“Convergent validity of the central sensitization inventory and experimental testing of pain sensitivity”

Introduction

Central sensitization (CS) is defined as “increased responsiveness of nociceptive neurons in the central nervous system to their normal or sub-threshold afferent input” [1]. Several mechanisms involved in the development of CS have been suggested, including an altered sensory processing in the brain [2], and a dysregulation in both ascending pain facilitatory [2, 3] and descending pain inhibitory pathways [4]. Central sensitization has been described in a number of chronic conditions, including chronic low back pain (LBP) [5], [6], [7], [8], and generally manifests as pain hypersensitivity in the form of lower pain thresholds, higher pain responses, and perturbations of pain modulation [9]. Despite extensive research into this phenomena, there is no general consensus on the definition of clinical criteria for CS, and a gold standard for assessing CS is lacking.

While direct physiological detection of CS in pain patients is not possible, psycho-physical measures of pain sensitivity, such as Quantitative sensory testing (QST), can be used to assess and evaluate the function of the somatosensory nervous system [10, 11], and is often used to infer perturbations in pain modulation, such as those seen in patients affected by CS [9, 11, 12]. Examples of such changes are lowered pain thresholds and elevated pain ratings for patients subjected to standardized painful stimuli in experimental pain testing [13, 14]. Limitations to QST are their time-consuming nature and requirement of expensive equipment, as well as expertise and adherence to a rigorous testing protocol. While standardization of QST has been proposed and implemented [15, 16], there is currently no universally accepted protocol for experimental pain testing for clinical subgroups, such as patients with LBP.

To aid clinicians in identifying CS, the Central sensitization inventory (CSI) was developed in 2011. The CSI is a self-reported questionnaire aiming to screen for symptoms related to CS [17]. The CSI has been shown to be associated with other self-reported instruments aimed at measuring psychological constructs, such as pain catastrophizing, fear avoidance, and anxiodepressive symptoms [18], [19], [20], [21], [22]. However, it remains uncertain whether CSI scores also correlate with more direct measures of pain sensitivity, such as QST. Only a few studies have examined the correlation between CSI and QST, and results have varied [18, 20, 23, 24]. Kregel et al. [23] examined PPT and CPM in a chronic spinal back pain cohort in a primary care setting, and found a weak correlation between CSI and PPT, but no correlation between CSI and CPM. The authors suggested that the CSI “does not reflect a direct measure of CS, yet is a representation of general distress, possible originating from CS symptoms”. Gervais-Hupé et al. [20] studied PPT, CPM and TS in a group of knee osteoarthritis (KOA) patients and found only weak or non-significant correlations between CSI and QST. They found that the CSI was “more strongly associated with psychological factors than psychophysical test results”. Coronado and George [18] examined the correlation between CSI and pain sensitivity in a shoulder pain population and reported similar findings. Furthermore, in a study by Mibu et al. [24], the association between CSI scores with PPT and TS was investigated in patients with either chronic LBP or KOA, finding no significant correlations in either group. In a more recent study, Zafereo et al. [25] investigated the association of CSI scores with thermal and pressure pain thresholds in patients with chronic musculoskeletal pain, finding only low-to-negligible correlations between CSI scores and QST variables.

Thus, when summarizing the literature, the correlation between QST and CSI remains uncertain. But does not appear strong. Previous studies have applied limited QST protocols and studied different pain populations, making conclusions less certain. Arguably, applying a broad QST battery consisting of relevant QSTs to a large group of LBP patients, all seen in the secondary care sector, could potentially shed further light on the association between CSI and QST.

The aim of the present study was to examine the convergent validity of the CSI questionnaire, by exploring the correlation of CSI with QST (construct validity) and self-reported psycho-social measures (concurrent validity), in a cohort of secondary-care sector patients with LBP. Furthermore, we explored the correlation between QST and self-reported psycho-social measures. A secondary objective was to examine the sensitivity and specificity of CSI when categorizing participants as sensitized or non-sensitized by QST. Thirdly, we determined the optimal cut-point on CSI for sensitized/non-sensitized categorizations in the present study population.

Methods

Results

Correlation between central sensitization inventory and quantitative sensory tests

Significant but weak correlations were found for six out of 21 variables included in the QST protocol, while two variables showed moderate correlation with CSI total score (seeTable 2). No strong correlations were found. Furthermore, a weak correlation was found between CSI and QST total scores (r=0.22, p-value=0.008). Scatterplots visualizing significant correlations are illustrated in Figure 1. Distribution of experimental pain test variables can be seen in Appendix 2. Only CSI, TS, CPM and QST total score were normally distributed.

