Network analysis of depressive symptoms in Hong Kong residents during the COVID-19 pandemic

The study addresses how depressive symptoms interrelate within a network framework during the COVID-19 pandemic in Hong Kong. Traditional psychopathology posits a common-cause model where an underlying disorder produces symptoms, often operationalized by summing scale items, implying interchangeability. Emerging network theory instead conceptualizes depression as arising from direct symptom-to-symptom interactions, where central symptoms can activate and maintain the syndrome. Given that depression’s presentation is shaped by sociocultural context and that most network studies were conducted in Western settings, there was a need to examine depressive symptom networks in Hong Kong, particularly amid COVID-19. The objective was to characterize the PHQ-9 depressive symptom network in a large community adult sample and to investigate central symptoms and gender differences in network structure.

Prior network analyses have mapped depressive symptom interrelations in Western countries, showing heterogeneity in symptom centrality and highlighting that central symptoms may be efficient intervention targets. Evidence indicates sociocultural and socioeconomic contexts shape depression’s clinical features, underscoring the need for country-specific network examinations. Network approaches can also identify bridges between syndromes. Few studies had examined depression networks during COVID-19, despite elevated comorbidity risk. Related work has identified symptoms like sad mood, guilt/self-dislike, and low energy as central in various populations, with some studies finding limited gender differences in global network strength.

Design and setting: Large-scale cross-sectional online survey conducted in Hong Kong from 24 March to 20 April 2020 using snowball convenience sampling via WhatsApp, WeChat, and Facebook. Eligibility: Hong Kong residents, living in Hong Kong during COVID-19, aged 18–59, able to read Chinese. Ethics approval: Human Subjects Ethics Sub-committee of the Hong Kong Polytechnic University (HSEARS20200227002-01). Electronic informed consent obtained; anonymity and helpline information provided. Participants: N=11,072 (men n=2,105; women n=8,815). Measures: Depressive symptoms assessed with the validated Chinese PHQ-9 (items: Anhedonia, Sad Mood, Sleep, Energy, Appetite, Guilt, Concentration, Motor, Suicide) rated 0–3 past two weeks; higher scores indicate more severe symptoms. Item preprocessing: Due to debates on modeling ordinal items in networks, items were dichotomized (0=absent; 1–3=present). Network estimation: Binary symptom network estimated using the Ising model (R package IsingFit 0.3.1), combining logistic regression with model selection via EBIC, with ELASSO regularization to yield a sparse, interpretable network. Visualization showed edge thickness proportional to association strength and color indicating direction. Centrality metrics: Strength, Betweenness, and Closeness computed (R packages IsingFit, networktools, qgraph; R 3.6.3). Accuracy and stability: Edge-weight accuracy assessed via nonparametric bootstrap 95% CIs; centrality stability assessed with case-dropping subset bootstrap to compute CS-coefficients; bootstrapped difference tests used to compare edges and node centralities (R package bootnet). Symptom severity/variability correlations: Spearman correlations between node strength and mean PHQ-9 item scores and between node strength and item SDs. Gender comparisons: Network Comparison Test (NCT; 1,000 permutations) compared male vs female networks in global strength, edge weight distributions, and individual edges with Holm-Bonferroni adjustment (R package NetworkComparisonTest 2.0.1). Covariate-adjusted network: Re-estimated controlling for age, education level, and marital status; compared edge magnitudes and node strength to the unadjusted network.

• Sample characteristics: N=11,072; men 19.0% (n=2,105), women 79.6% (n=8,815). Mean age 39.07 (SD 8.83). PHQ-9 total mean 9.60 (SD 6.04). After dichotomization for Ising model, PHQ-9 total mean 0.66 (SD 0.22). - Symptom prevalence/means (dichotomized presence): Highest for Energy (87.2%) and Anhedonia (86.3%); lowest for Suicide thoughts (22.5%) and Motor problems (47.2%). - Network structure: Central nodes with highest centrality (strength, betweenness, closeness) were Guilt (PHQ-9 item 6), Sad Mood (item 2), and Energy (item 4). Lower centrality observed for Concentration, Suicide, and Sleep. Strong positive associations included Anhedonia–Sad Mood, Guilt–Suicide, Concentration–Motor, Energy–Appetite, Sad Mood–Guilt, and Sleep–Energy. - Accuracy and stability: Bootstrapped edge weights were consistent with the original network, with larger-weight connections showing particularly stable estimates; centrality indices demonstrated acceptable stability. - Symptom severity/variability vs centrality: Node strength was not correlated with mean item severity (r=0.15) nor with item variability (SD; r=0.21), indicating centrality was not an artifact of symptom severity or variability. - Gender network comparison: No significant differences between male and female networks in global strength (males 3.81 vs females 3.77; difference=0.03, p=0.259), edge weight distribution (M=0.06, p=0.73), or individual edges (all p>0.05 after Holm-Bonferroni). However, mean symptom levels differed: females reported higher Sleep problems and Appetite (P<0.01), while males reported higher Sad Mood, Guilt, Motor, and Suicide thoughts (P<0.01). Strength was not related to symptom mean levels or variability within genders (female r=0.12 and 0.22; male r=0.25 and 0.21, respectively). - Covariate-adjusted network: Adjusting for age, education, and marital status yielded an almost identical network to the original (edge magnitude r=0.995, 95% CI [0.991, 0.997], P<0.01; strength r=0.984, 95% CI [0.931, 0.997], P<0.01).

The study identified Guilt, Sad Mood, and Energy as central symptoms in Hong Kong residents’ depression network during COVID-19, suggesting these nodes may activate or maintain broader depressive syndromes and thus represent efficient intervention targets. The centrality of Guilt aligns with literature linking guilt-related constructs (self-blame, worthlessness, helplessness) to depressive states. Energy’s centrality is consistent with clinical trial findings highlighting energy improvements as key outcomes. Sad Mood’s centrality is congruent with its status as a hallmark depressive symptom and with previous network studies in general and clinical populations. Despite known gender differences in depression prevalence and certain symptom profiles in epidemiological literature, network structures did not differ significantly between men and women in this sample, paralleling prior network analyses. Potential explanations include methodological differences between total-score approaches and symptom-network modeling, the disproportionate gender ratio of the sample, and pandemic-related stressors potentially affecting both genders similarly. The findings are contextualized by Hong Kong’s co-occurring large-scale stressors (social unrest and COVID-19), which have been associated with increased mental distress and may influence symptom interrelations. Overall, mapping central symptoms advances understanding of depression’s structure in a specific sociocultural and pandemic context.

This first network analysis of depressive symptoms in a large community sample in Hong Kong during COVID-19 found Guilt, Sad Mood, and Energy to be central nodes. These symptoms may serve as primary targets for psychosocial and biological interventions, potentially yielding broader symptom improvements. The depressive symptom network was robust across gender and to adjustment for demographic covariates. Future research should test targeted interventions on central symptoms, explore longitudinal network dynamics, and examine bridge symptoms with comorbid syndromes to inform prevention and treatment, including psychological and neurobiological investigations.

Findings on gender differences should be considered preliminary due to the disproportionate gender ratio (male n=2,105 vs female n=8,815).