key: cord-0870779-8tr0422y
authors: Yang, Fan; Fu, Mingqi; Huang, Ning; Ahmed, Farooq; Shahid, Muhammad; Zhang, Bo; Guo, Jing; Lodder, Paul
title: Network analysis of COVID-19-related PTSD symptoms in China: the similarities and differences between the general population and PTSD sub-population
date: 2021-12-06
journal: European journal of psychotraumatology
DOI: 10.1080/20008198.2021.1997181
sha: a72da3107cea6b25d4025c4687640ebaf18b160f
doc_id: 870779
cord_uid: 8tr0422y

BACKGROUND AND OBJECTIVES: Prevalent Post-traumatic Stress Disorder (PTSD) negatively affected individuals during the COVID-19 pandemic. Using network analyses, this study explored the construct of PTSD symptoms during the COVID-19 pandemic in China to identify similarities and differences in PTSD symptom network connectivity between the general Chinese population and individuals reporting PTSD. METHODS: We conducted an online survey recruiting 2858 Chinese adults. PTSD symptoms were measured using the PCL-5 and PTSD was determined according to the DSM-5 criteria. RESULTS: In the general population, self-destructive/reckless behaviours were on average the most strongly connected to other PTSD symptoms in the network. The five strongest positive connections were found between 1) avoidance of thoughts and avoidance of reminders, 2) concentration difficulties and sleep disturbance, 3) negative beliefs and negative trauma-related emotions, 4) irritability/anger and self-destructive/reckless behaviours, and 5) hypervigilance and exaggerated startle responses. Besides, negative connections were found between intrusive thoughts and trauma-related amnesia and between intrusive thoughts and self-destructive/reckless behaviours. Among individuals reporting PTSD, symptoms such as flashbacks and self-destructive/reckless behaviours were on average most strongly connected to other PTSD symptoms in the network. The five strongest positive connections were found between 1) concentration difficulty and sleep disturbance, 2) intrusive thoughts and emotional cue reactivity, 3) negative beliefs and negative trauma-related emotions, 4) irritability/anger and self-destructive/reckless behaviour, and 5) detachment and restricted affect. In addition, a negative connection was found between intrusive thoughts and self-destructive/reckless behaviours. CONCLUSION: Our results indicate similarly positive connections between concentration difficulty and sleep disturbance, negative beliefs and negative trauma-related emotions, and irritability/anger and self-destructive/reckless behaviours in the general and PTSD-reported populations. We argue that self-destructive/reckless behaviours are a core symptom of COVID-19 related PTSD, worthy of more attention in future psychiatric programmers.

Coronavirus disease 2019 (COVID-19) was defined a pandemic by the World Health Organization on 11 March 2020 and is one of the greatest public health challenges of the 21st century. Its impact on societies worldwide is multidimensional , ranging from widespread economic consequences (Pan & Yue, 2021) to substantial psychological problems such as post-traumatic stress disorder (PTSD), depression, anxiety, etc. (Haderlein, Wong, Yuan, Llorente, & Washington, 2020; Liang et al., 2020; Rutherford et al., 2021) . The disease was defined as a collective trauma by some researchers (Horesh & Brown, 2020; Tang et al., 2020; Venkateswaran & Hauser, 2020) , with an impact persisting in the community beyond that of the individuals experiencing symptoms. The increased presence of PTSD-related symptoms during the COVID-19 outbreak has been established both in survivors of COVID-19 as well as in health-care workers (see Salehi et al. (2021) for a recent meta-analysis). However, much remains to be learned about how individual PTSD symptoms relate to each other. Insight into which symptoms play a key role during the COVID-19 pandemic may facilitate targeted intervention for individuals with PTSD due to COVID-19. Therefore, it is crucial to investigate the structure of PTSD symptoms among populations exposed to the COVID-19.

PTSD consisted of a series of symptoms, including avoidance, intrusion, excessive arousal, emotional numbing, etc. (Deja et al., 2006) . Previous studies have often used reflective models that assume the existence of a latent PTSD variable that explains individual differences in the observed PTSD item scores (Fried, 2015; Schmittmann et al., 2013) . As a result, these analyses are typically focused on a total PTSD score, thereby ignoring causal interactions between individual symptoms. Indeed, a reflective latent variable model assumes that individual symptoms do not causally influence each other when conditioning on the presence of PTSD (Hofmann, Curtiss, & McNally, 2016) . Recent advanced in psychometrics have brought forwards a different approach to modelling psychopathology, called network analysis. According to network theory, psychological constructs can better be seen as networks of symptoms that causally influence each other (Borsboom, 2017) . Network analysis can discover connections among symptoms in the same network and identify each symptom's centrality. Thus, it could be an ideal method to explore the construct of psychological problems (Klein, Harris, Björgvinsson, & Kertz, 2020; Levinson et al., 2017; Montazeri, de Bildt, Dekker, & Anderson, 2020) .

Indeed, some have argued that PTSD can best be conceptualized as a system of causally interacting symptoms (McNally, 2012; McNally et al., 2015) . In addition, earlier studies have indicated that the PTSD construct can better be seen as a continuum rather than a discrete category (Ruscio, Rusciob, & Keane, 2002) . Studying the connections among PTSD symptoms could help us understand which specific symptoms contribute to individual differences on this PTSD continuum. This further highlights the importance of using network analysis to investigate both the PTSD symptom centrality and network connectivity (Epskamp & Fried, 2016; Fried, 2015) .

