Relations between PTSD symptom clusters and pain in three traumaexposed samples with pain

Objectives: Little is known about how the individual PTSD symptom clusters relate to intensity and interference of pain and whether these relationships differ across clinical groups. The present study examines relations between PTSD symptom clusters and pain in three trauma-exposed, unique clinical groups: 1) adults seeking treatment for chronic pain with current symptoms of PTSD, 2) trauma affected refugees seeking treatment for PTSD and chronic pain; and 3) individuals identified at admission to the emergency ward after whiplash injury. Methods: Network analysis was used to assess unique relations between pain intensity, pain interference, re-experiencing, avoidance, numbing, hyperarousal, depression, and anxiety separately in each sample. Links between PTSD clusters and pain were then compared within and between samples. Results: No within-group differences were identified for the linksbetweenpainandanyof PTSDclusters in the chronic pain and refugee groups. In the whiplash group, hyperarousal was more strongly related to pain than re-experiencing, avoidance, and numbing. Between group comparisons revealed a more pronounced relationship between hyperarousal and pain in the whiplash group, with no between-group differences between the chronic pain and refugee groups. Conclusions: The findings suggest that when depression and anxiety are accounted for, few unique associations are found between pain and the PTSD symptom clusters in trauma-exposed samples with pain, with the exception of a link between pain and hyperarousal in individuals with whiplash-related PTSD symptoms.


Introduction
There is a large body of evidence indicating that posttraumatic stress disorder (PTSD) is common among individuals with chronic pain and vice versa. Among the highest rates of comorbid PTSD and chronic pain are found in refugees (up to 87 %), particularly those exposed to physical torture [1,2]. Separate from this group, the prevalence of PTSD in those with chronic pain appears to vary considerably depending upon the index trauma and the type and origin of the pain, ranging from 0.69 % in individuals with chronic low-back pain (where the pain often has no traumatic cause), to 46.7 % in those with chronic pain from whiplash injuries or other injuries after motor vehicle accidents, and to 50.1 % in veterans with chronic pain tied to an injury during their index trauma [3].
The presence of chronic pain and PTSD (at the symptom or syndrome level) is associated with greater symptom severity and a more chronic course for both conditions, a poorer response to pain and PTSD -focused treatments, and poorer overall functioning [4][5][6][7][8][9][10]. These findings have led some to suggest: 1) that the high rates of comorbidity may reflect shared vulnerabilities beyond trauma exposure; and 2) that the increased severity/persistence of the two conditions reflects mutually exacerbating and maintaining interactions, e.g. via biased attention to threat cues, avoidant coping styles, or anxiety sensitivity [11,12]. However, there remains a significant gap in the literature with respect to evidence in support of mutual maintenance models. A recent meta-analysis found no evidence of a main effect of PTSD diagnostic status on pain perception in trauma exposed individuals [13]. The authors did find small to moderate effects suggesting that trauma type may affect pain thresholds.
Network analysis (NA) is an exploratory technique that can be used to investigate interrelations among symptoms within or between diagnostic categories. NA has been used to investigate the associations between the symptoms of PTSD [14][15][16], and between PTSD symptoms and those of anxiety and depression [17,18], between chronic pain and affective disorder symptoms [19], and between depressive symptoms in individuals presenting with chronic pain [20]. NA was developed to identify unique associations among difficulties that may cause or mutually reinforce each other, but to date only one study has used NA to examine the link between individual PTSD symptoms and pain [21]. In an inpatient sample of adults with severe PTSD tied to childhood abuse (n=655), the authors found that the strongest cross-disorder links were between PTSD hyperarousal symptoms and pain.
The majority of studies examining the relationship between PTSD and chronic pain have involved traumatically injured whiplash patients [22][23][24][25], leaving other clinical populations relatively unexplored. Often, such studies have also examined PTSD at the syndrome level (i.e., overall diagnostic or symptom severity). However, PTSD is comprised of 20 symptoms organized into four symptom clusters: re-experiencing, avoidance, negative alterations in cognitions and mood, and increased arousal and reactivity [26]. The relationship between these four symptom clusters and chronic pain remains less clear, with a single NA study suggesting strong links between hyperarousal and pain in child abuse survivors [21]. Further studies are needed, including those involving comparisons of the PTSD and pain networks found in groups identified either because of the presence of chronic pain or PTSD, as variations in rates of pain-PTSD comorbidity and the severity of each condition has been observed across these groups [1][2][3].
