Temporal Network of Depressive Symptoms across College Students with Distinct Depressive Trajectories during the COVID-19 Pandemic

Ma, Zijuan; Zhao, Jingbo; Chen, Huilin; Tao, Yanqiang; Zhang, Yifan; Fan, Fang

doi:https://doi.org/10.1155/2023/8469620

Depression and Anxiety

On this page

Abstract Introduction Methods Results Discussion Conclusion Data Availability Conflicts of Interest Authors’ Contributions Acknowledgments Supplementary Materials References Copyright Related Articles

Research Article | Open Access

Volume 2023 | Article ID 8469620 | https://doi.org/10.1155/2023/8469620

Temporal Network of Depressive Symptoms across College Students with Distinct Depressive Trajectories during the COVID-19 Pandemic

Zijuan Ma,¹Jingbo Zhao,²Huilin Chen,³Yanqiang Tao,⁴Yifan Zhang,¹and Fang Fan¹

Academic Editor: Carlos Carona

Received28 Nov 2022

Revised07 Jun 2023

Accepted17 Jun 2023

Published15 Jul 2023

Abstract

Background. There are marked differences in how individuals respond and adapt to depressive symptoms over time during the strain of public health emergencies; however, few studies have examined the interrelations between depressive symptoms in distinct depressive trajectories from the COVID-19 outbreak period to the COVID-19 control period. Therefore, this study conducted cross-lagged panel networks to investigate the temporal relationships between depressive symptoms across distinct depressive trajectories from the COVID-19 outbreak period (T1) to the COVID-19 control period (T2). Methods. A total of 35,516 young participants from the College Students’ Behavior and Health Cohort during the COVID-19 pandemic were included in the current study. Depressive symptoms were self-reported using the nine-item Patient Health Questionnaire. Unique longitudinal relationships between symptoms during the COVID-19 pandemic were estimated using a cross-lagged panel network. Results. Longitudinal relationships across distinct depressive trajectories were unique during the COVID-19 pandemic. Specifically, suicidal ideation at T1 in the chronic- and delayed-dysfunction groups was most predictive of other symptoms at T2, whereas “sleep” at T1 in the recovery group and “lack of energy” at T1 in the resistance group may be strongly related to the remission of other depressive symptoms at T2. Conclusions. These exploratory findings demonstrate the directionality of relationships underlying individual symptoms in the youth and highlight suicidal ideation, sleep, and energy as potential influencers of other depressive symptoms across distinct depressive trajectories. Targeting those symptoms during the outbreak period of COVID 19 would theoretically have been beneficial in preventing and/or reducing the likelihood of spontaneous depression during the subsequent control period.

1. Introduction

Depression is the leading cause of ill health and disability worldwide, affecting 4.4% (320 million) of the world’s population [1]. With the explosive spread of coronavirus disease 2019 (COVID-19) across the globe, one of the biggest global crises in generations, depression quickly became more common due to restriction of activities, transfer of educational mode to virtual learning, and changes in social life [2–5]. Correspondingly, recently published meta-analyses have pinpointed that the pooled prevalence of depression during COVID-19 among populations was 35% (95% confidence interval (CI): 33–38; [5]). Further, a meta-analysis conducted on college students during the COVID-19 pandemic reported the prevalence of depression to be 37% (95% CI: 32–42; [2, 3]). Notably, a systematic review of severe acute respiratory syndrome (SARS), COVID-19, and other infectious diseases also revealed that the pooled prevalence of depression (43%, 95% CI: 27–60) was significantly higher among university students than in other populations, including the general public, health workers, older adults, infected patients, infected survivors, and pregnant women [5]. Given the elevated prevalence of depression among college students, preventing and treating depression during the COVID-19 pandemic is of great significance.

Owing to the inherent unpredictability of the onset of depression, the duration of episodes, the number of episodes over a lifespan, and the pattern in which they occur are variable [6]. Indeed, evidence has suggested marked differences in how individuals respond and adapt to traumatic experiences over time [7]. For instance, Bonanno et al. [7] used latent class analysis to evaluate psychological functioning over time in 997 hospitalized survivors of the 2003 SARS epidemic in Hong Kong. They identified four latent classes: chronic dysfunction, delayed dysfunction, recovery, and resilience. In addition, a growing body of longitudinal studies has demonstrated the existence of heterogeneous time-course trajectories of depressive symptoms following traumatic experiences such as earthquakes [8] and the COVID-19 pandemic [9, 10], highlighting the significance of learning more about individuals who exhibit resilience and those who experience chronic difficulties [11]. Researchers have theorized four common trajectories as follows [7–10]: (1) chronic dysfunction, which is characterized by consistent moderate to severe depressive symptoms over time after the traumatic event; (2) delayed dysfunction, with initially no or minimal depressive symptoms followed by the postponed outbreak of moderate to severe symptoms; (3) recovery, which is defined as initial moderate to severe depressive symptoms but followed by gradual relief; and (4) resilience, which refers to the ability to maintain relatively stable levels of normal healthy functioning over time. Although the factors related to depressive trajectories among college students have been examined during the COVID-19 pandemic [9, 10], the substantial variation in individual symptoms and the associations among symptoms were not usually emphasized in previous studies. Recent network theory holds that psychiatric symptoms result from the causal interplay between symptoms, possibly involving feedback loops [12–14]. However, the intersymptom interplay across heterogeneous patterns of depressive symptoms remains relatively unclear. Therefore, it is vital to utilize less restrictive and more sophisticated analytic approaches to examine the dynamic relationships between depressive symptoms across distinct patterns.

