Network theory and psychological functioning

Network theory (NT) has been used to describe the structure and functioning of dynamic complex systems by using Graph Theory [1,2]. The basic idea of NT is that systems can be represented as patterns of non-overlapping elements (represented as nodes) which are interconnected (represented as edges). The graph summarizes the pattern of relations among the elements in a network’s topology [3].

Current conceptualizations of brain functioning use NT to describe the complex functioning of neural circuitries and their connection to different type of data (e.g. performance in cognitive tasks) which has opened the field of ‘network neuroscience’ [4]. Within this framework, it is nowadays widely assumed that brain operates as a network and such organization underlies information processing, emotions, sensations, or thoughts [5–7]. However, in standard current network approaches, such mental representations are still considered as independent entities for which, at least for some aspects of them (e.g. fear reaction, reward and episodic memory), neuroscience research has found underlying networks of neural activity [8,9]. Thus, unfortunately, psychological constructs are still far from being perceived as a psychological network itself. For instance, whereas attentional tasks are linked to ventral- and dorsal-attention networks [10], the connection between attention, and other cognitive components (e.g. working memory), or emotional components (e.g. anxiety) is almost entirely unknown [10,11].

NT and network analyses have also been recently used as an innovative framework to understand psychopathology [12]. The literature of NT in psychopathology has grown very rapidly in the last few years and has being applied to explore a relatively large variety of psychological problems as shown in a recent systematic review [13]. From a network perspective, psychological disorders are not entities leading to symptoms (as it has been commonly assumed in traditional causal models of mental disorders). Rather, psychological disorders are understood as networks of elements (basically, symptoms and signs) which are pairwise associated forming a dynamically complex system with potential mutual casual influences. Psychological disorders, according to this point of view, could be nothing else that the very network of interconnected symptoms [14,15]. The radical departure of network theory from other current diagnostic approaches is that it defines mental disorders as conditions consisting of strongly connected symptom networks, with no assumption of a latent entity subsumed under the symptoms [14]. Nevertheless, some network theorists have recently suggested that dwelling excessively on the lack of interest of latent variables may misguide the focus from what is more promising in conceptualizing disorders from a NA perspective (i.e., the idea of causally interconnected elements) [16,17]. In this sense, the use of psychological constructs (e.g., traits assessed by questionnaires), and not only elements like symptoms or signs, is beginning to be used in NA (e.g., [18–20]) and it could also be possible that, in the future, hybrid models, using not only elements like signs and symptoms but also latent variables, could shed light on the connections between psychological elements [21,22].

NT provides several advantages over previous models of psychopathology. Firstly, network analyses allow to empirically identify which nodes have a central role within the network, which departs from the Diagnostic and Statistical Manual’s unproved assumption that all symptoms have the same diagnostic weight [23]. Secondly, and even more important, when longitudinal data are gathered, network analyses allow to explore potential causal links between the elements of the network [12,24], thereby identifying etiological pathways between symptoms (e.g. a sleep problem may lead to fatigue which leads to lack of energy and, finally, to anhedonia). Thirdly, network analyses also allow to explore time-related changes in the configuration of a given network which underlines the dynamic nature of psychological problems. These changes, that can be analyzed with mathematical tools [25], might be due to the mere passage of time or, more importantly, being the product of an intervention like the automatization of a cognitive task [26] or a psychological intervention [27,28]. Thus, analysis of different layers of networks along a temporal dimension (which may vary from milliseconds to months or years, depending on the type of study) allows to explore dynamic reorganizations of the elements of a network [25] which, in turn, may allow to identify key elements that can be targeted to promote changes in the network [29].

In sum, NT provides unique conceptual and analytic tools to understand in richer ways than in current prevailing models, the nature and etiology of mental disorders [15,30]. Expanding this approach to psychological models, in general, NT could provide a useful innovative paradigm to understand normal psychological functioning and its structure in a true holistic view [31].

