Ann Rheum Dis 67:794-800 doi:10.1136/ard.2007.071977
  • Extended report

Magnetic resonance imaging findings in 84 patients with early rheumatoid arthritis: bone marrow oedema predicts erosive progression

  1. E A Haavardsholm1,2,
  2. P Bøyesen1,2,
  3. M Østergaard3,
  4. A Schildvold4,
  5. T K Kvien1,2
  1. 1
    Department of Rheumatology, Diakonhjemmet Hospital, Oslo, Norway
  2. 2
    Faculty of Medicine, University of Oslo, Norway
  3. 3
    Department of Rheumatology, Copenhagen University Hospitals at Hvidovre and Herlev, Copenhagen, Denmark
  4. 4
    Department of Radiology, Diakonhjemmet Hospital, Oslo, Norway
  1. Dr E A Haavardsholm, Department of Rheumatology, Diakonhjemmet Hospital, Box 23 Vinderen, N-0319 Oslo, Norway; e.a.haavardsholm{at}
  • Accepted 22 October 2007
  • Published Online First 2 November 2007


Objectives: To examine the spectrum and severity of magnetic resonance imaging (MRI) findings in patients with early rheumatoid arthritis (RA), and to investigate the predictive value of MRI findings for subsequent development of conventional radiographic (CR) damage and MRI erosions.

Methods: 84 consecutive patients with RA with disease duration <1 year were enrolled. Patients were treated according to standard clinical practice, and evaluated at baseline, 3, 6 and 12 months by core measures of disease activity, conventional radiographs of both hands and wrists and MRI of the dominant wrist. MR images were scored according to the OMERACT rheumatoid arthritis magnetic resonance imaging score (RAMRIS), and conventional radiographs according to the van der Heijde modified Sharp score.

Results: MRI findings reflecting inflammation (synovitis, bone marrow oedema and tenosynovitis) decreased during follow-up, while there was a small increase in MRI erosion score and CR damage. The proportion of patients with erosive progression at 1 year was 48% for conventional radiography and 66% for MRI. Baseline MRI bone marrow oedema (score >2 RAMRIS units) was identified as an independent predictor of both CR (odds ratio = 2.77 (95% confidence interval (CI) 1.06 to 7.21)) and MRI erosive progression (B = 0.21 (95% CI 0.08 to 0.34)).

Conclusions: MRI findings were common in early RA, and MRI bone marrow oedema was an independent predictor of radiographic damage. These results suggest that MRI scans of the dominant wrist may help clinicians to determine which patients need early and aggressive treatment to avoid subsequent joint damage.

Rheumatoid arthritis (RA) is a chronic, inflammatory polyarthritis which frequently leads to progressive joint destruction. Many prognostic markers have been identified for RA, including demographic, genetic, environmental, clinical, immunological and radiographic factors. The most consistent and reliable prognostic factors of radiographic damage are high initial erythrocyte sedimentation rate (ESR) or C-reactive protein (CRP), the presence of rheumatoid factor (RF) or anti-cyclic citrullinated peptide (anti-CCP) at baseline, the presence of shared-epitope alleles and early radiographic erosions.1 However, there is still a need for development of measures that can better distinguish between patients at risk for radiographic progression and those with a more favourable prognosis. Magnetic resonance imaging (MRI) is a highly sensitive method for detecting early inflammatory and structural abnormalities in RA joints.28 Although studies from smaller cohorts of 25–42 patients have shown encouraging data, less is known about the predictive value of MRI synovitis and bone marrow oedema for future erosive progression on conventional radiographic (CR) damage, and more knowledge from larger cohorts is needed in this field, especially in early RA.9 10

The aim of this prospective 1-year follow-up study was to examine the spectrum and severity of MRI findings in patients with RA of <1 year’s duration. We also wanted to examine the association between MRI findings and clinical measures, and investigate the predictive value of early MRI findings for subsequent development of CR damage and MRI erosions.


