Introduction

Sleep or nocturnal rest is critical for growth and development in human beings (Nunn et al., 2016). Human health, well-being, and functioning are directly impacted by sleep deprivation that does not provide adequate rest (Itani et al., 2016; Tomaso et al., 2021). Lack of sleep, difficulty falling asleep, and/or remaining asleep are typical symptoms of sleep disturbances (American Psychiatry Association, 2013) and are highly associated with depression and anxiety disorders (Zhou et al., 2020). Significantly, the outbreak of COVID-19 has placed a considerable burden on the mental health of the entire population (Altena et al., 2020), and a systematic review revealed that the prevalence of sleep disturbance was 17.65–81% in the general population (Lin et al., 2021). For Chinese university students, the prevalence of poor sleep quality and sleep disturbances has been calculated as 30% (Li et al., 2020) and 25.7% (Li et al., 2021). Previous theoretical studies have shown that identifying bridge symptoms in psychopathology networks can reveal fecund information because it promotes clarifying symptoms that put patients at higher risk of comorbidity (Borsboom & Cramer, 2013). For instance, Wang et al. (2018), a CS-C should not be lower than 0.25 and is preferably above 0.5. Finally, differences in network properties (i.e., edge weights, node strengths) were evaluated by bootstrapped difference tests (Epskamp et al., 2018).

In addition, predictability (i.e., R2), which reflects how well all its neighboring nodes predict a specific node, was estimated using the R package mgm (version 1.2–12) (van Borkulo et al., 2022). Furthermore, to explore bridge symptoms in the network that played essential roles in connecting two or more psychiatric disorders (Haslbeck & Waldorp, 2020), the bridge function in the R package networktools (version 1.4.0) was used (Chen & Chen, 2008; Jones et al., 2021). Bridge centrality indices assessed bridge symptoms, including bridge strength, bridge closeness, bridge betweenness, and bridge expected influence (1-step). Following a previous study (Robinaugh et al., 2016), bridge symptoms were selected using the index of bridge expected influence (1-step), including both positive and negative associations.

Comparisons based on sex

Following previous studies (Li et al., 2022; Wu et al., 2020), sex differences in network characteristics were assessed using the Network Comparison Test (NCT) in the R package NetworkComparisonTest (version 2.2.1) (van Borkulo et al., 2022). This test is employed on subsamples (female vs. male) with 1000 permutations to compare the global network strengths (absolute sum of all edge weights) and network structures (edge weight distributions) between the two networks. Furthermore, the strength of each edge connecting the two networks was determined using Holm–Bonferroni correlations for multiple comparisons.

Results

Descriptive statistics and items check

The means and standard deviations of the PHQ-9, GAD-7, and YSIS-8 items are reported in Table 1. Before performing network analysis, we checked for mean item level, item informativeness (i.e., item standard deviation), and item redundancy. All items were above the informativeness threshold (i.e., ± 2.5 SD below the mean level, PHQ-9: MSD = 0.59 ± 0.10; GAD-7: MSD = 0.54 ± 0.06; YSIS-8: MSD = 0.87 ± 0.10), and no item was considered statistically redundant (i.e., > 25%) by using the Fruchterman–Reingold algorithm. Hence, all items were included in the analysis.

Table 1 Basic information of scales and descriptive item statistics
Full size table

Network estimation structure and local network properties

The network estimated on the PHQ-9, GAD-7, and YSIS-8 is shown in Fig. 1(Part A) and Fig. S1. Several points are worth mentioning. First, through the weighted adjacency matrix (see Table S1), we found that poor sleep quality (#YSIS-1; i.e., quality of sleep during the past month) with sleep dissatisfaction (#YSIS-2; i.e., satisfaction with sleep during the past month) and DMS (#YSIS-4; i.e., waking up frequently during the night) with EMA (#YSIS-5; i.e., waking up earlier and cannot return to sleep) had the first and second strongest associations. Second, local property analysis showed that symptoms of sleep dissatisfaction (#YSIS-2; i.e., satisfaction with sleep during the past month) were the most central node, followed by poor sleep quality (#YSIS-1; i.e., quality of sleep during the past month) and uncontrollable worry (#GAD-2; i.e., unable to stop or control worry). The symptoms with the lowest centrality included appetite (#PHQ-5; i.e., poor appetite or overeating), DMS (#YSIS-4; i.e., waking up frequently during the night), suicide (#PHQ-9; i.e., thoughts about killing or hurting yourself), EMA (#YSIS-5; i.e., waking up earlier and cannot get back to sleep), and concentration (#PHQ-7; i.e., trouble concentrating).