Some earlier studies have applied network analysis to PTSD symptoms and found strong connections between hypervigilance and exaggerated startle response and between flashbacks and nightmares (Armour, Fried, Deserno, Tsai, & Pietrzak, 2017; Birkeland & Heir, 2017; Bryant et al., 2017; Cao et al., 2019; Spiller et al., 2017) . However, these studies did not result in definitive conclusions regarding which symptoms were most central in the network. The identified connections varied across studies due to differential exposures (Asmundson, Stapleton, & Taylor, 2004) . Recently, a cross-sectional study indicated that the network of PTSD symptoms related to COVID-19 differed from the structure observed during previous studies in the general population (Jiang, Ren, Yu, Tan, & Shi, 2020) . Specifically, this study revealed that avoidance of thoughts and avoidance of reminders had a strong connection and the most central symptom was self-destructive/reckless behaviour in a PTSD network related to COVID-19. However, this study was limited by its relatively small sample (N = 338), possibly resulting in falsely suggesting the absence of the smaller symptom connections in the network. Their sample also primarily involved women. Furthermore, their network of PTSD symptoms was estimated in a sample of healthy individuals, making it unclear whether this structure generalizes to those suffering from PTSD.

Notably, though psychological research often involves specific hypotheses that are stated in the introduction section of the study, the current network analysis method was not used to test hypotheses, but rather an exploratory study to explore the PTSD symptom network structure after COVID-19 exposure, both in the general population and a PTSD subgroup. Thus, we did not provide a concise hypothesis and our study aimed to use network analysis to discover the potential structure of COVID-19 related PTSD symptoms. In our study, we were both interested in the edge weights as well as the centrality index node strength. The edge weights showed the exact connections among symptoms, which could help us figure out which two symptoms had strong connections after adjusting for other symptoms in the network. And the centrality could help us discover which symptom was the most central, which means a symptom was most strongly connected with other symptoms in the network. By relating these two types of findings to other studies, we can evaluate whether our results are consistent with current theories in the PTSD field. For similar analysis methods please see some references in our study, i.e. works done by Armour et al. (2017) and Wang and Tang (2020) , etc. We investigated the PTSD symptom network structure in 2858 individuals exposed to the COVID-19 outbreak and presented our results in detail in the Results section and discussed them in the Discussion section instead. Moreover, our analysis compares the structure found in the general population to the structure found in a PTSD subgroup (determined using the PCL-5 and DSM-5 criteria). This allows us to investigate whether the network connectivity and symptom centrality generalizes to that from the general population to the PTSD patient population. Studying differences in these dynamics may benefit future prevention, intervention, or treatment of PTSD related to COVID-19.

Participants were recruited from an online study 'the psychological status of Chinese adults during the COVID-19', which was conducted from February 1st to 10th, 2020. At the time of collecting data (1 February 2020), China reported 11,791 confirmed cases of COVID-19, 260 deaths due to COVID-19 and 247 recovered cases of COVID-19. At that time, lockdowns and stay-at-home requirements were in place in China. Therefore, the sample was collected using a web-based platform (https://www.wjx.cn/app/sur vey.aspx). Convenience sampling and snowball sampling were used to recruit participants. Initially, several key contact participants in specific groups such as a chief nurse, class tutor, or company manager, were recruited. Then, these contact participants were requested to distribute the questionnaire to their contacts through WeChat (a widely used communication tool in China). All participants completed an online questionnaire that assessed various sociodemographic, psychiatric, COVID-19 related, and health-focused variables. The surveying platform collected only participants who could answer all the questions in this questionnaire. Therefore, there is no missing data in our study. Participants met our inclusion criteria if they were aged 18 years or older and provided online informed consent. Exclusion criteria were any conditions that affected the quality of the questionnaire response, for example, less than 10 minutes of response time, confusion of logic, etc. In total, 2858 valid samples were collected, and 558 participants met the diagnostic criteria of DSM-5. The characteristics of participants were shown in Table 1 . According to previous studies (Armour et al., 2016; Jiang et al., 2020) , covariates such as depression, anxiety, suicide ideation, social support, sex, and quality of life had no significant impact on the network of PTSD symptoms. Therefore, we did not consider them during the analysis stage. Of note, there was no question associated with skip-structures, which was mentioned as a requirement by Burger et al. (2020) in their developed reporting standards of network analysis. Nevertheless, we reported our results according to these standards. Based on this, a diagnosis of PTSD currently requires that trauma survivors endorse a minimum of six symptoms (at least 1 IN, 1 AV, 2 NACM, and 2 AAR), except reporting significant functional impairment and the persistence of symptoms for more than one month (American Psychiatric Association, A. P., & American Psychiatric Association, 2013). Therefore, this study identified two target groups, i.e. the general population and the PTSD subgroup (as shown in Table 1 ) by these criteria. Of note, there was one question in our questionnaire, 'during the past year, did you experience severe trauma like a car accident, nature disaster, death of a family member?' Based on this question, we conducted analysis to assess whether the previous exposure would have impacted on the PTSD symptoms and the results showed no significant influence (see in Supplementary for details). Also, before participants responded to these questions, we indicated in our instruction that they should 'assess the impact of COVID-19 event on your life during the past month', thus the criterium A of PTSD was the COVID-19 event in this study.