Network analysis, based on network theory, is a novel approach to understanding the strength and nature of associations among symptoms [15], which is popular and prominent in clinical and psychiatric research domains. The network approach to psychopathology proposes that mental illnesses result from dynamic interactions between symptoms rather than a nonobservable common cause [13, 15]. Although numerous studies have been published on interacting associations between depressive symptoms over the past half-decade [16–18], they were unable to discern the direction of relationships based on undirected cross-sectional networks. The cross-lagged panel network (CLPN) was recently developed by Rhemtulla et al. and first utilized by Bernstein et al.; it can unravel the dynamic causality between symptoms over time using longitudinal data [19, 20]. Despite the fact that the CLPN has autoregressive routes, it may be confounded by stable individual differences of a trait-like or time-invariant nature [20]. However, it is well suited for identifying temporal effects between individual elements of a construct in panel data by computing within-timepoint (undirected) and between-timepoint (directed) associations [21]. Within a CLPN graph, circles represent autoregressive feedback loops, and edges depict cross-time effects; edge thickness reflects the strength of the effects, and arrows indicate the direction of prediction.

Researchers have recently evaluated the cross-lagged relationships between individual symptoms of depression among adolescents and adults [18, 22]. For instance, Rubin et al. found that “feelings of failure” were highly likely to lead to future “suicidal ideation” at least over ten months in the middle adolescent CLPN, while suicidal ideation at time 1 was eight times more likely to precede movement dysregulation at time 2 among the early adolescents [22]. However, the above studies did not provide insight into the interrelations across heterogeneous patterns of depression on a more micro level from the COVID-19 outbreak period to the COVID-19 control period. Therefore, we conducted CLPN analysis to investigate causal interactions among depressive symptoms across distinct patterns of depressive trajectories from the COVID-19 outbreak period to the COVID-19 control period.

2. Methods

2.1. Study Background

The development trend of the COVID-19 pandemic in China is shown in Figure 1. Specifically, the Chinese government announced a nationwide lockdown to control the spread of the pandemic on January 24, 2020, and we conducted the first online survey nine days after the nationwide lockdown. The total confirmed cases of COVID-19 in Mainland China increased from 17,205 to 42,638 during the first survey. During the outbreak period, the cumulative COVID-19-confirmed cases in Guangdong Province ranged from 1000 to 9999 before March 1, 2020, in a pandemic moderate-risk area as assessed by the World Health Organization in early 2020 [23]. With the effective control of the COVID-19 pandemic, China has achieved great success since April 8, 2020, and the Ministry of Education in Guangzhou Province permitted students to return to school in batches starting May 11, 2020 [2, 3]. As of 24:00 on June 1, 2020, Guangdong Province reported 1596 confirmed cases of COVID-19.

2.2. Participants and Procedure

Data were extracted from the College Students’ Behavior and Health Cohort during the COVID-19 pandemic, sampling from 22 colleges/universities in Guangdong Province, China. Figure 2 shows the geographical distribution of participating colleges and universities, and more details on sampling and data collection have been described elsewhere (Wang et al. [9, 10, 24]). During the COVID-19 pandemic, three cross-sectional online surveys were conducted from February 3 to February 10 (the outbreak period; 164,101, 88.3% valid questionnaires), March 24 to April 3 (the initial remission period; 148,343, 95.4% valid questionnaires), and June 1 to June 15 (the control period; 166,052; 88.0% valid questionnaires). After matching the three waves of data, 35,516 college students completed all three surveys and provided valid data on all measures. The current sample included two-wave surveys: the COVID-19 outbreak period (Time 1 (T1): February 3 to 10, 2020) and the COVID-19 control period (Time 2 (T2): June 1 to 15, 2020). Among the 35,516 participants, 9244 (26.0%) were male students. Detailed sample information is shown in Supplementary Table 1. The information on demographic and pandemic-related factors was collected at T1, and depressive symptoms were assessed at T1 and T2.

The current study was approved by the Human Research Ethics Committee of South China Normal University (ethics no.: SCNUPSY-2020-01-001) and carried out in line with the Helsinki Declaration, as revised in 1989. The local education bureau strongly supported our survey, and all participants or their guardians (if necessary) provided electronic informed consent before the study. In addition, they were informed of the right to withdraw from the survey at any time.

2.3. Measures

2.3.1. Demographic and Pandemic-Related Factors

All participants completed brief demographic questions (sex and age) and pandemic-related items at T1. Pandemic-related factors included COVID-19 epidemic severity in the province of residence, confirmed or suspected cases in the community or village, relatives or friends being infected with COVID-19, and exposure to media coverage of COVID-19.

2.4. Depressive Symptoms

The Patient Health Questionnaire (PHQ9) was used to estimate symptoms of depression in the past two weeks at T1 and T2 [25]. Each item is rated from 0 (not at all) to 3 (nearly every day), with higher scores indicating more severe symptoms. The Chinese version of the PHQ9 has shown good psychometric properties, with Cronbach’s alpha of 0.86 in the general population in China, with a recommended cut-off score of 7 or more [26, 27]. Therefore, we chose 7 as a cut-off score in this study, consistent with previous studies [9, 10, 28]. In the present study, Cronbach’s alpha was 0.87 at T1 and 0.91 at T2.

2.5. Statistical Analyses

2.5.1. Trajectories of Depressive Symptoms

We adopted two steps to generate trajectories of depressive symptoms. First, students were classified as having probable or no depressive symptoms, with a cut-off score of 7. Second, based on a previous study [9, 10], we further clustered four trajectories of depressive symptoms: chronic dysfunction, delayed dysfunction, recovery, and resistance. More specifically, individuals in the chronic-dysfunction group showed depressive symptoms at both time points. Students who reported no depressive symptoms at T1 but symptoms at T2 were assigned to the delayed-dysfunction group. Participants with depressive symptoms at T1 but without depression at T2 were included in the recovery group. For the resistance group, depressive symptoms were absent at both time points.

2.5.2. Temporal Network

The R-package glmnet was used to construct CLPNs [29], and the R-package qgraph was applied to visualize all graphs of CLPNs [30, 31]. A CLPN model comprises autoregressive and cross-lagged pathways [20]. In terms of autoregressive pathways, a symptom at T1 predicts itself at T2 after controlling all other symptoms at T1. Regarding cross-lagged pathways, a symptom at T1 predicts a different symptom at T2 after adjusting for all other symptoms at T1. Moreover, to shrink small regression coefficients to 0, regression coefficients were regularized using LASSO (least absolute shrinkage and selection operator) with a 10-fold cross-validation tuning parameter selection [21].

