Introduction

Depressive and anxiety symptoms (depression and anxiety hereafter) are common mental health problems, which are increasing globally in the past decade [1]. The presence of either depression or anxiety often increases the risk of having the other. For instance, a meta-analysis revealed that depression and anxiety are bidirectional risk factors for one another [2]. In addition, depression and anxiety often occur concurrently, such as in a study on UK college students, 29.8% of females and 13.9% of males screened positive for both anxiety and depression [3].

The coronavirus disease 2019 (COVID-19) outbreak that started in early 2020 have resulted in an increase in common depression and anxiety across many populations [4,5,6]. After the COVID-19 outbreak was largely controlled in some countries such as China, studies found that the large scale public health measures (e.g., quarantine, self-isolation, and business and school closures) resulted in long-term stress and psychological distress in many populations [19].

Researchers have explored characteristics of the anxiety and depression network in various populations. For example, “fatigue” was identified as the central and bridge symptom in migrant Filipino domestic workers, which may increase the risk of comorbidity between anxiety and depression [33]. The CS-C means the maximum cases that could be dropped from the sample, in which the centrality indices from the subsamples are correlated with the indices from the original sample at a value of r = 0.7 [33]. Generally, the value of CS-C needs to be above 0.25 and is preferably above 0.5 [33]. A nonparametric bootstrap procedure was used to assess the edge weights stability based on the 95% confidence intervals (95% CIs). Edge accuracy was assessed by 95% CIs, with a narrower CI indicating a more trustworthy network [33, 39]. Additionally, to evaluate the differences between two edges or between two nodes strength, bootstrapped tests were conducted based on 95% CIs, which indicated that there were statistical differences between two edges or two nodes strength if zero was not included in the CIs [33]. All analyses in network stability were performed by the R package bootnet (Version 1.4.3) [33].

Network comparison

The Network Comparison Test (NCT) in the R-package NetworkComparisonTest (Version 2.2.1) was used to examine the three invariance measures (i.e., network structure invariance, edge invariance, and global strength) [40]. Network structure means the maximum difference of pairwise edges between two networks, edge invariance indicates the difference of individual edge weight between two networks, and global strength refers to the sum of all edges of each network. Holm-Bonferroni correction for multiple comparisons at the level of individual edge between two networks was adopted. Considering the moderating effect of gender [41], academic major [42] and living area [42, 43] on anxiety and depression among college students, network structure invariance, edge invariance, and global strength were compared between different subgroups (e.g., between females and males, between health-related major and others, and between rural and urban residents) based on a permutation test (n = 1000) [40].

Results

Descriptive statistics

Out of the 3075 college students invited to participate, 3,062 agreed and completed the assessment, giving a response rate of 99.58%. Of the 3,062 college students included in this network, the mean age was 19.8 (standard deviation (SD) = 2.0) years, 2,068 (67.5%) were females, 1563 (51.0%) were rural residents, and 1722 (56.2%) majored in health-related subjects (Table S1). The mean PHQ-9 and GAD-7 rating score was 0.21 and 0.80, respectively (Table 1), and the distributions of the responses to PHQ-9/GAD-7 items are shown in Table S2.

Network structure

The network of anxiety and depressive symptoms is shown in Fig. 1 and the corresponding partial correlation matric is presented in Table S3. The edge Nervousness-Uncontrollable worry (GAD1-GAD2) shows the strongest association, followed by the edge Uncontrollable worry-Excessive worry (GAD2-GAD3), Excessive worry-Trouble relaxing (GAD3-GAD4), Restless-Feeling afraid (GAD5-GAD7), Sleep-Fatigue (PHQ3-PHQ4), Motor-Suicide (PHQ8-PHQ9), Anhedonia-Sad Mood (PHQ1-PHQ2), and Concentration-Motor (PHQ7-PHQ8).

Fig. 1: Network structure of anxiety and depressive symptoms in college students.
figure 1

The left panel shows the visualization of the network structure; the right panel shows the value of strength in order.

Full size image

In Table 1 and Fig. 1, Fatigue (PHQ4) has the highest node strength in the anxiety and depression network among college students, followed by Excessive worry (GAD3), Trouble relaxing (GAD4), and Uncontrollable worry (GAD2). The item Excessive worry (GAD3) had the highest predictability in the network (Table 1) and an average of 56.3% of variance could be potentially accounted for by each node’s surrounding nodes (Mpredictability = 0.563 ± 0.091). In terms of bridge symptoms, Motor (PHQ8) showed the highest bridge strength, followed by Feeling afraid (GAD7) and Restlessness (GAD5) (Fig. 2).

Fig. 2: Network structure of anxiety and depressive symptoms showing bridge symptoms in college students.
figure 2

The left panel shows the visualization of the network structure of bridging symptoms; the right panel shows the value of bridge strength in order.