The proposal of a ‘psychonectome’

Most of the psychology, psychiatry, and clinical and cognitive neuroscience experimental studies are based on gathering information from questionnaires or tasks to measure skills, knowledge, abilities, attitudes or personality traits with the aim to identify cognitive or clinical deficits and to find all type of behavioral, neurobiological, or contextual correlates. The way in which typically operates this approach of doing science is by assuming that scores from these tests somehow quantify the status of a particular psychological construct [32–34]. For example, it is assumed that a test measuring memory skills taps a construct relatively independent than another test measuring attention. Yet, despite this implicit assumption, the nature of mental functions and mental representations is more likely to be one of interconnectedness where each component operates in relation to others having complex dynamics of mutual relations. These mental functions and representations do not exist as absolute constructs that operate as independent modules. As it has happened in the history of neuroscience in relation to the understanding of the nature of neural activity (for a review see [35]), localizationism should be abandoned in psychology in favor of considering mind as a complex network of psychological constructs.

The network perspective has been fruitfully used in fields as diverse as the understanding of brain structures and functioning, mental disorders, microbiota or social interactions, to name a few [36–38]. Yet, NT could also be expanded to grasp a more complex view on the mutual dependencies of elements of psychological functioning. In the psychological realm, expanding the recent attempts to understand psychological disorders using NT, it could be hypothesized that psychological functioning might be modeled as a complex network of psychological variables or modules that may [39], or may not [40], be interconnected. Furthermore, using appropriate designs and analyses, that network might reveal that some elements causally influence others.

A network of psychological elements, following what Guloksuz et al. [41] have called the wave of the ‘-omics’ approach (e.g. connectomics, genomics, or even ‘symptomics’ [42]), could be labelled as psychonectome. In general, a functional ‘connectome’ is the name given to those networks that reflect the activity of neural elements along a temporal dimension connecting neurons or regions (Friston, 2011; Korzeniewska et al., 2011). This psychonectome could be defined as a complex ensemble of dependences between psychological constructs (e.g. visuospatial memory, selective attention, or emotion regulation skills). That network might be, in turn, connected to more basic layers of elements belonging to the individual level (e.g. neural activity networks, signs and symptoms) or even elements external to the individuals (e.g. stressors, environmental circumstances) but still with the ability of activating some elements of other layers.

One of the most relevant features of the proposed psychonectome is that, as it happens in any network approach, the focus of interest is transferred from the individual variables or constructs (e.g. a behavioral response, socio-demographic information, or the score obtained from a psychological or clinical test) to the relation between them. The strength of a network is based on its decentralization and the synergy between its components is stronger than the sum of them. This new conceptualization of how systems work is being used not only to understand the functioning of complex systems but could also be fruitfully applied to understand the dynamic interconnections between psychological constructs as conceptualized in the psychonectome proposal.

A psychonectome approach to a mindfulness-based intervention

A good case where the idea of psychonectome could be applied is the field of mindfulness- based interventions (MBI) as they tap a variety of psychological constructs. Mindfulness (MF) is defined as a state of consciousness with a present-orientated attention, on purpose, and non-judging [43]. Although mindfulness meditation has its origin over 2500 years ago in the ancient Buddhist traditions [44,45], the incorporation of several secular practices of MF in the western world is quite recent [46] and has generated great interest in the scientific community and the general public alike [47].

Despite the wide range of research on meditation benefits to date [48], relatively few studies have attempted to examine the action mechanisms through which meditation produces its effects (for reviews see [49–54]. Most of the theoretical models published to date emphasize the central role of attention regulation, which is thought to underpin emotional and cognitive flexibility [52], which in turn improves the emotion regulation processes [55], and the ability to maintain non-judging awareness of thoughts, feelings and experiences. Theoretical models also emphasize the importance of body consciousness and changes in self–perspective [51,56], as well as the role of self-compassion [50] in the promotion of changes due to MBI practices.

The most widely evaluated MBI is the pioneer Mindfulness-Based Stress Reduction (MBSR; [43,57]). MBSR is a treatment program originally developed within a hospital context for the management of stress caused by chronic pain, and subsequently applied in an extensive variety of problems like the reduction of comorbid symptoms in other health problems such as fibromyalgia [58], mood changes in patients with cancer [59] or multiple sclerosis [60].