A set of questions was used to assess the COVID-19 exposure: 1) were you infected with COVID-19, 2) have you had close contact with the infected person, 3) were your family members infected with the COVID-19, 4) did one or more of your family members have close contact with the infected person, 5) were your relatives or friends infected with COVID-19, 6) did your relatives or friends had close contact with the infected person of COVID-19, 7) did your neighbours infected, 8) Is there an infected person in your community. All eight questions contain two response options: 'Yes' and 'No.' We regarded the option 'Yes' as the exposure to COVID-19 and summed all items up to assess the level of traumatic exposure (range = 0-8). In our network analysis, we used the total scores of all items as a measure of COVID-19 exposure.

All data analysis was conducted by R (version 4.1.0). There were three main reasons why we used the network analysis method: first, it could take all the symptoms into one estimation procedure and discover the connections between PTSD symptoms; second, it could indicate which symptom was the most central, which could also be regarded as the key symptom and should be paid more attention; in the end, it could visualize the analysis result, which might be easier to be understood. Due to this method, we could discover the potential connections among symptoms and compare these results with the previous study (network study or another study), which might help us better know the associations among symptoms and we might use these results to develop new ways of prevention or treatment (i.e. we could target key symptoms for treatment, in this way the treatment might be more effective). However, we admit the implications of our findings to clinical practice could be limited because our network analyses are based on a sample with specific characteristics (i.e. COVID-19 exposures, nationality, age, gender, etc.). Despite this, future systematic reviews or meta-analyses could be used to compare our results with those of other studies performed on different samples to identify patterns that apply broadly across different contexts and to evaluate how exactly our study contributes to the PTSD field.

We estimated the structure of networks via two stages with the R-package qgraph (Epskamp & Fried, 2016) . In the first phase, all analyses were conducted on the general population sample. First, we included all 20 PTSD symptoms in the estimation procedure (A1, Figure 1 ). Then, we added the exposure to COVID-19 total score as an additional node in the network (A2, Figure 2 ). These two figures visualize the multivariate dependencies of the data. In this process, a Gaussian Graphical Model was used to estimate the 190 pairwise association parameters in A1 and 210 pairwise association parameters in A2. These connections or edges between nodes range from −1 to 1 and can be interpreted as partial correlation coefficients, indicating the relationship between nodes A and B after controlling for the influence of all other nodes in the network. We used the least absolute shrinkage and selection operator (LASSO) (Tibshirani, 1996) that sets very small edges to zero, thus minimizing falsepositive connections. Besides, the networks were estimated based on polychoric correlations because PTSD symptoms and exposure to COVID-19 were measured on an ordered-categorical scale. Last of all, we removed the covariate COVID-19 exposure from the adjacency matrix of the 21-node network (A2), which results in the connections among the 20 PTSD symptoms controlling for the exposure to the COVID-19 (A2*). A2* consisted of the matrix of the 20-node network. Then, we subtracted this modified adjacency matrix A2* from the adjacency matrix of A1 (the 20 PTSD symptoms, not controlling for the covariate COVID-19 exposure) to derive a delta network (A3, see in the Supplementary Materials), which could be compared to 'A1 minus A2*'. At last, this delta network, A3, contains the change of A1 upon including covariate; thus, we could examine the impact of the covariates on the associations between 20 PTSD symptoms. Of note, the exact numbers of these edges in A3 were meaningless to some degree. The aim of this analysis was to visualize the differences between networks, a method also used by other researchers (McNally et al., 2015) .

In the second network analysis phase, we repeated the procedure above on the PTSD sample. We got another three networks here: P1 (the 20 PTSD symptoms, Figure 3 ), P2 (the 20 PTSD symptoms and exposure to the COVID-19, Figure 4 ), and P3 (delta network, see in the Supplementary materials). At the end of the second phase, we examined the correlation between the A1 and P1 networks (correlation coefficient: 0.64) to test if the PTSD symptoms network among the general population and the PTSD subgroup were differential and correspondingly formed the delta network AP1 (see the Supplementary materials).

Centrality estimation was used to investigate the strength of each node, indicating which symptoms were most informative for PTSD. Often, there were three graphtheoretical centrality measures (Opsahl, Agneessens, & Skvoretz, 2010) : node strength, closeness centrality, and betweenness centrality. Node strength calculates all edges of one symptom with all other symptoms, estimating the strength of a node connected with the network directly. Closeness centrality sums the inverse of all shortest path lengths of one node with all other nodes, indicating how strongly a node is connected with the network indirectly. And betweenness centrality represents the frequency of one node lying on the shortest paths of any other two symptoms, indicating the ability of one node of connecting all other nodes in the network.

Of note, recent studies indicated that the betweenness centrality and closeness centrality estimates tend to be unstable (Bringmann et al., 2019; Hallquist, Wright, & Molenaar, 2021). Therefore, we focused on the node strength and its accuracy and stability in this study.

To cope with the unclear stability and accuracy in edge weight estimates (Epskamp & Fried, 2016) , we bootstrapped (1000 iterations) the edge weights , further allowing a test for significant differences between edge weights based on between the bootstrapped 95% confidence intervals. Also, we took the correlation stability (CS)-coefficient to assess the strength centrality index (Costenbader & Valente, 2003) . A CScoefficient should be at least 0.25 for the centrality to be stable, preferably above 0.5 (Epskamp, Borsboom, & Fried, 2018) .

