Chronic pain domains and their relationship to personality, abilities, and brain networks : PAIN

Journal Logo

Research Paper

Chronic pain domains and their relationship to personality, abilities, and brain networks

Pinto, Camila Bonina,b; Bielefeld, Jannisa,b; Barroso, Joanaa,b; Yip, Byronc; Huang, Lejiana,b; Schnitzer, Thomasc,d,e; Apkarian, A. Vaniaa,b,c,d,*

Author Information
PAIN 164(1):p 59-71, January 2023. | DOI: 10.1097/j.pain.0000000000002657

1. Introduction

Chronic pain remains a leading cause of disability in the United States and one of the major reasons patients seek medical care.21 Besides its high prevalence, chronic pain is notoriously difficult to treat, and most patients lack long-term pain relief.67,87,101 Part of the problem lies in the complex evaluation of the pain experience because pain is expressed by multiple factors not limited to its intensity, duration, and location.41 Chronic pain also encompasses physical functioning, emotional, and cognitive changes.69 Other than the intensity of pain, these multiple factors play a critical role in medical management decisions and in evaluating therapeutical effects and risks of different treatments.

Behavioral and neuroimaging studies have started to uncover neural processes associated with chronic pain; there is convincing evidence of specific brain properties and patients' psychosocial and physical traits being primarily affected in chronic pain patients.7–9 Several groups have studied changes in functional brain networks between healthy and chronic pain patients, showing specific network architecture properties of chronic pain conditions.8,10,11 Moreover, these traits were previously shown to have an impact on chronic pain. For instance, pain catastrophizing49,85 and fear of pain are considered strong predictors of chronic pain, whereas pain acceptance was associated with lower pain intensity, less anxiety, depression and pain avoidance.44,59

However, these factors are often studied in isolation and mostly in association with the pain intensity,9,10,12,37,98 which contributes to the present lack of integration between psychosocial factors, brain networks, and the underlying mechanism of chronic pain. Earlier studies of our group show relationships between functional brain networks and psychosocial factors—including personality, psychology, social interaction, and ability—affecting chronic back pain (CBP) patients.92 These findings support the use of multimodal approaches taking into consideration the relationship between brain connectivity and psychosocial factors with the general clinical domains of chronic pain.28,48 We believe that the study of multiple pain outcomes with different psychosocial modalities may help to (1) understand how pain characteristics—including intensity, quality, and pain-related disability—can be partially explained by these psychosocial properties; (2) relate brain regions and biological processes to the observed top-down sensory, cognitive, and affective responses commonly associated with chronic pain; and (3) point to new strategies for pain management.

Therefore, the first aim of this study was to identify a set of domains based on CBP patients' pain experience, focusing on clinical pain outcomes. Six core outcomes were defined by the Initiative on Methods, Measurement, and Pain Assessment in Clinical Trials (IMMPACT), 3 of them (pain intensity, physical functioning, and emotional functioning measurements) were included in the pain domains analysis.69 The second aim of this study was to relate these generalized pain domains to psychosocial factors in CBP patients. Finally, we aimed to identify brain networks that are reliably associated with our chronic pain domains.

2. Methods

2.1. Study design and procedures

This study is the analysis of an ongoing randomized clinical trial: d-Cycloserine for the Treatment of Chronic, Refractory Low Back Pain (clinicaltrials.gov: NCT03535688). The trial was reviewed and approved by the Northwestern University Institutional Review Board (IRB ID: STU00205398). Written informed consent was obtained from all participants before any trial procedures. Only data obtained before treatment initiation are studied here.

2.2. Study population

This analysis includes all participants who underwent screening and baseline visits from the beginning of the trial in March 2018 until November 2020. One hundred seven participants were included in the pain domains analysis and in the biopsychosocial & ability (biopsy&ab) prediction models, and 71 participants were included in the functional magnetic resonance imaging (fMRI) analysis (not all participants underwent the MRI procedure).

Participants were eligible if they were between 18 and 75 years old, have had back pain for at least 6 months, experienced an average pain score of ≥4 (on a 0-10 pain numeric rating scale [NRS]) over a 5- to 7-day period before the beginning of the trial, had no inflammatory arthropathy or fibromyalgia, and had no history of seizures or other neurological diseases. Moreover, all participants completed a clinical health physical examination in addition to the MRI safety screening requirements.

At visit 1 (screening visit), participants were screened for eligibility. Those who were qualified were asked to record their pain levels on an NRS scale twice daily using a phone app and return for a baseline visit in 10 to 14 days. During screening and baseline visits, participants were asked to complete a battery of questionnaires used in this analysis (Supplementary Tables 1 and 2, available at https://links.lww.com/PAIN/B627).

2.3. Chronic pain domain variables: questionnaire and tests

All participants included in this study completed the standard NIH Toolbox (ability) assessments and a set of personality questionnaires. The questionnaires included in this analysis are part of the ongoing randomized clinical trial.

We first extracted the chronic pain domains from a set of measurements and questionnaires that pertain to the IMMPACT core outcomes. The IMMPACT recommends the assessments of 6 core outcomes: (1) pain, (2) physical functioning, (3) emotional functioning, (4) participant ratings of improvement and satisfaction with treatment, (5) symptoms and adverse events, and (6) participant disposition.69 Note that core outcomes 4 to 6 relate to subjects being exposed to a treatment. Thus, they do not apply to this observational cross-sectional study92 (Supplementary Table 1, available at https://links.lww.com/PAIN/B627). Data from these questionnaires were scored according to their references. The following questionnaires and subscores were included in this initial analysis: (1) NRS pain intensity45; (2) Short-Form McGill Pain Questionnaire66; (3) PainDETECT Questionnaire30; (4) The negative affect subscore from the Positive and Negative Affect Schedule99; (5) Beck Depression Inventory14; (6) SF-12 Health Survey (SF-12)98; (7) Oswestry Disability Index27; (8) the pain intensity, depression, pain interference, social interaction, and physical health subscores from the Patient-Reported Outcomes Measurement Information System (PROMIS-57)18; and (9) the pain interference subscore from the NIH Toolbox.23

For NRS pain intensity, all participants completed pain ratings on an NRS scale during screening and baseline visits. In addition, the daily pain intensity (referred in this article as pain diary) was collected by our phone app between screening and baseline (approximately 12 ± 4 days).

We performed a principal component analysis (PCA) on these questionnaires to reduce the dimensionality of the data while retaining the underlying trends and patterns. Parallel analysis was performed to determine the final number of components to retain from the PCA.29 Finally, we varimax-rotated the components to reduce overlap of the loadings for easier interpretation. We call these components pain domains and components were labeled based on their loadings. Figure 1 summarizes these steps and show a methodological overview of the article.

F1
Figure 1.:
Methodological overview. (A) Data-driven approach: pain domains. Twenty-tree pain-related outcomes questionnaires were selected based on the IMMPACT core outcomes. A PCA was performed to reduce the dimensionality of the data, and components were labeled based on their loadings. (B) Independent variables: (1) biopsychosocial & abilities (biopsy&ab): 77 questionnaires scores and subscores; (2) brain connectivity: after preprocessing pipeline, linear Pearson correlations were performed on time courses averaged within each of the 256 ROIs generating a 256 × 256 correlation matrix for each subject. A total of 32,640 independent variables (per-subject connections) were filtered based on the correlation between connections and the dependent variables—here the pain domains from the PCA. Features correlating below the significance threshold of P-thresh = 0.05 were kept and fed into the regression model as independent variables. (C) Modeling approach: data were randomly split into training (70%) and testing datasets (30%). A L1-regularized linear model optimizing the L1-parameter using a 5-fold cross-validation was fitted. The goodness-of-fit of the optimal model was then obtained on the testing sample. This entire procedure was repeated 20 times, this also allows for confidence interval estimation. Model coefficients are averaged between those 20 results to obtain the final models were reported one for each pain domain. (D) Neurosynth decoder. We first generated 3 brain images (one for each pain domain model), in which the ROIs were weighted based on the model coefficients values. The brain images were used in the Neurosynth decoder to retrieve a list of terms and correlation values. The 30 most associated nonanatomical terms were represented using word clouds. (E) Mediation analysis: for this analysis, we linearize our models by comparing the pain predicted by both models; therefore, we only tested mediation for the final models obtained for each domain. We used a simple bias-corrected nonparametric bootstrap method.8 Our statistical models come with a multitude of variables. The mediation analysis was performed for both possible mediators—functional brain connectivity or biopsy&ab predictions. The exact choice is irrelevant for this analysis; thus, directionality is not determinable. In both cases, there was only a partial mediation effect. PCA, principal component analysis; ROIs, regions of interest.