Fig. 1
figure 1

A, The network structure of anxiety, depressive symptoms, and sleep disturbance among college students during the COVID-19 pandemic. B, Standardized centrality indices (i.e., ExpecteInfluence) of the network structure of anxiety, depressive symptoms, and sleep quality (Z scores)

Full size image

Bridge symptoms of anxiety, depressive symptoms, and sleep disturbance

The bridge expected influence was estimated in Fig. 1 and S2. The strongest bridge symptoms (Z score above 1) were sleep (#PHQ-3; i.e., trouble falling or staying asleep or slee** too much), feeling afraid (#GAD-7; i.e., feeling afraid as if something awful might happen), guilt (#PHQ-6; i.e., feeling bad about yourself or that you are a failure or have let yourself or your family down), restlessness (#GAD-5; i.e., being so restless that it is hard to sit still) and irritability (#GAD-6; i.e., becoming easily annoyed or irritable). The bridge expected influence of these items was significantly correlated with expected influence values (rs = 0.88, CI [0.74; 0.95]).

Network stability and accuracy

The predictability index indicated that the neighboring nodes could account for 60% of the variance in each node (MR2 = 0.60 ± 0.11), on average. SD (i.e., #YSIS-2, R2 = 0.83), PSQ (i.e., #YSIS-1, R2 = 0.82), uncontrollable worry (i.e., #GAD-2, R2 = 0.73), excessive worry (i.e., #GAD-3, R2 = 0.71), nervousness (i.e., #GAD-1, R2 = 0.70), trouble relaxing (i.e., #GAD-4, R2 = 0.70), feeling afraid (i.e., #GAD-7, R2 = 0.69), sad mood (i.e., #PHQ-2, R2 = 0.63), DIS (i.e., #YSIS-3, R2 = 0.62), restlessness (i.e., #GAD-5, R2 = 0.62), US (i.e., #YSIS-7, R2 = 0.61), DFI (i.e., #YSIS-8, R2 = 0.61), SI (i.e., #YSIS-6, R2 = 0.60) and energy (i.e., #PHQ-4, R2 = 0.60) had the highest predictability indices within the network (Table 1). Although, on average, half of each symptom’s variance could potentially be explained by the other nodes, this also implied that the remaining variance was not explained by only considering the interplay among symptoms.

The expected influence and predictability were unrelated to PHQ-9 item variability (PHQ-9: rs = 0.17, CI [− 0.56; 0.75] and rs = 0.32, CI [− 0.44; 0.81], respectively) and item mean (PHQ-9: rs = 0.30, CI [− 0.46; 0.80] and rs = 0.46, CI [− 0.29; 0.86], respectively). In contrast, predictability was positively related to GAD-7 item variability (GAD-7: rs = 0.86, CI [0.30; 0.96]) and item mean (GAD-7: rs = 0.82, CI [0.18; 0.97]); however, expected influence was unrelated to GAD-7 item variability (GAD-7: rs = 0.61, CI [− 0.26; 0.93]) and item mean (GAD-7: rs = 0.51, CI [− 0.40; 0.91]). Expected influence and predictability were positively related to sleep item mean (YSIS-8: rs = 0.77, CI [0.14; 0.96] and rs = 0.74, CI [0.08; 0.95], respectively). The expected influence and predictability were unrelated to the PHQ-9 and were unrelated to the sleep mean item (YSIS-8: rs = 0.41, CI [− 0.42; 0.86] and rs = 0.3, CI [− 0.51; 0.83], respectively).

The edge weights in the current sample were consistent with the bootstrapped sample, especially the connections with larger weights, indicating that the current network structure was stable (Part A of Fig. 2). The bootstrapped difference tests revealed that a large proportion of the comparisons among edge weights were statistically significant (Supplementary Fig. S3, S4).

