Pain catastrophizing is associated with pain thresholds for heat, cold and pressure in women with chronic pelvic pain

2.1 Study design and sample

This is a secondary analysis of the data from a cross-sectional observational comparative study that was conducted between December 2013 and June 2016 at the Department of Obstetrics and Gynaecology at a central hospital and a university hospital in southeast Sweden [10]. The study was approved by the Regional Ethics Board of Linköping University (Reg.no. 2013/19-3).

In summary, pain thresholds for heat, cold and pressure were prospectively measured in 37 women with CPP referred for diagnostic laparoscopy due to symptoms that could indicate endometriosis, and in 55 healthy women without CPP. All participants filled in the PCS form, the Pain Sensitivity Questionnaire (PSQ) and the Hospital Anxiety and Depression Scale (HADS) form. A detailed description of the study is presented in the original study [10].

2.2 Measurements

2.3 Statistical analyses

Statistical analyses were conducted with the software Statistica v 13.1 (Dell Software, 5 Polaris Way, Aliso Viejo, CA 92656, USA) and the SIMCA-P software version 15 (Umetrics, Sartorius Stedim Biotech, Umeå, Sweden). Data are presented as median and interquartile range (IQR) or frequency (number and percent).

A between-group comparison of demographic characteristics, clinical characteristics, pain thresholds and questionnaire data was conducted using a Mann-Whitney U-test for continuous data and Pearson’s chi-squared test or Fisher’s exact test for nominal data. The level of statistical significance was set at p<0.05 for two-sided tests.

Due to the risk of downplaying the interrelationship among factors and thus reaching incorrect conclusions when using classic statistical methods (for instance linear regression), and the obvious risk of multicollinearity problems when using psychological variables, we used advanced multivariate data analysis (MVDA) as elaborated more in detail elsewhere [16]. The MVDA in the present study consists of principal component analysis (PCA) to detect outliers, and orthogonal partial least square regressions (OPLS) for the multivariate regressions. OPLS discriminant analysis (OPLS-DA) is used to show which variables have the largest discriminatory power for group separation (i.e. CPP vs. controls). Methods such as multiple linear regression assumes that the independent variables (x-variables, regressors) are not strongly intercorrelated i.e. multicollinearity is not present. Based on previous research we had good reasons to suspect the presence of multicollinearity. OPLS is instead based upon the assumption that the independent variables may be intercorrelated (in unknown ways) and takes advantage of this multicollinearity pattern.

The data is not required to be normally distributed when applying these methods of MVDA [37]. Basically, the MVDA R² describes the goodness of fit – the fraction of sum of squares of all the variables explained by a principal component. Q² describes the goodness of prediction – the fraction of the total variation of the variables that can be predicted by a principal component using cross-validation methods [38].

A check for outliers was conducted using score plots of the PCA in combination with Hotelling’s T², and distance to model in X-space [38]. No extreme outliers were detected in the present study. PCA can be regarded as a multivariate correlation analysis. Instead of performing multiple bi-variate correlations a PCA analysis is performed and the risks associated with multiple testing is markedly reduced. Besides checking for multivariate outliers PCA’s were made in order to understand the correlation pattern among the variables used as regressors (x-variables). Graphic presentations implemented in SIMCA-P software were used to facilitate the understanding of the correlation pattern.

OPLS was used to explore the relative roles of PCS subscales, HADS subscales and PSQ together with background data (age, BMI, and smoking) to explain the variations in the pain thresholds for each stimulus [38]. A variable influence on projection (VIP) ≥ 1.0 was considered significant if the VIP value had a 95% jack-knife uncertainty confidence interval non-equal to zero [38]. P(corr) depicts the loading of each variable scaled as a correlation coefficient, thus standardising the range from −1 to +1. An absolute p(corr)>0.4–0.5 is generally considered significant [37]. VIP values are specific for a certain regression but can be compared within a regression i.e. between the x-variables (regressors). P(corr) is suitable for comparisons between regressions but require that the same dependent variable (Y-variable) is used. For each regression, we report the R², Q², and the result (i.e. p-value) of a cross-validated analysis of variance (CV-ANOVA). In the present study we required a significant CV-ANOVA for a regression to be significant. A certain variable was considered a significant variable when VIP>1.0 and absolute p(corr)≥0.50.

3.1 Catastrophizing – group comparisons

Catastrophizing scores were significantly higher in the CPP women than in the healthy control women according to the three aspects (subscales) captured by PCS (Table 1).

3.2 Regressions of pain thresholds

3.2.4 Intercorrelations among the independent variables (the regressors)

We also investigated the intercorrelation pattern among the variables used as regressors (x-variables) in the regressions above (Tables 2 and 3). Bivariate correlation analyses both for all subjects together and separate for the two groups clearly indicated intercorrelations between the independent variables (Supplementary Table S1). However, the interpretations of these are complicated and therefore multivariate correlation analyses by means of PCA were performed to investigate whether the regressors represented one or several groups of variables (latent variables or components).