2.4. Independent variables: biopsychosocial questionnaires and ability tests

The biopsy&ab-independent variables are general measures of emotion, character, personality, social interaction, and psychological measures.

We included 77 biopsy&ab-independent variables: 31 variables were included within the NIH Toolbox cognition, motor, sensory, and emotion domains in addition to 20 psychological measures including personality tests, emotion, social interaction, and character with their respective subscore, resulting in 44 questionnaire scores, in addition to age and gender. Note that most scores have multiple subscores, which increased this number past the total number of questionnaires (full list in Supplementary Table 2, available at https://links.lww.com/PAIN/B627). These subscore and biopsy&ab variables were classified according to their main questionnaire goal to facilitate interpretation of results.

Performance ability was assessed by the NIH Toolbox measures, only the visual acuity and language were not assessed. We used z-scored “uncorrected” (no gender or age corrections) standard scores except for the grip strength and dexterity measures for which raw scores were used. All these NIH Toolbox scores are standardized considering the normative, healthy dataset.32

Missing data were mean-imputed; however, the impact thereof was small: In the NIH Toolbox measures, only 0.39% of the data were missing. In the remaining questionnaires, only 0.55% of the data were missing and, again, were mean-imputed. No outliers were found for each questionnaire score; thus, no subjects were excluded.

Questionnaires were scored according to their references. They included the following assessments: (1) Positive and Negative Affect Schedule99; (2) Pain Catastrophizing Scale (PCS)84; (3) Pain Sensitivity Questionnaire75; (4) Pain Anxiety Symptoms Scale (PASS-20)60; (5) Multidimensional Assessment of Interoceptive Awareness65; (6) Patient-Reported Outcomes Measurement Information System (PROMIS-57)18; (7) Chronic Pain Acceptance Questionnaire62,63; (8) Emotional Regulation Questionnaire36; (9) Attentional Control Scale26; (10) Emotional Attentional Control Scale13; (11) Five-Facet Mindfulness Questionnaire6,7; (12) NEO Five-Factor Inventory24; (13) Live Orientation Test—Revised76; (14) Loss Aversion Questionnaire8; and (15) NIH Toolbox (all scores found in Supplementary Table 2, available at https://links.lww.com/PAIN/B627).38 These were all z-scored before performing statistical analyses (PASS scores already come z-scored).

2.5. Brain imaging protocol, acquisition, and processing

Study participants were scanned with a 3-T S Magnetom Prisma whole-body scanner using a 64-channel head/neck coil. The GRAPPA sequence,35 used to obtain T1-anatomical brain images, was run with the following parameters: a voxel size of 1 × 1 × 1 mm3, a repetition time/echo time (TR/TE) = 2.3 seconds/2.4 ms, a flip angle of 9°, the in-plane resolution = 256 × 256, using 176 slices per volume, and a field of view of 256 mm.

The resting-state functional magnetic resonance imaging (fMRI) images were acquired on the same day using the following parameters: a voxel size of 2 × 2 × 2 mm3, a TR/TE = 555/22 ms, a flip angle of 47°, the in-plane resolution = 96 × 104, using 1110 volumes, a multiband accelerator = 8, and 64 slices acquired with interleaved ordering, covering the entire brain from the cerebellum to the vertex.

The preprocessing was performed using the FMRIB 5.0.8 software library (FSL),43 MATLAB2018a and in-house scripts. For each patient, after the first 20 volumes were removed to eliminate saturation effects and achieve steady-state magnetization, 1090 volumes remained and the following steps were performed: motion correction, intensity normalization, nuisance regression of 6 motion vectors, signal-averaged overall voxels of the eroded white matter and ventricle region, and global signal of the whole brain. After this, to obtain the low-frequency fluctuations of the resting-state fMRI signal, the blood oxygenation level–dependent time series were band-pass filtered to 0.008 to 0.1 Hz by applying a Butterworth filter.

All preprocessed fMRI data were registered to the MNI152 2-mm template by using a 2-step procedure, in which the mean of the preprocessed fMRI data was registered with a 7-degrees-of-freedom affine transformation to its corresponding T1 brain; transformation parameters were computed by nonlinearly registering individual T1 brains to the MNI152 template (FNIRT2). Combining the 2 transformations by multiplying the matrices yielded transformation parameters normalizing the preprocessed fMRI data to the standard space.

Finally, the brain was divided into 256 spherical region of interests (ROIs), as described by Power et al.75 with 5-mm-radius ROIs, located at coordinates showing reliable activity across a set of tasks and displaying plausible functional structure. Blood oxygenation level–dependent signal of each ROI was then extracted. Eight cerebellar ROIs were excluded from the analysis because of the scanner field-of-view used during acquisition.

2.6. Functional connectivity analysis

Ninety-two participants completed the fMRI data acquisition. To ensure data quality and detect outliers, we calculated individual participants' mean functional connectivity across the 256 ROIs. Three participants were excluded based on their mean functional connectivity being more extreme than that of the entire population by more than 2 SDs. Seventy-one participants had both fMRI and full questionnaire data; these data were then used to study brain function in association with pain domains.

To obtain the connectivity matrices, linear Pearson correlations were performed on time courses averaged within each of the 256 ROIs generating a 256 × 256 correlation matrix for each subject. All analyses were performed with both positive and negative connectivity values. A total of 32,640 independent variables (per-subject connections) were filtered before being fed into our regression model. This feature selection was based on the correlation between connections and the dependent variables—here the pain domains from the PCA. Features correlating below the significance threshold of pthresh=0.05 were kept. We obtained around 1700 separate features for each of the 3 dependent variables. Note that we also tested different pthresh  values without significantly changing the results. This step allowed us to obtain 3 single, sparse connectivity matrices per patient (one for each dependent variable), which sped up L1 model convergence. To account for the effect of age and gender, we fitted those 2 variables against each brain connectivity matrix and subtracted fitted results—residual values are used.

The Harvard-Oxford cortical and subcortical atlases, provided with FSL, were used to anatomically label individual ROIs within the 264 areas by Power et al.,43,80 which defines no labels (abbreviations can be found at Supplementary Table 3, available at https://links.lww.com/PAIN/B627).

2.7. Modeling approach

We used the biopsy&ab questionnaires and functional brain connectivity to develop linear regression models attempting to predict each pain domain. All statistical analyses were performed using in-house python code. All predictive models, both for relating questionnaires and functional brain connectivity, were fitted and evaluated using the same code: data were randomly split into training and testing datasets, training comprising 70% of the entirety. We fitted a L1-regularized linear model optimizing the L1 parameter using a 5-fold crossvalidation. The L1 procedure keeps models sparse by setting insignificant features' coefficients to zero. The goodness-of-fit (r2) of the optimal model was then obtained on the testing sample.

This entire procedure was repeated 20 times to reduce sampling bias, given the relatively small size of our dataset. Specific train-test splits can yield false-positive results, which we avoided with our ensemble approach. This also allows for confidence interval estimation. Model coefficients are averaged between those 20 results to obtain the final models reported herein—one for each pain domain.