Patient selection and ethics

Between February 2002 to June 2004 we consecutively enrolled 84 patients with RA with disease duration of <1 year. Disease onset was recorded as the date when the American College of Rheumatology (ACR) 1987 classification criteria were fulfilled.11 Median (interquartile range) age was 58 (47–67) years, disease duration was 107 (77–188) days, symptom duration was 353 (222–482) days and 65 (77.4%) patients were female. The regional ethics committee evaluated the study, the storage of data was licensed from the data inspectorate, approval for the collection of biological material was obtained from the Department of Health, and all enrolled patients gave informed consent.


The patients received treatment according to clinical practice. Disease-modifying antirheumatic drugs (DMARDs) were used by 77.4% of the included patients (57.1% methotrexate monotherapy, 8.3% sulfasalazine monotherapy, 7.1% hydroxychloroquine monotherapy, 3.6% DMARD combination therapy). Anti-tumour necrosis factor α treatment was used by one patient (1.2%), and 60.7% of patients received corticosteroids. At 1 year follow-up DMARDs were used by 91.8% of the patients, anti-tumour necrosis factor α drugs by two patients (2.6%) and corticosteroids by 49.3%.

Study design

The patients were assessed at baseline and after 3, 6 and 12 months. Examinations included a comprehensive clinical assessment, patient-reported health status by questionnaires, laboratory analyses and imaging procedures. The same protocol was employed at every examination, and is described in detail below.

Clinical examination

A trained research nurse performed 28-swollen joint count (28-SJC) and 28-tender joint count (28-TJC). Perceived pain, fatigue, patient and investigators’ global assessment were recorded on 100 mm visual analogue scales. Disability was assessed using the Modified Health Assessment Questionnaire and the hand and finger domain from the Arthritis Impact Measurement Scale 2 (AIMS2).12 13 Grip strength was measured for both hands (kg) on a hand dynamometer. The 28-joint count Disease Activity Score (DAS28) was computed based on ESR.14 Self-reported questionnaires in combination with interview performed by the study nurse were used to obtain information about medication, comorbidity, smoking habits, previous fractures, weight, height, education and civil status.

Laboratory analyses

ESR was measured by an “in-house” Westergren method, with range from 0 mm/1st h to 140 mm/1st h. Serum samples were obtained at all examinations and stored at −70°C for later analyses. Anti-CCP were analysed by a second-generation enzyme-linked immunosorbent assay (ELISA; INOVA Diagnostics Inc, San Diego, California, USA) with a range from 1 U/ml to 251 U/ml and considered positive at a cut-off value of 25 U/ml. IgM RF and IgA RF were measured by an “in-house” ELISA, the main modification being the use of alkaline phosphatase-labelled goat F(ab')2 anti-IgA (Sigma-Aldrich, St Louis, Missouri, USA) and alkaline phosphatase labelled rabbit F(ab')2 anti-IgM (Jackson ImmunoResearch, West Grove, Philadelphia, USA) as secondary antibodies. The range of measurement was 2–300 U/ml, and considered positive at a cut-off value of 25 U/ml. CRP was measured by high sensitivity CRP nephelometry (Dade Behring, Deerfield, Illinois, USA) with lowest detectable limit 0.15 mg/l.

Acquisition and assessment of MRIs

MRI of the dominant wrist was performed using a GE Signa 1.5 Tesla MRI scanner (General Electric (GE) Signa, Milwaukee, WI, USA) with a dedicated high-resolution wrist phased array coil. The same scanner and wrist coil were used for all examinations. The hand was placed in the wrist coil at the patient's side with the coil anchored to the base tray to reduce motion artefacts.

The MRI sequences in this study included the OMERACT recommended MRI core set of sequences plus an additional three-dimensional SPGR sequence for more detailed assessment of cartilage and bony changes.15 Details of the sequences have previously been described elsewhere.16 Experienced technicians reviewed the images immediately after acquisition, and a sequence was reobtained if the quality was not acceptable.