In the absence of prior studies comparing the network structures of PTSD and pain in different groups, the aim of this study is to examine relations between PTSD symptom clusters and pain in three trauma-exposed, unique clinical groups: 1) adults seeking treatment for chronic pain with current symptoms of PTSD, 2) trauma affected refugees seeking treatment for PTSD and chronic pain; and 3) individuals identified at admission to the emergency ward after whiplash injury. As proponents of the shared vulnerability/mutual maintenance model of chronic pain and PTSD have argued that studies examining the relationship between these conditions should attempt to control for the influences of both anxiety and depression, given the symptom overlap between these conditions and PTSD and the high rates of comorbidity between these conditions and chronic pain and PTSD [11,12], we control for the influences of self-reported anxiety and depression severity in the network analyses.

Participants
This study included three different trauma-exposed patient groups with pain. Sample 1 (Chronic pain sample) was described in a previous study and consisted of adults with a history of traumatic exposure and mixed chronic pain complaints (lasting for a minimum of three months), including fibromyalgia, neck-related pain, low back pain, and myalgia, referred for assessment to the Pain Rehabilitation Unit at Skåne University Hospital in Sweden [27]. In total, 71.1 % were female with a mean age of 41.0 years (SD: 11.1), and 40.3 % fulfilled the DSM-IV criteria for probable PTSD based on the Posttraumatic Diagnostic Scale (PDS). The mean PTSD symptom severity (using an adjusted PDS scale to align with the Harvard Trauma Questionnaire (HTQ)) was 2.96 (SD: 0.72) with the means for the different clusters being 2.83 (SD: 0.81) for re-experiencing, 2.82 (SD: 0.97) for avoidance, 3.27 (SD: 0.91) for hyperarousal, and 2.89 (SD: 0.83) for numbing. The index traumas were (in order): other traumatic event (31.4 %), accident (23.8 %), multiple traumas (14.6 %), sexual assault (13.0 %), non-sexual assault (8.9 %), life threatening illness (8.3 %). The average pain intensity was 7.41 (SD:1.59), the average pain interference (Multidimensional Pain Inventory (MPI)) was 4.76 (SD: 0.92), and the mean number of pain locations was 15.52 (SD: 9.06). The mean anxiety was 11.20 (SD: 4.86) and the mean depression was 10.24 (SD: 4.61) (Hospital Anxiety and Depression Scale (HADS)). Cases with more than two missing values on the study variables were removed which resulted in a final sample of 312 participants.
Sample 2 (Refugee sample) consisted of traumatized adult refugees with a pre-treatment PTSD-score, seeking treatment at four different specialized outpatient trauma clinics in Denmark (The PTSD and Anxiety Clinic, Aarhus University Hospital, Aarhus; The Rehabilitation Center for Trauma-affected, Haderslev; OASIS, Copenhagen; and DIGNITY (Danish Institute Against Torture), Copenhagen) of whom 431 (44.8 %) were female with the overall mean age for the sample being 41.9 years (SD: 9.8). In total, 89 % fulfilled the DSM-IV cluster criteria for probable PTSD at pre-treatment assessment, based on the HTQ. The mean PTSD symptom severity (HTQ) was 3.14 (SD: 0.49), with the following means for the specific clusters: 3.28 (SD: 0.63) for re-experiencing, 3.14 (SD: 0.77) for avoidance, 3.36 (SD: 0.53) for hyperarousal, and 2.83 (SD: 0.64) for numbing. The sample had been subjected to torture and other organized violence, including systematic beatings, sexual assault, and witnessing others being tortured or killed. Many had also experienced war related traumasincluding bombings, killings, kidnappings, loss of family, deprivation of food, shelter, and health care. The mean pain intensity was 6.51 (SD: 2.01), the mean pain interference (Brief Pain Inventory (BPI)) was 7.34 (SD: 2.11), and the mean number of pain locations was 16.81 (SD=10.27). The mean anxiety was 3.06 (SD: 0.57) and the mean depression was 3.05 (SD: 0.55) both taken from the Hopkins Symptom Checklist (HSCL-25). Again, cases with more than two missing values on the study variables were removed which resulted in a final sample of 822 participants.
