Module control in youth symptom networks across COVID-19

Abstract

The COVID-19 pandemic exposed young people to a prolonged and evolving societal stressor, yet it remains unclear whether symptom networks were reorganized or whether control was redistributed across a conserved modular scaffold. Here we analysed repeated cross-sectional data on 47 self-reported mental-health symptoms from 14,181 U.S. young adults aged 18-24 years across five COVID-19 phases between 2020 and 2023. For each phase, we estimated Gaussian graphical models, identified symptom communities, and characterized minimum-dominating-set-based module control. Symptom networks showed broadly conserved community organization across phases, indicating a stable mesoscale scaffold despite marked temporal variation. By contrast, intermodule control shifted from an early configuration centered on stress-related symptoms to a later, more distributed pattern spanning emotional, cognitive and social domains. Resampling analyses showed high stability for node strength and moderate stability for module-to-module control, whereas average within-module control was less robust. These findings suggest that prolonged crisis may preserve the modular architecture of youth psychopathology while redistributing control across symptom domains, and they identify intermodule control as a comparatively robust mesoscale feature for cross-phase comparison.

Figures (6)

• Figure 1: Analytical workflow for phase-stratified symptom-network and module-control analysis. (A) For each pandemic phase, Mental Health Quotient (MHQ) item scores (1--9) were used to estimate a Gaussian graphical model with graphical LASSO ($\alpha=0.40$). The precision matrix $\boldsymbol{\Theta}$ was transformed into partial correlations $\boldsymbol{\rho}$, and a weighted undirected network was constructed from $|\rho_{ij}|$, with edges retained whenever the penalized partial correlation was nonzero. (B) Community structure was identified using the weighted Louvain algorithm ($\gamma=1.0$; fixed seed), and replicate partitions were aligned to the phase-specific full-sample partition by Hungarian matching on node--module overlap. (C) On the unweighted support of the estimated graph, all minimum-cardinality dominating sets (MDSs) were enumerated exactly. Node control frequency, $\mathrm{CF}(v)$, was defined as the proportion of MDSs containing node $v$ and normalized to $[0,1]$. (D) Module-level control was summarized by average control frequency (ACF) within each community and by average module control strength (AMCS), where $\mathrm{MCS}_{i\to j}$ denotes the proportion of nodes in community $C_j$ dominated by drivers in community $C_i$ across all MDSs; row-normalized AMCS defines a directed module control network (MCN). (E) For cross-phase interpretation, nodes were mapped a priori to four domains---emotional regulation (EMO), stress response (STR), self-perception and physiological function (SPF), and cognitive and social function (CSF). Communities were named by majority vote (Mixed when purity was $<60\%$), and community-level control matrices were further aggregated into a $4\times4$ domain-level MCN.
• Figure 2: Conserved modular organization across pandemic phases. (A) Phase-specific symptom networks for the Early, First Wave, Second Wave, Omicron and Post-Omicron periods. Nodes are shown according to the detected community structure and aligned to a unified four-domain interpretation comprising stress response (STR), emotional regulation (EMO), cognitive and social function (CSF), and self-perception and physiological function (SPF). (B) Domain-level summaries of classical network topology, shown as phase-averaged standardized values for strength, betweenness, closeness, eigenvector centrality, clustering coefficient and k-core. (C) Node-level standardized profiles of the same network metrics, averaged across phases, illustrating heterogeneous local prominence despite broad conservation of the mesoscale scaffold.
• Figure 3: Redistribution of intermodule control across pandemic phases. (A) Domain-level module control networks (MCNs) for each pandemic phase, derived from exact enumeration of all minimum dominating sets. Circular sectors represent domain-labeled modules; in the First-wave phase, a small SPF-dominant physiological subcluster (PHY; 6 nodes) is shown as a separate module label for visualization, while domain-level aggregation remains in the fixed four-domain (EMO/STR/SPF/CSF) space. Ribbon widths indicate normalized average module control strength (AMCS) from a source domain to a target domain. (B) Domain-level average control frequency (ACF) across phases, summarizing the normalized participation of symptoms from each domain in minimum dominating sets. (C) Domain-level MCN in-degree, quantifying the amount of incoming control received by each domain. (D) Domain-level MCN out-degree, quantifying the amount of outgoing control exerted by each domain. Across phases, the control configuration shifts from a more stress-centered pattern in the early pandemic to a broader cross-domain allocation in later phases.
• Figure 4: Persistent backbone nodes and phase-sensitive boundary mixing. (A) Comparison of the Early and Post-Omicron symptom networks, with nodes grouped by domain and shaded according to normalized control frequency. The later phase shows greater boundary mixing while retaining a subset of consistently influential high-control symptoms. (B) Mean control frequency (CF) plotted against the coefficient of variation (CV) across the five pandemic phases for each symptom. Median-based reference lines and tolerance bands are used to distinguish backbone nodes, defined by high average control and low variability, from liaison nodes, defined by high average control and high variability. (C) Sankey summary of cross-phase control reallocation across domains. Flows indicate how high-control symptoms are redistributed over time without evidence of wholesale restructuring.
• Figure 5: Robustness of network and control metrics. (A) Equal-size bootstrap distributions of the number of edges in the Early and Post-Omicron networks. Dashed red lines indicate the corresponding full-sample estimates. (B) Equal-size bootstrap distributions of the number of minimum dominating sets in the Early and Post-Omicron networks, with dashed red lines marking the full-sample values. (C) Case-dropping stability curves for node strength, average module control strength (AMCS), and average control frequency (ACF) in the Early and Post-Omicron phases, showing replicate-to-baseline correlations as the retained sample proportion decreases. (D) Phase-wise distributions of minimum dominating-set size across resamples. (E) Phase-wise distributions of the number of detected modules across resamples.