Pain perception in chronic knee osteoarthritis with varying levels of pain inhibitory control: an exploratory study

1 Introduction

Knee osteoarthritis (KOA) is a form of arthritis that impairs functional abilities and decreases quality of life in patients by producing pain, stiffness, and limitation in range of motion of the knee joint [1], [2]. Recent epidemiology data suggest that KOA has been a leading contributor to global disability [3], [4] affecting approximately 27 million American adults [5] and an estimated 10% of the world’s population over the age of 60 years [6]. KOA also poses a significant economic burden accounting for more than $27 billion in health care expenditures annually in the US [7].

Pain is the most disabling symptom in KOA [8] and its underlying mechanisms have for long been poorly understood [9], [10]. Contemporary research have suggested that it may result from a combination of joint nociception (e.g. structural changes and synovitis), increased responsiveness of peripheral nociceptors during inflammation of the knee (i.e. peripheral sensitization), pathological neural signals from the joint causing central nervous system changes (i.e. central sensitization), in addition to having specific mediators and being influenced by contextual aspects such as psychosocial factors [11]. Clinical research reveals that the alterations of excess nociceptive ascending and deficient inhibitory descending control present in KOA, influences in the development and maintenance of central sensitization and may contribute to the transition from acute to chronic pain [12], [13], [14], [15], [16]. Central and peripheral sensitization can result in increased pain response and sensitivity to nociceptive stimuli at sites distant from the knee leading to relevant limitations [17], [18], [19]. These changes characterize impairments in the endogenous central pain control mechanisms, such as altered Conditioned Pain Modulation (CPM), that affects descending pain regulatory pathways and are present in knee osteoarthritis patients [20], [21].

CPM can be defined as an endogenous analgesic mechanism that occurs “when a painful stimulus (test stimulus) is inhibited by a second painful stimulus (conditioning stimulus) delivered at a different body location” [22], [23]. This neural control is triggered by noxious stimulation coming from widespread areas of the body and exert an inhibitory influence on wide-dynamic-range neurons [24]. This mechanism has been reported in many clinical studies and can be properly quantified mainly by two dynamic test paradigms that involve pain assessments, namely temporal summation and CPM testing [25], [26], [27]. Abnormal CPM has been demonstrated in many chronic pain syndromes including fibromyalgia, KOA, rheumatoid arthritis, neuropathy, and headaches [28]. Evidence suggests that peripheral nociceptive signals (e.g. originating from KOA) are strongly correlated with dysfunction of the descending pain modulatory pathways and reduced pain inhibition [15], [18], [20], [28].

Previous research have showed that differences in CPM have been consistent predictors of clinical pain and health-related variables in healthy and in musculoskeletal pain cohorts [29], [30], [31], [32]. In KOA, variability in CPM can also predict response to treatment [31]. However, questions remain whether different CPM levels are clinically relevant in characterizing chronic pain patient characteristics. Therefore, the aim of this study is to explore the association between clinical variables with pain perception in individuals with KOA with varying levels of endogenous pain inhibitory control as indexed by CPM function. We hypothesize that the pain perception associate with distinct clinical variables for those who have limited CPM response as compared to those who have normal CPM response.

2 Materials and methods

3 Assessments

This study implemented a set of assessments for KOA subjects with currently acceptable levels of sensitivity, reliability and validity based on their applicability to the non-invasive brain stimulation trial that we were conducting and/or as recommended by the ACTTION and IMMPACT [33], [34] guidelines for conducting chronic pain trials. Blinded raters performed the following test procedures.

4 Statistical analysis

Descriptive statistics were used to characterize the sample. Dichotomous and categorical variables were described in frequency and respective percentages. Continuous variables central tendency and variability were described by means and standard deviation or medians and interquartile ranges according to data distribution. Normality of distribution was checked by a combination of visual inspection of the histogram of data points and the Shapiro-Wilk normality test [48].