Table 2:

Bivariate correlations between Central Sensitization Inventory and experimental pain test variables in the Quantitative Sensory Pain Test protocol.

	n	Pearson’s (r) and Spearman’s (rs) correlation coefficient	p-value
Cuff algometry
CPPTdominant leg	148	rs=−0.01	0.927
CPTTdominant leg	148	rs=0.04	0.670
CPPTnon-dominant leg	148	rs=−0.02	0.839
CPTTnon-dominant leg	148	rs=0.03	0.718
CTS	148	r=0.17*	0.043
CCPM	145	r=0.08	0.351
Weighted pinprick pressure
WPPTlower back	167	rs=−0.13	0.104
WPPThand	167	rs=−0.18*	0.018
Pressure algometry
APPTlower back	167	rs=−0.33***	<0.001
APPTforearm	167	rs=−0.24**	0.002
Spring-loaded probe
SPPTlower back	167	rs=−0.37***	<0.001
SPPTforearm	167	rs=−0.22**	0.005
SPPIlower back	167	rs=0.22**	0.004
SPPIforearm	167	rs=0.09	0.244
TS1	167	rs=0.10	0.222
TS2	167	rs=0.09	0.224
Cold pressor test
CPMlower back	167	rs=−0.02	0.804
CPICPT		rs=0.17*	0.032
TCPT		rs=−0.13	0.101
AUCCPT	167	rs=0.02	0.823
Quantitative sensory test
Total score	142	r=0.22**	0.008

1. CPPT, cuff algometry pressure pain detection threshold; CPTT, cuff algometry pressure pain tolerance threshold; CTS, cuff algometry temporal summation; CCPM, cuff algometry conditioned pain modulation; WPPT, pinprick pain threshold; APPT, algometry pressure pain threshold; SPPT, spring-loaded probe pressure pain threshold; SPPI, spring-loaded probe pressure pain intensity; TS₁, temporal summation before cold pressor test; TS₂, temporal summation after cold pressor test; CPM_{lower back}, conditioned pain modulation in the cold pressor test; CPI_CPT, maximum cold pain intensity in the cold pressor test; T_CPT, time to achieve maximum cold pain intensity in the cold pressor test; AUC_CPT, area under the curve for the cold pressor test. Significance Levels: *Correlation is significant at the 0.05 level (2-tailed). **Correlation is significant at the 0.01 level (2-tailed). ***Correlation is significant at the 0.001 level (2-tailed).

Figure 1:

Scatterplots visualizing bivariate correlations of CSI total scores with experimental pain test variables included in the QST-protocol.

A: QST Total Score; B: CTS; C: WPPT_Hand; D: APPT_{lower back}; E: APPT_forearm; F: SPPT_{lower back}; G: SPPI_{lower back}; H: CPI_CPT. A linear regression line as well as 95% confidence interval is displayed in all scatterplots. CSI=central sensitization inventory. R=correlation coefficient.

ROC-curve analysis and Fisher’s exact test

Excluding observations containing missing data, 142 complete observations were available for Fisher’s exact test and ROC-curve analysis. Dichotomizing CSI and QST scores, 24.6% (n=35) of the participants were categorized with CS by the CSI (score ≥40), and 16.2% (n=23) by QST (mean + 1SD). Fisher’s exact test demonstrated a statistically significant association between participants scoring ≥40 on CSI and participants categorized as sensitized by experimental pain tests (p-value=0.03) (Table 4). The positive predicative value of CSI using a cut-point of 40 points was low (PPV=28.6%), while the negative predicative value was high (NPV=87.9%).

Table 4:

Receiver Operator Characteristic (ROC) values and Fisher’s Exact Test for suggested cut-points.

	AUC	Cut-point (sensitivity, specificity %)	Positive and negative predicative value	Fisher’s exact test p-value
CSI	0.656	40 (Sn=43.5%, Sp=76.5%)	PPV=28.6%, NPV=87.9%	0.033
		44 (Sn=39.1%, Sp=87.4%)	PPV=37.5%, NPV=88.1%	0.045

1. The Fisher’s exact test p-value denotes the significance of the deviation from the null hypothesis in the Fisher’s exact test of independence. Results of the Fisher’s exact test using different CSI cut-points is shown in the rightmost column of the table. Abbreviations: CSI, central sensitization inventory; AUC, area under curve; Sn, sensitivity; Sp, specificity; PPV, positive predicative value; NPV, negative predicative value.

