Depression and Daytime Dysfunction Centralize the Fatigue–Sleep Cascade in Island Firefighters: A Symptom Network and Bayesian DAG Study

Yudan Liu^1,2Zhihong Li^1,2Qiong Xiang^1,2Xue Zhang²Runhua Bai²Chenjing Sun^2*Jianguo Liu^2*

• ¹School of Medicine, South China University of Technology, Guangzhou, China
• ²Sixth Medical Center of PLA General Hospital, Beijing, China

Background: Sleep disturbances, fatigue, and psychological distress are prevalent among island-based firefighters, a high-risk occupational group. However, the interactions and mechanisms underlying these factors remain unclear. This study investigated relationships among fatigue, sleep disturbances, psychological distress, and psychological resilience using symptom network analysis and exploratory Bayesian Directed Acyclic Graph (DAG) modeling. Methods: We surveyed 570 male island-based firefighters (cross-sectional) using the PSQI, FSS, SCL-90, and CD-RISC. Variables were residualized for covariates and z-standardized. An EBICglasso Gaussian Graphical Model (γ = 0.50) quantified centrality and predictability. Robustness was tested via sensitivity (γ = 0.25–0.75), bootstrapping, and Network Comparison Tests (sleep status/work type). Exploratory Bayesian DAG modeling in SD used Tabu/Hill-Climbing (BIC scoring, bootstrapped aggregation) for a CPDAG. Results: Sleep disturbance prevalence was 46.0% (262/570). In the full network, depression (S4) and daytime dysfunction (P7) were among the most central nodes (EI = 1.938 and 1.613), and the fatigue total (F0) showed the highest predictability (R² = 0.176). In SD, hostility (S6, EI = 1.913) and anxiety (S5, EI = 1.462) emerged as potential affective hubs; tenacity (C1) was positioned upstream (Strength = 1.961; EI = −1.315) in relation to sleep and depression. Compared with SN, SD showed lower density and global strength (both P < 0.01). Between SW and NS, overall network structure differed (P = 0.014) whereas global strength did not (P = 0.694). Sensitivity analyses indicated high agreement of non-zero edges and minimal fluctuations in density/global strength across γ = 0.25– 0.75. The DAG/CPDAG suggested a potential path from subjective sleep quality → fatigue → depression → hostility → somatization, with C1 potentially influencing sleep and depression; directionality warrants further longitudinal validation. Conclusion: Depression (S4) and daytime dysfunction (P7) may serve as key nodes linking sleep and affective processes; fatigue may relate to psychological distress via sleep; and tenacity (C1) could play an upstream protective role. Sleep status and shift work may reorganize network structure without necessarily altering global connectivity. Targeted interventions may consider subjective sleep perception and psychological resilience in island-based firefighters.