RELATIONSHIP BETWEEN DEPRESSION, ANXIETY, AND CHRONIC PAIN (CP)
Empirical research has shown strong associations among CP and depressive and anxiety symptoms. The percentage of patients with depression is double in individuals with CP (30% vs. 15%) compared with no-CP controls, and the prevalence of anxiety disorders in CP patients is 50% higher than in controls (21% vs. 14%).1 In addition, there is a significant correlation between lifetime prevalence of depression and pain incidents.2 Pain intensity, number of pain locations, and the amount of days with pain are significantly associated with both depression and anxiety severity3 and with the likelihood of being diagnosed with depression.4 In addition, being diagnosed with a mood or anxiety disorder is related to an increase in the likelihood of having pain syndromes concurrently.2,5 Furthermore, a diagnosis of a CP disorder increases the risk of being diagnosed with an anxiety or depressive disorder in the future.6,7 Conversely, a previous diagnosis of depression or an anxiety disorder is significantly associated with a future diagnosis of CP.8
SYMPTOMS NETWORK ANALYSIS (NA)
Although the link among mood and anxiety disorders and CP has been well established in previous research, we lack empirical research to sufficiently understand the specific associations between symptoms of CP, anxiety, and depression. NA is a methodology that might enhance our understanding, providing a more fine-grained analysis at the symptom level that may have direct clinical implications, such as the identification of the most important symptom in a network that should be given priority in the treatment of CP.
Recent research in depression has emphasized the potential importance of a symptom level instead of a syndrome-level perspective.9 For example, individual symptomatology analyses have improved the accuracy of mainstream summary strategies on the basis of a broader amount of symptoms (eg, analysis of overall severity) to predict the onset of a depression.10 For other diagnostic groups, NA at the symptom level have already helped to better understand psychopathologic comorbidity patterns (eg, depression and anxiety,11 social anxiety and eating disorders,12 eating disorders, depression, and anxiety,13 generalized anxiety, and depression14).
The network approach in clinical psychology is founded on an innovative psychopathologic perspective.15,16 Historically, mental disorders have been considered latent (ie, unobservable) entities that cause an observable constellation of symptoms. For example, if someone experiences a combination of depressed mood, anhedonia, apathy, abulia, self-reproach tendencies, insomnia, fatigue, and concentration problems, this might be because the person a major depressive disorder.16,17 This psychopathologic conceptualization implies, on the basis of the classical medical model of disease causation, that the association between the observed symptoms can be explained by the existence of an underlying, latent mental disorder.16 As a consequence, symptoms need to be independent from each other and their association (ie, its joint occurrence) should disappear if the effect of the latent condition is controlled for.17 However, the independence of symptoms cannot be supported by psychopathology research.17,18 For example, in case of a depressed person, it is very likely that the experience of a depressed mood may also increase self-reproach behaviors, whereas rumination may contribute to problems in sleeping, which in turn may increase the person’s fatigue level, which may lead to concentration problems.19
To overcome these shortcomings, several authors recently proposed an alternative model for explaining the co-occurrence of certain symptoms within and across disorders (ie, comorbidity). In these models, symptoms are conceptualized as being interdependent and jointly forming a psychopathologic mental condition instead of assuming an underlying latent disease or disorder that explains the presence of a number of independent symptoms.15,19,20 Thus, the network of associations among the symptoms constitutes the syndromes and not the other way around.15 From a clinical point of view, this perspective might have fundamental implications. It means that changes in one symptom might produce changes in the other symptoms (and in the whole network), and interventions may try to produce changes in the whole network structure by focusing on specific central symptoms.15,21
To analyze systematic associations among symptoms, NA provides a visual and statistical tool that paves the way for gathering evidence of the importance of each single symptom in the complex network of symptoms.22 We will provide a more detailed explanation of NA in the methods section.
NA OF DEPRESSIVE AND ANXIETY SYMPTOMS IN CP PATIENTS
In our literature search, we found only few and very recent examples of studies analyzing depressive and anxiety symptoms in patients with CP using NA. McWilliams et al23 used this method to study the association among depressive symptoms in patients with CP. In this study, the most central symptoms (in terms of their connectivity within the network24) were depressive mood, difficulty concentrating, loss of interest, and fatigue. Depressive mood was the symptom with the highest strength (ie, a measure of centrality based on how strong the connections with the other symptoms are). Difficulty concentrating was the symptom with the highest closeness (ie, measure of centrality based on distance between the symptoms) and betweenness (ie, measure of centrality based on how important a symptom is to connect others). After depressive mood and difficulty concentrating, loss of interest and fatigue symptoms were the symptoms with the higher centrality strength, closeness, and betweenness within the network.
More recently, Thompson et al25 used complex NA to explore the association of anxiety and depression severity with pain severity indices (ie, pain intensity and pain interference with daily life) and fear avoidance. Results showed that overall depression severity seemed to play a fundamental role within the network, being the construct with the highest strength, and that the associations of anxiety within the whole network were mediated by depression symptoms. Furthermore, perceived pain interference was the construct with the greatest closeness and betweenness, followed by fear avoidance.
Although the studies by McWilliams et al23 and Thompson et al25 made meaningful contributions to our understanding of the complex associations among depressive, anxiety, and pain symptoms, the current study builds upon this evidence and tries to further our knowledge in several ways. First, McWilliams et al23 carried out fine-grained analyses of depressive symptoms in CP patients, but they included neither the core pain symptomatology nor anxiety symptoms in their analyses. Second, Thompson et al25 incorporated both mood and anxiety symptoms and pain severity indices, but used scale-level information, without exploring the specific associations between single mood, anxiety, and pain symptoms. Consequently, although this study supported the idea of interrelations among these constructs, focusing on specific associations between single symptoms and their relevance might increase our understanding of the complex interplay among depression, anxiety, and pain symptoms.
