Deconstructing chronicity of musculoskeletal pain: intensity-duration relations, minimal dimensions and clusters of chronicity

2.1.1 Patients of regional pain clinics

For the present study, initial assessment data of all n=261 patients eligible for the multicenter study suffering from clinically relevant cUBP or widespread muscle pain were considered. Of the 261 data sets 60 were incomplete or inconsistent leaving 201 patients qualifying. Further 16 patients met additional exclusion criteria so that 185 patients (107 cUBP, 78 FMS) entered the final analysis (CONSORT flow chart in the Supplementary material). Additional exclusion criteria were relevant drug taking or change in medication within 3 months prior to data collection; cardiovascular disease or hypertension not treatable with drugs, and renal insufficiency requiring dialysis assessed by doctor’s checklist (Supplementary material, Fig. S5). Entry assessment for mental disorders was done with the Structured Clinical Interview for DSM IV Axis I Disorders (SKID-I; [7]). Patients with major depression or anxiety disorders remained in the sample because affective comorbidity was a research question of the source projects. Comorbidity relations of chronicity as such were not subject of the present analysis and will be reported elsewhere.

cUBP criteria required that pain in upper or lower back was the primary problem and was not related to acute trauma, inflammatory or neurologic disease; radicular and neuropathic signs were also excluded on final medical investigation (Fig. S5: Doctor’s checklist in Supplementary material). Chronic widespread pain criteria corresponded to earlier ACR fibromyalgia (FMS) criteria based solely on muscle pain (11 of 18 tender points; [8]). This left 107 patients with cUBP and 78 with FMS diagnoses. Of the latter 63 matched also the FMS criteria suggested in 2010/2011 [9].

2.1.2 Employees at risk for musculoskeletal pain

Data of employees currently at work were acquired according to an adapted protocol of the patient study. The cooperating occupational health service centers collected the data, guaranteeing full anonymity of individualized data against employers as well as study partners. Employees were eligible for participating when in jobs requiring a seating or standing activity with high musculoskeletal load and established risk and prevalence of work-related musculoskeletal pain (detailed information on field of work and work schedule in Supplementary material, Table S1). Initially, nine companies in the program were interested and allowed contacting employees. One-hundred-and-forty employees fulfilling the inclusion criteria German language comprehension, age 18–65 years, actively working and reporting musculoskeletal pain at present, continuing or repeatedly during the last years were recruited this way; another 32 participants were acquired through advertisements in the press and brochures displayed in local practitioners’ offices. After exclusion of two persons without pain related to the musculoskeletal system 170 employees remained in the analysis (Table 1).

2.2.1 Pain assessment

Assessment of pain and related variables comprised the West Haven-Yale Multidimensional Pain Inventory (German version, MPI-D; [11]), the German Pain Questionnaire (Deutscher Schmerzfragebogen, DSF; [12], [13]) and the questionnaire for the Mainz Pain Staging System (MPSS; [14], [15]). The latter was only applied in the employee sample; for pain patients, MPSS variables were recoded from corresponding items of the DSF.

2.2.2 Chronicity measures

Chronicity was first coded according to Axis IV of the IASP Taxonomy of Pain [3], [16] using current pain intensity derived from the MPI-D [11]; item #1: present pain intensity) and the duration parameter of the DSF ([13]; item #25: time since onset). Secondly, von Korff’s Chronic Pain Grade questionnaire (CPG, German version; [17]) and the Mainz Pain Staging System (MPSS; [5]) were included as global chronicity indices and analyzed at item and subscale level. Only the results of the latter are reported here for brevity.

The CPG consists of item response theory based subscales [18], [19], [20] comprising three items on pain intensity (present, average, worst), three items on disability (interference with daily, recreational, social and family and work-related activities) using 11-point Likert scales to derive a disability score. An additional question concerns the number of days the person was not able to perform at work or carry out other relevant activities due to pain.

