The course of pain hypersensitivity according to painDETECT in patients with rheumatoid arthritis initiating treatment: results from the prospective FRAME-cohort study

Pain in rheumatoid arthritis (RA) has typically been regarded as nociceptive, that is, related to ongoing peripheral inflammation [1]. However, during the last decade, where focus has been on early diagnosis and aggressive treatment strategies in the treat-to-target regime [2], it has become clearer that in a subgroup of RA patients, pain can become an entity in its own right, probably elicited by, but not directly related to, ongoing inflammation [3]. A substantial proportion of RA patients in stable clinical remission continue to report moderate to severe pain levels [4] and studies have indicated that RA leads to widespread pain in 10–20% of patients [5]. Such observations have led to the contention that changes in the peripheral and central nervous system through processes of neural plasticity and central sensitization may play an important role [6]. As a rule, sensitization phenomena would be expected to extinguish as the tissue heals and inflammation subsides. However, a state of induced hypersensitivity of the pain system may persist in subsets of patients and lead to chronic pain states in which pain is no longer coupled to the presence of ongoing peripheral inflammation [7]. In such patients, persistent pain hypersensitivity may lead to continuous high reports of tender joints and poor global health; subcomponents of the commonly used composite disease activity score of 28 joints (DAS28-CRP) and thus overestimation of inflammatory activity. If the anti-inflammatory treatment is intensified on this background, little change in DAS28-CRP can be expected. Conversely, if inflammatory RA is left un- or not sufficiently treated it will lead to joint destruction and loss of function [8]. Identification of underlying pain mechanisms therefore has potential importance when prognosticating the effect of medical treatment on inflammation and pain.

The painDETECT questionnaire (PDQ) is a self-administered pain classification instrument originally developed to differentiate neuropathic (non-nociceptive) from non-neuropathic (nociceptive) pain [9]. It has been increasingly used in patients with osteoarthritis and fibromyalgia to assess clinical pain features indicative of central sensitization [10‐12] and has recently been introduced in studies assessing pain mechanisms in patients with RA and spondyloarthritis [13‐15].

Magnetic resonance imaging (MRI) is an objective and sensitive method to assess joint inflammation. The most common scoring system in the wrist and metacarpophalangeal (MCP) joints is the OMERACT (outcome measures in rheumatoid arthritis clinical trials) RA MRI scoring (RAMRIS) system [16]. Dynamic contrast-enhanced (DCE) MRI is a technique where the sequences are acquired rapidly and sequentially before and during contrast infusion. DCE-MRI has been shown to correlate better than conventional MRI with the histologic findings of synovitis [17, 18].

In this study, we hypothesized that a high PDQ score would serve as an indicator of central sensitization and thus a prognostic factor for a poorer treatment outcome (DAS28-CRP change) in patients with RA initiating or intensifying anti-inflammatory treatment. A possible statistical interaction between central sensitization (high PDQ score) and inflammatory load (baseline synovitis defined by hand MRI RAMRIS score) was considered as part of the hypotheses. In the exploratory part of the study, we hypothesized that DCE-MRI would capture change in inflammation and thus a possible relation to inflammatory pain mechanisms (low PDQ score) better than conventional MRI.

Figure 1 illustrates the flow of patients. In all, 151 patients fulfilled the inclusion criteria. Of these, 48 patients were excluded. In total, 103 patients received a baseline assessment; however two were excluded post hoc. In total, 101 patients completed the PDQ at baseline; 47 initiating csDMARD and 54 initiating bDMARD. Of these, 75 patients completed the PDQ and had an MRI scan performed at baseline and of these, 71 patients had an MRI scan performed at follow-up. This discrepancy was primarily caused by administrative delay resulting in no patients receiving MRI-scan during the first 2 months of the study period. Exclusion and reasons for dropout are further described in the figure text. Due to 3 patients not wishing to receive contrast, 72 patients with complete PDQ were included in the exploratory DCE-MRI study at baseline (not shown in Fig. 1). Of these72 patients, in all 65 completed the follow-up scan with contrast.

Baseline characteristics are described in Table 1 (some data has previously been published [36]). The distribution of patients across the PDQ classification groups were n = 66, n = 23, n = 12 for PDQ score < 13, 13–18, > 18, respectively. Statistically significant differences across the three PDQ classification groups were observed for tender joint count (TJC), tender point (TP) count, DAS28-CRP, physical function (HAQ-DI), VAS-fatigue, VAS-pain, VAS-GH, anxiety (GAD-10), depression (MDI), SF-36 PCS, and SF-36 MCS. A TP count ≥11 was found for higher proportions of patients with higher PDQ classification groups; however, this was not statistically significant. No differences were found across the groups for the biochemical and imaging variables.