Positive edges were printed in blue, whereas negative ones were in red. Besides, the stronger a connection was, the thicker and more saturated it would be. We used the Fruchterman-Reingold algorithm (Fruchterman & Reingold, 1991 ) that places nodes with stronger (and/or more) connections more closely together. We set the maximum edge value across all networks to 0.45 (Armour et al., 2016) , the strongest edge identified across networks. Thus, the saturation and thickness of edges could be compared directly throughout networks. Also, we used a minimum edge weight of 0.03 in all networks to enhance the interpretability of the graphs.

To compare the networks between the general population and the PTSD subgroup population, we conducted the network comparison test to compare the networks on four aspects: network invariance, global strength invariance, edge invariance and centrality invariance (Borkulo et al., 2017) . The network invariance and the global strength invariance were used to assess the differences between the whole network structure, while the edge invariance was applied to test the differences of edges and the centrality invariance was used to discover the invariance of the strength centrality between networks. All analysis was conducted by R-package 'Networkcomparisontest' and details about this method could be found in Borkulo et al.'s work (2017) .

Of note, all analytic codes of our study were available in Supplementary named 'analytic codes'.

Participants ranged in age at the time of assessment from 18 to > 70 years old and the females accounted for 53.6% of the sample. The average PCL-5 score was 17.15 (SD = 17.64), and a total of 558 participants (19.52%) had reached the criteria of PTSD according to PLC-5 and DSM-5. Among individuals reporting PTSD, the average PCL-5 score was 46.74 (SD = 12.57). 29.3% (N = 837) of the general population had been exposed to the COVID-19, whereas 33.2% (N = 224) of the individuals with PTSD had an exposure history. Table 2 demonstrated the details about each of the PTSD symptoms. thoughts (B1) and self-destructive/reckless emotions (E2) and that between intrusive thoughts (B1) and trauma-related amnesia (D1), both of which were negative, all other edges were positive. Some symptoms belonging to the same symptom group showed a strong connection: the avoidance of the thoughts (C1) and the avoidance of the reminders (C2), the difficulty concentrating (E5), and the sleep disturbance (E6), for example. Also, the avoidance of the thoughts (C1) and physiological cue reactivity (B5) showed considerable connection though they did not belong to the same clinical group. The five strongest edges in the network emerged between avoidance of thoughts (C1) and avoidance of reminders (C2); difficulty concentrating (E5) and sleep disturbance (E6); negative beliefs (D2) and negative trauma-related emotions (D4); irritability/anger (E1) and selfdestructive/reckless behaviour (E2); hypervigilance (E3) and exaggerated startle response (E4). Figure 2 showed the PTSD symptom network structure in the general population sample after adding exposure to COVID-19 to the network (A2). Though the exposure showed a positive connection with some symptoms, such as irritability/anger (E1) and a negative connection with self-destructive/reckless emotions (E2), the coefficient between network A1 and A2* was 0.9986, indicating high correlation. In addition, the sum of edges in this network reduced from 10.83 to 10.76 after controlling for exposure. Besides, the delta network A3 (available in the Supplementary materials) was nearly empty and featured few very weak edges, with the strongest edge weight being only 0.013. In conclusion, the estimated PTSD symptoms network structure is largely unaffected by whether COVID-19 exposure is included in the network, indicating the connections might be robust and convincible in the general population sample.

Similar to the results of the general population sample, Figure 3 represents the PTSD symptom network structure in the PTSD group (P1) and Figure 4 showed the network structure after adding exposure to COVID-19 to the network (P2). Though some connections were of equal sign and strength as those in network A1 (e.g. the difficulty concentrating (E5) and the sleep disturbance (E6)), some connections changed their strength (e.g. the avoidance of the thoughts (C1) and the avoidance of the reminders (C2)), or even became totally different (e.g. the strength between intrusive thoughts (B1) and trauma-related amnesia (D1) is 0 in P1). Although exposure to COVID-19 showed different connections with the symptoms from A1 as shown in P2, the correlation between networks P1 and P2* is 0.9981, and the sum of edges of 20 symptoms was reduced from 10.01 to 9.96 once controlling for exposure, indicating the exposure 'explained away' only about 0.50% of the connectivity of the PTSD symptoms. In sum, the estimated PTSD symptoms network structure is largely unaffected by whether exposure to COVID-19 is included in the network, indicating the connections might be robust and convincible in the PTSD subgroup sample.

Furthermore, considering the network is robust and COVID-19 exposure had little influence on the network, we examined the correlation between network A1 and P1 to investigate the differences between the general population and the PTSD population. The correlation was 0.6356 and the delta network's (AP1) sum of edges of 20 symptoms was 7.11, with an average edge weight of 0.037 and a maximum edge weight of 0.38 (the connection between the avoidance of the thoughts (C1) and the avoidance of the reminders (C2)). These results indicated that there were considerable differences between the network in the general group and the PTSD group. PTSD symptom network (see above Networks), we used network A1 to investigate the centrality. The centrality estimates were substantially interrelated: correlation of 0.70 between closeness and betweenness, correlation of 0.56 between closeness and node strength and correlation of 0.77 between node strength and betweenness. Because of these high intercorrelations and as mentioned above that closeness and betweenness estimates tend to be more unstable (Bringmann et al., 2019; Hallquist et al., 2021) , we focused our interpretation of the most relevant symptoms on node strength centrality on behalf of this study. As shown in Figure 5 , the five nodes with the highest node strength were self-destructive/reckless behaviour (E2), intrusive thoughts (B1), flashbacks (B3), negative trauma-related emotions (D4) and exaggerated startle response (E4), while, the least central node, which had the few and weak connections with other nodes, was detachment (D6). All these results fitted the network in Figure 1 . Similarly, Figure 5 (b) presents the standardized node strength centrality estimates in the PTSD subgroup. In this group, centrality estimates were also substantially interrelated: correlation of 0.72 between closeness and betweenness, correlation of 0.61 between closeness and node strength, correlation of 0.78 between node strength and betweenness. As Figure 5 (b) showed, the five nodes with the highest node strength were flashbacks (B3), self-destructive/reckless behaviour (E2), intrusive thoughts (B1), negative trauma-related emotions (D4), and difficulty concentrating (E5), while, the least two central nodes, which had the few and weak connections with other nodes, were nightmares (B2) and sleep disturbance (E6). All these results fitted the network in Figure 3 .