Sample 3 (Whiplash sample) consisted of participants consecutively collected at admission to a large Danish emergency ward after whiplash injury (grade I-II: 95 % and grade III: 5 %; 62.2 % female, mean age 36.5 years (SD: 13.7). The sample has been described in a previous study [28]. Head injury and unconsciousness as well as other serious injuries led to exclusion from the study. The participants had sub-acute whiplash and 35 % did not recover from pain six months post injury. In total, 15 % fulfilled the DSM-IV cluster criteria for probable PTSD at assessment, mean 21 days (SD=13.68), after the whiplash trauma. For all participants whiplash was rated as the index trauma. The average PTSD symptom severity (HTQ) was 1.72 (SD: 0.62), with the following means for the specific clusters: 1.81 (SD: 0.70) for re-experiencing, 1.59 (SD: 0.79) for avoidance, 1.96 (SD: 0.81) for hyperarousal, and 1.48 (SD: 0.63) for numbing. Level of average pain intensity was 3.39 (SD: 2.75) and the average pain interference (Pain Disability Questionnaire (PDQ)) was 34.02 (SD: 35.48). The mean anxiety and depression (HADS) levels were 13.10 (SD: 4.57) and 11.13 (SD: 4.44) respectively. Again, cases with more than two missing values on the study variables were removed which resulted in a final sample of 326 participants.

Measures
Symptom severity of PTSD: Participants in Sample 1 completed the 49-item Posttraumatic Diagnostic Scale (PDS; [29]). The first 21 items assess aspects of the (index) trauma; followed by 17 items corresponding to the 17 PTSD symptoms listed for the disorder in DSM-IV, and 11 items assessing impairment from and duration of symptoms. The symptom items are rated on a 0-3 frequency scale over the past month (0=Not at all or only one time; 3=5 or more times a week/almost always). A total score as well as subscale scores based on the DSM-IV symptom clusters are calculated. The PDS has been found to have satisfactory psychometric properties [4,29].
Participants in Samples 2 and 3 completed the Harvard Trauma Questionnaire (HTQ; [30]) a self-report measure of traumatic exposure and the index trauma (Parts 1 and 2), head injury (Part 3), and 30 symptoms of posttraumatic distress (Part 4). Of the 17 symptoms of PTSD identified in the revised edition of DSM-IV [31], Part 4 of the HTQ assesses 16 items compounding symptoms B4 and B5 of DSM-IV into one item. Respondents rate how much each symptom has bothered them over the past seven days using a 4-point scale (1=Not at all, 4=Extremely). A mean score at the item level is computed for the 16 DSM-IV PTSD symptoms from Part 4, with a cut-off of 2.5 suggestive of a current DSM-IV diagnosis of PTSD. The HTQ has acceptable psychometric properties [30]. Sample 2 and 3 were only assessed with part 4 of the HTQ (PTSD).
To make possible direct comparisons between Sample 1 where the PDS measuring 17 DSM-IV PTSD symptoms was administered, and Samples 2 and 3 where the HTQ measuring 16 DSM-IV PTSD was administered, and to align with previous research [16], two PDS items measuring emotional and physiological reactivity were combined into a single item, and the PDS's 0-3 scoring were re-scored to match the 1-4 scoring of the HTQ.
Pain intensity: All participants rated their average pain intensity on different versions of the Numerical Rating Scale (NRS), including a single item scored on an 11-point scale (0=no pain; 10=worst possible pain). Sample 1 rated their pain during the past week, sample 2 during the past 24 h, and sample 3 during the last month. The NRS has satisfactory psychometric properties [32].