The R-package qgraph was used to calculate the following centrality indices: in expected influence (in-EI) and out expected influence (out-EI) [32]. The in-EI quantifies the degree to which each symptom at T2 is predicted by other symptoms at T1 (i.e., the sum of the values of incoming edges related to a symptom), whereas the out-EI signifies the degree to which each symptom at T1 predicts other symptoms at T2 (i.e., the values of outgoing edges connected to a symptom) [18, 21, 22]. Successfully targeting a symptom with high out-EI is likely to result in the resolution of other symptoms in a longitudinal network [33, 34]. Moreover, three statistics of in-EI and out-EI are calculated: overall effects, which include all variables; cross-lagged effects, which exclude the autoregressive path of the node of interest; and cross-constructs, which exclude paths connecting nodes within the same community.

Two bootstrapping approaches implemented in the R-package boonet were used to measure the accuracy and stability of the CLPNs [30, 31]. First, edge weight accuracy was assessed by calculating 95% CIs around each edge weight value with nonparametric bootstrapping (1000 iterations). Second, correlation stability (CS) coefficients were evaluated using case-drop bootstrapping to determine the stability of the rank order of centrality indices. A CS value should not be less than 0.25 and preferably above 0.5 [30, 31]. Finally, edge weight difference tests were calculated to test whether edge weights differed significantly from each other.

3. Results

3.1. Depressive Trajectories

Figure 3 presents the change pattern and four trajectories of depressive symptoms from the COVID-19 outbreak period to the COVID-19 control period among 35,516 college students. More specifically, 5380 students (15.1%) with PHQ9 scores above the cut-off of 7 at T1 and T2 were classified in the chronic-dysfunction group. The delayed-dysfunction group, characterized by PHQ9 scores below the cut-off at T1 but equal to or above at T2, comprised 17.7% of the sample (). The recovery group (, 6.6%) presented an initial high level of depressive problems at T1, but no depressive symptoms at T2. Students in the resistance group (60.5%, ) showed no depressive symptoms at T1 and T2.

(a)

(b)

3.2. Temporal Networks for Distinct Depressive Trajectories

3.2.1. Network Structures

Figure 4 shows CLPNs from the COVID-19 outbreak period to the COVID-19 control period across four depressive trajectories. All edge weights are presented in Supplementary Tables 2–5. The CLPNs of depressive symptoms (9 nodes) revealed that 63 of 72 edges in the chronic-dysfunction group, 28 of 60 edges in the delayed-dysfunction group, 13 of 52 edges in the recovery group, and 63 of 69 edges in the resistance group were estimated to be above zero. Symptoms with the greatest autoregression coefficients were “suicidal ideation” (PHQ9) in the chronic- () and delayed-dysfunction groups () and “guilt” (PHQ6) in the recovery group () and the resistance group (). Additionally, the edge “lack of energy” (PHQ4) ⟶ “anhedonia” (PHQ1) was the strongest cross-lagged edge in the chronic-dysfunction () and resistance groups (). The edges “suicidal ideation” (PHQ9) ⟶ “motor” (PHQ8; ) and “suicidal ideation” (PHQ9) ⟶ “lack of energy” (PHQ4; ) displayed the strongest cross-lagged connections in the delayed-dysfunction and recovery groups, respectively. Table 1 displays the correlation of the network structures between the four distinct trajectories. As shown in Table 1, we found correlations in the network structures between the chronic- and delayed-dysfunction groups (, ) and the recovery and resistance groups (, ).

(a)

(b)

(c)

(d)

Figure 4

Cross-lagged panel networks (left) from the COVID-19 outbreak period to the COVID-19 control period (a: chronic-dysfunction group; b: delayed-dysfunction group; c: recovery group; d: resistance group). A threshold of all edge weights was manually set to 0.025 to make the figures more interpretable. Note: each curved arrow “loop” reflects an autoregressive association; white nodes indicate depressive symptoms and blue lines indicate positive relations, whereas orange lines signal negative relations, and line thickness and boldness reflect the strength of associations.

3.2.2. Network Inference

As shown in Figure 5 and Table 2, centrality estimates indicated that the symptoms of “depressed mood” (PHQ2; ) in the chronic-dysfunction group, “difficulty concentrating” (PHQ7; ) in the delayed-dysfunction group, “suicidal ideation” (PHQ9; ) in the recovery group, and “anhedonia” (PHQ1; ) in the resistance group had the highest in-EI. The symptoms “suicidal ideation” (PHQ9) in the chronic- () and delayed-dysfunction groups (), “sleep” (PHQ3; ) in the recovery group, and “lack of energy” (PHQ4; ) in the resistance group had the highest out-EI. Given the availability and effectiveness of clinical interventions for depression [34], priority should be paid to focusing on the symptoms with high out-EI at T1 that activate other depressive symptoms at T2.

