whereas social support was a stable protective factor. Machine learning results showed that LightGBM achieved the best overall performance. The overall risk profiles were similar across the two outcomes; however, childhood harmful exposure contributed more to depressive states, whereas problematic communication and family conflict were more prominent in anxiety states. Serial mediation analysis further showed that a sequential statistical association structure, composed of negative family interactions, social support, and sleep status, existed between childhood harmful exposure and anxiety/depressive states, with sleep problems showing the largest indirect effect. Conclusion: Anxiety and depressive states in university students exhibited a sequential statistical association structure that progressively approached emotional outcomes, extending from early adverse experiences and current negative family interactions, through insufficient social support, to proximal sleep impairment. Sleep problems were a shared key proximal correlate of both outcomes. In addition, although anxiety and depression shared highly similar risk profiles, anxiety was more closely related to current relational tension and real-life stress, whereas depression was more closely related to long-term cumulative adverse experiences. Prioritizing assessment and stratified intervention around the three key domains of sleep, negative family interactions, and social support may help improve the specificity and efficiency of mental health services in universities.