Thus, the aim of the current study is to analyze the network structure of the associations among depressive symptoms, anxiety symptoms, and pain severity indices in CP patients. We will first focus on analyzing patients’ baseline network of associations among these symptoms (aim #1). Then, we will analyze the symptoms network at posttreatment, comparing it with the one identified at baseline (aim #2). Finally, we will compare the network structures of responders and nonresponders to treatment (aim #3). To define responders, we followed the suggestions by the Initiative on Methods, Measurement, and Pain Assessment in Clinical Trials (IMMPACT26). In their consensus statement for CP patients, the IMMPACT recommended considering a change as clinically relevant if the patient presents a reduction in the subscale pain interference with daily life of the Brief Pain Inventory (BPI) that is equal to or greater than half a SD of the subscale.26
The results of this study might help to enhance treatments for CP by identifying relevant anxiety and depression symptoms that might be targeted by focused interventions within a multimodal treatment.
MATERIALS AND METHODS
Four hundred fifty-four inpatients diagnosed with CP disorder with somatic and psychological factors (F45.4127) who were consecutively admitted into an interdisciplinary multimodal treatment program at a tertiary psychosomatic university clinic in Switzerland were included in the sample.
On average, participants were 45.50 years old (SD=15.04), mostly women (60.9%), married (42.6%) or in a relationship (12.7%), and had at least a high school degree (79.2%). Most of the participants were either Swiss by birth (58.7%) or naturalized Swiss (27.4%). The majority of the sample reported pain symptoms for more than a year (84.6%), and more than one quarter of the participants for >10 years (26.1%). Approximately 60.4% of these individuals were unable to work.
Inpatient treatment consisted of a 3-week program that comprised different multifaceted interventions and therapies. Besides medical interventions and pharmacotherapy, the program entailed individual and group physiotherapy. Also, patients were enrolled in ergotherapy, individual psychotherapy (which consists of cognitive- behavioral therapy and biofeedback), group psychotherapy (focused on pain management, vertigo/dizziness management, communication group), and relaxation training (ie, mindfulness).
Depression and Anxiety Symptoms
To assess depression and anxiety symptomatology, we used the German version of the Hospital Anxiety and Depression Scale (HADS28), a measure designed to detect manifestations of anxiety and depression in patients with physical illnesses or somatoform symptoms. The HADS is a self-report instrument, rated on a Likert scale from 0 (not at all) to 3 (most of the time). As this measure excludes anxiety and depression items that are strongly associated with physical symptoms, it is recommended for CP populations.29 The HADS assesses the degree of depressive and anxiety symptoms during the past week with 2 subscales (ie, anxiety and depression) that have 7 items each. As in this study, for the descriptive analyses, we used the mean subscale score (ie, total score/number of items responded), the theoretical range of each subscale was from 0 to 3. The anxiety subscale includes items such “I feel tense or ‘wound up’,” whereas the depression subscale includes items as “I feel as if I am slowed down.” Some items included in the HADS represent positive emotions or states (being cheerful, being able to laugh, being able to enjoy things, feel relaxed, etc.). In those cases, items were reverse scored, assuming that the lack of those positive features represents either depressive or anxiety symptoms. The HADS showed adequate psychometric properties with good internal consistency, sensitivity and specificity, external validity, and construct validity (showing a 2-factors structure).30 In the current sample, the HADS subscales showed adequate internal consistency with Cronbach α of 0.77 and 0.82 for anxiety and depression subscales at baseline, respectively.
For pain assessment, we used the German version of the BPI.31 Originally designed to assess cancer pain, the BPI is now frequently used as a generic pain questionnaire for various CP conditions.32 This self-report measure has 2 subscales to evaluate pain: pain intensity and pain-related interference in the individual’s life. The pain intensity subscale has 4 items measuring the severity of the pain experience that are rated on an 11-point numeric rating scale from 0 (no pain) to 10 (most imaginable pain). For example, 1 item of the pain intensity subscale is “rate your pain by marking the number that best describes your pain at its worst in the last 24 hours.” The pain interference subscale has 7 items rated also on an 11-point numeric rating scale from 0 (does not interfere) to 10 (completely interferes). This subscale includes items such as “mark the number that describes how, during the past 24 hours, pain has interfered with your general activity.” In terms of timeframe, 3 items of the pain intensity subscale refer to the last 24 hours (ie, worst pain, least pain, and average pain), 1 item of the pain intensity subscale refers to the current moment (ie, current pain), while all the items of the pain interference subscale refer to the last week. For the descriptive analyses, we used the mean subscale scores (ie, total score/number of items responded). Both the scores of subscales for pain intensity and pain interference had the same theoretical range (although they have different amount of items) from 0 to 10. The German version of the BPI has shown good psychometric properties with good internal consistency, test-retest reliability, convergent validity, and construct validity (consistent with a 2-factor structure).33 In the current sample, the BPI subscales showed adequate internal consistency, with Cronbach α of 0.82 and 0.89 for intensity and interference subscales, respectively.
As a standard procedure, patients completed a battery of questionnaires including the HADS and BPI at intake and discharge with the help of instructed research assistants. At intake, all participants provided informed consent for the use of their data for research purposes. The Ethics Committee of the Canton of Bern approved the study (project ID 2018-00493).