More specifically, “suicidal ideation” (PHQ9) in the chronic-dysfunction group during the COVID-19 outbreak period may have strongly motivated eight depressive symptoms during the COVID-19 control period, such as “guilt” (PHQ6; ), “depressed mood” (PHQ2; ), and “motor” (PHQ8; ). Regarding the delayed-dysfunction group, “suicidal ideation” (PHQ9) during the COVID-19 outbreak period positively predicted seven depressive symptoms during the COVID-19 control period, such as “guilt” (PHQ6; ) and “motor” (PHQ8; ). In terms of the recovery group, “sleep” (PHQ3) during the COVID-19 outbreak period positively predicted six depressive symptoms during the COVID-19 control period, such as “lack of energy” (PHQ4; ). “Lack of energy” (PHQ4) in the resistance group during the COVID-19 outbreak period positively predicted seven depressive symptoms during the COVID-19 control period, such as “anhedonia” (PHQ1; ), “appetite” (PHQ5; ), and “sleep” (PHQ3; ).

n-EI,

3.2.3. Accuracy and Stability of Network Parameters

The accuracy plots of the four CLPNs show small-to-moderate confidence intervals around the edge weights, suggesting good accuracy for the baseline to follow-up networks across the four depressive trajectories (Supplementary Figure 1). Likewise, the case-drop bootstrapping results revealed that the rank order of in-EI and out-EI had moderate to strong stability across the four CLPNs (Supplementary Figure 2). Specifically, CS coefficients of in-EI and out-EI were as follows: chronic-dysfunction group: and ; delayed-dysfunction group: and ; recovery group: and ; and resistance group: and . In addition, the edge weight difference tests (Supplementary Figure 3) revealed that these edges were significantly stronger than most other edges, and centrality difference tests (Supplementary Figures 4 and 5) indicated that these symptoms displayed significantly higher out-EI and in-EI compared with other symptoms in the CLPNs.

4. Discussion

To the best of our knowledge, this is the first study using CLPN analysis to investigate the progression of depressive symptoms across distinct depressive trajectories from the COVID-19 outbreak period to the COVID-19 control period among college students. We summarized four depressive trajectory groups: chronic dysfunction, delayed dysfunction, recovery, and resistance. We found that longitudinal relationships across distinct depressive trajectories were unique during the COVID-19 pandemic. Specifically, the predictive ability of “suicidal ideation” at T1 for other depressive symptoms at T2 appeared to be strongest in the chronic- and delayed-dysfunction groups, while “sleep” at T1 in the recovery group and “lack of energy” at T1 in the resistance group may be strongly related to the remission of other depressive symptoms at T2. These findings offer new insights that may help in understanding the development, maintenance, and treatment of depression.

In the present study, four depressive trajectories were identified from the COVID-19 outbreak period to the COVID-19 control period: chronic dysfunction (11.6% of the whole sample), delayed dysfunction (17.7%), recovery (6.6%), and resistance (55.8%), which is slightly inconsistent with our previously published study because data that were collected during the initial remission period of COVID-19 were excluded from the current study [9, 10]. Additionally, the proportions of chronic dysfunction and recovery in our study were lower than the rates (chronic dysfunction: 42%; recovery: 10%) estimated in a sample of hospitalized survivors of the 2003 SARS epidemic in Hong Kong, while the rates of delayed dysfunction and resistance were higher than their rates (delayed dysfunction: 13%; resistance: 35%; [7]). Notably, Bonanno [35] held that only 5% to 10% of those who are exposed to loss or potential trauma develop chronic dysfunction, which aligned with our current findings for the rate of chronic dysfunction. In addition, the epidemiological data analyzed by Bonanno et al. were collected among hospitalized survivors with extreme levels of exposure to SARS [7]; in contrast, our data were collected from college students who were in home quarantine and had a relatively low risk of being infected by COVID-19. This could explain the high rate of the resistance trajectory in our study.

“Suicidal ideation” had the highest out-EI in the chronic- () and delayed- () dysfunction groups, meaning that it most strongly predicted other depressive symptoms at T2 after adjusting for all other symptoms at T1, and demographic and pandemic-related covariates, among individuals who were depressed during the COVID-19 control period. These findings are consistent with findings from longitudinal network analysis in both early adolescence [22] and adults [18], which indicate that suicidal ideation may be particularly strongly connected with other symptoms over time. In particular, if an early adolescent reported suicidal ideation at time 1, they were eight times more likely to report movement dysregulation at time 2 [22]. Also, our findings are supported by previous cross-sectional studies in different populations [36–38]. For instance, Ballester et al. [36] suggested that prior suicidal ideation () was the strongest predictor for the persistence of depression among university students. In addition, suicidal ideation in our study may more strongly motivate symptoms of “guilt” and “motor,” which are partly consistent with other studies [22, 38]. For example, Bryan et al. [38] found that the mean level of guilt was significantly higher among military personnel with a history of suicidal ideation. The above-mentioned findings indicate that suicidal ideation could be the predictor of other depressive symptoms. This finding can be understood by the theory of suicidal trajectories and interpersonal theory of suicide, which suggest that although suicidal ideation may be strong, it is at the bottom of the pyramid; the step toward suicidal planning and attempting requires the ability to reach the peak of the pyramid (i.e., suicide; [39, 40]). Thus, most individuals who stop at the stage of suicidal thoughts might activate negative cognitions selectively, including “myself as a burden to others,” “I cannot tolerate this pain anymore,” and “I have nothing to look forward to”; these negative cognitions produce a profound sense of defeat with a sense of entrapment, which further leads to many depressive symptoms, such as “guilt” and “depressed mood” [41, 42]. This provides further evidence that suicidal ideation should be actively monitored and targeted when present. Treatments targeting suicidal ideation such as cognitive therapy for suicide prevention may provide greater reductions in suicidal ideation and prevent the co-occurrence of other depressive symptoms or decrease these symptoms [41, 43].

“Sleep” had the highest out-EI () estimate in the recovery group, indicating that remission in this symptom over the follow-up period was most associated with remission in all other depressive symptoms at the end of follow-up. Little is known about sleep problems and depressive symptoms in longitudinal networks; however, existing studies suggest that insomnia may increase depressed mood and other depressive symptoms and directly cause a depressive episode [44, 45]. Additionally, a meta-analysis identified insomnia as a significant predictor of the onset of depression [46]. Based on the above evidence, we speculate that the monitoring and treatment of sleep problems during the COVID-19 outbreak period, especially insomnia, may have mitigated the incidence or diminished the severity of depression during the COVID-19 control period. Cognitive behavioral therapy for insomnia, the most widely used and widely studied nondrug treatment for insomnia, should be used to intervene and prevent sleep problems during a virus outbreak period to reduce the possibility of depression developing over time [47].