The OMERACT MRI group consensus on MRI definitions and the EULAR OMERACT RA MRI reference image atlas were applied.15,17 Images were scored according to the semiquantitative rheumatoid arthritis magnetic resonance imaging score (RAMRIS),15 as detailed below.

Bone erosions

Each wrist bone (carpal bones, distal radius, distal ulna, metacarpal bases; total 15 sites) was scored separately. The scale was 0–10, based on the proportion of eroded bone compared with the “assessed bone volume”. In addition the number of erosions at each of the 15 sites was noted.

Bone oedema

Bone oedema was scored on a scale of 0–3 by the volume of oedema compared with the “assessed bone volume” (each wrist bone scored separately). It should be emphasised that oedema was scored as a proportion of the estimated original bone volume, not of the remaining bone, when erosion and oedema were concurrently present.


Synovitis in the wrist was assessed in three regions (the distal radioulnar joint; the radiocarpal joint; the intercarpal and carpometacarpophalangeal joints). A score of 0 is normal (no synovitis), while scores of 1 to 3 (mild, moderate, severe) reflect the tertiles of enhancing tissue in the synovial compartment relative to the presumed maximum volume (using the EULAR OMERACT atlas as reference).


Both flexor and extensor tenosynovitis were evaluated at the level of the wrist, according to a new tenosynovitis scoring system.17 Tendon sheath abnormalities were graded semiquantitatively from grade 0 to grade 3, reflecting the maximum width (in mm) of post-contrast enhancement on T1-weighted axial images.

Acquisition and assessment of conventional radiographs

Digital conventional radiographs of the hands in posteroanterior view were scored according to the van der Heijde modified Sharp score (vdHSS) by a trained observer (PB).18 In each hand 16 joint areas were scored for erosions (score range for each area 0–5) and 15 areas were scored for joint space narrowing (score range 0–4 for each joint area), giving a possible maximum score of 280 units.

Image evaluation and assessment of reliability of readers

All MRIs and conventional radiographs were read on large-screen (55 cm (21″)) radiological workstation monitors in known chronological order using a standard PACS software program (SECTRA IDS5, Sweden). This software package provides advanced features of image viewing, allowing the reader to adjust window/level settings; and for MRIs the ability to zoom in/out, to use a localiser that allows the accurate placement of specific lesions in two planes (axial and coronal) and the opportunity to measure distances and areas accurately. The assessors were unaware of the clinical, laboratory and imaging findings.

To assess intrareader reliability of CR scoring, the same observer assessed the conventional radiographs of 10 patients twice at two different time points. The MRI images were read by one reader (EAH) who has a documented high inter-reader agreement with experienced readers.16

Statistical methods

All statistical analyses were undertaken using the Statistical Package for the Social Sciences for Windows, version 14 (SPSS, Chicago, IL, USA). Correlations were assessed by the Spearman test of rank correlation. Group comparisons were performed using Mann–Whitney U tests and within-group changes by the Wilcoxon signed ranks test. The relationships between radiographic progression and sex, baseline synovitis and bone marrow oedema are depicted in cumulative probability plots, where the radiographic change is presented on the y-axis, and the cumulative percentage of patients on the x-axis. These plots provide a visual overview of individual data and univariate associations.

We applied multivariate logistic regression analyses to examine possible predictors of CR progression. Univariate analyses of clinically important covariates were performed, and covariates with a p value <0.25 were included in the multivariate analyses. We were able to build a prediction model with only significant covariates by using backward stepwise elimination of the least significant covariate. Non-linear covariates were dichotomised using the median as the cut-off value. Covariates excluded during modelling were reintroduced in the final model to test for confounding, and the remaining covariates in the final model were tested for possible interactions. Finally, we applied the Hosmer and Lemeshow goodness-of-fit test to assess the overall model.