Furthermore, ROC-curve analysis revealed the CSI had a low sensitivity (Sn=43.5%) and acceptable specificity (Sp=76.5%) in identifying CS. The CSI resulted in an ROC-AUC of 0.656 in the present population. The ROC-curve of CSI and QST total scores suggested an optimal cut-point of 44 (Sn=39.1% Sp=87.4%) in the present population (Table 4 and Figure 2).

Figure 2:

ROC-curves and sensitivity-specificity plots for cut-points on CSI.

A: ROC-curve for the optimal cut-point on CSI, B: Sensitivity-Specificity plot for the optimal cut-point on CSI, C: ROC-curve for cut-point of ≥40 on CSI, D: Sensitivity-specificity plot for cut-point of ≥40 on CSI.

Discussion

This is the first study investigating the convergent validity of CSI with experimental pain measures obtained through an extensive QST protocol, and psycho-social factors, in a cohort of LBP patients. It is only the second study to report ROC-characteristics of CSI when compared to psychophysical tests in determining the presence of CS in a study sample [20]. We found significant correlations between nine out of 21 experimental pain tests and the CSI score, but coefficients were mostly weak. Furthermore, when categorizing participants as centrally sensitized as identified by QST, probability of agreement between CSI and QST was low. These findings suggest that the CSI and experimental pain testing quantify different underlying constructs. For significant correlations, p-values were generally very small, suggesting that the weak correlations were unlikely to be a chance finding. Conversely, when statistical significance was not reached, correlation coefficients were generally of magnitude ∼0.1, suggesting that it is unlikely that strong correlation coefficients were missed by chance. While it is possible that different experimental pain tests measure different domains of pain sensitivity, no clear pattern was observed between the types of QST procedures and correlations with CSI scores.

We explored the association between CSI and other psycho-social measures and found CSI to be significantly correlated with six out of ten psycho-social variables, with strength of correlations ranging from weak-to-strong. It is noteworthy that CSI correlated strongly with symptoms of exhaustion disorder (KEDS). Exhaustion disorder would expectedly impact other psycho-social measures, and it is thus hardly surprising that moderate correlations were also seen with disability (ODI). While psycho-social variables were associated with CSI scores, only negligible associations were found with QST scores. Our findings thus contrast those of Zafereo et al. [25], who found moderate correlations between QST and pain interference, and low correlations between QST and anxiety, fatigue and depression.

Our findings are consistent with results from an earlier study [51] examining the correlation between a comparable QST composite score and psycho-social profile, finding no significant correlations. As such, our findings support the notion that the CSI most likely quantifies an underlying psychological or social construct, as opposed to perturbations in the function of the somatosensory nervous system.

Psycho-social and cognitive factors have been suggested as a contributing factor to the development and persistence of CS [55], and the CSI has been shown to correlate with both social and cognitive behavioral factors in patients with chronic pain [21, 22]. The authors of the CSI have in fact made it clear that the aim of the CSI is not to assess CS as a neurophysiological phenomenon, but rather to clinically screen for CS-related symptoms [56]. In other words, the CSI is intended to measure symptoms theorized to stem from CS, but which may in fact result from a combination of psychological, social and physiological factors. The association between CSI and psycho-social measures found in the present study align with findings of earlier studies [18, 20, 23] showing the CSI to be more closely related to psycho-social measures than experimental pain tests. This raises questions regarding the construct validity of the CSI: do CSI scores relate meaningfully to an individual’s neurophysiological changes in pain modulation, or rather, does it measure a different construct, such as psycho-social factors which may (or may not) be related to chronic pain? Furthermore, questions can be raised regarding the content validity of the CSI, i.e., whether items on the CSI are fully representative of all relevant aspects of central sensitization. Our results would suggest that the CSI does not adequately capture unique signs of pain sensitization or changes in pain modulation.

Our findings echo those of previous studies of the convergent validity of CSI and experimental measures of pain sensitivity. Thus, when summarizing the literature, it seems that the clinical use of CSI as an indicator of CS as a neurophysiological phenomenon should be challenged. This should come as no surprise, as the individual questions which make up the CSI, at least on face validity, are concerned with psychological and social factors [17, 53]. Seeing as the CSI and QST appear to measure different constructs, the use of CSI as a self-report instrument to infer the presence of CS as a neurophysiological phenomenon is thus questionable.