Several meta-analysis and a growing body of robust empirical evidence from randomized controlled trials show that MBI is a promising treatment for a variety of mental health problems, including anxiety disorder [61], stress [62], depression symptoms [63] and depression relapses [64], substance abuse [65], and eating disorders [66], among others.

In addition to reducing psychological symptoms, mindfulness practice has been also shown to have positive effects on psychological well-being in healthy participants [67], quality of life improvements [62], empathy, compassion and prosocial behaviors [68] and cognitive functioning [49,51].

The present study

Given the complex nature of mindfulness interventions and the well-studied cognitive, emotional, and psychopathological components that have been delineated in current theoretical models of mindfulness [51,52,69], it was considered that psychological changes associated to the practice of mindfulness would be an excellent proof of concept of the psychonectome idea.

As far as we know, NA has not been applied yet to explore the relationships among different psychological variables before and after a standardized MBSR intervention. Thus, the aim of this study was to examine the MBSR’s impact on the network dynamics between mindfulness, compassion, well-being, psychological distress and emotional and cognitive control constructs, and how these constructs are reorganized after the intervention. Although this is likely the first study on network analysis applied to the field of mindfulness, we tested several inter-related hypotheses based on the extant evidence on the effects and mechanisms of MBSR. All the hypotheses are framed under two theoretical assumptions: The first one is that network analysis may reveal complementary information about the relation between psychological measures that cannot be inferred by standard univariate statistics (i.e. comparing pre and post scores on selected measures). The second assumption is network analysis is able to find the dependences between constructs rather than the study of the constructs per se.

In sum, the following hypotheses were set up: 1) as supported by the literature review [48,50,61,62], the MBSR would yield significant changes, in an adaptive direction, in mindfulness, compassion, psychological distress, psychological well-being, and emotional and cognitive control variables; 2) after the MBSR program the networks of these constructs would become topologically reorganized (as measured by network paths and topological parameters) expecting a higher connectivity, clustering and efficiency; 3) based on theoretical models of MBSR functioning [52] it was also hypothesized that mindfulness, emotion regulation and well-being constructs would increase their centrality in the resulting psychonectome after the MBSR; and, 4) it was expected that these constructs would regroup after the MBSR in psychologically meaningful sub-networks as calculated by means of community analysis.

Participants

A sample of 258 adults, enrolled in a standardized 8-week MBSR program between April and December 2017, were invited to participate in this research (75.2% accepted to participate and fulfilled the inclusion criteria described below). After applying all exclusion criteria see Fig 1), data from a total of 182 individuals were included in all analyses. Participants mean age was 43.7 (S.D. = 9.77), 70.9% women, 95.7% had higher education, 40.1% married, 72.5% employed, 13,7% physical illness, 56% previous meditation experience and the meditation years average was 4.19 (S.D. = 5.72). The research study was approved by the university ethics committee prior to participant recruitment.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Participation flow diagram.

https://doi.org/10.1371/journal.pone.0219793.g001

Inclusion criteria for MBSR program were as follows: 1) 18 years of age or more; 2) not having any current of serious psychological disorder or substance abuse / dependence. Statistical analyses were conducted only for data from participants who completed pre-assessment and attended a minimum of 6 sessions (i.e. 75% of the program). A precise description of the participation flow diagram is presented in Fig 1.

Procedure and materials

The study followed a pre-post design where the participants were blind to the aims of the study. Participants were invited to participate at the moment they registered in the official website offering the MBSR course. Those who accepted to participate, were administered a brief online screening questionnaire on demographics and inclusion criteria and were asked to sign an informed consent. Then participants were asked to complete online (via Qualtrics software) the set of questionnaires described in detail in the next section. The online assessment was completed the week before starting of the program (baseline assessment) and during the week after the end of the MBSR (post-assessment). When necessary, reminders were scheduled for those participants who hadn’t completed the questionnaires. Each pre-post online evaluation lasted approximately 45 minutes. After completing the post-treatment assessments, participants were rewarded with an individualized report of their questionnaires scores.