“Lack of energy,” with the highest out-EI () in the resistance group, predicted other depressive symptoms at follow-up, which is consistent with the findings from cross-sectional network analyses in Hong Kong residents during the COVID-19 pandemic [48]. Although there is a lack of evidence from the longitudinal network, the above finding implies that improvement in “lack of energy” over time was associated with subsequent improvement in most other depressive symptoms during the COVID-19 pandemic. “Lack of energy” refers to the exhausting nature of the condition and its associated stress states [49], which were caused by the COVID-19 outbreak and its confinement measures, including contact restrictions, home quarantine, and closure of schools, colleges, and universities. In addition, “lack of energy” is closely associated with lack of motivation and cognitive dysfunction [50], and severe cognitive dysfunction (i.e., negative evaluative bias) will support the persistence of depressed mood and other depressive symptoms [42]. Therefore, exercise therapy should be implemented to enhance energy to disrupt indirectly the persistence of depressed mood [51].

Over the past decade, the application of the network approach in the study of psychopathology has developed increased popularity and prominence [52]. It holds that mental disorders result from interactions between the symptoms of the specific mental illness [12–15]. Despite calls to investigate mental illnesses at the symptom level rather than the disorder level, most investigations on this topic have been cross-sectional [16, 48], precluding study of the dynamic interplay of depressive symptoms. Our study employed the CLPN model to corroborate work on new developments in the temporal associations between symptoms hypothesized by network theory. Compared to previous CLPN networks [18, 22], advancements in our study have led to the development of temporal sequences and changes in depressive symptoms over time across distinct depressive trajectories. In addition, similar to the finding that suicide ideation was directly related to most depressive symptoms [16], our study further found that suicidal ideation was a key symptom in both the chronic- and delayed-dysfunction groups and, therefore, was common to distinct depressive trajectories. These findings indicate that future researchers might apply CLPN models to heterogeneous groups to provide more detailed information regarding dynamic effects on psychological or mental symptoms.

Clinically, identifying which interventions directly target which symptoms in which populations may be useful. Our findings have provided information on which symptoms in which populations were important during the COVID-19 pandemic by identifying the key symptoms in networks of distinct depressive trajectories over time. Specifically, “suicidal ideation” had the strongest value in predicting other depressive symptoms in the chronic- and delayed-dysfunction groups; in contrast, “sleep” in the recovery group and “lack of energy” in the resistance group were highly associated with the remission of other depressive symptoms. Hence, brief interventions targeting “sleep” and “lack of energy” may promote the spontaneous remission of depressive symptoms in the acute phase, while treatments targeting “suicidal ideation” may help minimize the subsequent development of other depressive symptoms in individuals with chronic and delayed dysfunction. Overall, in order to achieve clinically meaningful predictions, future network models should be aimed at examining the network structure of subgroups rather than focusing on the whole sample only.

Our findings should be interpreted with consideration of several limitations. First, sampling bias should be noted in our study and may be reflected in several aspects as follows: (1) All participants were recruited from colleges/universities, which limits the generalizability of our findings to other populations, including COVID-19 patients, healthcare workers, medical professionals, and older populations. (2) The surveys were conducted in Guangdong Province, which is one of the richest parts of China. Therefore, there may be bias in our findings since the resources available in response to COVID-19 varied across China. (3) Our study was similar to previous studies in that female students may be more willing to participate in mental health surveys [53–56], but a high proportion of females might affect estimates of depression. Second, self-reported depressive symptoms may be prone to reporting bias, which could hinder symptom-level analysis, particularly if an individual cannot precisely distinguish a given depressive symptom out of the many. In addition, although a growing body of longitudinal studies has indicated that most people’s long-term psychological reactions can be reliably captured by four prototypical outcome patterns or trajectories (chronic dysfunction, delayed dysfunction, recovery, and resilience) across time [7, 57, 58], response bias may have resulted from simply using the sum score of depressive symptoms to divide students into four subgroups without a clinical diagnosis. Third, although we included several demographic characteristics (sex and age) and pandemic-related factors as covariates in the CLPN model to mitigate confounding effects from these sources, we did not consider other variables (e.g., history of mental illness and psychiatric drug usage) that might differentially influence or moderate the temporal interrelationships of depressive symptoms, and we did not collect any COVID-19-related information in T2. Fourth, although de Ron et al. [59] suggested that Berkson’s bias is a considerable and underappreciated problem when applying the Gaussian Graphical Model and the Ising Model to clinical populations [59], there have been few studies to investigate the effect of Berkson’s bias on the performance of the CLPN in clinical populations and other populations. To pay more attention to Berkson’s bias, future studies should focus explicitly on the impact of Berkson’s bias on other network models (i.e., the CLPN model) in different populations. Fifth, similar to other CLPN networks [21, 22, 33, 60, 61], our study also found that each item of the PHQ9 was not normally distributed according to the Kolmogorov–Smirnov test (see Supplementary Table 6). Unfortunately, we did not find an approach to estimate high dimensional directed graphical models efficiently and robustly. Future studies could pay attention to the nonparametric correlation in CLPN models. Finally, this study provides insight into predictive and potentially causal interplay among depressive symptoms across a four-month lag; however, it is still unclear what time lags are appropriate or optimal to capture relationships between individual symptoms [21]. Therefore, future research should explore various expected time lags when deciding sampling frequency.

5. Conclusion

With panel data from a large and population-based cohort of youths, our study applied the novel methodology of CLPN to identify unique longitudinal relationships between depressive symptoms across groups with distinct depressive trajectories over time, overcoming the limitations of past studies that failed to explore the dynamic effects of depressive symptoms during the COVD-19 pandemic. Our findings indicated that targeting symptoms of “sleep” and/or “lack of energy” during the outbreak period of public health emergencies, such as COVID 19, would theoretically be beneficial in preventing and/or reducing the likelihood of spontaneous depression during the subsequent control period. More importantly, the disappearance of “suicidal ideation” would theoretically have been accompanied by lower overall connectivity in the depressive network during the COVID-19 control period, thus promoting the rehabilitation of depressed individuals. Hence, individualized intervention was needed for young people who reported suicidal ideation during the COVID-19 outbreak.