For our analysis, the CPM variable was transformed into a dichotomous variable that was used to discriminate KOA subjects with distinct levels of endogenous pain inhibitory control as indexed by conditioned pain modulation (CPM) function, namely subjects with “normal CPM response” and with “limited CPM response”. The absolute standard error of the measurement (SEM) value calculated from the pre-conditioning PPT trials was used to classify the continuous CPM variable into the dichotomous variable. This is a valid distribution-based statistical approach used to determine true change in CPM that has been used in previous studies [49], [50] to identify inhibitory CPM effects. An inhibitory CPM effect was defined as a CPM value greater than 1SEM. Subjects with CPM>1SEM were categorized as the “normal CPM response” group, referring to those who were able to modulate the pain at least above the SEM. Subjects with CPM<1SEM defined the “limited CPM response” group, referring to those who were not able to modulate pain above the SEM and consequently having a more limited or absent ability to modulate pain after the conditioning stimulus. The SEM was calculated using the formula (SD * √1-r) where r is the reliability coefficient and SD is the standard deviation of all data from the 3 PPT trials. Independent sample t-test and Mann-Whitney U test were used to check for differences in the descriptive statistics between the two subgroups. All the subjects were blinded to their CPM scores or CPM response classification.

A multiple linear regression analysis was used to investigate the association between the clinical variables and pain perception in each of the two KOA subgroups. VAS for Pain was used as the dependent variable while all other variables were used as the exploratory independent variables. All the variables were treated as continuous variables.

Outliers were defined as cases with values greater than 3 SDs above or below the mean scores of the dependent and independent variables. For each independent variable, the assumption of linearity was assessed by visually comparing the scatterplot of the variable against the plot of a superimposed regression line. The assumption of homoscedasticity was checked by visually comparing the scatterplots of the standardized predicted values and those of the standardized residuals [51]. Normality of the residuals was checked by a combination of visual inspection of the histogram of the residuals and the Shapiro-Wilk normality test [48]. Durbin Watson estimate, tolerance values, Cook’s distance, variance inflation factor and Standardized DFBeta values were requested for analysis of regression diagnostics such as multicollinearity and influential cases.

To define the best models, univariate linear models were created with each independent variable to determine the significant covariates based on the regression coefficient significance p-value using a cutoff of p<0.2. If variables were not significant, they would be automatically excluded from the final model. Using a backward elimination approach, we included all variables that were significant in the univariate regression analysis in the full models. The regression coefficients were then checked for significance in the full model and those with a p-value <0.05 were excluded from the model. The Akaike’s information criteria was also used to determine the model with the best fit. In addition, if any variables revealed to be significantly different among the KOA subgroups in the descriptive analysis, such variable would be considered for the final models. Cohen’s ƒ2 effect size was calculated for each model as it is an appropriate effect size measure for multiple regression [52]. All analysis were performed using statistical software STATA version 15 (StataCorp. 2017. Stata Statistical Software: Release 15. College Station, TX, USA: StataCorp LLC).

5 Results

Sixty-three subjects with KOA participated in this exploratory analysis study. The SEM for PPT based on the pre-conditioning PPT trials was 1.45 pound-force/square centimeter (64.5 kPa) which is similar to SEM values reported in the literature [49]. An inhibitory CPM effect (>+1SEM) was elicited in 36.5% (23) of subjects in response to cold stimulus (normal CPM response group) while 63.5% (40) of subjects had limited CPM response. Tables 1 and 2 summarize the sample.

The analysis did not show any significant difference between the two subgroups of normal CPM response versus limited CPM response, except for the CPM variable (p<0.05). CPM was statistically significantly different between the two subgroups (median=2.8 and median=−.3 for the subgroup with normal CPM response and limited CPM response, respectively), U=920, z=6.567, p<0.001. This significant difference supports that our subgroups are indeed distinct on their CPM function.

Two different multiple linear regression models were generated, one for the KOA subjects with normal CPM Response (model 1) and one for KOA subjects with “limited CPM response (model 2). No outliers were identified. Visual analysis of the scatterplots as well as the residuals confirmed the assumption of linearity and homoscedasticity for all models. For multicollinearity, analysis of the correlation matrix between the predictors showed no values above 0.7 (strong correlation). Also, analysis of the variance inflation factor and tolerance values obtained from the collinearity diagnosis output showed no values >10 or <0.2, respectively, assuring the assumption of no multicollinearity. For influential cases, we found no Cook’s distance values >1 and no DFBeta values >|1|, indicating no influential cases. The Durbin Watson statistic showed no value <1 or >3 for all 2 models, indicating that the assumption of independent errors was met for all 2 models.

