Performance of an MRI scoring system for inflammation of joints and entheses in peripheral SpA: post-hoc analysis of the CRESPA trial

Abstract

Objectives

The aim of this study was to investigate the reliability, validity, and sensitivity to change of a novel MRI scoring system in early peripheral SpA (pSpA).

Methods

MRI of the pelvis and lower extremities was performed before initiation of the TNF inhibitor golimumab in 56 patients and repeated in 46 patients who achieved sustained clinical remission after 24, 36 or 48 weeks. Three readers applied a semi-quantitative MRI scoring system for lower-extremity joint and entheseal inflammation. Four lesion types were assessed: entheseal osteitis, entheseal soft-tissue inflammation, joint osteitis, and joint synovitis/effusion. MRI response was defined as a decrease in MRI lower-extremity inflammation index (sum of scores from 75 sites, each scored 0–3) above the smallest detectable change (SDC).

Results

At follow-up, the MRI index decreased in 34 of 46 patients (74%), and 15 (33%) patients achieved MRI response, i.e. a decrease above SDC of 2.8. When restricting the analysis to patients with clinical involvement of lower-extremity sites that were assessed by MRI, 13 of 28 (46%) achieved MRI response. Interreader reliability was very good, with an average-measure intraclass correlation coefficient of 0.92 (95% CI: 0.85–0.95) for status scores and 0.89 (0.80–0.94) for change in scores. The MRI index correlated with other measures of disease activity, including CRP [Spearman’s rho 0.41 (0.23–0.56)], swollen joint count of 6 joints [0.47 (0.27–0.63)], tender enthesis count of 14 entheses [0.32 (0.12–0.50)] and pain score [0.28 (0.08–0.46)], all P < 0.05.

Conclusion

The proposed MRI lower-extremity inflammation index demonstrated reliability, validity, and sensitivity to change in patients with early pSpA.

Trial registration

Clinicaltrials.gov, http://clinicaltrials.gov, NCT01426815.

Introduction

Patients with SpA may be subdivided into groups with predominantly axial disease (axSpA) and predominantly peripheral disease (pSpA), although overlap exists [1]. Enthesitis is found in 33–59% of patients and synovitis in 17–91% of patients, with higher prevalences in patients with non-psoriatic pSpA, PsA, or combined forms of SpA, and lower prevalences in patients with axSpA. Importantly, these manifestations are associated with a high disease burden [2]. It is well established that MRI allows an objective assessment of inflammation in the SI joints and spine and is therefore used routinely in clinical trials and practice [3]. However, MRI also allows an objective assessment of signs of inflammation in peripheral joints and entheses, including sites that may be difficult to assess clinically, e.g. hip joints, pelvic entheses, and midfoot joints [4–7]. Therefore, a measure for the overall level of inflammation of peripheral joints and entheses as assessed by MRI is of interest as a possible outcome measure in clinical trials and to assess disease activity in routine care.

MRI has demonstrated decreased inflammatory load at peripheral sites during treatment and can discriminate between treatment groups in randomized controlled trials in axSpA patients [8–13]. However, to our knowledge, evidence of the reliability, validity, and sensitivity to change of semi-quantitative MRI scoring of peripheral joints and entheses in patients with pSpA has not previously been reported.

The aim of this study was to investigate the interreader reliability, and validity compared with known measures of disease activity and sensitivity to change of a semi-quantitative lower-extremity MRI scoring system in early pSpA patients.

Methods

Setting

The CRESPA trial (clinicaltrials.gov identifier: NCT01426815) included 60 patients with newly diagnosed early pSpA, defined as a symptom duration of <12 weeks. Patients were randomized 2:1 to receive golimumab or placebo for 24 weeks, followed by an open-label phase with golimumab from week 24 to week 48 for all patients [14–16]. All patients fulfilled the Assessment of SpondyloArthritis international Society (ASAS) criteria for pSpA [17]. The main results of the trial have already been published [14–16]. In this post-hoc analysis, we include data from 56 patients with available MRI images before treatment initiation (baseline). Forty-six patients achieved sustained clinical remission and had repeated MRI at follow-up. Only four of those were in the placebo arm. Follow-up MRI was only performed if criteria for sustained clinical remission, defined as complete absence of peripheral arthritis, enthesitis and dactylitis on clinical examination at two visits with a 12 week interval being fulfilled at weeks 24, 36 or 48. Thus, the analyses presented here do not compare the two treatment groups but combine data from both groups at two time points: baseline and sustained clinical remission at either weeks 24, 36 or 48.

