Family atmosphere is associated with symptom levels and change but not with symptom network structure in a cohort of Chinese medical students: a 3.5-year longitudinal study

Background Population-level studies suggest that the network organization of psychological symptoms among university students is stable over time, but repeated cross-sectional designs cannot show whether the network of a cohort followed through training stays the same, or whether environmental factors relate to network structure or only to overall symptom level. Methods We followed 635 undergraduates at one traditional Chinese medicine university from enrolment (September 2022) to the eve of clinical internship (April–May 2026; mean interval 3.5 years), administering the Symptom Checklist-90 (SCL-90) at both waves. Cohort-level regularized partial correlation networks of the nine SCL-90 dimensions (one per wave, between-person data) were compared with the Network Comparison Test; accuracy and stability were assessed by bootstraps. Family atmosphere, a single non-validated follow-up item, was examined in relation to symptom levels, change and network structure (harmonious, n = 343, vs. average, disharmonious or poor, n = 292). Results No difference in network structure was detected between waves under the primary response-time criterion (global strength p = 0.337; structure p = 0.121; no edge or centrality difference after false discovery rate correction), but the structure test was significant under three of four alternative thresholds, so invariance is not established. Depression, anxiety and interpersonal sensitivity were most central at both waves (CS-coefficient = 0.89). Eight of nine dimension scores declined, whereas depression increased. A less harmonious family atmosphere was associated with higher scores on all nine dimensions at follow-up (all q ≤ 0.003; d = 0.24–0.44) and with less improvement: the harmonious group improved (Δ GSI = −0.10), the others did not (+0.01; time × family atmosphere F (1, 633) = 8.67, p = 0.003), and persisted after adjustment for baseline symptoms, sex, left-behind experience, economic status, family structure and place of origin (beta = 0.35, p < 0.001). Networks did not differ between family atmosphere groups. Conclusion At the cohort level, no structural difference was detected under the primary criterion, though threshold-sensitive. Family atmosphere, measured once by a single item, was associated with symptom severity and observed change; direction of association cannot be determined, and the single-site, predominantly female sample limits generalizability.