Construct Overlap and Threat-Arousal Bridges Across Post-Traumatic Stress, Insomnia, and Anxiety in Ukrainian Adults During War: An Ordinal Network Analysis

Abstract

Background. Post-traumatic stress, insomnia, and anxiety often co-occur during armed conflict, but total scores cannot show which symptoms connect the domains.

Objective. To estimate an item-level symptom network in Ukrainian adults during war, identify cross-domain bridges, and test their robustness to estimation, overlap, and sample choices.

Methods. This secondary cross-sectional analysis used responses from 241 Ukrainian adults recruited online in December 2024. Predefined communities comprised seven Insomnia Severity Index (ISI), seven Generalized Anxiety Disorder-7 (GAD-7), and 20 PTSD Checklist for DSM-5 (PCL-5) items. A rank-based Gaussian-copula matrix was regularized with graphical lasso and the extended Bayesian information criterion (EBIC; γ = 0.50); bridge expected influence quantified cross-community connections. Accuracy was evaluated with 500 fixed-tuning bootstraps, case-dropping, and four sensitivity analyses.

Results. Of 561 possible edges, 223 were nonzero, 106 of them cross-scale. PCL-5 sleep difficulty had the largest bridge expected influence (1.059; 95% bootstrap interval 0.871–1.207), followed by GAD-7 fear that something awful might happen (0.682; 0.488–0.888), GAD-7 irritability (0.603), ISI sleep-onset difficulty (0.591), and PCL-5 irritability/aggression (0.566). The strongest cross-scale edges joined parallel sleep items (0.460) and irritability items (0.473). Rankings changed little under polychoric estimation, lower EBIC γ, and N = 242 (Spearman ρ = 0.993–1.000). Without the two overlapping PCL-5 items, concordance remained substantial (ρ = 0.874), and fear, startle, relaxation, concentration, restlessness, and hypervigilance became prominent.

Conclusion. Bridges reflected shared item content and a broader threat-arousal pattern. Sleep, fear, irritability, startle, relaxation, and concentration should be assessed together; cross-sectional centrality is not a causal treatment target.