The accuracy and stability of the 20 PTSD symptoms network (A1) in the general group were examined and the results are presented in Figure 6 . Figure 6(a) shows the bootstrapped edge weights, which revealed that 95% confidence intervals for the edge weights were mostly overlapping, indicating considerable accuracy was estimated in the network. And Figure 6 (b) represented the subset bootstrap and the CS-coefficient was 0.75 for node strength. As mentioned in the Methods section, the coefficient could not be below 0.25 and preferably above 0.5. Therefore, this network showed great stability. To get details about accuracy and stability analyses for this network, readers could view the Supplementary material, including edge weights significance tests (testing for significant differences for all edges) and centrality difference tests (testing for centrality differences for all nodes). Similarly, the accuracy and stability of the 20 PTSD symptoms network (P1) in the PTSD group were examined and results were presented in Figures 6(c and  d) . Figure 6 (c) showed that the P1 had a considerable accuracy and Figure 6 (d) indicated the P1, with a CScoefficient of 0.36 which was beyond the 0.25, had considerable stability as well.

Visual inspection of networks A1 and P1 revealed that the network estimated in the general population might be denser than the PTSD subgroup -for the density of network of general population was 0.61 and was 0.53 of PTSD subgroup. Also, the negative connection between intrusive thoughts (B1) and trauma-related amnesia (D1) was absent in the PTSD subgroup network. The global node strength of the general group was 9.53, which was a bit higher than 9.40 of the PTSD subgroup. However, these differences could also be due to the larger sample of the general group as the power to detect an edge is higher in the general population sample than in the PTSD subgroup. Nevertheless, the results still indicated that the difference between those networks was statistically significant (S = 0.13, p = .006).

Besides, the omnibus test on invariance of network structure was conducted to investigate the differences in edges. Results indicated significant differences between the edges of the two networks (M = 0.29, p = .001). Therefore, we performed post hoc testing of all edges to investigate which edges(s) varied in networks. It had to be noted that we tested them all without applying a correction for multiple comparisons for our analyses were exploratory and had no a priori hypotheses. Results showed that 14 edges differed significantly between the two networks (See details in Supplementary) . Furthermore, we investigated the invariance of strength centrality and it turned out that three nodes differed significantly between the two networks: Flashbacks (B3), Traumarelated amnesia (D1) and Detachment (D6).

In this study, we applied a network analysis to the DSM-5 PTSD symptom scores of 2858 Chinese individuals exposed to the COVID-19 pandemic. We compared the networks between the general population and a PTSD subgroup (those who matched the criteria for PTSD according to . Our results showed a negligible influence of COVID-19 exposure on the PTSD symptom structure, both in the general population and in the PTSD subgroup. Note that this does not mean that COVID-19 exposure has no impact on PTSD because our results merely suggest that COVID-19 exposure did not change the relationship among the PTSD symptoms in the network.

In the current study, PTSD symptoms were generally positively connected to each other. The five strongest edges in the network emerged between avoidance of thoughts (C1) and avoidance of reminders (C2); difficulty concentrating (E5) and sleep disturbance (E6); negative beliefs (D2) and negative trauma-related emotions (D4); irritability/anger (E1) and self-destructive /reckless behaviour (E2); hypervigilance (E3) and exaggerated startle response (E4). And the negative connections were between intrusive thoughts (B1) and traumarelated amnesia (D1); intrusive thoughts (B1) and selfdestructive/reckless behaviour (E2).

For these connections, the strong connection between hypervigilance (E3) and exaggerated startle response (E4) was consistent with previous studies (Armour et al., 2017; Birkeland & Heir, 2017; Bryant et al., 2017; Jiang et al., 2020; McNally et al., 2015; Spiller et al., 2017) . Earlier research suggested a feedback loop between hypervigilance and exaggerated startle response (McNally et al., 2015) , which is supported by a Sensitization Model of PTSD, indicating that survivors may become sensitive to the threat and show an exaggerated startle response after exposure to traumatic occurrences (Stam, 2007) . In addition, the connection between difficulty concentrating (E5) and sleep disturbance (E6) may be explained by sleep disturbance arousing a series of adverse outcomes, including lack of concentrating (Harvey, 2009; Pagel, Forister, & Kwiatkowki, 2007) . Earlier research (Nixon & Kling, 2009 ) indicated that trauma-related emotions might act as a buffer to negative emotions, and may indirectly lead to a restructuring of negative beliefs, which could explain the connection between negative beliefs (D2) and negative trauma-related emotions (D4). The connection between irritability/anger (E1) and self-destructive/reckless behaviour (E2), may be explained by reckless behaviour being a way to release anger. Furthermore, anger is associated with self-control (Jensen-Campbell, Knack, Waldrip, & Campbell, 2007) , suggesting that the lower selfcontrol of angry individuals may also result in reckless behaviour. It is important to note that these connections matched the classification of the Hybrid Model of PTSD (Armour et al., 2015) , suggesting that our study's findings are consistent with Armour et al.'s work.