MRI acquisition

For this analysis, coronal and axial T1-weighted images and short tau inversion recovery (STIR) images of pelvis and hips, sagittal T2-weighted fat-saturated images of the knees, and sagittal T2-weighted fat-saturated images of the ankles, hindfeet and midfeet were obtained as previously described using body flexed array coils, knee coils and ankle coils. The MRI unit was a 1.5 T Magnetom Avanto, Siemens Healthineers, Erlangen, Germany [16].

MRI assessment

After an initial calibration exercise, three readers [an experienced musculoskeletal radiologist (L.J.), an experienced rheumatologist (M.Ø.) and a PhD student (S.K.)] independently assessed all MRI images of the pelvis and hips (except the SI joints), knees, ankles, hindfeet and midfeet, blinded to chronology and all clinical data. We applied a methodology for assigning semi-quantitative MRI scores that was largely like the proposed OMERACT MRI in Arthritis Working Group MRI-WIPE scoring system [18–21]. However, in the current study, MRI of upper extremities, anterior chest wall, and forefeet were not included, and we therefore used a modified scoring system limited to the pelvis, hips, knees, ankles, hindfeet and midfeet. We also decided to assess osteitis related to the anterior and posterior cruciate ligaments of the knees, even though this was not included in the OMERACT MRI-WIPE scoring system. Joints were scored 0–3 (none/mild/moderate/severe) for effusion/synovitis (10 sites) and 0–3 for bone marrow oedema/osteitis (22 sites). Entheseal sites were scored 0–3 for soft-tissue inflammation (19 sites) and 0–3 for bone marrow oedema/osteitis (24 sites). Grading was based on an overall apprehension of intensity and extent of the lesion (Fig. 1).

The following joints were assessed for effusion/synovitis and osteitis (Fig. 2): hip joint (osteitis was assessed separately for acetabulum and femoral head), knee joint (osteitis was assessed separately for patella, medial femoral condyle, lateral femoral condyle, medial tibial condyle, and lateral tibial condyle), ankle joint (osteitis was assessed separately for talus and the crural bones), the subtalar joint, and the talocalcaneonavicular-calcaneocuboid joints (assessed as one joint).

The following entheseal sites were assessed for soft-tissue inflammation and osteitis (Fig. 2): iliac crest, posterior iliac spine, greater trochanter, ischial tuberosity, symphysis (osteitis was assessed separately for left and right pubis), quadriceps femoris tendon insertion into patella, patellar tendon insertion into patella, patellar tendon insertion into tibial tuberosity, anterior cruciate ligament, posterior cruciate ligament, Achilles tendon, and plantar fascia insertion into calcaneus. The anterior and posterior cruciate ligaments were not assessed for soft-tissue inflammation; for each ligament, the femoral and tibial insertions were combined into one score for osteitis.

MRI indices

The MRI lower-extremity inflammation index was defined as the sum of scores from all 75 sites, with a theoretical range of 0–225. It was subdivided by lesion type into the MRI joint inflammation index [consisting of the sum of the MRI joint synovitis index (encompassing both synovitis and effusion) and the MRI joint osteitis index] and the MRI enthesis inflammation index (consisting of the sum of the MRI enthesis soft-tissue inflammation index and the MRI enthesis osteitis index). Another subdivision into three anatomical regions was also applied: the MRI pelvis and hips inflammation index, the MRI knees inflammation index, and the MRI ankles and feet inflammation index.

Clinical variables and PROMs

A tender joint count of 8 joints (TJC-8) and a swollen joint count of 6 joints (SJC-6) were derived by counting the number of involved joints among hip joints, knee joints, ankle joints and midfoot joints bilaterally (hips were not assessed for swelling) to match the joints that were assessed by MRI. Similarly, a tender enthesis count of 14 entheses (TEC-14) was derived by counting the number of tender entheses among the Achilles tendon and plantar fascia insertions into calcaneus, iliac crest, anterior superior iliac spine, posterior superior iliac spine, quadriceps tendon insertion into the superior pole of patella, and patellar tendon insertion into the inferior pole of patella or into the tibial tuberosity, to match the entheses that were assessed by MRI. In addition, CRP, ESR, pain score, patient global score, BASDAI and BASFI were used for these analyses.