We carried out all the analyses using R.34 Participants included in the analysis had completed at least 1 of the 2 measures used in this study at baseline. To deal with missing data, we used a random forest nonparametric multiple imputation strategy using the package missForest.35
Aim #1: Baseline Network of Depression, Anxiety, and Pain Symptoms
On the basis of the imputed sample, we ran an NA using the R-package bootnet.36 In NA, the network represents a pattern of relationships among the included symptoms. The different symptoms are named nodes and their relationships are called edges.16 The edges could be positive or negative, implying direct or inverse associations between the elements of the network. Graphically, the nodes are represented by circles and the edges by lines connecting the circles. The thickness of the lines represents the strength of the association between the nodes. Furthermore, the color of the edges represents the nature of the association between the nodes in terms of direct relationships or inverse relationships.
Mathematically, the network is formed based on calculating correlations between symptoms while controlling for the effects of other symptoms (ie, partial correlations37). We used common regularization techniques to reduce the amount of weak edges and statistical criteria to select the network that best fits the data.37 Thereby, the likelihood of spurious associations can be limited (ie, false positives).22,37 Thus, to estimate the partial correlations within the network, we used a regularization technique named ‘graphical least absolute shrinkage and selection operator’ (glasso).38 As the glasso method provided several networks,22 we used the Extended Bayesian Information Criterion to select the optimal network.39 The Extended Bayesian Information Criterion has a hyperparameter (γ) set by the researcher to establish the degree, to which the criteria will prioritize sparser or more parsimonious networks. This value ranges from 0 to 0.5, with higher values implying simpler networks.40 For this analysis, we set the γ parameter at 0.5, prioritizing the most parsimonious networks.40
Once we had identified the optimal network using this method, we calculated indices of symptom centrality within the network. Centrality is a measure of the connectivity of a symptom within the network, describing how well the symptom is associated with others in the network by estimating the likelihood that the presence (or intensity) of a symptom will trigger a systemic activation of the whole network.41,42 Three main indices characterize centrality: strength, closeness, and betweenness. Strength assesses how strong the connections among symptoms are.43 A symptom with stronger connections has stronger associations with other symptoms and might have a greater impact upon them (both positively and negatively). Closeness is a measure of distance between a specific symptom and all other symptoms in the network.22 It is defined as the inverse of the sum of the shortest distance of one node from all the other nodes in the network.43 Thus, a higher closeness means that a symptom has a shorter distance to other symptoms in the network. The effects of changes in symptoms with a high closeness are likely to spread more easily within the network.22 Finally, Betweenness is a measure of how important a symptom is in terms of its connections with other symptoms.43 A symptom with a high betweenness will often be in on the shortest path between other symptoms, linking one symptom to another. Betweenness is an important feature to identify symptoms that might function as bridges between symptoms or groups of symptoms within a network.44
We then checked the stability of the centrality indices by using bootstrapping procedures (2500 samples) and progressively dropping cases from the sample. Correlation stability (CS) indices (ie, proportion of the sample that can be dropped maintaining a correlation of at least 0.70) above 0.50 are recommended.36
Considering recent concerns regarding the ability of centrality indices to identify relevant symptoms within a network,45 we also calculated measures of nodes expected influence on the basis of the formulas presented by Robinaugh et al.46 This method allows establishing 2 measures of expected influence: a 1-step expected influence index that assesses symptoms influence with their direct neighbors and a 2-step expected index that measures both the immediate and the secondary influence of the nodes through their neighbors’ further influences.46
To further establish symptom relevance within the network, we also ran a minimum spanning tree (MST) analysis.47 The MST produces a reduced network that connects all the edges with their minimum amount needed, providing insights into nodes’ topological relevance and hierarchical organization.47,48 We run the MST analysis using the ape package.49
Aim #2: Baseline and Posttreatment Network Comparison
For comparison purposes, we estimated an additional network with the symptoms measured at discharge and compared both networks’ general structure (ie, the matrices of the networks’ connection strength) and global strength invariance (ie, the sum of all the weighted absolute edges’ strength indices in the network) using the package NetworkComparisonTest.50
Aim #3: Comparison Between Networks of Responders Versus Nonresponders
We identified responders and nonresponders to treatment and estimated a network for both samples at baseline and at posttreatment using the IMMPACT26 criteria described above. Once we identified the sample of responders and nonresponders by this method, we again used the NetworkComparisonTest package to compare (1) the baseline networks of responders with nonresponders and (2) the baseline versus posttreatment networks in the group of patients who responded to treatment.
Sample Descriptive Statistics
The mean and SD of all variables are shown in Table 1 both at baseline and at posttreatment. The comparison of baseline and posttreatment levels by paired t tests showed significant differences in 3 of the pain intensity items, and in all pain interference, anxiety, and depression items (all suggesting a reduction in patients’ symptom severity). When adjusting the α level due to multiple comparisons using a Bonferroni correction, 20 of the 25 items presented significant reductions during treatment. Of these, 1 anxiety symptom (ie, restless) and 2 depression symptoms (ie, appearance and enjoy TV) did not change significantly. On the basis of the IMMPACT criteria, 229 (50.44%) presented a clinically relevant change during treatment (ie, a reduction of half a SD [0.90 U] in the BPI pain interference subscale).