However, in contrast to previous studies, there were several negative connections between PTSD symptoms in our general population network. The negative connection between intrusive thoughts (B1) and trauma-related amnesia (D1) might be explained that PTSD could result in a deficit in the inhibitory processes in the memory (Cottencin et al., 2006) . Cottencin and colleagues conducted a directed forgetting study, indicating that PTSD patients could have a deficit in the inhibitory processes in the memory which could both reduce amnesia and contribute to the intrusive thoughts to some degree. Birkeland and Heir (2017) indicated that intrusive thoughts (B1) could bring patients to the trauma memory repeatedly, which could reduce the severity of the trauma-related amnesia (D1). In addition, Salkovskis and Campbell (1994) claimed that thought suppression might induce the occurrence of intrusive thoughts and as mentioned above, the reckless behaviour might release the emotions (Jensen-Campbell et al., 2007) and reduce these negative thoughts, which could explain the negative connection between intrusive thoughts (B1) and selfdestructive/reckless behaviour (E2).

Assessing the robustness of symptom centrality revealed that the most central symptoms did not differ substantially from each other in their centrality and should be considered roughly equally important. The most central symptoms were self-destructive/reckless behaviour (E2), intrusive thoughts (B1), flashbacks (B3), negative trauma-related emotions (D4), and exaggerated startle response (E4), and we concluded that they could be of greatest clinical significance in the normal population who experienced COVID-19 outbreak. It is noteworthy that these symptoms were similar to Jiang et al.'s study (the self-destructive/reckless behaviour is of highest node strength; Jiang et al., 2020) but different from Armour et al.'s study (the most central symptoms were negative trauma-related emotions, flashbacks, detachment, and physiological cue reactivity; 2017), which might result from the different type of trauma (Armour & Shevlin, 2009 ). This may be because Jiang et al.'s study and our study were both aimed at COVID-19 whereas Armour et al.'s study focused on trauma resulting from war. Moreover, differences between studies in the time interval between the traumatic event and moment the study was conducted may have contributed to this discrepancy (Jiang et al., 2020) and some networks of depression symptoms could vary across different stressful life events as well (Cramer, Borsboom, Aggen, & Kendler, 2012; Fried, 2015) . Notably, the interventions of the potential central symptoms might facilitate the relief of most PTSD symptoms.

In this study, PTSD symptoms were generally positively connected to each other. The five strongest edges in the network emerged between: difficulty concentrating (E5) and sleep disturbance (E6); intrusive thoughts (B1) and emotional cue reactivity (B4); negative beliefs (D2) and negative trauma-related emotions (D4); irritability/anger (E1) and self-destructive/reckless behaviour (E2); detachment (D6) and restricted affect (D7). And the negative connection was between intrusive thoughts (B1) and selfdestructive/reckless behaviour (E2).

For these connections, the strong connections between difficulty concentrating (E5) and sleep disturbance (E6), negative beliefs (D2) and negative trauma-related emotions (D4), irritability/anger (E1) and self-destructive /reckless behaviour (E2), and intrusive thoughts (B1) and self-destructive/reckless behaviour (E2) could be explained as above. In addition, an explanation for the connection between detachment (D6) and restricted affect (D7) may be that the PTSD patients disengage not only from negative emotions related to trauma but also from positive emotions after trauma due to the detachment symptoms, which might limit the effects on mood (Litz, Orsillo, Orsillo, & Weathers, 2000; Shepherd & Wild, 2014) . Besides, Armour et al. (2016) indicated that though the connection was only strong between nightmares and flashbacks, most re-experiencing symptoms were linked either directly or indirectly through other symptoms (e.g. nightmare was linked to emotional cue reactivity through flashbacks). Also, Jiang et al. (2020) discovered strong connections between intrusive thoughts and nightmares, and flashbacks and emotional cue reactivity. These results suggest a high correlation among intrusive symptoms, which might explain the connection between intrusive thoughts (B1) and emotional cue reactivity (B4) in our study. More importantly, such connections revealed a latent structure of PTSD symptoms related to COVID-19, which was consistent with the Hybrid Model of PTSD (Armour et al., 2015) , indicating mutual evidence between our study and Armour et al.'s work.

As for as the centrality strength was concerned, the most central symptoms were flashbacks (B3), self-destructive /reckless behaviour (E2), intrusive thoughts (B1), negative trauma-related emotions (D4), and difficulty concentrating (E5). The differences among studies could result in different trauma or the different intervals between exposures and studies as mentioned above as well (Cramer et al., 2012; Fried, 2015; Jiang et al., 2020) .