In all subjects taken together, the PCA resulted in one significant component (R²=0.45; Q²=0.32) (Fig. 1). Figure 1 shows the two HADS scales and the three PCS scales as highly positively intercorrelated (they had the same sign and had high p-values according to the first component [p1=the horizontal axis]). PSQ also had a relatively high loading on p1 but was more distant to the six psychological variables and was less strongly correlated with them. Smoking, BMI, and age had low absolute loadings (<0.22) upon p1 and thus were not so important. Note that the second component p2 (vertical axis) in Fig. 1 was not significant. In Supplementary Figure S1 the three pain thresholds are included in the PCA.

Fig. 1:

PCA of the independent variables (X-variables) used in the regression of pain thresholds (cf. Table 2) in all subjects taken together (n=92). Note that the second component p2 was not significant (only shown in order to improve interpretation). BMI=body mass index; Smoking=currently smoking (i.e. dummy variable: smoking=1, non-smoking=0); HADS-Depr=Hospital Anxiety and Depression Scale; =depression scale of HADS; HADS-Anx=anxiety scale of HADS; PCS=Pain Catastrophizing Scale; PCS-helpless=helplessness scale of PCS; PCS-rumin=rumination scale of PCS; PCS-magnif=magnification scale of PCS; PSQ=Pain Sensitivity questionnaire – total index

Also, in the CPP (Fig. 2), PCA resulted in one significant component (R²=0.31, Q²=0.11). The pattern of variables was very similar to what is shown in Fig. 1. Hence, in CPP the two HADS scales and the three PCS scales were highly intercorrelated (positively, i.e. they had the same sign and had high absolute p-values according to p1). The PSQ had a relatively high absolute loading on p1 but was more distant to the six psychological variables (not so strongly correlated with them). Smoking, BMI, and age had low absolute loadings (<0.15) upon p1. Note that the second component p2 is only shown to improve interpretation. It was not possible to obtain a significant PCA in the control group of healthy women. In Supplementary Figure S2 the three pain thresholds are included in the PCA.

Fig. 2:

PCA of the independent variables (X-variables) used in the regression of pain thresholds (cf. Table 3) in CPP (n=37). Note that the second component p2 was not significant (only shown in order to improve interpretation). BMI=body mass index; Smoking=currently smoking (i.e. dummy variable: smoking=1, non-smoking=0); HADS-Depr=Hospital Anxiety and Depression Scale -depression scale; HADS-Anx=anxiety scale of HADS; Pain dur=pain duration; PCS=Pain Catastrophizing Scale; PCS-helpless=helplessness scale of PCS; PCS-rumin=rumination scale of PCS; PCS-magnif=magnification scale of PCS; PSQ=Pain Sensitivity questionnaire – total index.

To summarise both the analysis of all subjects and the separate analysis of the CPP group (Figs. 1 and 2) showed that scales of HADS and PCS were strongly intercorrelated and that PSQ correlated positively with these variables. Hence, we found no evidence that the regressors in Tables 2 and 3 represented several groups of variables.

3.3 Relative importance of psychological variables and pain thresholds for group differentiating

Group membership (y-variable) was regressed using the background variables, psychological variables, subjective pain sensitivity (PSQ) and pain thresholds as regressors (x-variables). From this significant OPLS-DA (Table 4) it can be concluded that the subscales of PCS and HADS were somewhat stronger regressors according to VIP and p(corr) than the three pain thresholds. PSQ, age, smoking and BMI were not significant regressors in this context (i.e. VIP<1.0). Hence, the psychological variables contributed somewhat better than the pain thresholds to differentiating between CPP and healthy pain-free controls.

This conclusion was further demonstrated in two additional analyses. When only including the psychological variables including PSQ together with the background variables, a model with higher explained variation was achieved compared to when the pain thresholds together with background data and PSQ were used as regressors of group membership (R²=0.62, Q²=0.60, CV-ANOVA p-value<0.001 versus R²=0.43, Q²=0.40, CV-ANOVA p-value<0.001). Hence, these two analyses confirm the conclusion that the psychological variables were somewhat more important than pain thresholds for group differentiation.

4.1 Summary of findings

Important results of the present study were:

1. Women with CPP reported more catastrophizing than healthy pain-free women.

2. In all subjects taken together, the three regressions of the pain thresholds had in common that PCS-helplessness, PCS-rumination and HADS-depression were significant regressors.

3. In the group of women with CPP, the rumination subscale of PCS, BMI, and PSQ were significantly associated with HPT and CPT.

4. The subscales of HADS and PCS were somewhat stronger regressors of group membership (control or CPP) than the three pain thresholds.

4.2 Interpretation of results in relation to current knowledge/literature

Our results conform with previous studies proposing high pain catastrophizing in populations with chronic pain [12], [22], [23], [24], 26], [27], [29], [30]. Catastrophizing is a significant variable for quality of life and treatment outcomes, where more catastrophizing leads to lower quality of life and less positive treatment outcomes. Furthermore, catastrophizing is a predictor for poor outcome after surgery [39]. The relationships show the importance of taking catastrophizing and other psychological aspects into account when treating people with chronic pain [24], [27], [29]. Also, the regression of group membership (Table 4) showed that catastrophizing aspects together with other psychological aspects were associated with the clinical presentation in CPP. However, the relationships between catastrophizing and pain thresholds is somewhat different in the literature and in the present study.