Changes from baseline stratified by PDQ group are presented in Table 2. Change in PDQ classification group is reported as classification consistency, i.e. number of patient that did not change classification group. Statistically significant differences in classification consistency were found for the low, intermediate, and high PDQ group; 97%, 83%, and 0%, respectively. For all clinical variables and PROs, a U-shaped curve change pattern was observed with greater change in the high and low PDQ classification group than in the intermediate. Remission at follow-up according to DAS28-CRP (< 2.6) was found in 41%, 22%, and 42%, respectively. The change in imaging variables did not display a distinct pattern. Statistically significant differences between the classification groups were found for delta change of CRP, DAS28-CRP, VAS GH, HAQ-DI, and SF36- MCS. However for CRP, no differences between PDQ score < 13 and > 18, and 13–18 and > 18 were found. For DA28-CRP and VAS GH no differences were found between PDQ score < 13 and PDQ score > 18. Regarding HAQ-DI and SF36-MCS there were no difference between PDQ score < 13 and 13–18. DCE-MRI variables are presented separately in Table 3. There were neither significant differences nor trends in the DCE-MRI variables across the groups; however, the variables IRExNvoxel (in ml) and MExNvoxels (in ml) for the wrist also displayed the U-shaped change pattern in line with the RAMRIS score (BME). Baseline characteristics and delta change by initiation group have previously been reported [36, 37].

In the multivariable regression models (Table 4) change in the predefined outcome variables was expressed as least square means. No interaction was found between PDQ classification group and RAMRIS synovitis neither in the protocolized analysis, nor between PDQ classification group and the DCE-MRI variables (IRExNvoxel (ml), MExNvoxel (ml)) in the exploratory analysis. All interaction statements were therefore left out of the models. No prognostic value of PDQ classification was found for any of the outcomes in any of the models, sensitivity analysis included (p = 0.44) (data for DCE-MRI is presented as additional material, Additional file 1).

Baseline DAS28 was found to be statistically significantly associated with change in all outcomes in both adjusted models; DAS28, RAMRIS, and VAS pain (p < 0.01), sensitivity analysis excluded. In this analysis, baseline VAS pain was found to be significantly associated with change in VAS pain (p < 0.001). Baseline RAMRIS synovitis score was found to be statistically significantly positively associated with DAS28-CRP change in the unadjusted model (p = 0.01) and of RAMRIS change in both the unadjusted and the adjusted model (p ≤ 0.02).

In the exploratory DCE-MRI study inter and intra-reader reliability showed good to excellent agreement (data is shown in Additional file 2).

To our knowledge, this is the first study to evaluate the prognostic value of pain classification by the PDQ score in relation to change in DAS28-CRP, VAS pain, and RAMRIS score in RA patients initiating or escalating anti-inflammatory treatment. It was hypothesized that patients in the high PDQ classification group (score > 18) would display constant high TJC and VAS GH as features of persistent pain hypersensitivity (central sensitization) and thus gain little or no change of DAS28-CRP following treatment initiation. In contrast however, this was not confirmed in the multiple regression analysis, we found that these patients (n = 12) experienced the greatest numerical change in DAS28-CRP, self-reported disease severity measures, and objective inflammatory parameters, including MRI (Table 2). This result was seen despite the fact that baseline inflammatory parameters were the same across the three PDQ classification groups, and that other baseline characteristics of the high PDQ classification group clinically could indicate presence of central sensitization (high frequency of female sex, a high number of tender joints and tender points, poor mental well-being, a high disability index, and high VAS scores).

In contrast to the patients in the low and intermediate PDQ classification -group, all patients in the high PDQ classification group changed pain classification following treatment initiation or escalation. These findings could indicate that in these particular patients, a high PDQ score might not have marked central pain mechanisms uncoupled to ongoing inflammation, but could have reflected reversible inflammatory driven pain hypersensitivity, i.e. normal neuroplasticity. However, regression toward the mean needs to be taken into account. Interestingly, the observed numerically changes of the variables were smallest in the intermediate PDQ group (score 13–18), total hand RAMRIS synovitis score excluded. Thus, patients with an unclear pain mechanistic background had the poorest response to medical treatment also reflected in the U-shaped DAS28 remission pattern across the PDQ classification groups, indicating that the uncoupling of pain mechanism from present inflammation could be found in this group.