Our network comparisons suggest that there were similarities and differences in the network structures of the general population and the PTSD subgroup. For example, equally strong connections were found in both groups for the symptom sets difficulty concentrating (E5) and sleep disturbance (E6), negative beliefs (D2) and negative trauma-related emotions (D4), and irritability/anger (E1), and self-destructive /reckless behaviour (E2). However, the presence of negative connections differed between the two groups: In the general population intrusive thoughts (B1) and trauma-related amnesia (D1) were negatively connected, as were intrusive thoughts (B1) and selfdestructive/reckless behaviour (E2). In the PTSD group, intrusive thoughts (B1) was negatively connected with self-destructive/reckless behaviour (E2).

In terms of centrality strength, the top four strongest central symptoms were the same in the two groups, whereas the order changed between groups: in the general population self-destructive/reckless behaviour (E2) was most strongly connected to other symptoms in the network, while in the PTSD group this was flashbacks (B3). Such differences may be due to differences between the two population in the position of individuals on the PTSD continuum (Ruscio et al., 2002) . Therefore, our results indicate that on the one hand, the same connections between symptoms were considerably stable across groups, and on the other hand, these differences in strength indicated that symptoms might have different impacts on the general population and PTSD sub-population.

It has to be admitted that there were some limitations in our study: First, the robustness analysis showed a moderate uncertainty in the PTSD group regarding the estimation of the edge weights and centrality parameters. Therefore, future research could study the PTSD network structure in an even larger sample of PTSD patients to investigate whether our centrality and network connection estimates can be replicated. Second, the current study was crosssectional, which inhibits conclusions regarding causal influences between symptoms. This makes it unclear whether a particular symptom truly causes the occurrence of another symptom in the network. Therefore, prospective studies are needed to solve this problem. Third, the data was collected through a web-based and self-report questionnaire, which could limit the validity and reliability of our findings (Fried & Cramer, 2017) . However, the PCL-5 is a well-validated questionnaire and the reliability estimated in our current sample was excellent. Nevertheless, future research could replicate our network analysis based on PTSD symptom measurements in terms of specialized and structured interviews. Fourth, the CS-coefficient of PTSD subgroup was 0.36, suggesting suboptimal stability of the strength centrality index. Although this coefficient was larger than the minimum requirement of 0.25, any coefficient lower than 0.5 should be treated cautiously. Finally, the exposure was calculated with no weights in our study. During our analysis, we did not take the weights of the eight questions about exposure to COVID-19 into consideration, which might limit the extent to which this score represents the true exposure. Future studies could assign weights to the different exposures and take the severity of COVID-19 in a participant's living situation into consideration as well.

Despite the limitations, our study has potential strengths and implications as well. One strength of our study is that we overcome the limitation of both Jiang et al. (2020) and Armour et al.'s (2017) work by focusing on the differences in network structure between the general and clinical population. A second strength is that the connections of symptoms in both populations are fully consistent with the Hybrid Model of PTSD (Armour et al., 2015) , which could classify the clusters of PTSD better than DSM-5 (Armour, Müllerová, & Elhai, 2016) , indicating that our findings are consistent with Armour et al.'s work and revealing the potential structure of PTSD related to COVID-19.

Our results shed light on the prevention and intervention of PTSD related to COVID-19 in the future. One implication of our work is that specific PTSD symptoms could be targeted when treating patients suffering from PTSD due to COVID-19. Our study found high positive connections between 1) avoidance of thoughts and avoidance of reminders, 2) difficulty concentrating and sleep disturbance, 3) negative beliefs and negative traumarelated emotions, 4) irritability/anger and self-destructive /reckless behaviour, 5) hypervigilance and exaggerated startle response in the general population. And negative connections were found in the general population between intrusive thoughts and trauma-related amnesia, and intrusive thoughts and self-destructive/reckless behaviour. Therefore, clinicians could target related symptoms in treatment planning at the same time and contribute especially to the treatment of some symptoms that might be difficult to intervene on, such as difficulty concentrating. More specifically, because of the high connection between difficulty concentrating and sleep disturbance, we could target sleep disturbance in treatment instead, and the alleviation of this symptom could break down other highly related symptoms like difficulty concentrating. In addition, the treatment of the most central symptom, self-destructive /reckless behaviour in this network, could result in alleviation of other related symptoms in this network, and particularly prevent the general population from developing PTSD disorder. The negative connections also showed great importance when concentrating on the intervention of one symptom like intrusive thoughts in the general population. Specifically, it is essential to pay attention to trauma-related amnesia and self-destructive/reckless behaviour, since they were negatively connected with intrusive thoughts and might get worse along with the treatment of intrusive thoughts. Secondly, similar to our discussion above, the five strongest edges in the network in the PTSD subgroup emerged between the symptom sets: 1) difficulty concentrating and sleep disturbance; 2) intrusive thoughts and emotional cue reactivity; 3) negative beliefs and negative trauma-related emotions; 4) irritability/anger and self-destructive/reckless behaviour; 5) detachment and restricted affect. In the PTSD group a negative connection was found between intrusive thoughts and self-destructive /reckless behaviour. Taking those connections of symptoms into consideration can be important when intervening or treating PTSD related to COVID-19 in the clinical population. For instance, flashbacks and self-destructive/reckless behaviour could be targets for conducting an effective treatment. Moreover, the network generated from the general population suggests that the prevention of PTSD could be done by intervening or identifying the central symptoms in the early period. The network generated from the PTSD subgroup revealed more effective ways for treatment as well. Nevertheless, we emphasize that the cross-sectional nature of our study prevents conclusions regarding causal influences between symptoms. It remains unclear whether our centrality indices can be used as a measure of symptoms that can best be intervened upon. Future research could use longitudinal network analyses to identify whether the symptoms discussed above turn out to be causally efficacious.