Pain interference: Participants in Sample 1 completed the pain interference scale from the Multidimensional Pain Inventory Version 2 (MPI-Interference; [33]). It is comprised of 11-items measuring interference from pain using a 7-point scale (0-6). A mean interference score is calculated with higher scores indicating greater functional impairment from pain. The MPI has satisfactory psychometric properties [34].
Participants in Sample 2 completed the Brief Pain Inventory (BPI; [35]). The 11-item, self-report BPI measures pain intensity (4 items) and pain interference (7 items) using an 11-point scale (0-10). Separate mean scores are obtained for the intensity and the interference items and the subscale focusing on interference was used. The BPI has satisfactory psychometric properties [35].
Participants in Sample 3 completed the Pain Disability Questionnaire (PDQ; [36]). The scale was developed to measure pain-related disability or pain interference related to chronic musculoskeletal conditions. Disability was measured on a 15-item numeric rating scale from 0 to 10, where 0 is no disability and 10 is the worst imaginable disability. The scale consists of two subscales measuring physical (9 items) and psychosocial (6 items) disability and has demonstrated acceptable psychometric properties [36].
Anxiety and depression: Participants in Sample 1 and 3 completed the 14-item Hospital Anxiety and Depression Scale (HADS; [37]). The HADS measures the severity of symptoms of anxiety (7 items) and depression (7 items) over the past week using a 4-point scale (0-3). Greater levels of anxiety/depression were represented by higher scores and the cut-off points are: 0-7 for non-cases, 8-10 for doubtful cases, and 11-21 for clinical cases. The HADS has been found to have acceptable psychometric properties [37].
Sample 2 completed the 25-item Hopkins Symptom Checklist (HSCL-25; [38]). Respondents rate how much each symptom has bothered them over the past seven days (1=Not at all, 4=Extremely). A total score is calculated based on the mean severity rating for all items, as well as subscale scores for anxiety (10 items) and depression (15 items). The clinical cut-off for both the depression and anxiety subscales is 1.75 and the HSCL-25 has been found to possess satisfactory psychometric properties [38].

Data analysis
PTSD symptom dimension estimation: To identify empirically supported symptom clusters of PTSD, we examined whether associations among the 16 PTSD symptom items in each of the samples were adequately explained by latent constructs. Three different models were used to test which latent factors were most supported: (a) a model where all items were used as indicators of a broad, unidimensional PTSD factor; (b) a 3-factor model as suggested in DSM-IV where items are used as specific indicators of re-experiencing, arousal, and avoidance; and (c) a 4-factor model as suggested in DSM-5 where items are used as specific indicators of re-experiencing, arousal, avoidance, and negative alterations in cognitions and mood. Because we did not use assessment tools that were based on the PTSD criteria in DSM-5, model c was tested according to the model outlined in [39], with four dimensions referred to as re-experiencing, hyperarousal, avoidance, and numbing. The fit of each model to the data was evaluated using confirmatory factor analysis (CFA) using the R library lavaan. Diagonally weighted least squares estimation was used and scaled fit indices computed because the items were ordinal. All participants who had item-level data on the PTSD assessment tools were included in these analyses and missing data were handled using pairwise deletion. We interpreted adequate model/data fit as CFI/TLI values >0.90, RMSEA values <0.06, and SRMR values below 0.08. A difference on CFI above 0.01 was used as an indicator of a relevant difference in model fit [40].
Network estimations: In network analysis, each variable is called a node and unique relations between nodes are called edges. All network analyses were conducted using the R library BGGM, which estimates edges (in this case, partial correlations) between each node pair using a Bayesian statistical framework [41]. All edges were estimated by accounting for all linear associations within the full set of nodes and 5,000 posterior samples for each edge were estimated. When estimating edges, 95 % credible intervals (CIs) were used to control for false positive rate. A 95 % CI is similar to a frequentist 95 % confidence interval and indicates where a population parameter (in this case, a partial correlation) will fall 95 % of the times. Only edges that had a 95 % CI (based on the 5,000 posterior samples of the estimate) that did not include zero were considered to be statistically significant and only such edges were included when plotting the networks. Partial correlation coefficients can take on a value between −1 and 1, with positive values indicating positive associations (depicted as green edges in the networks), and negative values indicating negative associations (depicted as red edges in the networks). The proportion of missing data was low (Chronic pain sample: 1.5 %; refugee sample: 6.5 %; whiplash sample: 1.5 %) and missing data were handled using multiple imputation within the network estimation model. Networks were plotted using the R package qgraph and the Fruchterman-Reingold algorithm that places strongly connected nodes centrally in the network, and strongly associated nodes closely to each other while also minimizing overlap of nodes and edges. When plotting the networks, we standardized the layout across samples to facilitate comparisons.