For the functional connectivity models, the number of independent variables were greater than the number of subjects in the training sample.  However, the amount of overfit was manageable: while, unsurprisingly, in-sample (IS) r2  were close to one, out-of-sample (OOS) results were still significant, and goodness-of-fit was high. This is caused by correlated individual features.

The brain areas associated to each of the pain domains were separately interpreted using Neurosynth.34,103 The Neurosynth decoder function maps brain areas into commonly associated terms discussed in the literature. We first generated 3 brain images (one for each model), in which the ROIs were weighted based on the model coefficients values. The brain images were used in the Neurosynth decoder to retrieve a list of terms and correlation values. The 30 most associated nonanatomical terms were represented using word clouds.

2.8. Mediation analysis

We conducted a mediation analysis focusing on the simple triangular relationship between biopsy&ab traits, functional brain connectivity, and pain domains to estimate the relative importance between the 2 sets of independent variables. We used a simple bias-corrected nonparametric bootstrap method.11 Our statistical models come with a multitude of variables. Here, however, we chose to linearize our models by comparing the pain predicted by both models: biopsy&ab variables (X) and functional brain connectivity (M) to the actual dependent variable (Y). Whichever model is selected as the mediator—functional brain connectivity or biopsy&ab predictions—is irrelevant for this analysis and could be reversed. This linearization allowed for the simple use of the original model which is made available in python.95

3. Results

3.1. Multidimensional data analysis reveals 3 chronic pain domains

Here, we show that chronic pain is characterized by 3 domains: (1) pain magnitude; (2) pain affect & disability; and (3) pain quality (Supplementary Figure 1, available at https://links.lww.com/PAIN/B627). These aggregate the broad battery of clinical outcome measures covering pain perception and physical and emotional function administered in this study.

In detail, to classify these clinical pain domains (23 measures), we used the set of test and surveys that pertain to the IMMPACT core outcomes.80 We evaluated a total of 107 chronic low back pain subjects with a mean age of 57.0±12.8 years, the female-to-male ratio was 1 (50 male and 57 female subjects), and the mean pain at screening was 6.5 (0-10 NRS rating). Principal component analysis on these variables revealed the combinations that most succinctly described this dataset. Three components explained 60% of the total variance, the remainder was made up of idiosyncratic noise (Supplementary Figure 1, available at https://links.lww.com/PAIN/B627). Figure 2A shows the correlation matrix of individual questionnaire scores, the components are outlined in purple.

F2
Figure 2.:
Correlations within and composition of 3 distinct domains of chronic pain perception. PCA determines questionnaire loadings onto the domains. Parallel analysis was used to obtain the maximum number of components above noise, 3 here (see Figure S1, https://links.lww.com/PAIN/B627). The 3 components were varimax-rotated to minimize overlap of loadings therein. (A) Covariance matrix of the 23 questionnaires used to extract underlying chronic pain characteristics. The 3 purple squares surround the questionnaires with the highest loadings onto the domains. (B) Actual questionnaire loadings onto the pain magnitude, pain affect & disability, and pain quality domains. The x-axis is shared among all 3 plots, and the data are standardized, so loadings can be compared directly. A cutoff of 0.2 excludes small loadings from the plots. The color representation of the 3 pain domains is maintained throughout the figures. PCA, principal component analysis.

The first component—pain magnitude—loaded onto questionnaires reflecting pain intensity (NRS, MPQ pain intensity), including data collected by our electronic diary. The second component, pain affect & disability, included positive loadings for depression, negative affect and disability scores, and negative loadings of physical and mental function. Finally, the third component, pain quality, included positive loadings for sensory and affective qualitative aspects of pain scores (Figure 2B). Higher scores always mean increased intensity of the pain experience.

The 3 pain domains correlate with each other (Person correlation coefficients: magnitude vs affect ρ=0.39, magnitude vs quality ρ=0.50, and affect vs quality ρ=0.30).25,40 Importantly, the 3 pain domains only correlate weakly with age and gender: The most significant, pain quality vs age has an r2=0.02 and is insignificant at P=0.12. The results in this research are therefore independent of age and gender.

Overall, we identified 3 chronic pain domains using clinical pain outcomes, reflecting the physical, emotional, and cognitive aspects of pain experience. These pain domains are used as dependent variables in the subsequent predictive models.

3.2. Relating chronic pain domains to personality, psychology, social interaction, and ability

Next, we modeled the relationship between each pain domain and general measures of emotion, character, personality, and psychological traits as well as ability performance. We refer to these measures as the biopsy&ab-independent variables and there are 77: 31 NIH Toolbox variables, 44 questionnaire scores, age, and gender (Table S3, https://links.lww.com/PAIN/B627).

Of the 3 pain domains, only 2 were statistically significantly modeled by these independent variables: pain magnitude and pain affect & disability. Pain magnitude is significant at P=0.007±0.01 with an out-of-sample (OOS) r2=0.32±0.13 (μ±σ). The average number of features selected was 27.5±8.7 (Table 1). The top independent variables are plotted in Figure 3, and full model results can be examined in Supplementary Table 4 and Supplementary Figure 3 (available at https://links.lww.com/PAIN/B627). In detail, PCS helplessness and the multidimensional assessment of interoceptive awareness not-worrying scores accounted for higher pain intensity, whereas the PROMIS-57 social, the Chronic Pain Acceptance Questionnaire pain willingness scores, list sorting performance and dexterity account for lower pain intensity (Figure 3A). Overall, pain magnitude relates to emotional control, attention, working memory, and openness to experience from the NEO Five-Factor Inventory test. These results confirm that CBP patients with higher rather than lower pain levels are more helpless, have lower satisfaction with their participation in social roles and activities, present less intellectual curiosity, and show a lower overall capacity to regulate and adapt behaviorally. Interestingly, higher pain scores were also correlated with higher tendency to ignore and regulate distress related to pain and discomfort—implying that avoidance and control are not effective to control pain in the long run.

Table 1 - Biopsy&ab and resting-state connectivity models describing the 3 pain domains.
Mean OOS r Mean OOS P Mean IS r Mean IS P Mean num indeps Std OOS r Std OOS P Std IS r Std IS P Std num indeps
Biopsy&ab models
 Pain affect & disability 0.886 0.0 0.947 0.0 19.7 0.022 0.0 0.006 0.0 2.3
 Pain magnitude 0.573 0.0057 0.864 0.0 27.45 0.115 0.011 0.079 0.0 8.6
 Pain quality 0.325 0.148 0.721 0.000001 15.95 0.168 0.207 0.130 0.0 12.4
Brain connectivity models
 Pain quality 0.687 0.0046 1.0 0.0 122.7 0.102 0.012 0.0 0.0 16.2
 Pain magnitude 0.640 0.0094 1.0 0.0 143.4 0.112 0.022 0.0 0.0 47.9
 Pain affect & disability 0.603 0.023 0.99 0.0 65.0 0.163 0.033 0.007 0.0 34.3
Mean values and SDs are sampled across 20 different train-test split samples. The number of independent components (indeps) is determined by the regularization parameter (not displayed).
biopsy&ab, biopsychosocial & ability; OOS, out-of-sample.

F3
Figure 3.:
Averaged linear regression model coefficients against (A) pain magnitude and (B) pain affect & disability (not-selected count as zero in average). The L1-regularized linear models get averaged on bootstrapped data samples. It chooses from 77 independent variables, including NIH Toolbox tests and DCS psychosocial questionnaire scores.

Pain affect & disability was related to the independent variables with a significance of P close to zero and an OOS r2=0.78±0.04. The number of independent variables selected in the final model is 19.4±2.0 (Table 1). This result is by far dominated by the PROMIS-57 social score, which measures the ability to participate in social activities, companionship, emotional support, and social satisfaction (Figure 3B). Anxiety and, again, PCS helplessness appear in the model as well as general feelings and attitudes about life. These results show that higher pain affect & disability in CBP is correlated with the disruption of physical and social functioning, which highlights the importance of social factors to pain-related emotional distress. Therefore, social functioning should be considered as an important factor for new intervention techniques.

