Exercise-induced pain changes associate with changes in muscle perfusion in knee osteoarthritis: exploratory outcome analyses of a randomised controlled trial

This is an exploratory sub-study of a previously published randomised controlled parallel-group-trial investigating the effects of a 12-week exercise program on the pain sensitivity in patients with knee OA focusing on the per-protocol population [7] performed at an outpatient’s clinic at a public hospital in Denmark (2012–2013). The study was approved by the Regional Health Research Ethics Committee of The Capital Region of Denmark (H-2-2011-159) and registered prior to commencement of the trial (www.​ClinicalTrials.​gov: NCT01545258).

A comprehensive description of the recruitment process and inclusion/exclusion criteria is published elsewhere [7]. Briefly, eligibility criteria included age ≥ 40 years, a clinical diagnosis of knee OA confirmed by radiography assessed by an experienced radiologist, and a body mass index between 20 and 35 kg/m². Exclusion criteria included participation in exercise therapy within the previous 3 months, having inflammatory and autoimmune diseases, and lower extremity joint replacement.

The sample size of the main trial was based on group differences in the primary outcome, pressure-pain threshold, and was calculated to 60 participants [7].

After baseline assessments, participants were randomly assigned (1:1 stratified by gender) to exercise therapy (ET) or to a control group (CG). A computer-generated list of random numbers was used and concealed from the researchers enrolling and assessing participants. Participants were aware of their group allocation; outcome assessors and data analysts were kept blinded to the allocation [7].

The ET group was offered facility-based, functional and individualised exercise therapy supervised by a trained physiotherapist 3 times weekly for 12 weeks. A full description of the.

exercise program is available in [7] and in the Additional file 1. Attendance at minimum 24/36 sessions was defined as protocol adherence. The CG received no attention during the 12 weeks and was requested not to engage in therapeutic exercise during study participation.

MRI of the most symptomatic knee at baseline was performed on a 3 T Siemens Verio® (3 T Magnetom Verio, Siemens, Erlangen, Germany) system using a 15-channel dedicated send/receive coil. In addition to a standard clinical MRI protocol, a sagittal DCE T1 VIBE sequence was performed with a temporal resolution of 9 s and 30 repetitions. During the third repetition an intravenous injection of 0.1 ml/kg body weight Gadolinium (Gd) contrast agent (Prohance®, Bracco Diagnostics Inc., Italy) was administered (2 ml/s). The total scanning time was 30 min with the DCE-MRI sequence performed after 20 min. The full MRI protocol is available in the Additional file 1.

The investigator analysing the DCE-MRI images (EB) was blinded to group allocation and clinical data and supervised by experienced radiologists (MB and RR). Dedicated software (Dynamika® version 4.2.2., Image Analysis Ltd., London, UK) was used for all DCE-MRI analyses [20, 21]: motion correction between temporal slices was applied, followed by determination of a baseline level of signal intensity (Fig. 1a), and regions of interests (ROIs) were manually drawn on all slices around the peri-articular muscle groups: extensors (vastus lateralis, vastus medialis) and flexors (biceps femoris, sartorius, gracilis, triceps surae, popliteus, semitendinosus, and semimembranosus) without delineation of individual muscles (Fig. 1c). Major vascular branches were avoided. Finally, the ROIs were summed and averaged into three volumes of interest (VOIs): Extensor VOI (extensor ROIs), Flexor VOI (flexor ROIs) and Total Muscle VOI (consisting of all ROIs).

As previously described in detail [15, 21], heuristic DCE-MRI analysis is based on changes in signal intensity over time in each voxel. The signal intensity changes are calculated relatively to the baseline signal intensity and changes over time can be plotted as time-intensity-curves (TICs) (Fig. 1a). Based on a robust classification scheme [21], the dedicated software automatically assigns each voxel, based on the shape of linear approximations of the TICs, to one of four enhancement patterns (Fig. 1b). The muscle perfusion parameters (see Table 1 and Additional file 1) were calculated on a voxel-by-voxel basis and averaged across each VOI [21, 22]. A pharmacokinetic parameter, K^trans, was calculated as described in Riis et al. 2016 [16] with the T1-relaxivity values of muscle and blood set at 1420 and 1400 ms, respectively.

Outcomes were measured at baseline and after the 12 weeks of intervention. For the purpose of this exploratory outcome analysis, we examined changes from baseline to follow-up, looking at the outcomes: muscle perfusion from DCE-MRI, patient-reported pain and function using the Knee Injury and Osteoarthritis Outcome Score (KOOS) questionnaire (0(worst)–100(best) [23], physical performance by knee extensor and flexor muscle strength test [24] and the six-minute walking test (6MWT) [25] (see the Additional file 1 for details).

DCE-MRI derived perfusion variables are listed and explained in Table 1. All have previously been used in knee OA- and rheumatoid arthritis studies [14‐17, 26, 27]; except Initial Rate of Enhancement and Maximal Enhancement Indices (for details see the Additional file 1).