Links between PTSD and pain within samples: Links between PTSD clusters and pain (intensity and interference) in each separate sample were examined in two ways: (1) By analyzing which PTSD cluster was most strongly linked to pain (intensity and interference combined). (2) By analyzing whether pain intensity or pain interference was most strongly linked to PTSD.
Of note, in all networks, we included one node representing selfreported depressive symptoms and one node representing self-reported anxiety symptoms. Thus, each unique edge between pain and PTSD was estimated by accounting for the influence of depression and anxiety, but depression and anxiety were not specifically examined in the present study.
For 1, we estimated how strongly each PTSD node was linked to pain by adding its partial correlations with pain intensity and pain interference. Because each PTSD node had one unique link with pain intensity and one unique link with pain interference, this broader estimate ranged between −2 and +2. We then examined whether the overall link to pain differed for the different PTSD nodes. The 95 % CI for the difference in the overall link to pain was computed by subtracting the 5,000 posterior samples for the overall link to pain for one PTSD node from the 5,000 posterior samples for the overall link to pain for another PTSD node. When the 95 % CI for the difference in the overall link to pain did not include zero, we interpreted the difference as being statistically significant. To increase interpretability, we also computed the posterior probability (PP). A PP indicates the probability of a prespecified event happening, for example, that re-experiencing is more strongly linked to pain than numbing. By keeping with the 95 % range, PPs below 2.5 % or above 97.5 % were considered statistically significant. Of note, the 95 % CI for a difference in the link to pain between two PTSD nodes corresponds perfectly to PPs and are simply different ways to express the same underlying difference. For 2, we computed how strongly pain intensity and pain interference, respectively, were to PTSD by adding together their edges to each of the PTSD nodes. 95 % CIs for the difference in the overall link to PTSD and PPs were again used to make inference.
Links between PTSD and pain between samples: To compare whether links between pain and PTSD differed between samples, we followed the exact method described above but instead compared these estimates between samples; for example, whether re-experiencing and pain were more strongly linked in refugees than in individuals with whiplash.

PTSD and pain intensity
Last, we reconducted the above analyses but excluded pain interference and thus only included pain intensity in the networks, to get a more specified analysis of the relationship between PTSD and pain intensity.

Fit of the different symptom dimension models
Model/data fit for the three competing PTSD models in each of the samples are presented in Table 1. The 4-factor DSM-5 model fitted best in all samples and generally showed adequate to good fit indices. The CFI difference between the DSM-5 and DSM-IV models (with the latter model showing the next best model/data fit) was also larger than 0.01 in all samples except in the whiplash sample (a difference of 0.008). Accordingly, all subsequent analyses were conducted using four PTSD symptom clusters (re-experiencing, hyperarousal, avoidance, and numbing). The zero-order correlations between these clusters were moderate to strong in each clinical group: chronic pain, rs=0.58 to 0.68; refugees, rs=0.43 to 0.68; whiplash rs=0.58 to 0.71.

Links between PTSD clusters and pain
The networks of the four PTSD clusters, pain intensity, pain interference, depression, and anxiety in all samples are presented in Figure 1. The partial correlation matrices, on which the networks are based, are presented for each sample separately in the Supplementary Table 1. Overall, few significant edges between PTSD and pain nodes emerged.