Aim #1: Baseline Network of Depression, Anxiety, and Pain Symptoms
In Figure 1, we present a graphical representation of the estimated network grouped by the subscales of the items. The figure shows several associations between the item groups of pain interference and depressive symptoms. Although the pain interference nodes have associations with the anxiety symptoms (eg, restless↔sleep, worrying↔mood, panic↔mood), they were weaker than the ones of depressive symptoms (eg, [lack of being] cheerful↔mood, [lack of being] cheerful↔enjoyment, [lack of] laugh↔enjoyment). However, as it might be expected, there were several associations between anxiety and depression symptoms (eg, relaxed↔slowed, tense↔slowed, worrying↔[lack of being] cheerful).
The pain intensity symptoms are presented apart from the whole network; they mostly do not have direct associations with anxiety or depression symptoms, but only with pain interference symptoms. The sleep node, from pain interference, is connected both with pain intensity (least and worst pain) and with anxiety nodes (being restless).
In summary, this means that most of the nodes associations were presented between pain interference and depression symptoms. Pain intensity nodes were only associated with pain interference nodes, being only indirectly associated with depressive and anxiety symptoms.
Stability analyses of the centrality measures showed adequate CS indices for strength (0.67) and closeness (0.52), but a low CS index for betweenness (0.21). The visual inspection of the CS plot (Supplemental Material 1, Supplemental Digital Content 1, http://links.lww.com/CJP/A613) showed consistent results, with stable correlations for strength and closeness (ie, high level and slow reduction of the correlation when dropping cases), but unstable correlations for betweenness (ie, low levels and steeper reductions in the correlations when dropping cases). In Figure 2, we present the distribution of the standardized centrality indices of strength and closeness (ie, the indices that showed reliable results) for all the nodes in the network. The nodes with the highest strength centrality were [lack of] laugh (depressive symptom), enjoyment-pi (pain interference symptom; please note that there is another enjoyment item within the depression symptoms, ie, enjoyment-d), and average (pain intensity symptom). These nodes were the ones that presented the strongest connections within the network, having greater impact upon the other network’s nodes. The highest closeness scores were associated with mood (pain interference symptom), enjoyment-pi (pain interference symptom), and [lack of being] cheerful (depressive symptom). (We replicate these findings running exactly the same models, but using standardized scores for the items, to control for any biased due to the differential ranges of the nodes.51 The network estimated with this variation presented the same 3 nodes with the greatest centrality strength [lack of] laugh, enjoyment-pi, and average] and closeness [mood, enjoyment-pi, and [lack of being] cheerful]. Thus, these nodes were the ones that presented the shortest distance with other nodes within the network, meaning that changes in them would spread more easily within the network.
Expected Influence Indices
We have presented the results of the expected influences indices in Table 2. Consistent with the results of the centrality indices, the nodes with the greatest expected influence (both in 1-step and in 2-step methods, ordered by ranking) were [lack of] laugh (depressive symptom), enjoyment-pi (pain interference symptom), and average (pain intensity symptom). Mood (pain intensity symptom) was the fourth node with the greatest expected influence with the 1-step method and the fifth node with the greatest expected influence with the 2-step method. The node [lack of being] cheerful (depressive symptom) was the fifth node with the greatest expected influence with the 1-step method and the fourth node with the greatest expected influence with the 2-step method. The results of these indices suggest that the symptoms [lack of] laugh, enjoyment-pi, average, mood, and [lack of being] cheerful are the ones with both greatest immediate influence (ie, with their direct neighbors) and secondary influence (ie, through their neighbors’ further influences).
In Supplemental Material 2 (Supplemental Digital Content 2, http://links.lww.com/CJP/A614), we present the results of the MST analyses. The MST distribution is consistent with the observed baseline network of Figure 1. In all cases, the items of the same subscales were grouped together. The main node connecting the branches of the MST (connecting with the depression, anxiety, and pain interference symptoms) and likely being central in the network is the [lack of being] cheerful item of the depression subscale. Overall, the MST showed that depression symptoms connected anxiety and pain symptoms (interference and intensity). Furthermore, pain intensity symptoms are connected to the network through pain interference symptoms. Thus, pain interference symptoms (especially enjoyment, mood, and sleep interference symptoms) also connected pain intensity and depression symptoms. Particularly, the pain interference symptom of sleep connects the pain interference symptoms with pain intensity nodes.
Aim #2: Baseline and Posttreatment Network Comparison
The test of network invariance between baseline and posttreatment data approached significance (M-statistic=0.21, P=0.065). In addition, the global strength invariance test showed a significant difference between the baseline and posttreatment networks (S-statistic=1.27, P<0.001). The nodes from the baseline network presented a lower global strength (mean=10.60) than the nodes from the posttreatment network (mean=11.86). Thus, the global strength of the network tended to increase from baseline to posttreatment.
Aim #3: Comparison Between Networks of Responders Versus Nonresponders
On comparing baseline networks of responders versus nonresponders, we did not find a significant difference in the network structure (M-statistic=0.21, P=0.35). However, there was a significant difference in networks’ global strength (S-statistic=2.65, P=0.003). The nodes in the nonresponders’ network showed a greater overall strength (mean=10.76) than the responders’ network (mean=8.11).
Then, we compared the networks of responders at baseline with the same patients’ networks at posttreatment. We did not find a significant difference between baseline and posttreatment network structure (M-statistic=0.20, P=0.54). However, there was a significant difference in the networks’ strength (S-statistic=4.20, P<0.001). The nodes from the responders’ network at posttreatment had a greater overall strength (mean=12.31) than the baseline responders’ network (mean=8.11).