The same strategy was applied for possible predictors of MRI erosive progression. However, we were not able to build a robust prediction model with significant covariates using multivariate logistic regression modelling (data not shown). The relationships between the baseline MRI findings and subsequent MRI erosions were then explored further using a linear regression model. The correlations between covariates and MRI erosions were measured by R2, non-correlated covariates were excluded from further analysis. R2 was also used to describe the proportion of MRI erosive progression explained by the final multivariate model. For other tests and comparisons p values <0.05 (two tailed) were considered to be significant.



Table 1 summarises the demographic characteristics, immunological status and clinical measures of disease activity at baseline and 12 months. DAS28, ESR and AIMS2 were significantly higher in women than in men at baseline (p<0.05). No other differences in baseline characteristics were observed between men and women. Statistically significant improvements in ESR, SJC, TJC, DAS28 and grip strength of the dominant hand were found after 1 year. At the 1 year follow-up six participants withdrew (one owing to pregnancy, one owing to cancer, four for unknown reasons).

Table 1 Demographic characteristics at baseline and 12 months.

MRI and CR findings

The intraobserver single measure intraclass correlation for the total vdHSS score was 0.83, and for MRI synovitis 0.88, MRI erosion 0.96, MRI bone marrow oedema 0.93 and MRI tenosynovitis 0.94.

MRI findings reflecting inflammation (synovitis, bone marrow oedema and tenosynovitis) decreased during follow-up, while there was a small increase in MRI erosion score and CR damage (table 2). Moderate significant correlations were found between baseline MRI synovitis and CRP(rs = 0.31, p<0.01), 28-SJC (rs = 0.32, p<0.01) and grip strength (rs = −0.25, p<0.05).

Table 2 Imaging findings at baseline, 3, 6 and 12 months in patients with early rheumatoid arthritis

Bone marrow oedema correlated moderately with 28-SJC (rs = 0.23, p<0.05), grip strength (rs = −0.31, p<0.01), MRI synovitis (rs = 0.43, p<0.01) and strongly with the total number of MR erosions at 1 year (rs = 0.63, p<0.01). Modest correlations were found between MRI tenosynovitis and CRP (rs = 0.23, p<0.05), and MRI synovitis (rs = 0.43, p<0.01).

The progression in damage is presented in a cumulative probability plots (fig 1A total vdHSS, Figure 1B RAMRIS erosion score). The mean (95% CI) change in erosion, joint space narrowing and total vdHSS were 0.91 (0.55 to 1.27), 0.42 (0.08 to 0.76) and 1.32 (0.74 to 1.91), respectively. The proportions of patients without/with radiographic progression at 1 year (increase in total vdHSS score ⩾1 unit) were 52%/48%. The mean (95% confidence interval (CI)) change in RAMRIS erosion score was 1.68 (1.16 to 2.21) and the proportions of patients without/with MRI erosive progression at 1 year (increase of ⩾1 RAMRIS erosion unit) were 34%/66%. There was a modest, but significant positive correlation between MRI erosions at all examinations and the total vdHSS at 12-months (rs = 0.26–0.28, p<0.05).

Figure 1 (A) Cumulative probability plot of individual 1-year conventional radiographic (CR) progression in patients with early rheumatoid arthritis (RA). (B) Cumulative probability plot of individual 1-year magnetic resonance imaging erosive progression in patients with early RA. (C) Cumulative probability plot of 1-year CR progression in patients with synovitis score greater than the median (•) and patients with synovitis score less than or equal to the median (o). (D) Cumulative probability plot of 1-year CR progression in patients with bone marrow oedema score greater than the median (•) and patients with bone marrow oedema score less than or equal to the median (o). RAMRIS, rheumatoid arthritis magnetic resonance imaging score; vdHSS, van der Heijde modified Sharp score.