Contrary to the other 2 components, pain quality could not be represented by these independent variables with an OOS P=0.15±0.2 (Table 1).

To evaluate the direct age and gender impact on these models, we measured their impact on the prediction by comparing their absolute coefficient values to the total of absolute coefficient values of the model (age and gender are included as independent variables). Their impact on the pain domain predictions is low—2.0% for pain magnitude and 0.6% for pain affect & disability, which reflects the pain domains' low correlations with age and gender.

We show that distinct biopsy&ab variables relate to distinct domains of the pain experience, further validating the existence of distinct clinical dimensions of the disease.

3.3. Chronic pain domains' unique brain network signatures

Finally, we investigated whether specific patterns of brain activity were related to each of the pain domains. We show that resting state functional connectivity reliably models all 3 pain domains.4,56 As independent variables, we used age- and gender-corrected connectivity matrices (see Methods for details).

Pain magnitude correlates with an OOS r2=0.41±0.14 to functional connectivity and is significant OOS at P=0.009±0.02 (asymmetric distribution/no negative P values). This predictive model includes 143±48 (Table 1) network edges on average and exhibited dense functional connectivity across and within all brain areas (Figure 4A–C).

F4
Figure 4.:
Pain domains brain connectivity linear regression model. Of 32,640 connectivities, the models chose ∼100 connectivities predicting the 3 pain components. Networks are split into positive (pain amplifying) and negative (pain mitigating) components. (A, D, and G) Positive coefficients: Circle plots display connectivity model coefficients above 0.2. The legend quantifies model coefficients. (B, E, and H): Negative coefficients: Circle plot displays connectivities with model coefficients smaller than −0.2. (C, F and I) The top 10 model connectivities by absolute coefficient size. Brain plots show the location, not the size, of the corresponding ROI. ROIs used in the modelling are smaller at 5 mm. Names are using the Harvard-Oxford atlas, abbreviations are found in Table S1, https://links.lww.com/PAIN/B627. (J) Corresponding networks for each ROI. ROIs, regions of interest.

The most important predictors (ie, biggest absolute coefficients in the final, averaged model) are displayed in Figure 4C: the connectivity between the right insular cortex and the left temporal fusiform cortex (posterior division); the right precentral gyrus and the right superior parietal lobule; the right frontal orbital cortex and the left frontal pole; and the left central opercular cortex and the left juxtapositional lobule cortex (formerly the supplementary motor cortex). In high-level networks, most of the links were within the sensorimotor (SMN) and the cingulo-opercular task control network (CON), followed by the default mode network (DMN), auditory network (AN) and visual network (VN). This suggests the importance of multisensory integration for pain severity prediction as well as attention control processes.39,51,79 These results are in line with previous literature showing that multiple systems contribute to chronic pain magnitude; here we show specific subsystems representing high and low pain magnitude.

The pain affect & disability domain correlated with an OOS r2=0.35±0.20 to functional connectivity and was significant OOS at P=0.035±0.057 (Table 1). The final model included 65±34 positive and negative weight network edges (Figure 4D–F). The most important predictive links included connectivity between the right angular gyrus to the right supramarginal gyrus (posterior division) within the frontoparietal task control network, the left paracingulate gyrus to the right frontal pole (both within the default mode network), and the left cingulate gyrus (anterior division) to the right middle frontal gyrus, both within the salience network (Figure 4F). The highest degree nodes are also present in the DMN, which plays an important role in regulating affective and motivational components of pain,9,19,74 followed by the SMN, VN, and task control network.

The pain quality domain was predicted by brain function at OOS r2=0.43±0.22, its OOS significance P=0.036±0.107.

The final model included 123±16 network edges. The most important edges to the model come from the connectivity from the left planum temporale to the right parahippocampal gyrus (posterior division); left insular cortex to the right central opercular cortex; and the right precentral gyrus to the left parahippocampal gyrus (posterior division). These connections mostly integrate the DMN, SMN, and the cingulo-opercular task control networks (Figure 4G–I).

Interestingly, no 5-mm ROI-to-ROI connectivity finds itself as part of all 3 pain domain models, which highlights how different the 3 pain domains brain representations are.

To help distinguish between our models, we combined the individual 5-mm ROIs into high-level brain regions using the Harvard-Oxford structural atlas (Figure 5A). As a second simplification, we display regions that edges connect rather than actual connectivity. Although, for example, the right precentral cortex plays a large role in all the 3 models, other regions such as the right insular cortex are more specific to pain magnitude. Moreover, more positive brain region coefficients predict increases in the pain domains (ie, higher pain magnitude) and negative ones vice versa.

F5
Figure 5.:
High-level similarities and differences of brain region importance in pain-predicting models. Each bar is split into the 3 pain components in addition to positive and negative contributions. These contributions are the sums of all positive and negative model coefficients (respectively) connecting into the corresponding ROIs. Displayed are the top 15 regions with the largest absolute model coefficient contributions. The Harvard-Oxford ROIs are bigger than the parcellations used for modelling (5-mm spheres), so multiple neighboring connectivity endpoints count toward the same ROI in this figure. Abbreviations are given in Table S1, https://links.lww.com/PAIN/B627. ROIs, regions of interest.

To further generalize and help to interpret the models, we used 12 large-scale functional brain networks (we aggregated the hand/mouth SMN) to visualize the sum of the predictive weights (model coefficients) of each model72 (Supplementary Figure 3, available at https://links.lww.com/PAIN/B627). We show that most model coefficients integrate within and across connectivity of the SMN, DMN, COP, VN, and FPN, in all 3 models (3 pain domains). Note how some networks equally correlate positively and negatively with the pain domains, whereas others correlate unequally, for example, the salience network with pain affect & disability.

Additionally, we used Neurosynth to provide a meta-analytic decoding of the neural map of each pain domain. For this, for each one of the 3 models, we multiplied the obtained ROI maps by the models' coefficients, thus creating a weighted representation of the brain given coefficient weights from the models. The top 30 terms of each model are represented in Figure 5B. Although pain magnitude integrates somatosensory aspects, pain affect & disability domain relates much more to memory aspects, whereas pain quality is a mixture of both.

3.4. Mediation between brain and questionnaire relationships

To test the relationship between the functional brain connectivity, the biopsy&ab models, and the pain domains, we performed a mediation analysis. We only tested mediation for the final models obtained for each domain, using a linear multiple-regression approach. We show that the resting-state functional brain connectivity only partially mediates the previously described relationship of biopsy&ab effects over the pain domains for pain magnitude and affect & disability.

These results suggest that the 2 models are complementing each other in predicting pain domains: original predictions are still significant in the presence of the mediator although the mediation pathway is a dominant path accounting for the total variance (Figure 6). For pain quality, mediation analysis does not apply as biopsy&ab traits do not relate significantly to this domain.

F6
Figure 6.:
Mediation analysis between the biopsy&ab traits, the resting-state fMRI, and the pain domains. The curved arrow with the corresponding numbers explains the indirect effect, taking both classes of independent components into consideration. The bottom arrow denotes the direct effect of biopsy&ab traits on the dependent variable. 2-
σ
confidence intervals are included to show that none of the paths are insignificant (including a zero-coefficient within its confidence intervals). Bootstrapped P-values are denoted by asterisks using *
P<0.05
, **
P<0.01
, and ***
P<0.001
. (A) Mediating the pain magnitude shows partial mediation. (B) Mediating pain affect & disability shows partial mediation. Both models are therefore enhanced using the fMRI data; it only partially co-aligns with the biopsy&ab traits. P-values are lower than those in the results section, because here we include all data points, train and test. biopsy&ab, biopsychosocial & ability; fMRI, functional magnetic resonance imaging.