Measures / Assessment.

Sociodemographic and health information. For the baseline assessment a custom-made brief questionnaire, which included information about age, gender, education, occupation, psychological and physical health problems, was administered.

Meditation experience. A 32-item questionnaire, gathering information about previous meditation practice, meditation types and meditation retreats, was specifically designed for this study.

Constructs associated to mindfulness training. Five different areas, associated to the main outcomes and variables studied in the MF literature [52,70] were included. The constructs included:

1. Mindfulness (i.e. facets and mindfulness, decentering, non-attachment, and bodily awareness);
2. Compassion (i.e. compassion towards oneself and others and empathy)
3. Psychological well-being (i.e. satisfaction with life, optimism, and overall well-being).
4. Psychological distress (i.e. anxiety, stress, and depression)
5. Emotional and cognitive control (i.e. emotional regulation, rumination, thought suppression and attentional control)

Table 1 shows a brief description of the measures and the internal consistency scores found in our study (Cronbach’s alpha based on the polychoric correlations).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Constructs and instruments used in the study.

https://doi.org/10.1371/journal.pone.0219793.t001

Mindfulness-based stress reduction program (MBSR)

The MBSR programs was implemented at a university-associated center specialized in MBI. MBSR instructors were highly experienced and all were certified by the University of Massachusetts Center for Mindfulness (https://www.umassmed.edu/cfm/). Each instructor had more than 6 years of teaching experience and had conducted a minimum of 30 MBSR programs.

The MBSR program [86] consisted of 32-hour training during eight weeks, including a 3-hour initial orientation session, 7 weekly 2.5-hour of face-to-face sessions, an 8-hour intensive day of practice, 45 minutes of daily home formal and informal practices and a final 3.5-hour session. Training was conducted in groups of 20–30 participants. During the program, different mindfulness practices are performed, including focused attention on the breath, open monitoring of awareness in body-scanning, prosocial meditation (i.e. loving kindness and compassion) and gentle hatha yoga. To support the practice, each participant was given a set of pre-recorded audio files to guide daily practices and a MBSR Workbook. Program adherence was supported through regular group supervision meetings.

Data preprocessing and univariate statistical test

Data pre-processing, analyses of missing data and imputation methods were conducted with the SPSS v. 22. All the network analyses were carried out with R v. 3.3.1 and Matlab R2017b.

Following CONSORT guidelines [87], Intent-To-Treat (ITT) analyses were carried out following Newman's guidelines [88]. Maximum Likelihood (ML) estimation was performed via Expectation Maximization imputation (EM) using SPSS v.22 software. We followed the procedure proposed by Hair and colleagues [89] to treat missing data. We first tested for both the recommended limits of missing measures [90], with a 10.8% of overall missing values and their random patterns using a Little MCAR test (χ2 (1674) = 356.45, p > .05), concluding that missing data were completely random. To examine whether missing data could be predicted by other variables (attrition bias) a logistic regression was carried out by including clinical (DASS-21 stress, anxiety and depression symptoms), demographic (gender, age, education), and meditation experience (meditation practice and years of experience). Only the stress score at baseline was weakly associated with greater likelihood of post missing data (R² de Cox-Snell = 0.07; R² de Nagelkerke = 0.10). However, the correlation between stress and likelihood of missing data wasn’t significant. None of the included variables predicted missingness. Finally, Maximum Likelihood estimation (ML) was performed, and Sensitivity Analysis compared the results of the completers to the estimated values was carried out, concluding that ML estimation would not lead to biased estimations.

All measures were statistically compared (pre versus post) using a Student t-test with SPSS as showed in Table 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Paired comparisons of pre-post measures in the constructs assessed in the MBSR program.

(Description of the variables and their acronyms is shown in Table 1).

https://doi.org/10.1371/journal.pone.0219793.t002

Network analysis

NA was conducted to analyse the multiple relations (edges) between different psychological constructs (nodes) simultaneously, and how those relations would be reorganized after the MBSR. The successive steps procedure proposed for network analysis in psychology [91] was followed. Analyses were conducted adapting this procedure to our objectives: 1) pre and post MBSR network estimation; 2) pre and post MBSR network inference (topological characterization); and 3) pre and post MBSR network node communities analysis.