Intra- and inter observer reproducibility of the DCE-MRI analyses were assessed for the Total VOI on a random subsample (n = 10) in which manual segmentations were repeated after a minimum of 4 weeks. The intra-observer reliability of the volume (cm³) of the Total Muscle VOI was ICC = 0.99 (95% CI: 0.98 to 0.99). ICC of the muscle perfusion variables was ≥0.99 (in each case, the lower limit of the 95% CI was ≥0.91′). Due to low reproducibility of the K^trans it was not included in the further analysis. Details of reproducibility of the DCE-MRI variables are presented in the Additional file 1.

The reliability of KOOS [28], muscle strength tests [24], and 6MWT has been demonstrated [29].

As pre-specified in the main trial protocol, the study focuses on individuals who adhere to the protocol. The per-protocol population was defined as participants with 12 or fewer incidences of non-attendance at exercise sessions in the ET group, and no exercise in the CG. The present exploratory analyses aim at investigating the effects of exercise on muscle perfusion in the peri-articular knee muscles; thus, only individuals included in the per-protocol population with complete DCE-MRI data sets at baseline and follow-up were included in the present analyses (Fig. 2).

Individual changes from baseline were calculated by subtracting the baseline value from the follow-up value for all outcomes. To assess group mean differences in the changes from baseline in all variables (purpose 1), analyses of covariance (ANCOVA) was used with change from baseline at follow-up as dependent variable, and baseline value of the dependent variable, treatment group, and gender (due to stratification by gender when randomising) as covariates [30]. For sensitivity purposes the analyses were repeated with further inclusion of age, weight and gender as covariates. All analyses were performed using SAS software (SAS Inc., Cary, NC, USA).

To explore whether changes from baseline to follow-up in the muscle perfusion variables were associated with changes in pain and function (KOOS) and physical performance measures (purpose 2), we assessed the rank correlation (Spearman’s rho) of changes from baseline between the DCE-MRI muscle perfusion variables, KOOS subscales, muscle strength, and 6MWT across the entire population.

Furthermore, we investigated the potential effect mediation of changes in the DCE-MRI perfusion variables for changes in KOOS pain in a post-hoc mediation analysis using a model-based approach with nonparametric bootstrap for variance estimation. The mediation models were adjusted for the same variables as the primary analysis (i.e. baseline KOOS pain and gender). We estimated an average causal mediation effect [31], i.e. the amount of the total effect that are mediated by a specific variable, as well as the proportion of the total effect that is mediated. The R software, version 3.5.1 (R Foundation for Statistical Computing, Vienna, Austria), with the R Package ‘mediation’ for Causal Mediation Analysis [32] was used.

At follow-up there were statistically significant group differences in changes from baseline in all muscle perfusion variables in favor of ET in the Total Muscle and Flexor VOIs showing a relative increase in the ET group compared with the CG.

Furthermore, in the Extensor VOI, the changes in Initial Rate of Enhancement Composite Score (IRExNvoxel) and Initial Rate of Enhancement Index (IRExNvoxel%) were statistically significantly different in favor of ET, showing similar patterns as for the Total Muscle and Flexor VOIs, with a relative increase in the ET group compared with the CG. The results are summarized in Table 3.

There was a statistically significant and clinically relevant group difference (10.7 KOOS points (95%CI 3.3 to 18.1)) in changes from baseline in KOOS pain in favor of ET similar to the per-protocol population in the main study [7]. No statistically significant group differences were observed in the other KOOS subscales (Table 3).

No statistically significant group differences were observed in changes from baseline to follow-up in the muscle strength variables and in 6MWT (Table 3).

The results were robust to sensitivity analysis, with only slight changes in the estimates (see the Additional file 1).

Changes from baseline in KOOS pain were positively correlated with changes in The Proportion of Highly Perfused Voxels (%) (Nvoxel%) and Maximal Enhancement Index (MExNvoxel%) in the Total and Flexor VOIs (r ≥ 0.36, P ≤ 0.042). Further, changes in KOOS pain were positively correlated with changes in IRE Composite score (IRExNvoxel) and ME Composite score (MExNvoxel) in the Flexor VOI (r ≥ 0.35, P ≤ 0.044). This indicates that pain reduction was associated with increase in the degree and area of perfusion of the peri-articular muscles (Fig. 3).

No statistically significant associations were found between changes in muscle perfusion variables and the remaining 4 KOOS subscales (r < 0.18, P > 0.317). Changes from baseline in isometric knee extensor muscle strength were the only muscle strength variable which correlated significantly with changes in the muscle perfusion variables (the highest Spearman’s rho = − 0.46, P = 0.028). No other statistically significant associations were found between changes in muscle perfusion variables and changes in the variables of physical performance (highest Spearman’s rho = − 0.38, P = 0.076) (See the Additional file 1 for correlation matrix).