The MPSS is a multiaxial system for staging pain chronicity used in the German speaking area. Three stages of pain chronicity are derived from a compound sum score ranging from 4 to 12 points out of four “axes” of 3-point items. Axis I evaluates “temporal characteristics” of pain (occurrence pattern, episode duration, changes in intensity). Axis II codes “spatial aspects” of pain (number of painful areas). Axis III evaluates “medication use” (drug intake, previous withdrawal treatments). Axis IV concerns current and previous “utilization of the health care system/patient career” (number of physician changes, pain-related hospitalizations, pain-related operations, pain-related stays in a spa, rehabilitation center or pain center). Scores of 4–6 points on these items correspond to pain chronicity stage I, 7–8 points code as stage II, and 9–12 as stage III. The higher the “pain stage”, the more persistent the pain symptomatology and the more intense therapeutic intervention is needed for complicating factors [21].

2.2.3 Control variables

Control variables were pain medication and psychological comorbidity assessed with the DSF [13], the Center for Epidemiological Studies Depression (German version: ADS; [22]) and the State-Trait Anxiety Inventory (German version: STAI-T; [23]).

2.3.1 Correlation and frequency analyses

Relations of intensities to duration, CPG grades and MPSS stages were calculated as nonparametric correlations (Spearman’s Rho, Kendall’s Tau) or contingency coefficients (Pearson’s contingency coefficient C_corr, adjusted for number of categories [25]). Specific intensity characteristics at different durations and chronicity levels were further explored by analyses of frequency distributions across single duration classes and chronicity grades. Differences of pain intensity-duration characteristics and global chronicity indices were assessed by non-parametric planned post-hoc tests. Effects of control variables were checked by correlation and median-split analyses. Significance levels were Bonferroni-Holm corrected for multiple testing, family-wise and separately for each chronicity index and dataset of patients and employees. The significance level was set at p<0.05 throughout; exact probabilities are reported where appropriate.

2.3.2 Dimensional analyses at scale and item level

The dimensional structure of the IASP Taxonomy Axis IV coding, CPG and MPSS was explored by principal component analysis (PCA) and principal axis factoring (PAF) with varimax rotation and pairwise exclusion of cases with missing data for both samples separately and combined. The Kaiser and scree criteria were applied to determine the number of components to be extracted. Confirmatory factor analyses (CFA) were conducted with the lavaan package in R [26] to evaluate the dimensions derived from exploratory factor analyses by descriptive fit indices [27].

2.3.3 Latent class analyses

To substantiate the dimensional relations found and to identify specific variable groupings possibly obscured in conventional factor analyses, hierarchical latent class analysis (LCA, R program pvclust; [28], [29], [30]) was employed as second structure finding method. LCA generates variable groupings by a maximum likelihood model. It is apt to support and inform the results of classic dimensional analysis from a different perspective operating on the same data set. In addition, pvclust provides tests of robustness of cluster solutions. Probability values were calculated for each cluster with non-parametric bootstrap probability (BP) and approximately unbiased (AU) p-values in % ranging from 0 (not robust) to 100 (highly robust). To reduce a type 2 error, AU and BP values were uncorrected for multiple testing and used only for descriptive assessment of cluster dendrograms, not for inferential difference testing. Correlations between observed IASP Axis IV, CPG and MPSS marker values entered the cluster analysis with the average linkage method. Data were permuted 1000 times to assess the stability of cluster solutions.

3.2.1 Correlations of intensity and duration

As expected, pain intensities were only weakly although significantly associated with duration for the combined sample of patients and employees (Spearman’s ρ=0.24; Kendall’s τ=0.20; coefficient of association, corrected for number of categories, C_corr=0.39; p<0.01, adjusted).

However, the low overall correlation is misleading because correlations differed qualitatively and in sign between the subgroups (Table 3). In patients, pain intensity correlated weakly negatively with duration (ρ=−0.156; τ=−0.136; n=170; p<0.05, adjusted) for both cUBP and FMS patients. In contrast, pain intensity correlated weakly positively although insignificantly with duration in employees with MSP (ρ=+0.135; p=0.079; τ=+0.109; p=0.069 adjusted; n=170; C_corr n.s.).