Pre-post psychological changes (Hypothesis 1)

To analyze the effects of the MBSR program on the constructs selected, a standard analysis of pre-post mean differences was performed. Table 2 lists the means score and standard deviation for each variable as well as the significant differences found. As it was hypothesized, there was a significant change in most of the dependent variables measuring mindfulness, compassion towards others and oneself, psychological well-being, and emotional and cognitive functioning, as well as a significant reduction of psychological distress. Thus, the MBSR program was effective in changing the psychological state of participating individuals as confirmed by standard statistical procedures, which is a common result in MBSR programs. The rest of the analyses conducted and reported in the next sections were aimed to go beyond this standard analytic strategy by using network analysis procedures.

MBSR network estimation (Hypothesis 2)

Before inferring the network reorganization after the MBSR, the network architecture in both pre- and post-intervention was analyzed. The pre and post MBSR regularized partial correlation networks are presented in Fig 2. Of all the possible 300 edges, 43.67% and 40.67% were estimated to be different from zero in pre- and post-assessments, respectively, which means that neither an all-to-all network nor a disconnected topology was found.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Network representations.

Network representations in pre (left panels A and C) and post (right panel B and panel D) MBSR intervention. A network is graphed by nodes (circles representing the constructs assessed in the MBSR program, described in Table 1 and named in Table 2) and edges (lines representing the statistical correlation between nodes as described in Methods section). Blue edges represent positive relationships and red edges represent negative relationships. The spatial position of nodes is chosen by the Fruchterman-Reingold algorithm to draw close those nodes with stronger and/or more connections while it places in the periphery nodes with low centrality. A) Regularized partial correlation networks pre-MBSR and B) post-MBSR intervention. The figure also shows paths, defined as a sequence of linked nodes that never visit a single node more than once, in pre-MBSR (panel C) and post-MBSR (panel D).

https://doi.org/10.1371/journal.pone.0219793.g002

A path is defined by the sequence of links between constructs and the corresponding strength of the dependence (shown here in parenthesis). Using a find path algorithm, three paths were identified in the pre-MBSR network (displayed in different colors in Fig 2, panel C and panel D):

1. A first path was found crossing transversely the network (shaded in yellow in Fig 2, Panel C). This path was formed by mindfulness and self-compassion measures: FFMQ-O—MAIA (0.33), MAIA—EQ (0.29), EQ—NAS (0.22), NAS—SCS-M (0.23), SCS-M—SCS-A (0.25), SCS-A—SCS-H (0.22).
2. A second path included clinical symptoms and rumination measures (shaded in red in Fig 2, Panel C): DASS-A–DASS-S (0.45), DASS-S–DASS-D (0.18), DASS-D–RRS-B (0.20), and RRS-B–RRS-R (0.21).
3. A third path included the rest of mindfulness measures and the self-reported attentional control measure (shaded in blue in Fig 2, Panel C): FFMQ-D–ACS (0.28), ACS–FFMQ-A (0.27), FFMQ-A–FFMQ-J (0.18).

Interestingly, some strongly related dyads (i.e. pairs of highly related constructs that are less dependent on the rest of the network) were also found. One dyad was related to self-compassion measures (IRI-E–CSP (0.34), whereas the rest were composed by measures related to distress and/or well-being: PHI–SWLS (0.42), DASS-S–DASS-A (0.45), WBSI—FFMQ-J (-0.20), and PHI—DASS-D (-0.19).