Over all, in the comparison between exercise and a no-attention control group, for changes in muscle perfusion variables (DCE-MRI) there was no statistically significant mediation of the pain-reducing effect of exercise in patients with knee OA, except for three muscle perfusion variables in the Flexor VOI. These were initial Rate of Enhancement Composite Score (IRExNvoxel), Maximal Enhancement Composite Score (MExNvoxel) and Maximal Enhancement index (MExNvoxel%) (Table 4), mediating 26% (95% CI, 0 to 1.24), 28% (95% CI, 1 to 1.71) and 28% (95% CI, 1 to 1.46), of the total effect of the pain-reducing effect of exercise, respectively. However, the upper limit of the 95% CI of the estimated proportions mediated suggest a potential mediation of effects on pain by all muscle perfusion variables (DCE-MRI) (Table 4).

The current study does not provide insight into the underlying mechanism of changed muscle perfusion. Exercise causes structural adaptations in the muscles, such as increased capillarisation or angiogenesis as a physiological response [9‐11]. Based on the present results, it could be speculated that exercise in knee OA prevents such loss of capillarisation in the muscle tissue. Exercise programs specifically aiming at physiological muscle adaptations (e.g. resistance or cardiovascular training) may prove even more effective in inducing changes in muscle perfusion and are also beneficial in terms of pain [33, 34].

Pain experience is affected by psychological variables, such as depression and quality of life [35, 36], and exercise is likely to change these [37‐39]. It is plausible that exercise-induced pain reductions may also be mediated through improvements in psychological well-being.

As exercise therapy cannot be blinded to the provider or to the patient, the observed clinical effects may be facilitated by effects associated with attention and study participation [8]. However, as the DCE-MRI assessment of muscle perfusion is a truly objective method, this study provides evidence of biological mechanisms associated with the pain-relieving effects of exercise therapy. Unlike radiographs, where there is discordance between radiographic findings and pain [40], and a poor short-term responsiveness to detecting change [41], muscle perfusion may be a potential objective marker of exercise-induced pain reduction (and potentially other treatments) on a structural level in knee OA.

Increased perfusion of intra-articular structures (e.g. synovium and the infra-patellar fat pad) are conventionally interpreted as signs of inflammation, and studies have repeatedly shown that intra-articular perfusion associates with a higher degree of pain [15‐17, 27]. Our current and previous results [12] differ from these observations as the pain–perfusion relationship in the peri-articular muscle tissue is opposite. This could suggest a possible trade-off between intra-articular and peri-articular perfusion in the response to inflammation or treatment with subsequent changes in pain. This proposed mechanism needs further exploration.

We assessed knee OA pain using the KOOS questionnaire and found clinically relevant group differences that exceed the minimal clinically important difference (MCID) of 8–10 KOOS points as suggested by the developers of the questionnaire (www.​koos.​nu). This increases the relevance of our findings. In contrast, the MCIDs in the muscle perfusion variables have not been established, which encumbers interpretation of our results. However, the observed differences in muscle perfusion variables were all beyond the smallest detectable changes, and the reliability of the image analyses was satisfactory, which altogether strengthens our confidence in the results.

When evaluating the effects of exercise in knee OA, patient-reported outcomes (PROs) and performance-based tests are recommended and frequently used [5, 42, 43]. PROs and performance-based tests are affected by patient-provider interaction [44, 45], which is an important source of bias when the intervention cannot be blinded. Furthermore, the validity of performance-based tests (e.g. muscle strength) may be challenged by pain or fear of pain during the test [46, 47]. This means that a high degree of participant cooperation is necessary, which increases the risks of measurement bias and underlines the need for objective and sensitive measures in the evaluation of the underlying effects of exercise. DCE-MRI is such an objective method, and we applied it in the analysis of muscle perfusion, which is novel within OA research.

The use of a rigorous randomised study design with pre-specified methods using standardised and validated image analyses with high reproducibility are major strengths of our study. On the other hand, the limited sample size and the many variables and statistical tests performed increase the chances for multiplicity. The fact that the study was designed and powered for another outcome are limitations and calls for replication in larger studies specifically designed for DCE-MRI. Nevertheless, we could detect clinically meaningful important signals associated with the effects of exercise, a universally recommended treatment for knee OA. Another important aspect to consider is the per protocol design of the study. This was done to investigate the underlying mechanisms of exercise and thus it is necessary to focus on the participants who had complete DCE-MRI assessments at both baseline and follow-up and who had adhered to the protocol, i.e. received the allocated intervention (exercise vs control). This is fundamentally different from intention to treat study designs and therefore this study cannot be used to inform clinical practice. An important limitation is that out of 14 potential mediation analyses (corresponding to KOOS subdomains, muscle strength measurements, and 6-min walk distance) we only analysed mediation of effects on KOOS pain as this was the only outcome with significant group differences. This analysis did not support that changes in muscle perfusion mediate pain improvements.