Reliability

Multiple-reader agreement plots were constructed to show agreement and discrepancies between readers. Two-way intraclass correlation coefficient (ICC) models by absolute agreement, both as single-measure (relevant if it is planned to use scores from one reader) and as average-measure (relevant if it is planned to use averaged scores from three readers) were used to estimate the ICC.

Validity

Spearman correlation analyses between the MRI lower-extremity inflammation index, MRI joint inflammation index and MRI enthesis inflammation index vs other known measures of disease activity, including serum markers and patient-reported outcome measures, both for status scores and change scores, were performed. Bootstrapping with 2000 replicates was used to derive 95% CIs. All measures of disease activity were expected to be higher at baseline compared with clinical remission; therefore, data from both time points were combined in the correlation analyses for status scores to allow the full range of observed values to be captured, and thus one patient could contribute with data on two separate occasions to this analysis. Relations between MRI lower-extremity inflammation index and established measures of disease activity, using data from both time points, were also investigated using scatter plots and locally estimated scatterplot smoothing (loess) curves with bootstrap 95% CIs.

The clustering of various measures of disease activity in patients with pSpA was investigated using hierarchical cluster analysis. Joint and enthesis counts, MRI indices, and serum markers were transformed using log(x + 1) to be closer to normal distribution, and thereafter all variables were scaled to mean 0 and standard deviation 1. Manhattan distance, where the pairwise discrepancies in two measures across all patients are summed, was used, and the average aggregation method, which takes all values into account during the clustering process, was applied. To assess the uncertainty of the clustering process, the bootstrap probability for each cluster being chosen was computed using 10 000 bootstrap samples [22].

Change in MRI lower-extremity inflammation index among PSpARC40 responders and non-responders was investigated using a paired t test. PSpARC40 has been defined as ≥40% improvement from baseline (and ≥20-mm absolute improvement on a visual analogue scale) in the patient’s global assessments of disease activity and pain, and ≥40% improvement in at least one of the following features: swollen joint and tender joint counts, total enthesitis count, or dactylitis count [23].

Sensitivity to change

Descriptive statistics by lesion type or anatomical region and standardized response mean (average change value divided by the standard deviation of the change values) were calculated.

The number of patients achieving sustained clinical remission with golimumab treatment that could be classified as MRI responders was investigated using various computational approaches: (1) improvement above the estimated smallest detectable change (SDC), (2) improvement in MRI index of at least 50%, (3) net number of patients with improvement in MRI index, defined as the number of patients with any improvement >0 minus the number of patients with any worsening >0, i.e. similar to an approach used to determine radiographic structural progression at the group level [24], and (4) the number of patients with improvement >0 as assessed concordantly by all three readers. The smallest detectable change using the average scores of three readers was estimated from a two-way analysis of variance with patients and readers as covariates by applying the formula 1.96 × √(residual mean square)/√(3) [25].

Ethics

This study complies with the Declaration of Helsinki and was approved by the Medical Ethics Committee of the Ghent University Hospital. Written informed consent was obtained from each patient before study-related procedures were performed.

Results

Reliability

Multiple-reader agreement plots showed that discrepancies between readers were numerically lower for patients with lower status scores and lower change scores. Discrepancies in change scores were relatively small in most patients but larger in a subset of the cases (Fig. 3, Panel A). ICC values were moderate to very good for the overall MRI lower-extremity inflammation index as well as for its subdivisions. The MRI lower-extremity inflammation index had a very good average measure ICC of 0.92 for status scores at baseline and 0.89 for change scores (Table 1).

Validity

The MRI lower-extremity inflammation index, MRI joint inflammation index and MRI enthesis inflammation index correlated significantly with other measures of disease activity, both when status scores and change scores were analysed (Supplementary Table S1 and S2, available at Rheumatology online). Change in the MRI peripheral enthesis inflammation index correlated most closely with change in TEC-14. Change in the MRI joint inflammation index correlated most closely with change in SJC-6 and TJC-8. The MRI indices were positively related to known measures of disease activity as visualized by scatter plots, although some residual variation was apparent (Fig. 3, panel B).