In regard to the network configuration after the MBSR, the three paths observed in the pre-MBSR network disappeared. Instead, the post-MBSR network appears reorganized in a different manner, emerging two visually differentiated sub-networks (see Fig 2) which were further mathematically explored (see the following Results section). A close inspection of the Fig reveals that in the upper half part of this post-MBSR network (shaded in red in Fig 2, Panel D), it appeared a subnetworks of nodes that are mostly related to psychopathological constructs [i.e. stress (DASS-S), anxiety (DASS-A), depression (DASS-D); thought suppression (WBSI); emotion suppression (ERQ-S); and rumination-brooding (RRS-B)]. In the lower area of the network (shaded in purple in Fig 2, Panel D), a second subnetwork emerged in which most of the nodes are adaptive psychological constructs (i.e. mindfulness, attentional control, self-compassion, compassion and psychological well-being). Thus, the overall reorganization of the networks indicates that the nodes of the different sets of measures were rather scattered or disconnected at pre-MBSR whereas, at post-MBSR, the nodes seem to be more closely reorganized.

Other six network features are noteworthy: 1) a psychologically interesting feature was that the “reappraisal” component of “Emotional Regulation Questionnaire” (ERQ-R; an adaptive emotional regulation strategy) was related to rumination (RRS-R), thought suppression (WBSI) and non-reactivity (FFMQ-R) at pre-MBSR; however, after the MBSR, reappraisal was related to mindfulness (EQ, FFMQ-R and FFMQ-O) and well-being measures (PHI); 2) in regard to the “suppression” component of “Emotional Regulation Questionnaire” (ERQ-S; a non-adaptive emotional regulation strategy–see [107]), which was with thought suppression (WBSI), depression (DASS-D) and Empathic Concern (IRI-E) after the MBSR, and maintained maintaining their negative relationships with compassion to others (CSP) and describing (FFMQ-D); 3) something similar happened with the “reflection” component of Ruminative Responses Scale (RRS-R; a short and mid-term adaptive strategy), which established positive relations with the describing facet of mindfulness (FFMQ-D) and attention control (ACS) after the MBSR; 4) another interesting feature was that the self-compassion measures (SCS) were quite disconnected from the well-being measures (PHI, SWLS and LOT) before the intervention but established strong connections after the MBSR; 5) in the same way, optimism (LOT), a variable that was quite disconnected from the rest of well-being measures (general well-being [PHI] and life satisfaction [SWLS]) before the intervention, increased its relations with them and with self-compassion after the MBSR; and 6) in post-MBSR network the thought suppression node (WBSI) established strong positive relations with other non-adaptive or psychopathological elements, such as emotional suppression (ERQ-S), brooding (RRS-B) and anxiety (DASS-A).

MBSR network inference (topological characterization) (Hypothesis 3)

The Expected Influence (EI) and strength of all psychological constructs were calculated (see Fig 3). The constructs with the highest EI scores, both at pre- and post-MBSR were: mindfulness measures such as non-react (FFMQ_R), interoceptive awareness (MAIA), nonattachment (NAS), decentering (EQ), self-compassion mindfulness (SCS_M) and the overall well-being measure (PHI). Only in four nodes [brooding (RRS_B), reflection (RRS_R), non-judgment (FFMQ_J) and emotional regulation-reappraisal (ERQ_R)], there was a discrepancy between EI and strength indicating that in those nodes there were many negative edges. No major changes were observed after the MBSR in either EI or strength.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Topological characterization of networks.

Topological characterization of pre- and post-MBSR intervention networks for Expected Influence and Strength. Constructs theoretically associated with MBSR are grouped according to classification given in Table 1. Each measure estimates the topological role of each node (x-axis, described in Table 2) in the network. A) Expected Influence: the sum of both positive and negative weights between a node and all other nodes in the network (nodes with high EI are positively strongly associated with other nodes in the network); B) Strength: sum of the absolute weights of all edges in the network involving that node (nodes with high strength are strongly associated with other nodes in the network); C) Degree: number of connections per node (the higher the degree, the more connected the network is); D) Clustering: closeness of a node to the other network nodes, sharing a clustering (high clustering means that a node’s neighbors are neighbors between them); E) Efficiency: average of inverse shortest path length (which is inversely related to the path length).

https://doi.org/10.1371/journal.pone.0219793.g003

In addition to EI, other centrality measures were calculated (see Fig 3): a) Degree: the analysis of hubs, defined as sets of nodes with the highest degree, indicated a pattern highly consistent with the results on EI and strength, with the exception of optimism (LOT), which showed a high number of relations (i.e. degree), but with low weights (i.e. low EI and strength); b) Clustering: nodes with the highest topological change were common humanity (SCS_H), compassion to others (CSP), depression (DASS_D), thought suppression (WBSI), and attentional control (ACS); c) Efficiency: nodes with the highest changes in efficiency were thought suppression (WBSI), attentional control (ACS), and reflection (RRS_R). It must be noted that nodes with highest changes in either clustering or efficiency are interpreted as having a key role in the network reorganization.