Fig. 2
figure 2

A, estimation of edge weight difference by bootstrapped difference test. Nonparametric bootstrapped difference test for strength. Grey boxes indicate no significant difference, whereas black boxes indicate a statistically significant difference (p < .050). Diagonal values represent the strength score of each node. B, the stability of the network structure by case drop** subset bootstrap**

Full size image

Regarding network stability, the case-drop** bootstrap procedure showed that the expected influence (i.e., CS-C = 0.75) and bridge expected influence (1-step) (i.e., CS-C = 0.28) values remained stable after drop** different proportions of the sample (Part B of Fig. 2).

Fig. 3
figure 3

A, the maximum difference in edge strength. B, the difference in global strength (right panel)

Full size image

Sex differences

We tested whether global strength, structure and single edges differed by sex. Overall, the two networks appeared similar (Fig. S5). There were significant sex differences in network global strength (females: 11.22 vs. males: 12.21; S = 0.99, p = .000) and the network structure distribution of edge weights (M = 0.14, p = .047) (all p > .05 after Holm–Bonferroni correction, Fig. S6), as shown in Fig. 3. Specifically, examining each node and the expected influence index shows that men’s strength was stronger than women’s strength for #PHQ-7 and #YSIS-7 (p < .05).

Finally, we tested whether males and females reported different symptom mean levels statistically. There were 20 of 24 symptoms with statistically significant (all adjusted p values ≤ 0.045) differences (Table S2). However, the magnitude of the differences was negligible to small (Cohen’s d ≤ 0.5). Hence, the level differences of symptoms in females and males should be interpreted with caution.

Discussion

Using network analysis, the present study examined the relationships among sleep disturbance, depression, and anxiety among a large sample of college students during the COVID-19 pandemic. All results showed stability and accuracy, suggesting strong correlations between the internal symptoms of these three mental disorders.

From the node centrality, sleep dissatisfaction (#YSIS-2; i.e., satisfaction with sleep during the past month), poor sleep quality (#YSIS-1; i.e., quality of sleep during the past month), and uncontrollable worry (#GAD-2; i.e., unable to stop or control worry) were the highest strength values (Z score above 1), indicating those in the depression, anxiety, and sleep disturbance network structure, which should be noted. Sleep dissatisfaction refers to both the quality and quantity of sleep that cannot meet one person’s demands, such as adequate and regular sleep duration, a healthy sleep pattern, and high-quality sleep (Varghese et al., 2020). For college students, depression, anxiety, substance abuse, day naps, and rest time are prefiguring sleep dissatisfaction (Wilsmore et al., 2013). During the pandemic quarantine period, college students’ daily lives were replaced with taking online teaching and staying at home. Moreover, without hierarchical supervision from teachers or schools, students can easily become addicted to video games or mobile devices, and without peers’ companionship, college students may rely on substances to evade negative affection (Cuong et al., 2021). Therefore, superficially, the pandemic changed students’ daily lives. Nevertheless, college students’ regular sleep rhythm was devastated by the unrestrictive usage of video games, mobile devices, and substances. To combat sleep dissatisfaction, regular physical activity might improve sleep quality (Baron et al., 2022).

In addition to sleep dissatisfaction, poor sleep quality is another conspicuous factor besetting college students. In its literal meaning, poor sleep quality denotes that one person evaluates his or her sleep quality as poor or bad in the last month, which is positively associated with a higher prevalence of mental disorders and a longer duration of insomnia symptoms (Ohayon et al., 1997). Our view maintains that college students’ daily routine changes disturb their regular sleep patterns, which cannot be recovered or appeased by long sleep or day naps. Currently, the mainstream treatment to improve sleep quality is via mindfulness to release psychological stress (Lau et al., 2018).

In addition to sleep problems, the other factor, uncontrollable worry, is deleterious to college students’ mental health. Uncontrollable worry is being unable to restrain or control worry cognitively (Hallion & Ruscio, 2013), which is also defined as pathological worry (Gorday et al., 2018). Our result is consistent with one previous study showing that alienation from peers is one etiology of uncontrollable worry (Curzik & Salkicevic, 2016). For college students, staying away from peers means alienation from emotional divulgence. Moreover, more fierce competition for jobs or further education and uncertainties followed by pandemics can drag college students into deep emotional turmoil. Unfortunately, college students are tyros in the confrontation of traumatic events, although, without peer companions, a warm and solid filial relationship may mollify worry.