Strengths of the present study include the large sample size, no inter-rater variability (one QST tester), and a relatively extensive QST protocol used, in line with current recommendations [57]. Methodological limitations include the use of several short-item questionnaires. While these short-item questionnaires have shown acceptable concurrent validity [38] when compared to the full-length reference standard from which they are derived [39], [40], [41], [42], [43], there is inherently a degree of uncertainty regarding their content validity. Another limitation to this study is that, while the CSI was cross-culturally adapted to Danish prior to this study, the Danish version of CSI is yet to be validated in a Danish context. However, the CSI-Dan was translated following guidelines [30] for cross-cultural adaptation of self-report measures, and the CSI has previously been cross-culturally adapted to a range of languages whilst maintaining its psychometric properties and validity [32], [33], [34], [35], [36]. As such, we expect the CSI-Dan to perform similar to the original and other adapted versions of the CSI.

A further limitation is the fact that there is no universally accepted cut-point for CS as indicated by QST. As such, we chose an arbitrary cut-point of one standard deviation above the sample mean. This cut-point was chosen as it has been shown that, while healthy human subjects have a bell-shaped distribution of pain sensitivity on QST [16], patients suffering from chronic pain conditions tend to fall on the right side of the curve, displaying significant secondary hyperalgesia and allodynia [12, 58, 59].

Further limitations include the use of custom-made tools in the experimental pain test protocol. While the custom-made spring-loaded probes used in the present study have been shown to be reliable in measuring pressure pain thresholds [60], the weighted pin pricks used are new and yet to be validated. The use of such tools limits the comparability of our results with data reported in studies using similar experimental pain measurements.

Lastly, due to exclusion criteria for some of the QST tests, there was a large number of missing data in the cuff algometry test. While missing data in the QST protocol limits the statistical strength of our analysis, we found there were no significant between group differences in CSI scores (Welch Two Sample t-test score=−0.08, p-value=0.9) when comparing participants who had complete QST data to participants with missing data. Furthermore, there was no significant between group mean difference in each of the experimental pain test variables. This suggests that the characteristics of participants with missing data was not significantly different from participants with no missing data, and that the inclusion of participants with missing data did not have a significant impact upon our statistical analysis.

To complement the psycho-social measures used in our study, we created a psycho-social composite variable, aiming to gauge the overall psycho-social profile of participants. This composite variable was created by combining scores across all psycho-social questionnaires, excluding the CSI. It should be noted that, while the intention of the composite variable was to reflect the psycho-social profile of participants, it is uncertain if in fact the composite variable adequately reflects this. However, the composite variable showed moderate correlation to CSI scores, and the strength of this correlation was similar to the strength of correlation between CSI and the other psycho-social measures (ODI and KEDS), suggesting the psycho-social profile of participants was reflected adequately by the psycho-social composite variable. Similarly, to gauge the overall performance of participants in the QST protocol, we created a composite variable by combining scores in each QST test. As with the psycho-social composite variable, to what extent the QST composite variable reflects the function of the somatosensory nervous system of participants, is uncertain.

To our knowledge, only one previous study [20] has reported sensitivity and specificity of CSI when categorizing participants as sensitized or not sensitized, as determined by QST. The optimal cut-points identified by Gervais-Hupé and colleagues yielded sensitivity levels ranging from 75.0 to 87.2%, while specificity ranged from 34.2 to 63.4%. In the present study, using the cut-point of ≥40 on CSI, only 43.5% of participants categorized as sensitized by CSI, were also categorized as sensitized by QST. Conversely, of the participants categorized as not sensitized by CSI, 76.5% were concordantly categorized as not sensitized by QST.

In our analysis, the ROC-AUC was used to assess the convergence between CSI and QST in categorizing participants as “CS” and “no CS” patients. The observed AUC of 0.656 in our study would suggest that the convergence of CSI and QST was low. It should be noted that only “Part A” of the CSI was used for the ROC-analysis. For this reason, we were not able to subgroup participants using “Part B” of CSI, and as such, did not know whether the participants were known with CSS, and if so, in what proportion and how this affects our reported correlations.

Conclusions

This study examined the convergent validity of CSI. We found small and not clinically meaningful correlations between CSI and QST, demonstrating poor convergent validity. The CSI was more closely correlated with psycho-social factors such as symptoms of ED and disability. These findings are consistent with previous studies, indicating that while the CSI is related to psycho-social factors associated with chronic pain conditions, CSI scores do not directly relate to perturbations in the function of the somatosensory nervous system. In the present cohort, only negligible correlations were found between QST variables and psycho-social factors, contrasting the findings of a previous study. Our findings suggest that the CSI and experimental pain testing quantify different underlying constructs, and clinicians should be cautious when drawing conclusions regarding CS in patients based solely on CSI scores.