In the multivariable regression analyses, the PDQ did not have prognostic value in relation to change of DAS28-CRP, RAMRIS score, or VAS pain. No interaction between PDQ score and baseline RAMRIS synovitis score was found in any of the regression analyses including the exploratory DCE-MRI analyses indicating no relation between ongoing inflammation and pain phenotype (pain classification group). However, in the light of the relatively small sub-sample of patients with a high PDQ score at baseline, it does not seem reasonable to reject any prognostic value of the PDQ at this point.

In contrast, DAS28-CRP at baseline was a significant predictor (p < 0.01) of change in the dependent variable (DAS28-CRP, RAMRIS synovitis score, VAS pain) in all adjusted models. Furthermore, RAMRIS synovitis predicted DAS28-CRP change in the unadjusted model (p = 0.01), and RAMRIS synovitis change both in the unadjusted and the adjusted model (p ≤ 0.02). The impact of RAMRIS synovitis score on clinical change is not well documented [38, 39]. Although our group previously has found correlations between RAMRIS synovitis score and DAS28-CRP in another cohort of RA patients (in review), this may be the first study to indicate a possible prognostic value of RAMRIS synovitis score in relation to DAS28-CRP change within 4 months. The clinical consistency and relevance of this finding needs further investigation.

Overall the DCE-MRI analyses did not add further information to conventional MRI.

The unexpected behavior of the PDQ may lead to speculations about whether the presents finding are due to the PDQ not being valid in this population. However, the PDQ has, in at least three cross-sectional studies [13, 14, 36], indicated involvement of non-nociceptive pain mechanisms in subgroups of RA patients. Although we failed to demonstrate a prognostic value of the PDQ, the implications may still be that in patients with clear indication of non-nociceptive pain mechanisms treatment strategies should involve management of chronic pain, including treatment with classes of medication that target central pain mechanisms [6, 7, 40‐43], and not only focus on medical treatment of the underlying disease.

Inflammation was not only assessed clinically but also by MRI of one hand to ensure objectivity. Assessing one hand as a target joint area to reflect general inflammation is a common procedure but can be a limitation to the regression analysis as it involves a risk of misestimating the inflammatory load especially in patients with a non-typical RA presentation. However, it has not yet been clarified how many joints that should be included in a prognostic image analysis to reflect overall inflammatory load.

The study was limited by the framework of the PDQ; the original development for classification of neuropathic pain among patients with various chronic pain conditions, the algorithm for cutoff points also being validated in this sample and the lack of a clinical ‘gold standard’ assessing augmented central pain mechanisms, against which the PDQ preferably should be tested. However, patients with augmented central pain mechanisms express the same pain features as patients with neuropathic pain, though in a generalized pattern [12, 44‐46]. In our opinion this sensitive but not specific feature of the questionnaire, association to tender point count and sign of central sensitization in quantitative sensory testing and functional MRI findings within other chronic pain conditions vouch for the use of the PDQ as indicator of augmented central pain processing [10‐12, 47‐49].

The relatively short follow-up time of 4 months is a limitation that may have reduced the treatment response for some patients. Further, the baseline visits including MRI could be up to 3 weeks after starting therapy for patients initiating a csDMARD and up to 1 week for patients initiating a bDMARD, which also may have reduced the observed treatment change from baseline to follow-up, particularly on DCE-MRI that is known to be very sensitive to fast inflammatory changes [31, 35, 50, 51]. Finally, a direct pain modulating effect of anti-tumor necrosis factor (TNF)-alpha treatment is indicated by the literature [52], and it could therefore be speculated that in some patients treated with anti-TNF blocker the allowed delay in baseline assessment may have influenced nociceptive signaling and thereby the study findings.

In this study, higher numerically changes of inflammatory and patient-reported outcomes, and change of PDQ classification-group at follow-up were observed in the group of patients with a high PDQ score at baseline, which may indicate that inflammatory pain can lead to pain hypersensitivity of a reversible character. Interestingly, overall reduced change in variables was found in patients in the intermediate PDQ classification group with unclear pain mechanism indicating irreversible pain mechanisms. In contrast to our hypothesis, a high PDQ score had no prognostic value in relation to treatment outcome specified as change in DAS28-CRP, RAMRIS score, DCE-MRI, or VAS pain 4 months after initiation or escalation of medical therapy, however, it was found that the RAMRIS synovitis score may have prognostic value in relation to DAS28-CRP response. Further large-scale studies are needed to clarify the prognostic value of the PDQ.