4. Discussion

In this study, we characterized CBP based on patient-reported outcomes. We identified 3 pain domains—pain magnitude, pain affect & disability, and pain quality—with 2 of these domains related to distinct biopsy&ab properties. Additionally, we show unique brain connectivity profiles for each domain, thus indicating that the cost of chronic pain in the brain extends beyond its nociceptive properties. All 3 domains involved highly distributed neural processes across functional brain networks, particularly involving the SMN, DMN, and CON. We used Neurosynth to provide a meta-analytic term-based interpretation for the functional connectivity identified for each pain domain. We show that pain magnitude brain functional connectivity is related mostly to sensory and motor terms; pain affect & disability circuitry with memory terms; and pain quality circuitry to both memory and sensory and motor terms. Finally, mediation analysis showed that the relationship between the biopsy&ab properties and pain domains is to a large part mediated by the brain functional connectivity networks.

4.1. Chronic pain domains

To define chronic pain domains, we used questionnaires that reliably yield clinical pain outcomes. The IMMPACT core outcomes are recommendations from an expert group, in which each questionnaire is classified based on their general objectives. Here we used a data-driven approach to uncover dimensionality of pain domains. As expected, standard clinical categorizations of pain align with our results because the questionnaires selected for this approach followed the original classification.89 However, we observed some differences. IMMPACT classifies emotional and physical pain responses as different core outcomes, whereas our data-driven analysis show that CBP patients' emotional and physical responses co-vary and can be better described as 2 manifestations of the same domain: pain affect & disability. Although separating these dimensions seems a logical approach, both represent the extent to which pain can cause distress, unpleasantness, and decreased quality of life. Moreover, the effects of chronic pain on physical and psychosocial disability are shown to be synchronized, and both affect treatment response.58,61,64 Summarizing this effect as one pain domain is therefore not surprising. Moreover, while in IMMPACT pain intensity and pain quality are clustered as “pain” outcomes, here we show that pain quality and measurements of pain intensity—pain intensity, pain variability and pain interference—are classified into different domains.

4.2. Chronic pain domains and biopsychosocial & ability variables

Biopsychosocial & ability traits and brain connectivity changes associated with chronic pain are frequently studied using univariate or simple correlational analyses.31,90,92 Here, by contrast, we use a multidimensional approach to take into consideration all the factors, accounting for possible influences among each other and their relative importance. For instance, other studies focus on personality changes—from the five-factor model of personality—linked to chronic pain,86,100 which we show to be less affected, and less predictive, as compared with other biopsy&ab measures. Overall, however, research points in the same direction as our findings22,47,77; for example, pain catastrophizing, lower social relationships, and anxiety indicate higher pain intensity and pain-related disability. Such traits have been repeatedly shown to contribute to chronic pain. Yet, across domain overlapping variables, pain catastrophizing reinforces its overall influence on the chronic pain state. In contrast, nonoverlapping factors highlight their domain specific contribution. For example, while pain magnitude is associated with mindfulness and awareness, social relationships dysfunction seems to be the major contributor for higher pain affect & disability domain scores.

Even though these 3 pain domains influence each other, their biopsy&ab characterizations highlight specific and common contributions. These relationships are consistent with the literature regarding the impact of treatments on chronic pain. For example, decreases in pain magnitude do not necessarily change pain quality trajectories,91 even though pain quality can be altered by different analgesic interventions.52 This outcome-independence is also experienced between pain intensity and affect & disability domains. Similarly, although pain magnitude and psychosocial deterioration frequently co-occur and influence each other, clinical trials are inconclusive regarding this relationship.46,83 Future studies are necessary to more systematically examine treatment influences on both domains and underlying biopsy&ab factors.

Importantly, we show a lack of predictivity between biopsy&ab traits and the pain quality domain. We hypothesize that this lack of association may be because of the putative stronger association of this domain with the injury characteristics (nociceptive inputs). For example, although capsaicin-induced skin and muscle pain lead to similar pain intensity scores, they differ in pain quality descriptors.20 Note that the independent variables evaluated here are more reflective of psychological, emotional, personality, and performance ability traits rather than the injury itself.3

4.3. Pain domain relationships to functional brain connectivity

Functional connectivity explaining pain magnitude domain integrated mainly the middle temporal gyrus, insula, supramarginal gyrus, and frontal orbital cortex. These areas are frequently associated with chronic pain9,12,33,41,78,105 and in healthy subjects associated with inhibition control, emotional regulation, language, and decision making.41 Our findings underline the importance of networks and regions associated with multisensory integration, such as the DMN and insula, predicting chronic pain magnitude.

For pain affect & disability domain, functional connectivity engaged mainly the lingual gyrus, frontal pole, and cingulate gyrus. These areas are known to process visual memories, high-order behaviors like decision making and arbitrating between competing goals, and emotion formation and processing.55 Moreover, these areas are affected by pain.5,25,53,96 We show that the frontal pole relation to the left paracingulate gyrus appears as an important contribution to this domain, linking it to emotion formation; similar relationship has also been found in other studies16,70,97 (Supplementary Figure 2, available at https://links.lww.com/PAIN/B627). Previously, our group outlined the interaction between chronic pain and emotional learning and memory mainly through the impact of pain on the prefrontal cortex, including the cingulate gyrus and frontal pole.53,88 Here, we show specific associations between memory processing brain areas with the pain affect & disability domain, implying that memory and emotional associations might be more related with suffering and unpleasantness than with pain intensity itself. Moreover, chronic pain patients with memory complaints also report a lack of family support and dissatisfaction with their social and sexual life as compared with patients with the same pain levels without memory complaints.42,57

For pain quality domain, functional connectivity engaged mainly the parahippocampal gyrus, frontal pole, and precuneus. Vachon-Presseau et al.94 show parahippocampal gyrus involvement in anticipatory anxiety and associative learning, traits that both determine pain quality experiences. The precuneus cortex is traditionally known to be involved with reflection and episodic memory,17,82,102 and volumetric studies also show plasticity therein related to chronic pain.54 Overall, connectivity relating to pain quality was more local than functional connectivity predicting the other pain domains, consistent with the notion that it may be better related to nociceptive characteristics.

4.4. Mediation of biopsy&ab with brain circuitry regarding pain domains

The mediation results show that brain and biopsy&ab parameters consistently explain more of the variance of the 2 pain domains where such analysis was possible. Yet, they also highlight existence of unaccounted additional factors. Perhaps, the latter would also be related to nociceptive characteristics of CBP.

4.5. Limitations

While the prediction of chronic pain has been studied elsewhere,1,15,109 this study focuses on back pain patients only, thereby constraining any analysis on variance of the pain state itself, as opposed to comparing it with a healthy control group. Any conclusions drawn in this study are therefore only applicable to CBP. Also, the extent of invariance in the identified relationships across different chronic pain conditions, many of which differ in their underlying biology (eg, somatic vs visceral conditions), is unknown. Moreover, the independent variables studied here were recorded cross-sectionally in the same CBP patients. Therefore, the cause and effect between the pain domains biopsy&ab factors and brain circuitry remain ambiguous.

In conclusion, our results establish the high-level dimensionality of chronic pain and show that distinct facets of the pain experience have different biopsy&ab and brain network correlates. Therefore, we highlight the complexity of chronic pain, identifying properties of constituent domains. It would be simpler to eliminate weaker factors from each explanatory model, but this is misleading and undoubtedly would fail to be replicated in future studies. The results also emphasize inclusion of measures that account for the natural pain variability, evaluating pain and related measures over weeks instead of one time point pain. How much of this profile is unique to the specific condition, CBP, varies across types of chronic pain and is shared between types of chronic pain remains to be pursued in the future. These findings suggest novel mechanistic relations and may provide better insights on treatment influences on constituent domains and underlying factors and brain circuits.