Data Availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Conflicts of Interest

All authors declare no conflict of interest.

Authors’ Contributions

Zijuan Ma and Jingbo Zhao contributed equally to this work.

Acknowledgments

The authors would like to thank all participants. The present study was funded by the National Natural Science Foundation of China (grant nos. 31871129 and 32271135).

Supplementary Materials

Detailed sample information is shown in Supplementary Table 1. All edge weights are presented in Supplementary Tables 2–5. Supplementary Figures 1–5 provide the results for accuracy (Supplementary Figure 1) and stability (Supplementary Figure 1) of the network, the edge weight difference tests (Supplementary Figure 3), and centrality difference tests (Supplementary Figures 4 and 5). (Supplementary Materials)

References

WHO, Depression and Other Common Mental Disorders: Global Health Estimates, World Health Organization, 2017.
C. Wang, W. Wen, H. Zhang et al., “Anxiety, depression, and stress prevalence among college students during the COVID-19 pandemic: a systematic review and meta-analysis,” Journal of American College Health, vol. 1-8, pp. 1–8, 2021.
View at: Publisher Site | Google Scholar
D. Wang, H. Chen, S. Zhai et al., “Is returning to school during the COVID-19 pandemic stressful? A study on immediate mental health status of Chinese college students,” Jourbal of Affective Disorders, vol. 287, pp. 261–267, 2021.
View at: Publisher Site | Google Scholar
WHO, World Mental Health Report: Transforming Mental Health for All, World Health Organization, 2022, https://www.who.int/publications/i/item/9789240049338.
K. Yuan, Y.-B. Zheng, Y.-J. Wang et al., “A systematic review and meta-analysis on prevalence of and risk factors associated with depression, anxiety and insomnia in infectious diseases, including COVID-19: a call to action,” Molecular Psychiatry, vol. 27, no. 8, pp. 3214–3222, 2022.
View at: Publisher Site | Google Scholar
G. S. Malhi and J. J. Mann, “Depression,” The Lancet, vol. 392, no. 10161, pp. 2299–2312, 2018.
View at: Publisher Site | Google Scholar
G. A. Bonanno, S. M. Ho, J. C. Chan et al., “Psychological resilience and dysfunction among hospitalized survivors of the SARS epidemic in Hong Kong: a latent class approach,” Health Psychology, vol. 27, no. 5, pp. 659–667, 2008.
View at: Publisher Site | Google Scholar
Y. Zhou, Q. Han, and F. Fan, “Latent growth curves and predictors of depressive symptoms among Chinese adolescent earthquake survivors,” Personality and Individual Differences, vol. 100, pp. 173–178, 2016.
View at: Publisher Site | Google Scholar
D. Wang, J. Zhao, B. Ross et al., “Longitudinal trajectories of depression and anxiety among adolescents during COVID-19 lockdown in China,” Jourbal of Affective Disorders, vol. 299, pp. 628–635, 2022.
View at: Publisher Site | Google Scholar
D. Wang, J. Zhao, S. Zhai et al., “Longitudinal trajectories of insomnia symptoms among college students during the COVID-19 lockdown in China,” Journal of Psychosomatic Research, vol. 157, article 110795, 2022.
View at: Publisher Site | Google Scholar
J. D. Osofsky, H. J. Osofsky, C. F. Weems, L. S. King, and T. C. Hansel, “Trajectories of post-traumatic stress disorder symptoms among youth exposed to both natural and technological disasters,” Journal of Child Psychology and Psychiatry, vol. 56, no. 12, pp. 1347–1355, 2015.
View at: Publisher Site | Google Scholar
D. Borsboom, “Psychometric perspectives on diagnostic systems,” Journal of Clinical Psychology, vol. 64, no. 9, pp. 1089–1108, 2008.
View at: Publisher Site | Google Scholar
D. Borsboom, “A network theory of mental disorders,” World Psychiatry, vol. 16, no. 1, pp. 5–13, 2017.
View at: Publisher Site | Google Scholar
A. O. Cramer, L. J. Waldorp, H. L. van der Maas, and D. Borsboom, “Comorbidity: a network perspective,” Behavioral and Brain Sciences, vol. 33, no. 2-3, pp. 137–150, 2010.
View at: Publisher Site | Google Scholar
D. Borsboom and A. O. Cramer, “Network analysis: an integrative approach to the structure of psychopathology,” Annual Review of Clinical Psychology, vol. 9, no. 1, pp. 91–121, 2013.
View at: Publisher Site | Google Scholar
M. W. M. Gijzen, S. P. A. Rasing, D. H. M. Creemers, F. Smit, R. Engels, and D. De Beurs, “Suicide ideation as a symptom of adolescent depression. A network analysis,” Journal of Affective Disorders, vol. 278, pp. 68–77, 2021.
View at: Publisher Site | Google Scholar
Y. Liang, Y. Chen, Q. Huang, Y. Zhou, and Z. Liu, “Network structure and temporal stability of depressive symptoms after a natural disaster among children and adolescents,” European Journal of Psychotraumatology, vol. 14, no. 1, 2023.
View at: Publisher Site | Google Scholar
K. Savelieva, K. Komulainen, M. Elovainio, and M. Jokela, “Longitudinal associations between specific symptoms of depression: network analysis in a prospective cohort study,” Journal of Affective Disorders, vol. 278, pp. 99–106, 2021.
View at: Publisher Site | Google Scholar
E. E. Bernstein, E. M. Kleiman, R. Bork et al., “Unique and predictive relationships between components of cognitive vulnerability and symptoms of depression,” Depression and Anxiety, vol. 36, no. 10, pp. 950–959, 2019.