We proceeded to compare how strongly linked each PTSD node was to pain (intensity and interference combined) in each of the samples. All results are in the top panel of Table 2. In the chronic pain sample and the refugee sample, no specific PTSD node was more strongly linked to pain than any other PTSD node and the overall links between pain and PTSD were small (the largest overall link was 0.05; see top panel of Table 2). In whiplash patients, hyperarousal was more strongly linked to pain than re-experiencing, avoidance, and numbing, and the overall link between hyperarousal and pain was 0.20, which was the strongest link found in any sample.
Within-sample results for the two pain variables are at the bottom panel of Table 2. Results showed that, in whiplash patients, pain interference was more strongly linked to PTSD (with links to all four PTSD nodes being considered) than pain intensity while no significant differences in how strongly linked pain intensity and pain interference was to PTSD were found in the chronic pain and refugee patient groups. However, graphical inspection of the networks revealed significant and moderate edges between numbing and depression in the refugee sample (partial r=0.44 [95 %CI: 0.38 to 0.50], see Figure 1) and between depression and pain interference in the chronic pain sample (partial r=0.36 [95 % CI: 0.26 to 0.46], see Figure 1).
We then examined whether the overall links between pain and PTSD differed between samples. No significant differences emerged when the refugee and chronic pain samples were compared. Hyperarousal was significantly more strongly linked to pain in whiplash patients than in refugee and chronic pain patients. Further, pain  interference was significantly more strongly linked to PTSD in whiplash patients than in refugee and chronic pain patients. Full results are in Table 3.

Links between PTSD and pain intensity
We reran all network models but excluded pain interference. Results are presented in Table 4. Very similar results emerged in that hyperarousal was the PTSD node most strongly linked to pain intensity in the whiplash group, while no within-group differences were found for links between the PTSD nodes and pain in the chronic pain and refugee groups. Similarly, for the between-group differences, hyperarousal was more strongly linked to pain intensity in the whiplash group than in the other groups.

Discussion
The present study aimed to help fill a gap in the literature with respect to the relations between the individual PTSD symptom clusters and pain (measured using both intensity and interference), while controlling for comorbid anxiety and depression. To our knowledge, this is the first study that examines links between pain and the individual symptom clusters of PTSD within and between unique clinical groups. Overall, the results suggest that, after controlling for the current severity of depression and anxiety symptoms, few unique relationships existed between pain and each of the four PTSD symptom clusters in individuals with current symptoms of PTSD presenting for treatment of chronic pain, trauma affected refugees presenting for treatment of both Table : Within-group posterior probabilities (PPs) and  % credible intervals for differences ( %CI, within parentheses) of the overall link between pain and PTSD in each of the samples. PPs below . % and above . % are highlighted with bold text. A PP indicates the probability that the variable in the first column is more strongly linked to pain (or PTSD) than the variable in columns -.

Avoidance
Numbing PPs are based on the assumption that the overall link to pain/PTSD for the variable in the first column is larger than the overall link with pain/PTSD for the variables in columns three to five. The difference scores are computed by subtracting estimates for variables in columns three to five from estimates for variables in column one; thus, a positive value indicates that the sum of partial correlations is larger for the variable in the first column. Estimates whose % CI does not include  are highlighted in bold and considered to be statistically significant.
PTSD and pain, and individuals presenting to an emergency room because of a whiplash injury. An exception to this pattern was the observation of a more pronounced relationship between the hyperarousal symptom cluster and pain in the whiplash sample. It has been suggested that the increased severity and persistence of pain and PTSD in individuals with both conditions reflects mutually exacerbating and maintaining interactions [11,12]. Surprisingly, the present study only identified one clear relation in support of a mutually maintaining relationship (the hyperarousal-pain Table : Between-group posterior probabilities (PPs) and  % credible intervals for the difference ( %CI) in partial correlations (within parentheses) for PTSD nodes being more strongly linked to pain intensity and interference (top) and pain intensity/interference being more strongly linked to PTSD (bottom). PPs below . % and above . % are highlighted in bold.