3.2.2 Frequency distribution analyses

The inconsistent intensity-duration correlations found are likely due to nonlinear and group-dependent relations in accordance with hypothesis 1. This was confirmed by significant differences in cumulative intensity distributions between particular duration classes in the total sample, i.e. between shorter and longer durations (p<0.01; 2-sample K-S, U-test). Further, intensity distributions across duration classes differed between patients and employees in specific ways (Fig. 1): In patients, lower, not higher intensities prevailed at longer durations above 5 years, whereas in employees, lower intensities dominated in shorter durations (p<0.001 and 0.05; 2-sample K-S, U-test; Fig. 1A and D).

Fig. 1:

Different of intensity-duration characteristics in clinic patients and employees. First and second rows: Histograms (A, B) and cumulative relative frequency distributions (C, D) of pain intensities per duration class in patients and employees (n=170, each). Duration classes d ]i … j]: i<d≤j months; no durations ≤6 months in patients. Differences: (1) Intensity histograms of patients (A) and employees (B) at all durations (p<0.001; K-S and U-tests, corrected). (2) Cumulative relative intensity frequencies at long durations >60 months shifted to lower intensities in patients (C, violet; p<0.05), which tend to dominate at short durations ≤6 months in employees (D, blue; p<0.08). Third row: Temperature plots of the relative intensity frequencies per duration class with marginal distributions controlled: color-coded differences between observed (f_obs) and expected (f_exp) frequencies; f_exp=f_i. * f_.d/N_total; f_i.=N of intensity i; f._d =n of duration d. (E) patients; (F) employees. Intensity-duration contours generated with statistical package R, function filled.contour. Note: (1) In patients, difference frequencies concentrated in the red-to-orange area starting from low to medium intensities at very long and long durations and decreasing to high and very high intensities at medium to short durations at the lower right. This indicates a general tendency of lower pain intensities at shorter times since onset consistent with the negative overall-correlation (Table 3 and text). (2) In employees, the difference plane shows a different picture concordant with the zero overall-correlation of intensity and duration. In addition to the low intensity main group with short durations (red, bottom left), subgroups with different intensity-duration relations appeared: one with pain intensity increasing with longer durations, another with decreasing intensity after 1 year since onset and a third subgroup again worse at durations longer than 2 years (bifurcation at pain intensity 2).

The two-dimensional temperature-plots of intensity-duration distributions corrected for base rates are apt to further clarify these specific relations in patients and employees (Fig. 1E and F): In patients, contrary to hypothesis 1, pain intensities decreased with increasing duration only up to 5 years and leveled out above (Fig. 1E). The picture differed completely in employees (Fig. 1F): Pain intensities first increased with duration at shorter durations (according to hypothesis 1) but decreased again at longer durations (see also in Supplementary material, Figs. S6 and S7).

Finally, contradicting hypothesis 2 and according to the correlational analysis above, there were neither significant syndrome-specific intensity-duration characteristics between cUBP and FMS patients nor significant differences in absolute intensities or durations. However, the negative result may be due to low power, particularly, in males and FMS subgroups.

3.3.1 Correlations of pain intensity and duration with CPG and MPSS

Pain intensity correlated moderately positively with both chronicity indices (Table 3). These associations were much lower when patients and employees were considered separately, and held also at the syndrome level for both cUBP and FMS patients. Table 3 illustrates that, contrary to expectation, duration correlated negatively with CPG and zero with the MPSS in patients, while CPG correlated zero and the MPSS correlated positively in employees. Similar inconsistent and weak correlations of pain duration with chronicity indices were repeated for the patient sample at the syndrome level, separately for cUBP and FMS patients. Corresponding intensity and duration frequency characteristics underlined these different relations to the chronicity indices in patients and employees. For instance, in patients, lower intensities were more frequent at grade III than at II and the intensity frequencies of grade II did not significantly differ from that of grade IV (Supplementary material, Figs. S8–S10). In employees, in contrast, intensity characteristics progressed with increasing CPG.

Complex relations of the chronicity indices held also for duration: In patients, contra-intuitively, longer durations were more frequent at the lowest CPG and shorter at the highest. In employees, however, duration characteristics across CPG grades resembled the expected sequence of longer durations with increasing grades more closely. The intensity and duration distributions for the MPSS showed a more systematic sequence from lower to higher intensities and from shorter to longer durations with progressive MPSS stages.