The unadjusted odds ratio (95% CI) for CR progression in patients with RAMRIS synovitis score greater than or equal to the median was 2.7 (1.0 to 7.0) (Pearson χ2 p = 0.04) and for patients with RAMRIS bone marrow oedema greater than or equal to the median 2.5 (1.0–6.1) (Pearson χ2 p = 0.05). These relationships are depicted graphically in cumulative probability plots (figs 1C and 1D).

Prediction of CR progression

Possible predictors of radiographic progression (baseline RAMRIS synovitis, erosion and bone marrow oedema, MRI tenosynovitis, baseline vdHSS, DAS28, anti-CCP status, ESR, 28-SJC, 28-TJC and CRP, as well as sex and age) were examined in a multivariate logistic regression model. Using backward elimination, we found that bone marrow oedema and male sex were independent predictors of radiographic progression (table 3). The final model correctly classified 76% of progressors, and the Hosmer and Lemeshow goodness-of-fit test revealed a p value of 0.692, suggesting that the fit of the model to actual data was quite good. The odds ratio for radiographic progression for a male patient with RA with a bone marrow oedema score >2 compared with a female patient with RA with a bone marrow oedema score ⩽2, is estimated to be 10.8 in this model.

Table 3 Multivariate logistic regression model for radiographic progression

Prediction of MRI erosive progression

Possible predictors of RAMRIS erosive progression were also examined in a logistic regression model, but we were unable to build a robust model with significant covariates (data not shown). A linear regression model was then applied to gain statistical power. Possible predictors for erosive changes were explored, and covariates not influencing the models adjusted R2 were excluded from the model (baseline RAMRIS synovitis, tenosynovitis, 28-SJC, 28-TJC, ESR, CRP, anti-CCP and RF). The covariates in the final model explained 91% of the RAMRIS erosion score at 1 year (R2 = 0.91). Baseline RAMRIS erosion and bone marrow oedema score were associated with subsequent 1-year RAMRIS erosion score corrected for age and sex (table 4).

Table 4 Linear regression model for MRI erosive progression


In this 1-year follow-up study of 84 patients with early RA we have demonstrated that MRI findings are common, and erosions were seen more frequently with MRI than on conventional radiography, even if the latter approach captures bilateral information from multiple joints. We identified bone marrow oedema as an independent predictor of progression of joint damage, evaluated by both MRI and conventional radiography.

A few other early RA cohorts have assessed patients by both MRI and conventional radiography. McQueen et al found that MRI and CR erosions were less frequent than in our study.5 10 19 Lindegaard et al found CR erosions in 24% and MRI erosions in 44% of patients; however, in that study a low-field MRI machine was used.9 We found a higher proportion of patients with erosive disease, and believe there are several reasons for this discrepancy. Although the inclusion criterion was disease duration of <1 year in our study, many of the patients had much longer symptom duration, and the median symptom duration was close to 1 year. Our technical equipment and radiology work stations with advanced imaging software allowed very detailed assessment of findings and recognition of small erosive changes that possibly would have been missed with regular MR-film assessment.

Modern treatment in early RA implies early initiation of aggressive treatment in subjects at risk for developing joint damage, and the treatment should ideally be started before erosive disease is detected.2022 Thus, it is crucial to have access to reliable prognostic markers for an early identification of these individual patients at risk for a severe disease course. Several factors have been identified as predictors of radiographic damage in RA, including clinical (high SJC), laboratory (raised ESR and CRP), immunological (anti-CCP and RF positivity), genetic (presence of shared epitope) and demographic (female sex) factors.1 The presence of CR erosions at baseline is a strong predictor of poor outcome, but is often not apparent at this early stage.1 23 However, even when all this information is taken into account, it is difficult to predict the disease course at the individual level.