PP for chronic pain  sample > refugee sample   PP for chronic pain  sample > whiplash sample   PP for refugee  sample > whiplash  Estimates whose % CI does not include  are highlighted in bold and considered to be statistically significant.  Estimates whose % CI does not include  are highlighted in bold and considered to be statistically significant. Åkerblom et al.: Relations between PTSD symptom clusters and pain link in whiplash patients). These results add to findings from a recent meta-analysis which found no evidence that PTSD diagnostic status was related to pain perception in traumaexposed individuals [13]. We identified no within-group differences for the links between pain and any of PTSD clusters in the chronic pain and refugee groups. Chronic pain and PTSD are multifaceted problems where a wide spectrum of biological, behavioral, psychological, and social factors is of importance to the development, maintenance, and impacts of these disorders [42][43][44]. Speculatively, the results of the present study may reflect the high level of heterogeneity in these groups, with varying types of pain and traumatic experiences, severity of PTSD, overall symptom load, and differing number of exposures. The increased severity of chronic pain and PTSD in these samples may also be partly due to other factors such as broader psychological distress (e.g., anxiety and depression). For example, we found an association between numbing and depression in the refugee sample and an association between depression and pain interference in the chronic pain sample, which may be worthy of further investigation.
In contrast, the whiplash sample was more homogenous in that all had experienced a recent single potential traumatic event in the form of a motor vehicle accident. This sample also had the lowest level of probable PTSD. In this sample, hyperarousal was more strongly linked to pain than re-experiencing, avoidance, and numbing, and pain interference more strongly linked to PTSD than pain intensity.
The more pronounced link between hyperarousal and pain among individuals with whiplash also became clear when we compared the whiplash network to the networks of the chronic pain and refugee samples. These findings are largely consistent with the findings of a network analysis study by Kratzer et al. [21], which found that some of the strongest cross-disorder associations in an inpatient sample of adults with PTSD tied to childhood abuse was between hyperarousal and pain, although the authors did not control for the presence of comorbid anxiety and depression [21]. The findings are also largely in accordance with a previous study on the relationship between PTSD and pain in individuals suffering from whiplash, where symptoms of hyperarousal and numbing were shown to better predict future neck pain-related disability [24]. The importance of hyperarousal to the severity of pain in whiplash patients has also been highlighted in other studies [22,23,25,45,46]. Ravn et al. [45] argue that hyperarousal drives the remaining symptom dimensions of PTSD and pain due to catastrophic misperceptions and negative interpretations of somatic sensations intensifying the fear of and the focus on pain sensations, which increases the likelihood and intensity of pain experiences [45].
The hyperarousal cluster overlaps with symptoms of chronic pain, anxiety, and depression and consists of symptoms such as having sleep difficulties, experiencing irritability or aggression, poor concentration, hypervigilance, and being easily startled [26]. The present results suggest that arousal symptoms may be an underrecognized target for interventions for individuals suffering from PTSD after a whiplash injury. However, the importance of the hyperarousal cluster can also be heightened by the fact that disrupted sleep, irritability, and concentration are cardinal symptoms of chronic pain and other mental health conditions [24]. In other words, the more pronounced relationship between the hyperarousal symptoms and pain could be a reflection of the pain experience itself.
Of importance to establishing whether a mutual maintenance model of pain and PTSD is based on symptoms that are specific to PTSD, we did not identify a significant unique relationship between pain and the reexperiencing in the separate samples. Aside from Kratzer et al. [21], no previous study has investigated the relationship between pain and the individual PTSD symptom clusters, and that study did not find particularly strong relationships between pain and re-experiencing relative to the other PTSD symptoms [21]. The present findings do not argue strongly in favor of a mutual maintenance model based on the interaction between pain and the symptoms of PTSD that are specific to that condition, e.g. that traumatic reexperiencing may heighten attention for and sensitivity to pain sensations or vice versa [11]. Thus, the present findings add to a small body of literature focused on individuals with whiplash-related PTSD that find no or weaker associations between re-experiencing symptoms and pain [22,24].