View at: Publisher Site | Google Scholar
M. Rhemtulla, R. V. Bork, and A. O. J. Cramer, “Cross-lagged network models,” Multivariate Behavioral Research, 2017, https://www.semanticscholar.org/paper/Cross-Lagged-Network-Models-Rhemtulla-Bork/fbb8ab2cea6c60c65b3254aeb2f0325009ac0f59.
View at: Google Scholar
C. J. Funkhouser, A. A. Chacko, K. A. Correa, A. J. E. Kaiser, and S. A. Shankman, “Unique longitudinal relationships between symptoms of psychopathology in youth: a cross-lagged panel network analysis in the ABCD study,” Journal of Child Psychology and Psychiatry, vol. 62, no. 2, pp. 184–194, 2021.
View at: Publisher Site | Google Scholar
M. Rubin, A. Bicki, S. Papini, J. A. J. Smits, M. J. Telch, and J. S. Gray, “Distinct trajectories of depression symptoms in early and middle adolescence: preliminary evidence from longitudinal network analysis,” Journal of Psychiatry Research, vol. 142, pp. 198–203, 2021.
View at: Publisher Site | Google Scholar
WHO, “Novel coronavirus (2019-nCoV) situation report-10,” 2020, https://www.who.int/docs/default-source/coronaviruse/situation-reports/20200130-sitrep-10-ncov.pdf?sfvrsn=d0b2e480_2.
View at: Google Scholar
Y. Zhang, D. Wang, J. Zhao et al., “Insomnia and other sleep-related problems during the remission period of the COVID-19 pandemic: a large-scale survey among college students in China,” Psychiatric Research, vol. 304, article 114153, 2021.
View at: Publisher Site | Google Scholar
K. Kroenke, R. L. Spitzer, and J. B. Williams, “The PHQ-9: validity of a brief depression severity measure,” Journal of General Interna Medicine, vol. 16, no. 9, pp. 606–613, 2001.
View at: Publisher Site | Google Scholar
N. Du, K. Yu, Y. Ye, and S. Chen, “Validity study of patient health questionnaire-9 items for internet screening in depression among Chinese university students,” Asia-Pacific Psychiatry, vol. 9, no. 3, article e12266, 2017.
View at: Publisher Site | Google Scholar
W. Wang, Q. Bian, Y. Zhao et al., “Reliability and validity of the Chinese version of the patient health questionnaire (PHQ-9) in the general population,” General Hospital Psychiatry, vol. 36, no. 5, pp. 539–544, 2014.
View at: Publisher Site | Google Scholar
Z. Ma, J. Zhao, Y. Li et al., “Mental health problems and correlates among 746 217 college students during the coronavirus disease 2019 outbreak in China,” Epidemiology and Psychiatric Sciences, vol. 29, article e181, 2020.
View at: Publisher Site | Google Scholar
J. Friedman, T. Hastie, and R. Tibshirani, “Regularization paths for generalized linear models via coordinate descent,” Journal of Statistical Software, vol. 33, no. 1, pp. 1–22, 2010.
View at: Publisher Site | Google Scholar
S. Epskamp, D. Borsboom, and E. I. Fried, “Estimating psychological networks and their accuracy: a tutorial paper,” Behavior Research Methods, vol. 50, no. 1, pp. 195–212, 2018.
View at: Publisher Site | Google Scholar
S. Epskamp, L. J. Waldorp, R. Mõttus, and D. Borsboom, “The Gaussian graphical model in cross-sectional and time-series data,” Multivariate Behavioral Research, vol. 53, no. 4, pp. 453–480, 2018.
View at: Publisher Site | Google Scholar
S. Epskamp, A. O. J. Cramer, L. J. Waldorp, V. D. Schmittmann, and D. Borsboom, “qgraph: network visualizations of relationships in psychometric data,” Journal of Statistical Software, vol. 48, no. 4, 2012.
View at: Publisher Site | Google Scholar
Y. Liang, L. Yang, J. Xi, and Z. Liu, “The unique role of sleep problems among symptoms of posttraumatic stress disorder: a cross-lagged panel network analysis,” Acta Psychologica Sinica, vol. 54, no. 10, p. 1206, 2022.
View at: Publisher Site | Google Scholar
R. J. McNally, “Can network analysis transform psychopathology?” Behaviour Research and Therapy, vol. 86, pp. 95–104, 2016.
View at: Publisher Site | Google Scholar
G. A. Bonanno, “Loss, trauma, and human resilience: have we underestimated the human capacity to thrive after extremely aversive events?” The American Psychologist, vol. 59, no. 1, pp. 20–28, 2004.
View at: Publisher Site | Google Scholar
L. Ballester, I. Alayo, G. Vilagut et al., “Predictive models for first-onset and persistence of depression and anxiety among university students,” Journal of Affective Disorders, vol. 308, pp. 432–441, 2022.
View at: Publisher Site | Google Scholar
I. C. H. M. Bogers, M. Zuidersma, M. L. Boshuisen, H. C. Comijs, and R. C. Oude Voshaar, “The influence of thoughts of death and suicidal ideation on the course of depression in older depressed patients,” International Journal of Geriatric Psychiatry, vol. 32, no. 8, pp. 882–891, 2017.
View at: Publisher Site | Google Scholar
C. J. Bryan, C. E. Morrow, N. Etienne, and B. Ray-Sannerud, “Guilt, shame, and suicidal ideation in a military outpatient clinical sample,” Depression and Anxiety, vol. 30, no. 1, pp. 55–60, 2013.
View at: Publisher Site | Google Scholar
K. Hawton and K. Van Heeringen, The International Handbook of Suicide and Attempted Suicide, John Wiley & Sons, Ltd., 2000.
View at: Publisher Site
K. A. Van Orden, T. K. Witte, K. C. Cukrowicz, S. R. Braithwaite, E. A. Selby, and T. E. Joiner, “The interpersonal theory of suicide,” Psychological Review, vol. 117, no. 2, pp. 575–600, 2010.
View at: Publisher Site | Google Scholar