The different relation of CPG and MPSS to the primary pain properties intensity and duration shown above raised the question of their relation to each other. Accordingly, correlations differed largely between subgroups and were not significant in patients (Table 3). Frequency distribution analyses specified this (Supplementary material, Figs. S11 and S12).

3.3.2 Exploratory factor analyses of the combined chronicity markers

When data of patients and employees were combined, exploratory factor analyses of pain intensity, duration and CPG/MPSS subscale values produced two principal components and principle axis factors accounting for 58.5% of the total variance (Table 4c). The dominant first component (46.3%) was characterized by disability (CPG disability score and disability days), pain intensity (CPG intensity scale and MPI-D), MPSS scales temporal characteristics and medication use. The second component (12.3%) related closely to chronic development itself indicated by duration and MPSS scales health care utilization/patient career and spatial aspects (number of painful areas). The remaining 41.5% of the variance were distributed over seven non-significant components.

The two-factors structure was replicated in the separate analysis for the employee sample (Table 4b); a general common chronicity factor was not detected. However, when the patient data were analyzed separately, three rather than two significant principal factors explaining 62.3% of the variance were necessary (Table 4a). Again, the dominant first component (32.3%) was best characterized by disability, pain intensity and medication use, but not by temporal characteristics of the pain symptoms. Moreover, pain intensity (CPG, MPI-D) loaded most on the second component (15.5%) together with temporal characteristics and spatial aspects (number of painful areas) independently of disability. The third component (14.5%) resembled the chronic development marked by duration (time since onset) and health care utilization/patient career similar to factor 2 in the pooled sample except for the spatial aspects (number of painful areas), which loaded mainly on the second component in the clinical sample.

The stability of these principal dimensions at the syndrome level was checked by separate analyses for cUBP and FMS patients excluding the MPSS scale spatial aspects because of maximal values (≥3) in the latter (Supplementary material, Table S2). The FMS data yielded three significant components explaining 69.2% of the variance mapping well onto the 3-factors structure for all patients. In contrast, the corresponding factor analysis for the cUBP group revealed four significant components accounting for 80.8% of the variance. Three of them were concordant with the previous analyses. An important exception consisted in disability scales forming a separate factor (extracted second) on their own.

The 3- and 2-factor models suggested by the exploratory factor analyses for patients and employees were tested by confirmatory factor analyses (CFA) using the three diagnostic markers with the highest PCA and/or PAF loadings (shaded cells in Table 4). The 3-factor model for patients was confirmed for non-orthogonal (correlated) factors with acceptable fit indices (corrected chi-square, χ²/df<3). The 2-factor model for employees showed excellent fit (corrected chi-square, χ²/df<2; RMSEA=0.00; CFI=1.00; Supplementary material, Table S3).

In summary, no common factor of chronicity was found and dimensional structures of MSP chronicity patterns differed between patients and employees. The composition of the third factor in patients and the second factor in employees suggests that the dimensional structure of the related chronicity was varying with duration. Syndrome-specificity within the patient sample was only partially supported by the 4- vs. 3-factor solutions for cUBP and FMS patient data, respectively.

Qualitative inspection of mutual variable distances in factorial space suggested three conspicuous and clinically meaningful tentative clusters of chronicity markers (Fig. 2A and B; Supplementary material, Fig. S13), which differed in important aspects between patients and employees. In patients, disability and intensity markers from the CPG and IASP Axis IV, on the one hand, and spatial and temporal characteristics from the MPSS, on the other hand, formed three separate groups of closely related variables. Duration (time since onset) and the MPSS variables medication use and health care utilization/patient career remained relatively isolated. In employees, in contrast, markers of the chronic development (duration, healthcare utilization/patient career) formed a cluster with spatial aspects (number of painful areas) while temporal characteristics grouped with medication usage. Interestingly, in the still active employees pain intensity clustered with the disability markers from the CPG (Fig. 2B, at the right).