Conflict of interest statement

The authors have no conflicts of interest to declare.

Appendix A. Supplemental digital content

Supplemental digital content associated with this article can be found online at https://links.lww.com/PAIN/B627.

Acknowledgements

The authors want to thank all the members of the Apkarian laboratory for their contributions and discussion in this work. The authors would also like to thank all study participants for their time and participation in this study A.V. Apkarian, C.B. Pinto, and J. Bielefeld designed the research. C.B. Pinto and J. Bielefeld performed the data analysis. C.B. Pinto and J. Bielefeld drafted the manuscript. A.V. Apkarian, J. Barroso, MB, and T. Schnitzer provided intellectual advice. A.V. Apkarian and J. Barroso revised the manuscript. BY coordinates the study. L. Huang preprocessed the fMRI data. T. Schnitzer provided clinical oversight over the study. All authors contributed to the article and approved the submitted version. This work was funded by Center for chronic pain and drug abuse 1P50DA044121-01A1, Department of Defense award # W81XWH-17-1-0426 and J. Bielefeld received funding from T32AR007611.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

References

[1]. Althaus A, Hinrichs‐Rocker A, Chapman R, Becker OA, Lefering R, Simanski C, Weber F, Moser KH, Joppich R, Trojan S. Development of a risk index for the prediction of chronic post‐surgical pain. Eur J Pain 2012;16:901–10.
[2]. Andersson JL, Jenkinson M, Smith S. Non-linear registration, aka Spatial normalisation FMRIB technical report TR07JA2. FMRIB Anal Group Univ Oxford 2007;2:e21.
[3]. Apkarian AV, Baliki MN, Geha PY. Towards a theory of chronic pain. Prog Neurobiol 2009;87:81–97.
[4]. Apkarian AV, Hashmi JA, Baliki MN. Pain and the brain: specificity and plasticity of the brain in clinical chronic pain. PAIN 2011;152(3 Suppl):S49.
[5]. Apkarian AV, Thomas PS, Krauss BR, Szeverenyi NM. Prefrontal cortical hyperactivity in patients with sympathetically mediated chronic pain. Neurosci Lett 2001;311:193–7.
[6]. Baer RA, Smith GT, Hopkins J, Krietemeyer J, Toney L. Using self-report assessment methods to explore facets of mindfulness. Assessment 2006;13:27–45.
[7]. Baer RA, Smith GT, Lykins E, Button D, Krietemeyer J, Sauer S, Walsh E, Duggan D, Williams JMG. Construct validity of the five facet mindfulness questionnaire in meditating and nonmeditating samples. Assessment 2008;15:329–42.
[8]. Baets SD, Buelens M. Development of the loss aversion questionnaire, 2012.
[9]. Baliki MN, Mansour AR, Baria AT, Apkarian AV. Functional reorganization of the default mode network across chronic pain conditions. PLoS One 2014;9:e106133.
[10]. Baliki MN, Petre B, Torbey S, Herrmann KM, Huang L, Schnitzer TJ, Fields HL, Apkarian AV. Corticostriatal functional connectivity predicts transition to chronic back pain. Nat Neurosci 2012;15:1117.
[11]. Baron RM, Kenny DA. The moderator–mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations. J Personal Soc Psychol 1986;51:1173.
[12]. Barroso J, Wakaizumi K, Reis AM, Baliki M, Schnitzer TJ, Galhardo V, Apkarian AV. Reorganization of functional brain network architecture in chronic osteoarthritis pain. Hum Brain Mapp 2021;42:1206–22.
[13]. Barry TJ, Hermans D, Lenaert B, Debeer E, Griffith JW. The eACS: attentional control in the presence of emotion. Personal Individual Differences 2013;55:777–82.
[14]. Beck AT, Ward CH, Mendelson M, Mock J, Erbaugh J. An inventory for measuring depression. Arch Gen Psychiatry 1961;4:561–71.
[15]. Boersma K, Linton SJ. Expectancy, fear and pain in the prediction of chronic pain and disability: a prospective analysis. Eur J Pain 2006;10:551–7.
[16]. Bromm B. The involvement of the posterior cingulate gyrus in phasic pain processing of humans. Neurosci Lett 2004;361:245–9.
[17]. Cavanna AE, Trimble MR. The precuneus: a review of its functional anatomy and behavioural correlates. Brain 2006;129:564–83.
[18]. Cella D, Choi SW, Condon DM, Schalet B, Hays RD, Rothrock NE, Yount S, Cook KF, Gershon RC, Amtmann D, DeWalt DA, Pilkonis PA, Stone AA, Weinfurt K, Reeve BB. PROMIS(®) adult health profiles: efficient short-form measures of seven health domains. Value Health 2019;22:537–44.
[19]. Chang PC, Centeno MV, Procissi D, Baria A, Apkarian AV. Brain activity for tactile allodynia: a longitudinal awake rat functional magnetic resonance imaging study tracking emergence of neuropathic pain. PAIN 2017;158:488–97.
[20]. Chang PF, Arendt-Nielsen L, Graven-Nielsen T, Svensson P, Chen AC. Comparative EEG activation to skin pain and muscle pain induced by capsaicin injection. Int J Psychophysiol 2004;51:117–26.
[21]. Control CFD, Prevention. Prevalence and most common causes of disability among adults-United States, 2005. MMWR 2009;58:421–6.
[22]. Cook AJ, Brawer PA, Vowles KE. The fear-avoidance model of chronic pain: validation and age analysis using structural equation modeling. PAIN 2006;121:195–206.
[23]. Cook KF, Dunn W, Griffith JW, Morrison MT, Tanquary J, Sabata D, Victorson D, Carey LM, Macdermid JC, Dudgeon BJ, Gershon RC. Pain assessment using the NIH Toolbox. Neurology 2013;80:S49–53.
[24]. Costa PT. NEO personality inventory-revised (NEO PI-R). Odessa, FL: Psychological Assessment Resources, 1992.
[25]. Dehghan M, Schmidt‐Wilcke T, Pfleiderer B, Eickhoff SB, Petzke F, Harris RE, Montoya P, Burgmer M. Coordinate‐based (ALE) meta‐analysis of brain activation in patients with fibromyalgia. Hum Brain Mapp 2016;37:1749–58.
[26]. Derryberry D, Reed MA. Anxiety-related attentional biases and their regulation by attentional control. J Abnorm Psychol 2002;111:225–36.
[27]. Fairbank JC, Pynsent PB. The oswestry disability index. Spine 2000;25:2940–52. discussion 2952.
[28]. Foster NE, Anema JR, Cherkin D, Chou R, Cohen SP, Gross DP, Ferreira PH, Fritz JM, Koes BW, Peul W. Prevention and treatment of low back pain: evidence, challenges, and promising directions. Lancet 2018;391:2368–83.
[29]. Franklin SB, Gibson DJ, Robertson PA, Pohlmann JT, Fralish JS. Parallel Analysis: a method for determining significant principal components. J Vegetation Sci 1995;6:99–106.
[30]. Freynhagen R, Baron R, Gockel U, Tolle TR. painDETECT: a new screening questionnaire to identify neuropathic components in patients with back pain. Curr Med Res Opin 2006;22:1911–20.
[31]. Gatchel RJ. Clinical essentials of pain management. Washington, DC: American Psychological Association, 2005.
[32]. Gershon R. NIH toolbox norming study. Harvard Dataverse, 2016. Available at: https://support.dataverse.harvard.edu/.
[33]. Glass JM, Williams DA, Fernandez-Sanchez ML, Kairys A, Barjola P, Heitzeg MM, Clauw DJ, Schmidt-Wilcke T. Executive function in chronic pain patients and healthy controls: different cortical activation during response inhibition in fibromyalgia. J Pain 2011;12:1219–29.
