Abstract
Background: Long COVID is a heterogeneous post-viral condition in which persistent neurological, autonomic, cognitive, and psychometric symptoms often occur without clear isolated biomarker abnormalities. This mismatch suggests that disease persistence may be driven not only by changes in individual markers, but by longitudinal reorganization of biological and clinical interactions. Materials and Methods: This observational longitudinal study evaluated patients with persistent symptoms after confirmed SARS-CoV-2 infection at 3 and 6 months. Clinical assessment included neurological examination, Hospital Anxiety and Depression Scale, Beck Depression Inventory, and COMPASS-31. Biomarkers representing hypoxia signaling, oxidative/redox stress, endothelial and renin–angiotensin system activity, glycation-related processes, and complement regulation were analyzed. Correlation analysis, association-level biomarker–clinical network modeling, and complementary Graphical LASSO-based sparse network estimation were used to compare network density, community organization, centrality, and edge rewiring between time points. Results: Conventional paired analysis identified HIF-1α as the only continuous variable with a statistically significant longitudinal change (Wilcoxon statistic = 610.0, p=0.000350), whereas association-level network analysis revealed a broader systems-level signal. The association-level biomarker–clinical network preserved a similar global size at 3 and 6 months, with 16 nodes, 27 versus 26 edges, and densities of 0.225 versus 0.217. However, this apparent stability concealed substantial rewiring: 19 edges were shared, 8 were lost, and 7 emerged. Complementary Graphical LASSO analysis with 1000 bootstrap resamples supported this pattern by identifying a conservative sparse conditional-dependency core, including seven shared conditional-dependency edges across time points and selective weakening of four early conditional dependencies. The C3–C4 relationship reversed from negative to positive correlation (r=−0.618 to r=0.618), indicating marked remodeling of complement-associated regulation. A psychometric–autonomic module involving Beck, HADS I, HADS II, and COMPASS-31 remained stable across both assessments. Conclusions: Long COVID progression was characterized by dynamic remodeling of immune, endothelial/RAS, oxidative-redox, hypoxia-related, autonomic, and psychometric interactions. Longitudinal network analysis identified a systems-level interaction structure that was not captured by isolated biomarker comparisons alone and that was further supported by complementary sparse conditional-dependency analysis.