Full size image

Network stability

In Fig. 3, the case-drop** bootstrap procedure shows that both CS-Cs of node strength and bridge strength were 0.75, which indicates that 75% of samples could be dropped, but the findings were still similar to the primary results (r = 0.7). The results of nonparametric bootstrap procedure show that most comparisons among edge weights and node strength were statistically significant (Figs. S1, S2). Additionally, bootstrapped 95% CIs were narrow, representing edges were trustworthy (Fig. S3).

Fig. 3: The stability of strength and bridge strength using case-drop** bootstrap.
figure 3

The x-axis indicates the percentage of cases of the original sample included at each step. The y-axis indicates the average of correlations between the centrality indices from the original network and the centrality indices from the networks that were re-estimated after excluding increasing percentages of cases.

Full size image

Network comparisons

As shown in Fig. S4, there was significant difference in network global strength (Urban: 7.655 vs Rural: 7.469, S = 0.186, p = 0.044) between rural and urban college students. In other two subsample comparisons, no significant differences were found in network global strength (Health-related major: 7.456 vs Other majors: 7.431, S = 0.025, p = 0.703; Females: 7.461 vs Males: 7.496, S = 0.035, p = 0.613). In terms of network structure and individual edge weight comparisons, there were also no significant differences between two networks in the three subsample comparisons.

Discussion

To the best of our knowledge, this was the first study that characterized the depressive and anxiety network in Chinese college students during the late stage of the COVID-19 outbreak. All the strongest edges were within the respective disorder, while none of the strongest edges linked anxiety and depressive symptoms, which are consistent with previous findings identified in network analysis of depression and anxiety [15, 15]. In this study, we found that certain anxiety symptoms, including “Excessive worry” (GAD3), “Trouble relaxing” (GAD4), and “Uncontrollable worry” (GAD2), also had high values of node strength, indicating these symptoms may also play important role in activating and maintaining the depression and anxiety network. This could be partly explained by the fear of contagion when students are faced with this novel and potentially fatal infectious disease, which can increase such anxiety symptoms [44]. Specific interventions could be adopted, such as cognitive behavioral therapy (CBT), applied relaxation and medications, the latter being considered for those with severe symptoms.

In this depression and anxiety network, the most influential bridge symptom was the depressive symptom of “Motor” (PHQ8), which is similar to that in a previous study in Chinese adults, where “Motor” (PHQ8) showed a high bridge centrality both during the COVID-19 peak and post-peak outbreak period [49]. In another study, the symptom of “Motor” was identified as the crucial priority due to its relation to “thought of death” in female nursing students [15], suggesting that this symptom should be a target of interventions to reduce depression and anxiety. Other influential bridge symptoms included the anxiety symptoms of “Feeling afraid” (GAD7) and “Restlessness” (GAD5), suggesting that these symptoms should also be targeted in treatment.

The predictability of each node in the network of depression and anxiety was calculated. There were no associations between predictability and mean values of each node (rs = −0.056, p = 0.837), suggesting that certain symptoms might have a high value of predictability in the depressive and anxiety network, although these symptoms appeared less frequently [39]. On average, 56.3% of the node variance could be explained by neighboring nodes, implying that the potential sources of the remaining variance (e.g., stress and insomnia symptoms) were not included by both the PHQ-9 and GAD-7. Previous studies found that certain factors, such as gender, living area (urban/rural), and study major, were associated with depression and anxiety at the disorder level [41,42,43]. In this study, network comparison test found that compared to those from rural areas, students from urban areas had a significantly higher global strength of the network, indicating that individual symptoms in the model of urban college students were strongly inter-connected. This finding was not found in the relevant studies using network analysis and should be explored in future studies. In other comparisons (such as health-related major vs. other majors, and female vs male), no significant differences were found.

The strength of this study included the large sample size and use of the network approach to visualize depressive and anxiety symptom patterns in college students, with stable results. However, several limitations should be noted. First, the cross-sectional data collected by snowball sampling method were used to construct depressive and anxiety symptoms network structure, which could not identify the causality between individual symptoms and had limited representativeness. Therefore, the findings should be confirmed in future longitudinal studies. Second, self-reported measures were used to assess depressive and anxiety symptoms, which may have recall bias and are limited to capture clinical phenomena [15]. Third, for logistical reasons, depressive and anxiety network prior to and in the early stage of the COVID-19 pandemic were not assessed. Hence, the psychological impact of the pandemic could not be evaluated. Finally, some relevant symptoms, such as post-traumatic stress and certain somatic symptoms, were not measured, which could partly explain the relatively low predictability in the network.

In conclusion, centrality symptoms (i.e., “Fatigue”, “Excessive worry”, “Trouble relaxing” and “Uncontrollable worry”) and bridge symptoms (i.e., “Motor”, “Feeling afraid” and “Restlessness”) were identified in this network of depressive and anxiety symptoms in Chinese college students. Monitoring college students’ mental health in the late stage of the COVID-19 outbreak and targeting interventions (e.g., CBT, applied relaxation and medications) for selective symptoms are important to alleviate the overall level of anxiety and depression in this population.