Hierarchical cluster analysis identified one cluster of outcome measures that could be labelled ‘joint inflammation and serum markers’, one cluster of outcome measures that could be labelled ‘entheseal inflammation’, and one cluster of outcomes measures that could be labelled ‘patient-reported outcomes’ (Fig. 3, panel C).

Among the 46 patients achieving a status of sustained clinical remission, 34 were PSpARC40 responders and 11 were PSpARC40 non-responders (1 had missing data). PSpARC40 responders had a larger mean improvement in the MRI lower-extremity inflammation index compared with non-responders [mean 3.4 (s.d. 4.7) vs mean 1.0 (s.d. 2.3), P = 0.03].

Sensitivity to change

Joints and entheses showed a significant improvement in the MRI lower-extremity inflammation index and its components between baseline and time of clinical remission. Numerically, the largest improvement was found in the MRI joint synovitis index (mean decrease of 1.8), followed by the MRI enthesis soft-tissue inflammation index (mean decrease of 0.9) and the MRI enthesis osteitis index (mean decrease of 0.4) (Table 2). The MRI joint osteitis index remained stable and thus did not contribute to the observed overall improvement. The largest standardized response mean was observed for the MRI joint synovitis index followed by the MRI enthesis soft-tissue inflammation index (Table 2).

All three anatomical regions (pelvis/hips, knees, ankles/feet) contributed to the overall improvement (Table 2 and Fig. 1).

The net number of patients with improvement in the MRI inflammation index was 24 of 46 (52%), the number of patients with improvement in the MRI inflammation index of ≥50% was 21 of 46 (46%), the number of patients with improvement in the MRI inflammation index as assessed by all three readers was 17 of 46 (37%) and the number of patients with an improvement larger than the estimated SDC of 2.8 was 15 of 46 (33%) (Table 3). The MRI joint synovitis index allowed a similar number of patients to be identified as responders as did the total MRI lower-extremity inflammation index, but all other subdivisions by lesion type or anatomical region allowed fewer patients to be classified as responders compared with the total MRI lower-extremity inflammation index (Table 3). In a sensitivity analysis in which patients were excluded if they had clinical involvement only in the upper-extremities or forefeet (i.e. no clinical involvement in the areas assessed by MRI), the observed MRI response rates were higher (Table 3).

Discussion

We successfully demonstrated the reliability and validity of a semi-quantitative MRI scoring system for inflammation in peripheral joints and entheses in the lower extremities in patients with early pSpA. The method was sensitive to change, had good reliability for status and change scores, and validity was established by correlation with other measures of disease activity.

The MRI lower-extremity inflammation index had a range of 0–225, but scores at baseline were only in the lower part of the range, with a mean score of 7.2 and an observed range of 0–32, consistent with an oligoarticular and oligoentheseal pattern, in which only a few sites had signs of inflammation, while the majority of sites had a normal appearance under MRI. The mean change score was −3.1, rather close in magnitude to the smallest detectable change (SDC) of 2.8, and only 15 of 46 (33%) patients had a decrease above the SDC. In a sensitivity analysis in which patients with clinical involvement of only the upper extremities or forefeet were excluded, since these parts were not covered by the MRI scans, a larger proportion of the patients was identified as responders and the standardized response mean (SRM) was higher. When data from each of the three anatomical regions, (i) pelvis and hips, (ii) knees, and (iii) ankles and feet, were analysed separately, numerically lower proportions of patients were identified as responders. Based on this, we consider it likely that an MRI response might have been detected in an even higher proportion of patients, if MRI of the upper extremities, anterior chest wall, and forefeet had been available.

The number of patients that could be identified as MRI responders was calculated using several approaches. Numerically, more patients could be identified as MRI responders when calculated as the number of patients with at least 50% improvement in the MRI inflammation index or when calculated as the number of patients for whom all three readers agreed upon as scoring an improvement, whereas fewer patients could be identified as MRI responders when calculated as the number of patients with an improvement larger than the SDC. Differences between readers tended to be larger in patients with larger average change scores, and readers may agree that a patient has a large improvement during treatment but disagree as to how large this improvement is numerically; such discrepancies between readers inflate the SDC and thereby decrease the ability to correctly classify patients with smaller average improvements as responders [25].