Furthermore, the perspective of bridge expected influence (1-step) measures the symptoms that play a crucial role in coalescing depression, anxiety, and sleep disturbance. Sleep (#PHQ-3; i.e., trouble falling or staying asleep or slee** too much) and feeling afraid (#GAD-7; i.e., feeling afraid as if something awful might happen) should be mentioned and discussed. The physical reaction to depression is mainly shown as difficulties initiating sleep, maintaining sleep, and fatigue. In a cognitive–emotion–behavior circulation (Weiner, 1980), we conjecture that sleep in the body directly reflects an irregular lifestyle and great pressure. When college students notice physical maladaptation without experience, college students may digress to either retain the same pattern or painstakingly revert to the original track. However, failures in lifestyle change may induce negative affection containing remorse, depression, surliness, or anger (Sarris et al., 2014). Additionally, cognitive bias may elicit college students to have a depressive status in which college students hold the belief that efforts to change are always in vain (Pang & Wu, 2021).

Referring to learned helplessness (Abramson et al. 1989), a person who endures unescapable negative events extensively will give up exerting efforts on the situation. College students are in a key developmental period to start a career, pursue further education, or cultivate a long-term intimate relationship. Nevertheless, in the pandemic, a profound adverse event, the main developmental task must be paused, and with chronic resurgence, containment measures become a quotidian. Ambitious college students are naturally fearful about future awful events that divest auspiciousness and omens. Furthermore, in terms of cognition–emotion–behavior circulation (Weiner, 1980), accumulated fearfulness manifests in sleep disturbance, depression, and anxiety.

After delineating the importance of symptoms, strong associations between two specific nodes should be mentioned, as strong associations indicate that alleviating either symptom in the association means that two symptoms are improved accordingly, which can shed light on practice. From our findings, poor sleep quality (#YSIS-1; i.e., quality of sleep during the past month) with sleep dissatisfaction (#YSIS-2; i.e., satisfaction with sleep during the past month) and DMS (#YSIS-4; i.e., waking up frequently during the night) with EMA (#YSIS-5; i.e., waking up earlier and cannot get back to sleep) are the two strongest associations, implying that sleep problems should be main targets in treatments to release college students’ mental distress. College students must sleep well to deal with daily routine work. In a cognition–emotion–behavior circulation (Weiner, 1980), college students have poor sleep quality behaviorally, including being unable to maintain sleep, sleep deeply or initiate sleep, which may hold antagonist affection toward sleep. Dissatisfaction with sleep, including distress, sadness, and hatred toward sleep, can decay sleep quality (Duran & Erkin, 2021). Hard-to-maintain sleep makes college students mentally excited after extended periods of mobile usage, playing video games, or watching television programs. Thus, excited college students cannot maintain deep sleep naturally, and it is difficult to get up earlier to resume activities (Mei et al., 2022).

Limitations

Based on a large sample plan, this study investigated the network characteristics of mental health symptoms of college students under close management during COVID-19. However, there are still several research limitations that need to be addressed. First, similar to most studies (Van den Bergh et al., 2021; Wang, 2022), this study was cross-sectional, which does not best explain longitudinal causality. Second, the study was conducted at a closed university, and thus, the results may not be generalizable to all groups in the new population.

Conclusions

In summary, network analysis provides new insights into the internal relationships among depression, anxiety, and sleep disturbance. The results highlight that specific symptoms can function as bridges among anxiety, depression, and sleep complaints, which helps to explain their comorbidity and reciprocal influences. In particular, sleep, guilt, restlessness, irritability, and feeling afraid play unique roles as bridge symptoms. Although the current network does not consider the directionality between symptoms, it still contributes to comprehending college students’ depression, anxiety, and sleep disturbance during the closed management of COVID-19. However, the network-comparison test results showed significant differences in global strength and edge weights between males and females in the network structure for depression, anxiety, and sleep disturbance. In future research, expanding the issue of directivity between symptoms is necessary. At the same time, it is also important to investigate the symptom network between closed management and after closed management and provide meaningful suggestions for clinical practices.