E. D. Klonsky, A. M. May, and B. Y. Saffer, “Suicide, suicide attempts, and suicidal ideation,” Annual Review of Clinical Psychology, vol. 12, no. 1, pp. 307–330, 2016.
View at: Publisher Site | Google Scholar
M. A. Lau, Z. V. Segal, and J. M. Williams, “Teasdale’s differential activation hypothesis: implications for mechanisms of depressive relapse and suicidal behaviour,” Behaviour Research and Therapy, vol. 42, no. 9, pp. 1001–1017, 2004.
View at: Publisher Site | Google Scholar
N. Slee, N. Garnefski, R. van der Leeden, E. Arensman, and P. Spinhoven, “Cognitive-behavioural intervention for self-harm: randomised controlled trial,” British Journal of Psychiatry, vol. 192, no. 3, pp. 202–211, 2008.
View at: Publisher Site | Google Scholar
L. Staner, “Comorbidity of insomnia and depression,” Sleep Medicine Review, vol. 14, no. 1, pp. 35–46, 2010.
View at: Publisher Site | Google Scholar
I. Vargas and M. L. Perlis, “Insomnia and depression: clinical associations and possible mechanistic links,” Current Opinion in Psychology, vol. 34, pp. 95–99, 2020.
View at: Publisher Site | Google Scholar
E. Hertenstein, B. Feige, T. Gmeiner et al., “Insomnia as a predictor of mental disorders: a systematic review and meta-analysis,” Sleep Medicine Review, vol. 43, pp. 96–105, 2019.
View at: Publisher Site | Google Scholar
R. Mirchandaney, R. Barete, and L. D. Asarnow, “Moderators of cognitive behavioral treatment for insomnia on depression and anxiety outcomes,” Current Psychiatry Reports, vol. 24, no. 2, pp. 121–128, 2022.
View at: Publisher Site | Google Scholar
T. Cheung, Y. Jin, S. Lam et al., “Network analysis of depressive symptoms in Hong Kong residents during the COVID-19 pandemic,” Psychiatry, vol. 11, no. 1, 2021.
View at: Publisher Site | Google Scholar
A. Hoffart, S. U. Johnson, and O. V. Ebrahimi, “The network of stress-related states and depression and anxiety symptoms during the COVID-19 lockdown,” Journal of Affective Disorders, vol. 294, pp. 671–678, 2021.
View at: Publisher Site | Google Scholar
P. Ferentinos, V. Kontaxakis, B. Havaki-Kontaxaki, D. Dikeos, and L. Lykouras, “Psychometric evaluation of the fatigue severity scale in patients with major depression,” Quality of Life Research, vol. 20, no. 3, pp. 457–465, 2011.
View at: Publisher Site | Google Scholar
J. Knapen, D. Vancampfort, Y. Morien, and Y. Marchal, “Exercise therapy improves both mental and physical health in patients with major depression,” Disability and Rehabilitation, vol. 37, no. 16, pp. 1490–1495, 2015.
View at: Publisher Site | Google Scholar
A.-M. Isvoranu, S. Epskamp, and M. W. L. Cheung, “Network models of posttraumatic stress disorder: a meta-analysis,” Journal of Abnormal Psychology, vol. 130, no. 8, pp. 841–861, 2021.
View at: Publisher Site | Google Scholar
J. Gao, P. Zheng, Y. Jia et al., “Mental health problems and social media exposure during COVID-19 outbreak,” PLoS One, vol. 15, no. 4, article e0231924, 2020.
View at: Publisher Site | Google Scholar
C. Gonzalez-Sanguino, B. Ausin, M. AngelCastellanos et al., “Mental health consequences during the initial stage of the 2020 coronavirus pandemic (COVID-19) in Spain,” Brain Behavior and Immunity, vol. 87, pp. 172–176, 2020.
View at: Publisher Site | Google Scholar
W. Tang, T. Hu, B. Hu et al., “Prevalence and correlates of PTSD and depressive symptoms one month after the outbreak of the COVID-19 epidemic in a sample of home-quarantined Chinese university students,” Journal of Affective Disorders, vol. 274, pp. 1–7, 2020.
View at: Publisher Site | Google Scholar
Y. Wang, Z. Hu, Y. Feng, A. Wilson, and R. Chen, “Changes in network centrality of psychopathology symptoms between the COVID-19 outbreak and after peak,” Molecular Psychiatry, vol. 25, no. 12, pp. 3140–3149, 2020.
View at: Publisher Site | Google Scholar
G. A. Bonanno, “Resilience in the face of potential trauma,” Current Directions in Psychological Science, vol. 14, no. 3, pp. 135–138, 2005.
View at: Publisher Site | Google Scholar
G. A. Bonanno, A. D. Mancini, J. L. Horton et al., “Trajectories of trauma symptoms and resilience in deployed US military service members: prospective cohort study,” British Journal of Psychiatry, vol. 200, no. 4, pp. 317–323, 2012.
View at: Publisher Site | Google Scholar
J. de Ron, E. I. Fried, and S. Epskamp, “Psychological networks in clinical populations: investigating the consequences of Berkson’s bias,” Psychological Medicine, vol. 51, no. 1, pp. 168–176, 2021.
View at: Publisher Site | Google Scholar
P. Schlechter, J. H. Hellmann, R. J. McNally, and N. Morina, “The longitudinal course of posttraumatic stress disorder symptoms in war survivors: insights from cross-lagged panel network analyses,” Journal of Traumatic Stress, vol. 35, no. 3, pp. 879–890, 2022.
View at: Publisher Site | Google Scholar
N. H. Zainal and M. G. Newman, “A cross-lagged prospective network analysis of depression and anxiety and cognitive functioning components in midlife community adult women,” Psychological Medicine, vol. 53, no. 9, pp. 4160–4171, 2023.
View at: Publisher Site | Google Scholar

Copyright

Copyright © 2023 Zijuan Ma et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

PDF Download Citation

Download other formats

Order printed copies

Views

428

Downloads

327

Citations