Pre-post MBSR predictability values for each node are presented in Supplementary materials (S2 and S3 Figs). The average predictability in both networks was similar, rating from 0.55 (pre-MBSR) to 0.57 (post-MBSR), indicating that an overall average of 56% of the variance of a node was predicted by its neighbors at both moments of assessment. As compared to other network studies in psychology, following the criteria suggested by Haslbeck & Fried [104], the overall predictability was high. Also, additional analyses on the pattern of correlations between pre- and post-MBSR interventions on strength and degree, clustering and efficiency (see Supplementary Materials, Network inference) seemed to indicate that that there was a genuine network reorganization of psychological constructs after MBSR intervention.

MBSR network node communities (Hypothesis 4)

Fig 4 shows the results of the Community Detection analysis. The Spinglass algorithm detected six node communities in pre-MBSR network and five node communities in post-MBSR network.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Community detection analysis.

Results of the community detection analysis performed on the network shown in Fig 1. The identified communities depict the variables that more strongly inter-correlated.

https://doi.org/10.1371/journal.pone.0219793.g004

Six meaningful communities or clusters emerged in pre-MBSR network:

1. The largest cluster (Cluster B in blue in Fig 4), included 8 nodes that reflected psychological distress (anxiety [DASS-A], stress [DASS-S], and depression [DASS-D]) and emotional and cognitive control measures (brooding [RRS-B], reflection [RRS-R], reappraisal [ERQ-R], emotion suppression [ERQ-S] and thought suppression [WBSI]).
2. The second largest cluster F, depicted in red, included 7 nodes which all reflected mindfulness (decentering [EQ], non-attachment [NAS], non-reactivity [FFMQ-R], non-judgment [FFMQ-J] and mindfulness [SCS-M]) and self-compassion measures (common humanity [SCS-H] and self-kindness [SCS-A]).
3. The third largest cluster A (depicted in orange) included 3 nodes which integrates the well-being measures (general well-being [PHI], life satisfaction [SWLS] and optimism [LOT]).
4. The fourth largest cluster C (depicted in green) also include 3 nodes with other mindfulness measures (acting awareness [FFMQ-A], describing [FFMQ-D] and attentional control [ACS]).
5. Finally, two small clusters were found:
   1. Cluster E (depicted in purple) composed by the rest of mindfulness measures (interceptive awareness [MAIA] and observing [FFMQ-O]).
   2. Cluster D (depicted in yellow) composed by compassion measures (compassion to others [CSP] and empathic concern [IRI-E]).

These six large clusters were reorganized at post-MBSR, generating a 5-cluster final organization:

1. The mindfulness and self-compassion measures that were scattered in clusters F, C and E in pre-MBSR network, were reorganized in two big clusters after the MBSR:
   1. Cluster B (in blue) included non-attachment [NAS], self-compassion [SCS], non-judgment [FFMQ-J] and acting awareness [FFMQ-A].
   2. Cluster E (depicted in purple) included decentering [EQ], interoceptive awareness [MAIA], non-reactivity [FFMQ-R], observing [FFMQ-O] and reappraisal [ERQ-R].
2. The Psychological distress and Emotional and cognitive control measures remained grouped in the same cluster D (depicted in yellow), with the exception of emotional reappraisal [ERQ-R].
3. Well-being measures remained all together in orange cluster A (general well-being [PHI], life satisfaction [SWLS] and optimism [LOT]).
4. Finally, a mixed cluster C (depicted in green) was found, including compassion to others measures [CSP and IRI-E], attentional control [ACS] and describing [FFMQ-D].