The aim of the current study was to analyze the network structure of depression symptoms, anxiety symptoms, and pain severity indices in patients with CP. We first ran a NA using baseline data from CP patients. The results from the NA showed the greatest centrality and expected influence in the nodes [lack of] laughing, enjoyment (pain interference), average pain, [lack of being] cheerful, and mood. Second, we compared the network structure of the whole sample at baseline versus at posttreatment, finding that the posttreatment networks present a significantly greater node’s strength compared with the baseline network. Thus, patients tended to increase their network strength during treatment. Third, we compared the baseline networks in responders versus nonresponders and, in patients who responded to treatment, comparing baseline and posttreatment networks. The nonresponders’ network nodes presented a significantly greater strength than the responders’ network. However, the responders’ network at posttreatment had a significantly greater strength than the baseline responders’ network. These results suggest that treatment nonresponders tend to have a more strongly connected network at baseline, whereas responders tend to have weaker associations within their network. Nevertheless, the patients who respond to treatment tend to increase the strength of nodes association from baseline to posttreatment. To facilitate the comprehension of the findings, we included a summary of the main results of the study in the Supplemental Material 3 (Supplemental Digital Content 3, http://links.lww.com/CJP/A615).
Baseline Network of Depression, Anxiety, and Pain Symptoms
Overall, the results of the network analyses at baseline showed that pain interference and depressive symptoms were most central. Consistently, in a NA in CP, Thompson et al25 found that on the aggregate, the nodes for overall depressive symptomatology and pain interference were the ones with greatest centrality. In the current paper, we used a more fine-grained symptom-level analysis rather than an overall subscale approach as the one of Thompson et al,25 which allowed us to disaggregate specific symptoms within the estimated network and specify their single relevance. The symptoms with the greatest centrality (ie, strength and closeness) in the network were the depressive symptoms of (not) laughing and (not) being cheerful, and the pain interference items of (lacking) enjoyment and experiencing (negative) mood. These results were replicated by the node’s expected influence analyses.46
From a clinical standpoint, symptoms with a high strength and a high expected influence could potentially be good candidates for therapeutic targets11,46 because a change in these symptoms is more likely to have a greater impact on reducing other symptoms within the network of depression symptoms, anxiety symptoms, and pain symptoms.24,52 For example, trying to help the patient to increase the number of enjoyable experiences in his or her life with a greater potential for (genuine) laughter may be a promising target associated with greater changes in pain interferences. Also, the findings on closeness centrality indices may have important clinical implications. Because changes in symptoms with high closeness are more likely to spread more quickly within the network, symptoms with high closeness may also be promising therapeutic targets (especially if they have a substantive strength) and may be associated with a quicker systemic change by targeted interventions.43 For example, as (not being) cheerful had high closeness indices within the network, interventions that help patients to experience more cheerful mood states are likely to have a more widespread beneficial impact on other network symptoms, particularly the pain interference experiences, compared with nodes with lower closeness scores.
Thus, the 4 items with greatest centrality (ie, [lack of] laugh, [lack of being] cheerful, enjoyment, and mood) may produce the strongest and quickly spreading effects on the whole network if they are changed. Therefore, these 4 symptoms might represent prime candidates for psychosocial interventions in the treatment of patients with CP. However, this study represents only a first step in the identification of potential depressive and anxiety and pain symptoms that might be relevant targets for the treatment of CP patients. Further research would be necessary to replicate these findings and identify specific symptom targets of clinical interventions for pain.
It is worth highlighting that the depressive symptoms presenting the greatest associations with pain interference and the greatest centrality and expected influence within the network describe the absence of positive psychological states (ie, cheerful and laugh), and not only the presence of negative states. These results are in line with a growing body of research associating resilience resources with better coping strategies and attitudes toward CP.53,54 Specifically, positive affect has been identified as a resilience mechanism associated with higher levels of coping responses, perceived self-efficacy, and overall mental health in patients with CP.54,55 Furthermore, these findings might support the importance of following a capitalization model when treating CP, which aims to work on activating the patients’ strengths as the main mechanism of change (eg, aiming to improve their capacities of being cheerful and having more humorous experiences), and not only a compensation model that focuses on modifying patients’ deficits or vulnerabilities.56,57
The depressive symptoms also presented strong connections with the anxiety symptoms, partially linking anxiety symptoms with the rest of the network. These results are also consistent with those of Thompson et al25 at a construct level. In addition, they are in line with a strong body of research indicating the close relationships between anxiety and depressive symptoms in terms of high rates of comorbidity,58 common etiopathogenesis59,60 and the growing importance of unified protocols to treat them.61,62
Pain intensity items turned out to be separated from the other nodes of the network, being only connected with depression and anxiety symptoms through pain interference items. Particularly, sleep problems, from the pain interference subscale, were associated with both pain intensity and anxiety symptoms. It might be possible that higher pain intensity is particularly interfering and that anxiety symptoms were related if there are sleep problems involved. This is consistent with previous research showing associations between the intensity of CP and sleep problems63 and evidence suggesting a reciprocal relationship between both phenomena.64 Thus, sleep problems might be a pivotal “connector” between pain intensity and anxiety, and may be potentially important for therapy or for prevention programs. These fine-grained symptom-level findings are also consistent with the results of Thompson et al,25 who reported that pain interference symptoms connect pain intensity with the whole network of symptoms. These findings might further support the relevance of interference pain symptoms for the multimodal treatment of CP.
The MST analysis generally confirmed the above results and provided additional information on the differential topological relevance of the nodes in the network. Besides confirming some of the relationships among the groups of items (ie, depression symptoms linking anxiety symptoms with the rest of the network and pain interference linking pain intensity with the whole structure), the MST highlighted the importance of (not being) cheerful as a core depressive symptom connecting pain interference, anxiety, and depression.