[34]. Gorgolewski KJ, Varoquaux G, Rivera G, Schwarz Y, Ghosh SS, Maumet C, Sochat VV, Nichols TE, Poldrack RA, Poline JB. NeuroVault. org: a web-based repository for collecting and sharing unthresholded statistical maps of the human brain. Front Neuroinform 2015;9:8.
[35]. Griswold MA, Jakob PM, Heidemann RM, Nittka M, Jellus V, Wang J, Kiefer B, Haase A. Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magn Reson Med Official J Int Soc Magn Reson Med 2002;47:1202–10.
[36]. Gross JJ, John OP. Individual differences in two emotion regulation processes: implications for affect, relationships, and well-being. J Personal Soc Psychol 2003;85:348.
[37]. Hashmi JA, Baliki MN, Huang L, Baria AT, Torbey S, Hermann KM, Schnitzer TJ, Apkarian AV. Shape shifting pain: chronification of back pain shifts brain representation from nociceptive to emotional circuits. Brain 2013;136:2751–68.
[38]. Hodes RJ, Insel TR, Landis SC. The NIH toolbox: setting a standard for biomedical research. Neurology 2013;80(11 Suppl 3):S1.
[39]. Hofbauer RK, Rainville P, Duncan GH, Bushnell MC. Cortical representation of the sensory dimension of pain. J Neurophysiol 2001;86:402–11.
[40]. Hoffman MD, Hoffman DR. Does aerobic exercise improve pain perception and mood? A review of the evidence related to healthy and chronic pain subjects. Curr pain headache Rep 2007;11:93–7.
[41]. Ichesco E, Schmidt-Wilcke T, Bhavsar R, Clauw DJ, Peltier SJ, Kim J, Napadow V, Hampson JP, Kairys AE, Williams DA. Altered resting state connectivity of the insular cortex in individuals with fibromyalgia. J Pain 2014;15:815–26. e811.
[42]. Jamison RN, Sbrocco T, Parris WC. The influence of problems with concentration and memory on emotional distress and daily activities in chronic pain patients. Int J Psychiatry Med 1989;18:183–91.
[43]. Jenkinson M, Beckmann CF, Behrens TE, Woolrich MW, Smith SM. FSL. Neuroimage 2012;62:782–90.
[44]. Jensen MP, Hakimian S, Ehde DM, Day MA, Pettet MW, Yoshino A, Ciol MA. Pain-related beliefs, cognitive processes, and electroencephalography band power as predictors and mediators of the effects of psychological chronic pain interventions. PAIN 2021;162:2036–50.
[45]. Jensen MP, Karoly P. Self-report scales and procedures for assessing pain in adults. In D.C. Turk & R. Melzack (Eds.), Handbook of pain assessment. New York: The Guilford Press, 1992. pp. 135–151.
[46]. Katz J, Melzack R. Auricular transcutaneous electrical nerve stimulation (TENS) reduces phantom limb pain. J Pain Symptom Manage 1991;6:73–83.
[47]. Kroenke K, Yu Z, Wu J, Kean J, Monahan PO. Operating characteristics of PROMIS four-item depression and anxiety scales in primary care patients with chronic pain. Pain Med 2014;15:1892–901.
[48]. Lee J, Gupta S, Price C, Baranowski A. Low back and radicular pain: a pathway for care developed by the British Pain Society. Br J Anaesth 2013;111:112–20.
[49]. Lewis GN, Rice DA, McNair PJ, Kluger M. Predictors of persistent pain after total knee arthroplasty: a systematic review and meta-analysis. Br J Anaesth 2015;114:551–61.
[50]. Loggia ML, Jensen K, Gollub RL, Wasan AD, Edwards RR, Kong J. The catechol-O-methyltransferase (COMT) val 158 met polymorphism affects brain responses to repeated painful stimuli. PLoS One 2011;6:e27764.
[51]. Lu C, Yang T, Zhao H, Zhang M, Meng F, Fu H, Xie Y, Xu H. Insular cortex is critical for the perception, modulation, and chronification of pain. Neurosci Bull 2016;32:191–201.
[52]. Malavera A, Silva FA, Fregni F, Carrillo S, Garcia RG. Repetitive transcranial magnetic stimulation for phantom limb pain in land mine victims: a double-blinded, randomized, sham-controlled trial. J Pain 2016;17:911–18.
[53]. Mansour A, Farmer M, Baliki M, Apkarian AV. Chronic pain: the role of learning and brain plasticity. Restorative Neurol Neurosci 2014;32:129–39.
[54]. Mao C, Wei L, Zhang Q, Liao X, Yang X, Zhang M. Differences in brain structure in patients with distinct sites of chronic pain: a voxel-based morphometric analysis. Neural Regen Res 2013;8:2981.
[55]. Maratos E, Dolan RJ, Morris J, Henson R, Rugg M. Neural activity associated with episodic memory for emotional context. Neuropsychologia 2001;39:910–20.
[56]. May A. Chronic pain may change the structure of the brain. PAIN 2008;137:7–15.
[57]. Mazza S, Frot M, Rey AE. A comprehensive literature review of chronic pain and memory. Prog Neuropsychopharmacol Biol Psychiatry 2018;87:183–92.
[58]. McCracken LM. Attention to pain in persons with chronic pain: a behavioral approach. Behav Ther 1997;28:271–84.
[59]. McCracken LM. Learning to live with the pain: acceptance of pain predicts adjustment in persons with chronic pain. PAIN 1998;74:21–7.
[60]. McCracken LM, Dhingra L. A short version of the Pain Anxiety Symptoms Scale (PASS-20): preliminary development and validity. Pain Res Manag 2002;7:45–50.
[61]. McCracken LM, Eccleston C. A prospective study of acceptance of pain and patient functioning with chronic pain. PAIN 2005;118:164–9.
[62]. McCracken LM, Vowles KE, Eccleston C. Acceptance of chronic pain: component analysis and a revised assessment method. PAIN 2004;107:159–66.
[63]. McCracken LM, Vowles KE, Eccleston C. Acceptance-based treatment for persons with complex, long standing chronic pain: a preliminary analysis of treatment outcome in comparison to a waiting phase. Behav Res Ther 2005;43:1335–46.
[64]. McCracken LM, Zhao-O’Brien J. General psychological acceptance and chronic pain: there is more to accept than the pain itself. Eur J Pain 2010;14:170–5.
[65]. Mehling WE, Acree M, Stewart A, Silas J, Jones A. The multidimensional assessment of interoceptive awareness, version 2 (MAIA-2). PLoS One 2018;13:e0208034.
[66]. Melzack R. The short-form McGill pain questionnaire. PAIN 1987;30:191–7.
[67]. Müller-Schwefe G, Morlion B, Ahlbeck K, Alon E, Coaccioli S, Coluzzi F, Huygen F, Jaksch W, Kalso E, Kocot-Kępska M. Treatment for chronic low back pain: the focus should change to multimodal management that reflects the underlying pain mechanisms. Curr Med Res Opin 2017;33:1199–210.
[68]. Odrcich M, Bailey JM, Cahill CM, Gilron I. Chronobiological characteristics of painful diabetic neuropathy and postherpetic neuralgia: diurnal pain variation and effects of analgesic therapy. PAIN 2006;120:207–12.
[69]. Patel KV, Amtmann D, Jensen MP, Smith SM, Veasley C, Turk DC. Clinical outcome assessment in clinical trials of chronic pain treatments. Pain Rep 2021;6.
[70]. Porreca F, Navratilova E. Reward, motivation, and emotion of pain and its relief. PAIN 2017;158:S43–9.