There were no significant between-group differences for the pain and refugee groups in the links between pain and PTSD domains, indicating that the clinical characteristics of these two samples may be comparable. Differences were clearer in relation to the whiplash patients where hyperarousal and pain were more strongly linked than in the other two patient groups. Looking at the whiplash sample more specifically, we can see that the trauma type, PTSD symptom severity, pain intensity, and the acuteness of the difficulties differed from the other samples. Generally, the individuals in the whiplash sample had symptoms of PTSD and pain that stemmed from the same traumatic event (motor vehicle accident). Their pain duration was short, their mean pain intensity was of mild intensity, and their mean PTSD symptoms below the clinical threshold for a probable diagnosis. This sample was in the early stage of possible traumatization and therefore with the expectation of remission in many cases. The traumatic events in the other samples were much more varied including multiple and diverse events, such as torture, sexual abuse, motor vehicle accidents, physical threats, and imprisonment, and included individuals where the pain and the PTSD symptoms stemmed from different events. The chronic pain and refugee samples also had longer pain duration, severe pain intensity, a mean PTSD symptom severity above the threshold for a probable diagnosis, indicating a higher overall clinical load.

Limitations
This study uses existing data sets, like the majority of previous network analyses on psychopathology [47]. When using existing data sets, the analysis depends on the limitations of previous databases and types of gathered data rather than being based on specific designs aimed at testing specific hypotheses. Accordingly, slight differences in questionnaires and in the temporal nature of variables were seen across samples in the present study, although these scales were measuring the same symptoms/clusters. Still, such data use includes the threat of possible confounding variables and future studies are needed where all samples are offered the same measures. We conducted most analyses in relation to pain intensity or pain as measured by both intensity and interference. Possibly, other unique relationships would have been found if we had focused more on pain interference only and this remains to be tested in future studies. We conducted conservative statistical tests, where a wide spectrum of variables (different types of pain, all PTSD clusters, anxiety, and depression) was included and controlled for, which could increase the risk of type II errors. Furthermore, the sample sizes differed across samples, which affects statistical power for identifying true effects. This study includes correlational data from patients in different real life services. Hence, these findings reflect clinical practice, but the study also lack an experimental control and cause and effect cannot be established. Consequently, the study included individuals with PTSD at the symptom and syndrome level. Participants reported varying symptom levels, the mean for the whiplash sample even below threshold for a diagnosis, which might pose a threat to the internal validity of the study. In addition, measures were based on self-report which include risks for response biases, do not allow for follow-up questions about ambiguous answers and can only result in a probable PTSD diagnosis. Hence, these results may not generalize to clinician-rated data. Although the present study was large and included data from different trauma samples recruited from several countries and clinics, these results might not be generalizable to other contexts, and this warrants further study.

Conclusions
The results suggest that when depression and anxiety are accounted for, few unique links between pain and each of the four PTSD symptom clusters emerge, at least in these three diverse clinical groups. An exception is the clear link between pain and hyperarousal in individuals with whiplash-related PTSD symptoms. Arousal symptoms may be an underrecognized target for interventions for this patient group. No within-or between-group differences were identified for the links between pain and any of the PTSD domains in the chronic pain and the refugee groups, which may be explained by the high level of heterogeneity in these groups and/or that other factors, such as broader psychological distress (e.g., anxiety and depression), may be partly responsible for the increased severity of chronic pain and PTSD in these samples.
Research funding: Authors state no funding involved. Author contributions: All authors have accepted responsibility for the entire content of this manuscript and approved its submission. Competing interests: Authors state no conflict of interest. Informed consent: Informed consent has been obtained from all individuals included in this study. Ethical approval: This study complied with all relevant national regulations, institutional policies and is in accordance with the tenets of the Helsinki Declaration (as amended in 2013). The data collection for the chronic pain sample was approved by the Regional Ethical Review Board in Lund, Sweden (2013/381). In Denmark, applying for ethical approval prior to initiating research projects is mandatory only for certain biomedical research projects and research projects involving laboratory animals. For all other types of academic research projects, formal ethical approval is not required by law. The data collection for the refugee sample was approved by the Danish Patient Safety Authority (3-3013-2550/1). The data collection for the whiplash sample was reviewed by the Research Ethics Committee, University of Southern Denmark .