The final regression equation for the subgroup of KOA subjects with normal CPM response (model 1) was: VASPain=7.247 – 0.046*SF36 General Health. The multiple regression model statistically significantly predicted VASPain, F(1, 21)=5.60, p<0.05, R-squared=0.2104. General Health measured by the SF36 General Health subscale was significantly associated to VASPain scores. Model coefficients suggested that as SF36 General Health score decreased (poorer general health), VAS Pain increased. The respective Cohen’s ƒ2 effect size was 0.27 (medium) [52]. Regression coefficients and standard errors are detailed in Table 3.

For KOA subjects with limited CPM response (model 2), the final regression equation was: VASPain=5.441 – (0.019*SF36 Role Limitation due to Emotional Problems)+(0.330* Stress VAMS). The multiple regression model statistically significantly predicted VASPain, F(2, 37)=7.02, p<0.01, R-squared=0.275. The SF36 Role Limitation due to Emotional Problems as well as the perception of stress at the moment of assessment measured by the Stress VAMS was significantly associated to VASPain scores. The direction of the coefficients suggested that as limitations due to emotional problems scores diminishes (meaning more disability), VASPain increases; and as Stress VAMS score increased, VASPain also increased. The respective Cohen’s ƒ2 effect size was 0.37 (large) [52]. Regression coefficients and standard errors are detailed in Table 4.

6 Discussion

The aim of this study was to explore the association of clinical variables and pain perception in KOA individuals with varying levels of CPM function. We have found that general health perception is associated with pain perception in KOA subjects with normal CPM response, while perception of limitations at work/other activities due to emotional problems and self-reported stress is associated with pain perception in those with limited CPM response. We believe this innovative approach can contribute to unveil the factors contributing to pain in KOA subjects and ultimately help guide treatment strategies.

The model for KOA subjects with normal CPM response (model 1) included only the SF36 General Health variable and was able to explain 21% of the variability in pain in that subgroup. There was a significant association between the model and the VASPain. This result supports the idea that in a functional endogenous pain inhibitory control system, pain perception is associated with general health perception. This association is not surprising as it is largely accepted that pain is a determinant of quality of life and general health due to its effects on mental health, well-being and activities of daily living [53], [54], [55]. Other [43] research has shown that general health status is explained partially by aspects of pain, for example, pain catastrophizing. Using a large community sample, this study showed that older people with chronic pain who report higher levels of pain tended to rate their general health status more negatively. This reinforces recent evidence that suggest a greater multidimensional negative impact of chronic pain on quality of life as compared to general population [56].

In model 2, the SF36 “Role Limitation Due to Emotional Problems” subscale variable was significant in the model. To thoroughly understand what this variable represents in our model, we must consider how it is determined. It is a combined result of three questions inside the SF36 questionnaire that asks if, over the past 4 weeks, the subject has spent less time at work or performing other activities, accomplished less than they would like and didn’t do work/other activities as carefully as usual due to any emotional problems. As these three questions specifically ask about the past 4 weeks, this variable likely represents the individual’s function prior to the pain variable being collected. Our results suggest that on those subjects who have limited CPM response, the perception of limitations at work/other activities due to emotional problems in the past month can significantly explain a portion of the variability in pain perception. The direction of this association reflects that the worst the subjects’ perception of how limited their work/other activities were in the past month because of emotional problems, the higher the pain. Although our results are based on analysis of cross-sectional data and no causality conclusions can be made, one may suggest this model to perhaps be considered as a potential prediction model due to how the questions were asked referring to the past 4 weeks. All the other collected variables such as Stress VAMS, Depression VAMS, Beck Depression Index, WOMAC, did not significantly add to the model.