As a novel approach, we calculated the net number of patients with improvement as the number of patients with any improvement minus the number of patients with any worsening. The net number of patients with improvement appeared to be the most sensitive method for capturing responders at the group level when change score was dichotomized into responders vs non-responders. This is similar to an approach that has been proposed to determine radiographic structural progression at the group level [24], but it may have other drawbacks and will need to be explored in future research.

It is a limitation that a repeat MRI was only performed in the case of sustained clinical remission, and not at follow-up in all patients. Therefore, we cannot compare the two treatment arms at week 24 to test whether the MRI indices were able to discriminate between golimumab treatment and placebo. Second, without follow-up MRI data for patients who did not achieve sustained clinical remission, we cannot compare change scores in clinical responders with change scores in clinical non-responders. Third, osteitis, synovitis and entheseal soft-tissue inflammation, as detected by MRI, are not specific for SpA, but are seen in a range of other conditions, including degenerative diseases, trauma, and overuse. Even in patients diagnosed with SpA, some of the MRI lesions may be unrelated to SpA. Therefore, these validity aspects could not be assessed and will need further investigation. We note that at least two ongoing trials in PsA (clinicaltrials.gov: NCT03783026 and NCT04108468) will use MRI inflammation in peripheral joints and entheses as an outcome measure. These trials are expected to provide further knowledge on the validity and usefulness of this technique.

Compared with other trials using WB-MRI inflammation as an outcome measure, the CRESPA trial used an intensive MRI protocol with dedicated coils. By excluding some parts of the musculoskeletal system, images of the lower extremities could be obtained in high image quality. We believe that this led to a more accurate evaluation of the joints and entheses of the lower extremities, which are of interest in pSpA, and this is one of the strengths of this study. After finishing the readout, the 10 most discrepant cases were reviewed and discussed by the three readers. In some of the cases, coil artefact was prominent at the anterior aspects of the knees and seemed to contribute to poor agreement. Therefore, we believe that continued attempts at reducing MRI artefacts and improving image quality may further improve reliability and accuracy. Despite these limitations, very good average measure ICCs for status scores and change scores were found.

We found that no patients had osteitis at the entheseal sites related to the anterior and posterior cruciate ligaments of the knees, which are also not included in the OMERACT MRI-WIPE scoring system [20, 21]. We could not discriminate between soft-tissue inflammation related to the cruciate ligaments and synovitis/effusion; therefore, we cannot exclude the possibility that enthesitis at these ligament insertions was a driver for knee inflammation, although we judge it unlikely based on the absence of osteitis.

In conclusion, a semi-quantitative MRI scoring system for inflammation in peripheral joints and entheses demonstrated sensitivity to change, reliability, and validity. Further validation and application in randomized clinical trials of similar semi-quantitative MRI scoring systems will provide further insights into using MRI of peripheral joints and entheses as an objective outcome measure.

Supplementary data

Supplementary data are available at Rheumatology online.

Data availability statement

The data underlying this article may be shared on reasonable request to P.C.

Funding

The CRESPA trial was supported by Janssen Biologics B.V. Janssen Biologics B.V. had no role in the study design or in the collection, analysis, or interpretation of the data, the writing of the manuscript, or the decision to submit the manuscript for publication. Publication of this article was not contingent upon approval by Janssen Biologics B.V.

Disclosure statement: S.K. has received research support from AbbVie, MSD and Novartis. D.E. has received research support and/or consultancy/speaker fees from AbbVie, Bristol-Myers Squibb, Eli Lilly, Galapagos, Janssen, Merck, Novartis, Pfizer and UCB. F.V.d.B. has received research support or worked as consultant for AbbVie, Celgene Corporation, Eli Lilly, Galapagos, Janssen, Novartis, Pfizer and UCB. M.Ø. has received research support and/or consultancy/speaker fees from AbbVie, Bristol-Myers Squibb, Celgene, Eli Lilly, Galapagos, Gilead, Hospira, Janssen, Merck, Novartis, Novo Nordisk, Orion, Pfizer, Regeneron, Roche, Sandoz, Sanofi and UCB. All other authors have declared no conflicts of interest.

Acknowledgements

The study design was by P.C., D.E. and F.V.d.B. Assessment of MRI was undertaken by L.J., S.K. and M.Ø. Analysis and interpretation of data was undertaken by S.K., T.R., P.C., L.J. and M.Ø. All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version.

References

Author notes