This study explores the construct of PTSD symptoms under the COVID-19 pandemic in China and reveals its similarities and differences in symptom connectivity between the general Chinese population and individuals reporting PTSD. Results indicate similar positive connections betw een concentration difficulty and sleep disturbance, negative beliefs and negative trauma-related emotions, and irritability/anger and self-destructive/reckless behaviours in both the general and PTSD-reported populations. We therefore argue that self-destructive/reckless behaviours are a core symptom of COVID-19 related PTSD, which needs more attention in future psychiatric programmes.

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy reasons. For data request, please contact the corresponding author by email: jing624218@163.com.

No potential conflict of interest was reported by the author(s).

Diagnostic and statistical manual of mental disorders: DSM-5

A network analysis of DSM-5 posttraumatic stress disorder symptoms and correlates in US military veterans

A systematic literature review of PTSD's latent structure in the diagnostic and statistical manual of mental disorders: DSM-IV to DSM-5

Testing the dimensionality of PTSD and the specificity of the dysphoria factor

Dimensional structure of DSM-5 posttraumatic stress symptoms: Support for a hybrid anhedonia and externalizing behaviors model

Are avoidance and numbing distinct PTSD symptom clusters

Making connections: Exploring the centrality of posttraumatic stress symptoms and covariates after a terrorist attack

Comparing network structures on three aspects: A permutation test

A network theory of mental disorders

What do centrality measures measure in psychological networks

Acute and chronic posttraumatic stress symptoms in the emergence of posttraumatic stress disorder: A network analysis

Reporting Standards for Psychological Network Analyses in Cross-sectional Data. preprint

Sex differences in global and local connectivity of adolescent posttraumatic stress disorder symptoms

The stability of centrality measures when networks are sampled

Directed forgetting in PTSD: A comparative study versus normal controls

The pathoplasticity of dysphoric episodes: Differential impact of stressful life events on the pattern of depressive symptom inter-correlations

Social support during intensive care unit stay might improve mental impairment and consequently health-related quality of life in survivors of severe acute respiratory distress syndrome

Estimating psychological networks and their stability: A tutorial paper

A primer on estimating regularized psychological networks. arXiv, arXiv Preprint:160701367

A tutorial on regularized partial correlation networks

Moving Forward: Challenges and Directions for Psychopathological Network Theory and Methodology

Problematic assumptions have slowed down depression research: Why symptoms, not syndromes are the way forward

Graph drawing by force-directed placement

Association of PTSD with COVID-19 testing and infection in the veterans health administration

Problems with centrality measures in psychopathology symptom networks: Why network psychometrics cannot escape psychometric theory

The adverse consequences of sleep disturbance in pediatric bipolar disorder: Implications for intervention. Child and Adolescent Psychiatric Clinics of North America

A complex network perspective on clinical science

Traumatic stress in the age of COVID-19: A call to close critical gaps and adapt to new realities

Do big five personality traits associated with self-control influence the regulation of anger and aggression

A network analysis of post-traumatic stress disorder symptoms and correlates during the CoViD-19 pandemic

A network analysis of symptoms of obsessive compulsive disorder and depression in a clinical sample

The core symptoms of bulimia nervosa, anxiety, and depression: A network analysis

The effect of COVID-19 on youth mental health

Emotional processing in posttraumatic stress disorder

Mental disorders as causal systems: A network approach to posttraumatic stress disorder

The ontology of posttraumatic stress disorder: Natural kind, social construction, or causal system

Network analysis of behaviors in the depression and autism realms: Inter-relationships and clinical implications

Treatment of adult post-traumatic stress disorder using a future-oriented writing therapy approach

Node centrality in weighted networks: Generalizing degree and shortest paths

Adolescent sleep disturbance and school performance: The confounding variable of socioeconomics

Multidimensional effect of covid-19 on the economy: Evidence from survey data

The latent structure of posttraumatic stress disorder: A taxometric investigation of reactions to extreme stress

The COVID-19 pandemic as a traumatic stressor: Mental health responses of older adults with chronic PTSD

The prevalence of post-traumatic stress disorder related symptoms in Coronavirus outbreaks: A systematic-review and meta-analysis

Thought suppression induces intrusion in naturally occurring negative intrusive thoughts

Deconstructing the construct: A network perspective on psychological phenomena

Emotion regulation, physiological arousal and PTSD symptoms in trauma-exposed individuals

Symptoms of posttraumatic stress disorder in a clinical sample of refugees: A network analysis

PTSD and stress sensitisation: A tale of brain and body: Part 1: Human studies

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

Regression shrinkage and selection via the lasso

Living with PTSD amid a global pandemic

Perceived psychosocial health and its sociodemographic correlates in times of the COVID-19 pandemic: A community-based online study in China

The PTSD checklist for DSM-5 (pcl-5). Scale Retrieved from the National Center for PTSD

This study was supported by China National Social Science Foundation (20VYJ042), and The Fundamental Research Funds for the Central Universities (Number: BMU2021YJ029).

The Ethics Committee of Peking University Medical Center approved this study. All participants were informed of the survey's purpose and their rights to decline their information. Also, electronic informed consent was obtained from these survey participants.