In the group with limited CPM response (model 2), the final model also included Stress VAMS. In this model, higher stress perception was correlated with higher levels of pain perception. The stress variable was collected by asking the subject to rate their stress level at the moment of the assessment, which would reflect the subject’s current or acute level of stress. Our findings contradicts the concept of stress-induced analgesia [57], [58], [59]. We attribute this disagreement to the fact that, although our Stress VAMS variable is collected by specifically asking subjects to rate their stress at the moment of evaluation using a visual analog scale, it is possible that subjects rated instead their overall stress. Thus, rather than acute stress, our data might reflect chronic stress which has been shown to increase pain sensitivity in both animal and clinical research [60], [61], [62], [63], [64]. Evidence that stress increases pain sensitivity has been found in several studies, and the reasons for suspecting such a relationship are well-founded theoretically [65]. Stress have also been found to be highly prevalent in subject with severe KOA who are waiting for knee replacement [66]. It has also been reported that therapies that focus on cognitive byproducts of stress reduce pain, which provides further evidence that these psychological factors play an important role in chronic pain [67].

The comparison of model 1 and 2 suggests that differently that in subjects with normal CPM response, in subjects with limited CPM response, the perception of limitations at work/other activities due to emotional problems and the stress felt at the moment of the assessment are relevant factors that may explain differences in pain levels. Determining if the impact of emotional problems on work/other activities is a significant predictor of pain for KOA subjects with normal CPM response but not for subjects with limited CPM response may have direct implications for clinical practice. Although this result may suggest that interventions targeting emotional problems and its impact on social/work activities may contribute to address pain perception on KOA subjects with a well-functioning pain modulatory system, we cannot infer that the same type of intervention would not affect pain perception on those with a limited-functioning pain modulatory system. On the other hand, general health perception was found to be significantly associated with pain level in subjects with normal CPM response but not in those with limited CPM response. This finding may suggest that general health perception may not have as much impact on explaining pain levels on subjects with limited CPM response compared to those with normal CPM response and perhaps should not be emphasized as an intervention in this subgroup of KOA subjects. To confirm the different characteristics of these subgroups of KOA patients may be relevant to better determine treatment plans and achieve better clinical results.

Previous research [29] has explored the relationship between clinical pain and CPM measures in healthy subjects; specifically, hierarchical regression models have shown that CPM response has the greatest impact on pain response, with greater CPM responses related to less pain. Our analysis did not consider CPM as an independent variable, but instead, used CPM response to identify distinct groups based on their CPM function. However, the two regression models developed in this study suggested that pain is associated differently according to the individual CPM response. Edwards et al. [29] found that younger participants showed adequate CPM response (suppression of pain rating with conditioning stimulus) more commonly than older participants. Our results are consistent with Edwards et al.’s results as, although not significant, age was found to be different between our subgroups of KOA subjects with the subgroup with normal CPM response being slightly younger that with limited CPM response. We tested is “age” was a confounding factor in our sample and found that it did not contribute significantly to VASPain scores in any of the separate models.

Our study has limitations. First, our data for pain and stress were collected using visual analog scales. Some may argue the subjectiveness of these types of scales and its limitations [68]. However, its vast use in research and clinically allow easy interpretation and direct comparison with related research. Second, our small sample size hampers the generalizability of our results. Although our sample has characteristics that are similar to those of other reported OA samples, it has a similar distribution between men and women, where KOA is typically predominant in women [69]. Thirdly, we have dichotomized the level of endogenous pain inhibitory control as indexed by CPM function and information on individual differences may have been lost in the dichotomization process causing other subgroups with dysfunctional CPM levels to not be detected. Previous research has discussed the impact of the dichotomization of quantitave measurements [70], [71]. Ultimately, this study has an exploratory nature and it is a secondary analysis of a clinical trial. Our analysis used only cross-sectional data and, although we discussed the possibility of some of the models to be seen as prediction models, causation inferences are not appropriate. All inferences about any results should consider these limitations and be interpreted accordingly.

7 Conclusions

Our study suggests that pain perception may be associated with different factors in KOA subjects that have different levels CPM function. This information may help understand the factors contributing to pain in KOA and assist clinical practice. The exploratory nature of our analysis implies that our inferences should be considered only with an understanding of the limitations of our methodology. Future research should explore other clinical variables to optimize the best predictor/association model for pain in KOA subjects with different levels of CPM function. We also encourage researchers to explore treatment strategies focused on subgroups of KOA subjects based on their different levels of endogenous pain inhibitory control.