Depression in adolescents involves complex interactions among depression, anxiety, sleep disturbances, and suicidal symptoms. Network theory offers insights into dynamic symptom relationships during recovery.

Of 797 adolescents initially enrolled, 649 with complete baseline data were included in the network analyses; 458 and 277 participants were retained at the 1-month and 3-month follow-ups, respectively. Cross-sectional Gaussian Graphical Models and Cross-Lagged Panel Network (CLPN) analyses examined relationships among nine symptom domains: depression, somatic/subjective anxiety, sleep quantity/quality, daytime insomnia, passive/active sleepiness, and suicidal ideation/tendency. Network centrality and bootstrap validation assessed parameter stability.

Cross-sectional networks showed structural invariance across timepoints (p>0.05). Subjective anxiety demonstrated highest centrality at T0-T1, while somatic symptoms dominated at T2. Depression maintained high closeness centrality throughout. Although betweenness centrality also suggested a central role for depression, its lower stability (CS < 0.5) necessitates a more cautious interpretation of this specific metric. CLPN revealed more predictive relationships during T0→T1 (76.5% significant edges) than T1→T2 (24.7%). Active sleepiness strongly predicted subsequent somatic anxiety (B=0.683) and depression (B=0.647). Suicide ideation-tendency showed stable bidirectional connections. Network stability was excellent (CS>0.5) except betweenness centrality.

Central symptoms evolved during recovery, with subjective anxiety initially dominant but somatic symptoms becoming central over time. The early post-treatment period showed heightened symptom network activity, with sleep disturbances identified as robust predictors of subsequent affective deterioration. Findings support dynamic, network-informed interventions targeting evolving symptom centrality and predictive pathways, particularly addressing sleep-related symptoms and suicide risk during critical recovery phases.