[71]. Power JD, Cohen AL, Nelson SM, Wig GS, Barnes KA, Church JA, Vogel AC, Laumann TO, Miezin FM, Schlaggar BL. Functional network organization of the human brain. Neuron 2011;72:665–78.
[72]. Power JD, Fair DA, Schlaggar BL, Petersen SE. The development of human functional brain networks. Neuron 2010;67:735–48.
[73]. Raja SN, Carr DB, Cohen M, Finnerup NB, Flor H, Gibson S, Keefe FJ, Mogil JS, Ringkamp M, Sluka KA, Song XJ, Stevens B, Sullivan MD, Tutelman PR, Ushida T, Vader K. The revised International Association for the Study of Pain definition of pain: concepts, challenges, and compromises. PAIN 2020;161:1976–82.
[74]. Ren W, Centeno MV, Berger S, Wu Y, Na X, Liu X, Kondapalli J, Apkarian AV, Martina M, Surmeier DJ. The indirect pathway of the nucleus accumbens shell amplifies neuropathic pain. Nat Neurosci 2016;19:220–2.
[75]. Ruscheweyh R, Marziniak M, Stumpenhorst F, Reinholz J, Knecht S. Pain sensitivity can be assessed by self-rating: development and validation of the Pain Sensitivity Questionnaire. PAIN 2009;146:65–74.
[76]. Scheier MF, Carver CS, Bridges MW. Distinguishing optimism from neuroticism (and trait anxiety, self-mastery, and self-esteem): a reevaluation of the Life Orientation Test. J Personal Soc Psychol 1994;67:1063.
[77]. Schütze R, Rees C, Preece M, Schütze M. Low mindfulness predicts pain catastrophizing in a fear-avoidance model of chronic pain. PAIN 2010;148:120–7.
[78]. Seminowicz DA, Moayedi M. The dorsolateral prefrontal cortex in acute and chronic pain. J Pain 2017;18:1027–35.
[79]. Senkowski D, Höfle M, Engel AK. Crossmodal shaping of pain: a multisensory approach to nociception. Trends Cognitive Sciences 2014;18:319–27.
[80]. Smith SM, Dworkin RH, Turk DC, McDermott MP, Eccleston C, Farrar JT, Rowbotham MC, Bhagwagar Z, Burke LB, Cowan P. Interpretation of chronic pain clinical trial outcomes: IMMPACT recommended considerations. PAIN 2020;161:2446–61.
[81]. Smith SM, Jenkinson M, Woolrich MW, Beckmann CF, Behrens TE, Johansen-Berg H, Bannister PR, De Luca M, Drobnjak I, Flitney DE, Niazy RK, Saunders J, Vickers J, Zhang Y, De Stefano N, Brady JM, Matthews PM. Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage 2004;23 Suppl 1:S208–19.
[82]. Stillesjo S, Nyberg L, Wirebring LK. Building memory representations for exemplar-based judgment: a role for ventral precuneus. Front Hum Neurosci 2019;13:228.
[83]. Sturgeon JA, Dixon EA, Darnall BD, Mackey SC. Contributions of physical function and satisfaction with social roles to emotional distress in chronic pain: a Collaborative Health Outcomes Information Registry (CHOIR) study. PAIN 2015;156:2627.
[84]. Sullivan MJ, Bishop SR, Pivik J. The pain catastrophizing scale: development and validation. Psychol Assess 1995;7:524.
[85]. Sullivan MJ, Thorn B, Haythornthwaite JA, Keefe F, Martin M, Bradley LA, Lefebvre JC. Theoretical perspectives on the relation between catastrophizing and pain. Clin J Pain 2001;17:52–64.
[86]. Suso-Ribera C, Gallardo-Pujol D. Personality and health in chronic pain: have we failed to appreciate a relationship? Personal Individual Differences 2016;96:7–11.
[87]. Tompkins DA, Hobelmann JG, Compton P. Providing chronic pain management in the “Fifth Vital Sign” Era: historical and treatment perspectives on a modern-day medical dilemma. Drug Alcohol Depend 2017;173:S11–21.
[88]. Tu Y, Jung M, Gollub RL, Napadow V, Gerber J, Ortiz A, Lang C, Mawla I, Shen W, Chan S-T. Abnormal medial prefrontal cortex functional connectivity and its association with clinical symptoms in chronic low back pain. PAIN 2019;160:1308–18.
[89]. Turk DC, Dworkin RH, Allen RR, Bellamy N, Brandenburg N, Carr DB, Cleeland C, Dionne R, Farrar JT, Galer BS. Core outcome domains for chronic pain clinical trials: IMMPACT recommendations. PAIN 2003;106:337–45.
[90]. Turk DC, Okifuji A. Psychological factors in chronic pain: evolution and revolution. J Consult Clin Psychol 2002;70:678–90.
[91]. Vachon-Presseau E, Abdullah TB, Berger SE, Huang L, Griffith JW, Schnitzer TJ, Apkarian AV. Validating a biosignature predicting placebo pill response in chronic pain in the settings of a randomized controlled trial. PAIN. 2022;163:910–22.
[92]. Vachon-Presseau E, Berger SE, Abdullah TB, Griffith JW, Schnitzer TJ, Apkarian AV. Identification of traits and functional connectivity-based neurotraits of chronic pain. PLoS Biol 2019;17:e3000349.
[93]. Vachon-Presseau E, Centeno M, Ren W, Berger S, Tetreault P, Ghantous M, Baria A, Farmer M, Baliki M, Schnitzer T. The emotional brain as a predictor and amplifier of chronic pain. J Dental Res 2016;95:605–12.
[94]. Vachon-Presseau E, Roy M, Martel MO, Caron E, Marin M-F, Chen J, Albouy G, Plante I, Sullivan MJ, Lupien SJ, Rainville P. The stress model of chronic pain: evidence from basal cortisol and hippocampal structure and function in humans. Brain 2013;136:815–27.
[95]. Vallat R. Pingouin: statistics in Python. J Open Source Softw 2018;3:1026.
[96]. Veer IM, Beckmann C, Van Tol MJ, Ferrarini L, Milles J, Veltman D, Aleman A, Van Buchem MA, Van Der Wee NJ, Rombouts SA. Whole brain resting-state analysis reveals decreased functional connectivity in major depression. Front Syst Neurosci 2010;4:41.
[97]. Vogt BA. Pain and emotion interactions in subregions of the cingulate gyrus. Nat Rev Neurosci 2005;6:533–44.
[98]. Ware J Jr, Kosinski M, Keller SD. A 12-Item Short-Form Health Survey: construction of scales and preliminary tests of reliability and validity. Med Care 1996;34:220–33.
[99]. Watson D, Clark LA, Tellegen A. Development and validation of brief measures of positive and negative affect: the PANAS scales. J Pers Soc Psychol 1988;54:1063–70.
[100]. Weisberg JN. Personality and personality disorders in chronic pain. Curr Rev pain 2000;4:60–70.
[101]. Weisberg MB, Clavel AL Jr. Why is chronic pain so difficult to treat? Psychological considerations from simple to complex care. Postgrad Med 1999;106:141–64.
[102]. Wirebring LK, Stillesjo S, Eriksson J, Juslin P, Nyberg L. A similarity-based process for human judgment in the parietal cortex. Front Hum Neurosci 2018;12:481.
[103]. Yarkoni T, Poldrack RA, Nichols TE, Van Essen DC, Wager TD. Large-scale automated synthesis of human functional neuroimaging data. Nat Methods 2011;8:665.
[104]. Yarnitsky D, Crispel Y, Eisenberg E, Granovsky Y, Ben-Nun A, Sprecher E, Best LA, Granot M. Prediction of chronic post-operative pain: pre-operative DNIC testing identifies patients at risk. PAIN 2008;138:22–8.
[105]. Zhuo M. Contribution of synaptic plasticity in the insular cortex to chronic pain. Neuroscience 2016;338:220–9.
Keywords:

Aspects of pain; Pain predictions; Human ability; Personality and character; NIH Toolbox; Brain functional connectivity

Supplemental Digital Content

© 2022 International Association for the Study of Pain