Fig. 2:

Dimensions of pain and chronicity markers in clinic patients and employees. (A) Patients: three principal components in relation to main markers and associated variables, symbols as in Table 4: PC 1, Direct Consequences, marked by “disability days” and “disability score (interference)” of the CPG; PC 2, Clinical Characteristics, marked by “pain intensity” and “temporal” and “spatial aspects” (number of painful areas) of the MPSS; PC 3, Chronic Development, marked by “duration (time since onset)” of the DSF and “health care utilization” from the MPSS. Descriptively, four clusters (elliptic frames) of variables may be identified by their distances in the 3D vector space including associated variables with moderate loadings on more than one principal component, labeled tentatively as (1) “intensity cluster”, (2) “temporo-spatial pattern cluster”, (3) “disability cluster” and (4) “chronic development”. (B) Employees: two principal components in relation to main markers and associated variables, symbols as in Table 4: PC 1, Direct Consequences & Clinical Characteristics, marked equally strong by pain intensity and disability variables of the CPG and MPI-D; PC 2, Chronic Development, marked by “duration (time since onset)” of the DSF and “health care utilization/patient career” from the MPSS as in patients. Three descriptively defined clusters of variables differing from those in patients: (1) intensity and disability variables now closely related except for “disability days” of the CPG, all mainly loading on PC 1; (2) “temporal pattern and medication usage”, also near PC 1; (3) a cluster “chronic development” including duration and health care utilization related to the MPSS variable “spatial aspects” (number of painful areas).

3.3.3 Hierarchical latent class analysis of chronic pain markers

The clinically meaningful clustering of variables apparent in the distance mapping of principal pain markers in two- and three-dimensional factor space was cross-examined by hierarchical latent class analyses (LCA) for patients and employees, separately and combined. In the LCA two super-clusters of variables could be distinguished in both patients and employees analyzed separately (Fig. 3) according to the 95%-AU criterion.

Fig. 3:

Clusters of chronic pain markers of clinic patients and employees at the subscale level. Dendrograms of variable clusters according to latent class analyses of all pain markers of present pain intensity, duration (time since onset) and chronicity scales (CPG, MPSS); multiscale bootstrap resampling technique [28]. Red and green numbers AU/BP (arbitrary unbiased/bootstrap probability) values of significant clusters (AU≥95% significant). Symbols as in Table 4 and Fig. 2; numbers in parentheses refer to tentative descriptive clusters in Fig. 2. (A) Patients: Two super-clusters representing (1) Chronic Development (left dendron) separated from (2) Pain Intensity, Clinical Characteristics and Direct Consequences which related more closely to each other (right dendra): The MPSS marker “healthcare utilization/patient career” (MPS-HC) clustered with “duration (time-since-onset)” (left dendron; AU/BP=99/84) as in the PCA (Fig. 2). Within the second dendron at the right sub-clusters of Pain Intensity and of Disability were detected (AU/BP=98/71). “Temporal characteristic” (MPS-TC) and “spatial aspects” (MPS-SA) were separated from all other variables of supercluster (2) (AU/BP=84/40). (B) Employees: Two super-clusters of variables appeared also in the data of employees similar to those in patients, but less clearly defined and some variables grouping differently: (1) As in patients, “duration (time since onset)” and “healthcare utilization/patient career” clustered strongly together, but “spatial aspects” (MPS-SA, left-most) complicated the picture. (2) The second dendron (right) represents a super-cluster of pain severity with Clinical Characteristics similar to that in patients as well as with Direct Consequences. The general cluster structure is less well defined, mainly because of instable groupings of MPSS variables, but disability and intensity variables remained closely related as in the factor analyses.

Cluster 1 comprised scales related directly to the chronic development as such (duration/time since onset, healthcare utilization/patient career), while Cluster 2 contained clinical characteristics (intensity, temporal and spatial aspects) and direct consequences of the pain (disability score and disability days) together with medication usage (Fig. 3A and B). Intensity and the direct consequences (disability) were strongly interconnected within a coherent sub-cluster itself connected only weakly with medication use. Temporal characteristics and spatial aspects formed a second less coherent sub-cluster (AU criterion > 80%). This cluster structure was replicated in single analyses for cUBP and FMS patients and for employees analyzed alone, although the cluster pattern for the latter was somewhat less differentiated and spatial aspects did not cluster (Supplementary material, Fig. S14).