Baseline and Posttreatment Network Comparison
Whereas symptom distress scores were significantly reduced after treatment and the network structure remained relatively stable from pretreatment to posttreatment, the general strength of the associations among the nodes in the network increased over the rather short inpatient treatment (on average, 3 weeks). The finding that the network structure remained stable from baseline to posttreatment has also been observed in other short-term interventions.11 This might imply that the overall symptom structure cannot be modified substantially during a short-term multimodal inpatient treatment of CP. The insignificant difference in the network structure between baseline and posttreatment symptoms is in line with this hypothesis, whereas this result should be interpreted extremely cautiously. The finding of an increase in network strength from baseline to posttreatment also replicates previous studies showing an increase of item intercorrelations (ie, higher connectivity in the network models) over time in different clinical settings.11,52 This pattern has been consistent between different ranges of time (ie, for 6 weeks to 2 years) and different sources of clinical information (patients self-reported and clinician-rated measures), and was also associated with a reduction of scale dimensionality and an increase in reliability within scales over time.52 Although the authors of these papers explored several statistical and psychometric explanations (eg, lack of normal distribution, floor or ceiling effects, spurious effects due to measurement bias), they refuted all of them because they did not fit their data, failing to find a likely cause to understand this finding.11,52 Further research might be needed to replicate the finding of an increased network strength from baseline to posttreatment and continue exploring different statistical, psychometric, and clinical explanations and test them empirically to fully understand the implications of this finding.
Comparison Between Networks of Responders Versus Nonresponders
The comparison between the baseline scores of responders versus nonresponders, and between the networks of responders at baseline versus posttreatment suggests additional clinical implication for the treatment of CP patients. On the one hand, patients who did not respond to treatment had a greater baseline strength of associations within the networks in comparison to patients who responded. This finding, consistent with previous theoretical and empirical studies,15,65 underlines the importance of evaluating patients’ baseline network of symptomatic associations as predictor of treatment outcome, considering that having a strongly interconnected symptom structure at baseline seems to impede the success of a multimodal inpatient treatment for CP. A more intense repeated-measures analysis of the associations between symptoms at baseline within each patient might not only provide personalized prognostic data (ie, how hard it would be to produce changes in the patient) but could also inform on likely profitable symptom targets for an individual patient,14 providing key information for personalized treatments of CP patients.
As noted above, the comparison between responders’ baseline and posttreatment networks showed an increase of the strength of the associations within the symptoms. In other words, whereas having looser symptom associations seems to facilitate symptom change during inpatient multimodal treatment for CP, the symptom structure of successfully treated patients seems to consolidate over treatment. Understanding the quality of less pathologic and more changeable symptom structures could provide further guidance to treatment development for CP.
Limitations and Future Directions
Several limitations characterize this study. First, the study analyzed only CP and depression and anxiety symptoms. Further psychopathologic phenomena might be included to evaluate CP syndrome networks in more detail (eg, past adverse, traumatic events, or personality disorder traits). The lack of a direct measurement of positive affect or resilience features related to CP symptomatology represents another limitation of the study. Future research might need to explore the associations among depression and anxiety symptoms, CP symptoms, and other psychopathologic and resilience phenomena in more detail. Second, we only analyzed cross-sectional data at 2 time points, that is, at baseline and posttreatment, limiting the potential for deriving causal inferences in terms of symptoms associations. As a consequence, the observed relationships between the symptoms rather need to be interpreted as being interdependent over time. Following Granger’s causality model,66,67 future studies might need to implement a more intensive repeated-measure assessment of depression and anxiety symptoms and CP symptoms, with methods like Ecological Momentary Assessment,68 evaluating cross-lagged associations among symptoms. Third, we only analyzed a short-term 3-week inpatient treatment for CP. The short-term nature of the treatment might limit the possibility of finding structural differences between baseline and posttreatment networks. Further research, analyzing treatments for CP that may also include day-hospital or outpatient treatments, might provide stronger evidence for the structural evolution of symptom networks during CP treatment. Fourth, the low stability of the betweenness centrality index limited us from drawing robust conclusions on relevant symptoms in terms of mediated relationships among symptoms. Finally, the NA was focused only on analyzing symptom-level and symptom-interference nodes. Whereas these data might inform clinical practice from a psychopathologic perspective (ie, symptoms and symptom interferences that are relevant within the network), it does not inform about other clinically and therapeutically relevant variables that are associated both with the network and change during treatment, for example, patient expectations or motivation to treatment.
Besides these limitations, this study provides relevant information on pain, depression, and anxiety symptoms associations in patients with CP. Furthermore, the findings have the potential to inform treatment development, particularly supporting a capitalization approach to CP treatment to improve depressive and pain interference features.
1. Benjamin S, Morris S, Mcbeth J, et al. The association between chronic widespread pain and mental disorder. Arthritis Rheum. 2000;43:561–567.
2. Larson SL, Clark MR, Eaton WW. Depressive disorder as a long-term antecedent risk factor for incident back pain: a 13-year follow-up study from the Baltimore Epidemiological catchment area sample. Psychol Med. 2004;34:211–219.
3. Gerrits MMJG, Vogelzangs N, van Oppen P, et al. Impact of pain on the course of depressive and anxiety
disorders. Pain. 2012;153:429–436.
4. Ho PT, Li CF, Ng YK, et al. Prevalence of and factors associated with psychiatric morbidity in chronic pain
patients. J Psychosom Res. 2011;70:541–547.
5. Williams LJ, Pasco JA, Jacka FN, et al. Pain and the relationship with mood and anxiety
disorders and psychological symptoms. J Psychosom Res. 2012;72:452–456.