The OPLS regressions of CPT and HPT were significant, while the regression of PPT was not. The reasons for this need further evaluation. Meinits et al. [40] explored catastrophizing as a mediator for pain sensation in patients with chronic low back pain and found that PPT was inversely associated with pain catastrophizing [40]. Also, studies of chronic neck pain reported such a negative association between PPT and catastrophizing [41], [42]. Partly in contrast, Walton et al. [43] analysed the phenotypes of individuals with neck pain in five international registers and found the association between PPTs and pain characteristics, including catastrophizing, to be conflicting [43].

The psychological impact of the pain experience in CPP has been the focus of many studies, of which some are synthesised in the recent review by Till et al. [25]. There is convincing evidence for the theory of pain and emotion regulation as a trans-diagnostic process [21], where anxiety, depression and catastrophizing are highly collinear and together may influence the pain experience, and vice versa [25]. Martinez-Calderon et al. [44] recently reported that the diagnosis of depression had a stronger association with pain hypersensitivity than pain catastrophizing in patients with chronic shoulder pain [44]. In our results, the influence of PCS-rumination was significant for the regressions of the two thermal pain thresholds and a similar tendency was found in the non-significant regression of PPT in the CPP group. The fact that PCS-rumination was consistently significant while depression and anxiety were not in the CPP group is interesting and indicates the important role of catastrophizing in the pain experience of CPP. The significance of the rumination scale of PCS may also suggest that rumination should be specifically targeted in the planning of new interventions. McPeak et al. [27] analysed the PCS subscales in relation to pain health-related quality of life. They found the strongest association between high helplessness and poor quality of life and suggested treatment strategies to manage pain catastrophizing [27]. The importance of taking a multi-disciplinary approach and thereby including these types of psychological aspects in the treatment and care of women with CPP has frequently been highlighted in the literature [19], [23], [25]. This conclusion is further supported by the results of the present study where emotional stress (i.e. symptoms of depression and anxiety) and catastrophizing aspects were increased in CPP and also constituted significant regressors for group membership. In order to understand the influence of psychological variables on the development of CPP, it may also be important to include pain catastrophizing.

Previous studies have shown an age-dependent factor in pain catastrophizing, where younger age was associated with higher pain catastrophizing intensity [22], [26]. We failed to demonstrate a significant age-dependent impact on pain catastrophizing. In contrast, BMI was a significant regressor for the two thermal pain thresholds but not for the pressure threshold. The scanty research on the influence of BMI upon pain sensitivity in women with CPP is ambiguous. Yosef et al. [29] found a positive association between high pain sensitivity and high BMI [29] while Gurian et al. [45] found higher pain thresholds among women with overweight compared with those with normal weight or with obesity [45]. This indicates that the influence of BMI upon pain thresholds merits further research.

The data pinpoints that the psychological variables i.e. symptoms of anxiety and depression and the three aspects of catastrophizing together with the three pain thresholds significantly explained group membership. Hence, both the psychological variables and the pain thresholds contribute to the understanding of the clinical presentation. Moreover, the five psychological variables were – according to VIP and p(corr) – somewhat more important than the three pain thresholds even though the three latter variables were also significant. From a clinical perspective, this information may be important since registration of pain thresholds is associated with short periods of pain and, moreover, is also considerably more time-consuming than filling out questionnaires on HADS and PCS. As reported by us previously, PSQ is associated with the three pain thresholds and higher in women with CPP [16] but in the present study with a relatively comprehensive set up of variables, it can be concluded that PSQ was not significant in the regression (OPLS-DA) of group membership. PSQ captures the overall perceived pain sensitivity while the pain thresholds – semi-objective measures – capture both stimuli specific information (mechanical pressure, cold or heat) and the degree of spatial spreading [10].

4.3 Strengths and limitations

A strength of this study was the use of validated instruments and methods that provided information on different aspects of the subject. All women with pain had experienced pelvic pain for at least 4 months, a period which is often set as the minimum for considering the pain as chronic [46]. Another strength was the use of MVDA, which is designed to use the properties of the data set in an optimal way. This approach differs from classical statistical methods, such as multiple linear regression, which have a tendency to quantify the level of relations of individual factors and at the same time disregard interrelationships among different factors [47]. We also avoided the use of multiple linear regression due to the risk of multicollinearity problems in our data set. Although parity differed significantly between the groups it was not possible to include this variable in the MVDA since it did not fulfil the predetermined criteria (i.e. it had a VIP value with a 95% jack-knife uncertainty confidence interval including zero). Even though the number of participants in the study was sufficient to enable PCA and OPLS, the sample size may be considered small and thus may be a limitation of the study. Due to the very few studies published on the subject and consequently the lack of reliable information in the literature, no power analysis with sample size estimation was performed when planning the study. The sample size was intended to be at least the same as the studies previously published in the field. Another limitation may be that the women filled in the PCS after the QST, which may have influenced their scoring, even though the results of the QST were not disclosed. The cross-sectional design of the study and thus the direction of causality between catastrophizing and pain thresholds cannot be analysed. The relationship may be bidirectional.