6. Dominick CH, Blyth FM, Nicholas MK. Unpacking the burden: understanding the relationships between chronic pain
and comorbidity in the general population. Pain. 2012;153:293–304.
7. Tsang A, Von Korff M, Lee S, et al. Common chronic pain
conditions in developed and developing countries: gender and age differences and comorbidity with depression
disorders. J Pain. 2008;9:883–891.
8. McWilliams LA, Cox BJ, Enns MW. Mood and anxiety
disorders associated with chronic pain
: an examination in a nationally representative sample. Pain. 2003;106:127–133.
9. Fried EI, Nesse RM. Depression
sum-scores don’t add up: why analyzing specific depression
symptoms is essential. BMC Med. 2015;13:72.
10. Boschloo L, van Borkulo CD, Borsboom D, et al. A prospective study on how symptoms in a network predict the onset of depression
. Psychother Psychosom. 2016;85:183–184.
11. Beard C, Millner AJ, Forgeard MJ, et al. Network analysis
symptom relationships in a psychiatric sample. Psychol Med. 2016;46:3359–3369.
12. Levinson CA, Warren C, Fernandez KC, et al. Social anxiety
and eating disorder comorbidity and underlying vulnerabilities: using network analysis
to conceptualize comorbidity. Int J Eat Disord. 2018;51:693–709.
13. Smith KE, Mason TB, Crosby RD, et al. A comparative network analysis
of eating disorder psychopathology and co-occurring depression
symptoms before and after treatment. Psychol Med. 2019;49:314–324.
14. Fisher AJ, Reeves JW, Lawyer G, et al. Exploring the idiographic dynamics of mood and anxiety
via network analysis
. J Abnorm Psychol. 2017;126:1044–1056.
15. Borsboom D. A network theory of mental disorders. World Psychiatry. 2017;16:5–13.
16. Borsboom D. Psychometric perspectives on diagnostic systems. J Clin Psychol. 2008;64:1089–1108.
17. Borsboom D, Cramer AOJ. Network analysis
: an integrative approach to the structure of psychopathology. Annu Rev Clin Psychol. 2013;9:91–121.
18. Cramer AOJ, Waldorp LJ, van der Maas HLJ, et al. Complex realities require complex theories: refining and extending the network approach to mental disorders. Behav Brain Sci. 2010;33:178–193.
19. McNally RJ. Can network analysis
transform psychopathology? Behav Res Ther. 2016;86:95–104.
20. Hofmann SG, Curtiss J. A complex network approach to clinical science. Eur J Clin Invest. 2018;48:e12986.
21. Rubel JA, Fisher AJ, Husen K, et al. Translating person-specific network models into personalized treatments: development and demonstration of the dynamic assessment treatment algorithm for individual networks (DATA-IN). Psychother Psychosom. 2018;87:249–251.
22. Hevey D. Network analysis
: a brief overview and tutorial. Health Psychol Behav Med. 2018;6:301–328.
23. McWilliams LA, Sarty G, Kowal J, et al. A network analysis
of depressive symptoms in individuals seeking treatment for chronic pain
. Clin J Pain. 2017;33:899–904.
24. Rodebaugh T, Tonge NA, Piccirillo M, et al. Does centrality in a cross-sectional network suggest intervention targets for social anxiety
disorder? J Consult Clin Psychol. 2018;86:831–844.
25. Thompson EL, Broadbent J, Fuller-Tyszkiewicz M, et al. A network analysis
of the links between chronic pain
symptoms and affective disorder symptoms. Int J Behav Med. 2019;26:59–68.
26. Dworkin RH, Turk DC, Wyrwich KW, et al. Interpreting the clinical importance of treatment outcomes in chronic pain
clinical trials: IMMPACT recommendations. J Pain. 2008;9:105–121.
27. World Health Organization. The ICD-10 Classification of Mental and Behavioural Disorders: Clinical Descriptions and Diagnostic Guidelines. Geneva, Switzerland: World Health Organization; 1992.
28. Herrmann-Lingen C, Buss U, Snaith RP. Hospital Anxiety and Depression Scale, Deutsche Version
(HADS-D). Bern, Switzerland: Huber; 2011.
29. Pincus T, Fraser L, Pearce S. Do chronic pain
patients ‘Stroop’ on pain stimuli? Br J Clin Psychol. 1998;37:49–58.
30. Bjelland I, Dahl AA, Haug TT, et al. The validity of the Hospital Anxiety
Scale. An updated literature review. J Psychosom Res. 2002;52:69–77.
31. Cleeland CS, Ryan KM. Pain assessment: global use of the Brief Pain Inventory. Ann Acad Med Singapore. 1994;23:129–138.
32. Poquet N, Lin C. The Brief Pain Inventory (BPI). J Physiother. 2016;62:52.
33. Radbruch L, Loick G, Kiencke P, et al. Validation of the German version of the Brief Pain Inventory. J Pain Symptom Manage. 1999;18:180–187.
34. R Core Team. R: A Language and Environment for Statistical Computing [Computer Program]
. Vienna, Austria: R Foundation for Statistical Computing; 2018.
35. Stekhoven DJ, Buehlmann P. MissForest—non-parametric missing value imputation for mixed-type data. Bioinformatics. 2012;28:112–118.
36. Epskamp S, Borsboom D, Fried EI. Estimating psychological networks and their accuracy: a tutorial paper. Behav Res Methods. 2018;50:195–212.
37. Costantini G, Epskamp S, Borsboom D, et al. State of the aRt personality research: a tutorial on network analysis
of personality data in R. J Res Pers. 2015;54:13–29.
38. Friedman J, Hastie T, Tibshirani R. Sparse inverse covariance estimation with the graphical lasso. Biostatistics. 2008;9:432–441.
39. Chen J, Chen Z. Extended BIC for small-n-large-P sparse GLM. Stat Sin. 2012;22:555–574.
40. Foygel R, Drton M. Extended Bayesian information criteria for Gaussian graphical models. Adv Neural Inf Process Syst. 2010;23:2020–2028.
41. Borgatti SP. Centrality and network flow. Soc Networks. 2005;27:55–71.
42. Freeman LC. Centrality in social networks conceptual clarification. Soc Networks. 1978;1:215–239.
43. Opsahl T, Agneessens F, Skvoretz J. Node centrality in weighted networks: generalizing degree and shortest paths. Soc Networks. 2010;32:245–251.
44. Langer JK, Tonge NA, Piccirillo M, et al. Symptoms of social anxiety
disorder and major depressive disorder: a network perspective. J Affect Disord. 2019;243:531–538.
45. Fried EI. How to interpret centrality values in network structures (not). Psych Networks[Tutorial]; 2018. Available at: https://psych-networks.com/how-to-not-interpret-centrality-values-in-network-structures/
. Accessed April 25, 2019.
46. Robinaugh DJ, Millner AJ, McNally RJ. Identifying highly influential nodes in the complicated grief network. J Abnorm Psychol. 2016;125:747–757.
47. Letina S, Blanken TF, Deserno M, et al. Expanding network analysis
tools in psychological networks: minimal spanning trees, participation coefficients, and motif analysis applied to a network of 26 psychological attributes. Complexity. 2019;2019:9424605.
48. Mantegna RN. Hierarchical structure in financial markets. Eur Phys J B. 1999;11:193–197.
49. Paradis E, Schliep K. ape 5.0: an environment for modern phylogenetics and evolutionary analyses in R. Bioinformatics. 2019;35:526–528.
50. van Borkulo CD, Epskamp S, Millner A. NetworkComparisonTest: Statistical comparison of two networks based on three invariance measures. R package version 2.0.1; 2016. Available at: https://CRAN.Rproject.org/package=NetworkComparisonTest
. Accessed March 4, 2019.
51. Bulteel K, Tuerlinckx F, Brose A, et al. Using raw VAR regression coefficients to build networks can be misleading. Multivariate Behav Res. 2016;51:330–344.
52. Fried EI, Epskamp S, Nesse RM, et al. What are ‘good’ depression
symptoms? Comparing the centrality of DSM and non-DSM symptoms of depression
in a network analysis
. J Affect Disord. 2016;189:314–320.
53. Karoly P, Ruehlma LS. Psychological “resilience” and its correlates in chronic pain
: findings from a national community sample. Pain. 2006;123:90–97.
54. Sturgeon JA, Zautra AJ. Resilience: a new paradigm for adaptation to chronic pain
. Curr Pain Headache Rep. 2010;14:105–112.
55. Alschuler KN, Kratz AL, Ehde DM. Resilience and vulnerability in individuals with chronic pain
and physical disability. Rehabil Psychol. 2016;61:7–18.
56. Flückiger C, Wüsten G, Zinbarg R, et al. Resource Activation: Using Clients’ Own Strengths in Psychotherapy and Counseling. Boston, MA: Hogrefe Publishing; 2010.
57. Cheavens JS, Strunk DR, Lazarus SA, et al. The compensation and capitalization models: a test of two approaches to individualizing the treatment of depression
. Behav Res Ther. 2012;50:699–706.
58. Kessler RC, Nelson CB, Mcgonagle KA, et al. Comorbidity of DSM-III-R major depressive disorder in the general population: results from the US National Comorbidity Survey. Br J Psychiatry Suppl. 1996;168:17–30.
59. Barlow DH. Unraveling the mysteries of anxiety
and its disorders from the perspective of emotion theory. Am Psychol. 2000;55:1247–1263.
60. Brown TA, Chorpita BF, Barlow DH. Structural relationships among dimensions of the DSM-IV anxiety
and mood disorders and dimensions of negative affect, positive affect, and autonomic arousal. J Abnorm Psychol. 1998;107:179–192.
61. Barlow DH, Allen LB, Choate ML. Toward a unified treatment for emotional disorders. Behav Ther. 2004;35:205–230.
62. Leichsenring F, Steinert C. Towards an evidence-based unified psychodynamic protocol for emotional disorders. J Affect Disord. 2018;232:400–416.
63. Morin CM, Gibson D, Wade J. Self-reported sleep and mood disturbance in chronic pain
patients. Clin J Pain. 1998;14:311–314.
64. Smith MT, Haythornthwaite JA. How do sleep disturbance and chronic pain
inter-relate? Insights from the longitudinal and cognitive-behavioral clinical trials literature. Sleep Med Rev. 2004;8:119–132.
65. Esfahlani FZ, Visser K, Strauss GP, et al. A network-based classification framework for predicting treatment response of schizophrenia patients. Expert Syst Appl. 2018;109:152–161.
66. Granger CW. Investigating causal relations by econometric models and cross-spectral methods. Econometrica. 1969;37:424–438.
67. Tschacher W, Ramseyer F. Modeling psychotherapy process by time-series panel analysis (TSPA). Psychother Res. 2009;19:469–481.
68. Ebner-Priemer UW, Trull TJ. Ecological momentary assessment of mood disorders and mood dysregulation. Psychol Assess. 2009;21:463–475.