The predictive value of early MRI findings for subsequent radiographic progression has previously been examined in a few other cohorts. MRI erosions and bone marrow oedema have been shown to be predictive of later CR progression in cohorts followed for up to 6 years.10 Another study has shown a clear relationship between baseline MRI synovitis and the development of subsequent MRI erosive damage, but this study did not assess the relationship with CR damage.24 In a study of patients with early RA using a low-field dedicated E-MRI unit, Lindegaard et al found that presence of MRI erosions in a radiographically non-eroded bone gave a more than fivefold increased risk of developing a CR erosion a this site within the next year.9

In our study we could not reproduce the predictive value of early MRI erosions for CR progression, but baseline MRI erosion score was an important predictor for MRI erosive progression. Baseline synovitis was a risk factor for subsequent radiographic progression in univariate analyses with an odds ratio of 2.7, and this association is depicted in the cumulative probability plot in fig 1C. However, this association disappeared in the multivariate analyses after adjustment for other possible predictors. This lack of prediction of erosions by synovitis might be explained by the statistical modelling itself since we include synovitis and bone marrow oedema which are intercorrelated (rs = 0.43). However, other MRI studies have also failed to demonstrate an association between synovitis and subsequent erosive progression.10

In multivariate analyses bone marrow oedema was the most important independent predictor for radiographic progression (fig 2). A baseline RAMRIS bone marrow oedema score >2 increased the risk of CR progression threefold, and contributed more than the other MRI findings (synovitis, tenosynovitis, erosions) and conventional measures (28-SJC, 28-TJC, ESR, CRP, DAS-28, anti-CCP) which were eliminated from the model. Bone marrow oedema has long been recognised as being important in RA pathology and has been assumed to represent inflammatory activity within the bone.25 The basic mechanism underlying the association between MRI bone marrow oedema and erosions has recently been discussed in studies published by Jimenez-Boj et al and McQueen et al.26 27 They suggest that MRI bone erosions and MRI bone marrow oedema are due to formation of inflammatory infiltrates in the bone marrow of patients with RA, and thus represent an additional target structure for anti-inflammatory treatment.

Figure 2 Magnetic resonance imaging (MRI) of the wrist, at baseline (A, B) and 1-year follow-up (C). (A) Baseline coronal STIR image showing bone marrow in the capitate (thick arrow) and the lunate and trapezoid (thin arrows). (B) Baseline T1-weighted image without MRI erosion in the capitate (arrow). (C) T1-weighted image at 1-year follow-up, showing erosive progression in the capitate (arrow). Erosive progression was also seen in the trapezoid and the lunate, even though not optimally displayed in the presented slices.

In several studies female sex has been found to be associated with poor outcome, and in a recent study of a population with early RA, female sex was associated with more severe disease and considerably lower remission rates than in men.28 Our finding of male sex as a predictor for radiographic progression is therefore somewhat surprising. There was a clear tendency for more progression in men than in women for both CR progression (65% vs 44%) and MRI erosive progression (82% vs 61%). However, the number of men in our cohort was rather small (n = 19), and the association between male sex and radiographic progression may be a spurious finding, and needs to be explored in larger numbers.

The total number of patients included in this cohort was, albeit at least twice the size of previous early RA MRI cohorts, rather small (n = 84). The comprehensive data collection combined with very complete data (92% completed 1-year follow-up) allowed a thorough examination of possible associations, although the power to perform logistic regression analyses was still rather low. This was a longitudinal observational study, and identification of predictors in such a “real- life” setting may be assumed to have high external validity for clinical practice. However, as a consequence of this study design, the patients were treated with DMARDs according to standard clinical practice. Continuing treatment with DMARDs and corticosteroids may have led to a reduction in the joint damage sustained during the 1-year follow-up. Also, the baseline levels and the relationship between synovitis and bone marrow oedema may have been influenced by these treatments. The combination of these possible interactions would tend to weaken the association between baseline bone marrow oedema and the radiographic damage at 1 year, and thus our analysis represents a conservative approach.

In this study of early RA we found that MRI findings were common and associated with erosive progression. In particular, we confirmed the role of MRI bone marrow oedema as an independent predictor of radiographic progression. These findings, taken together with results from previous smaller studies, suggest that MRI scans of the dominant wrist may assist the clinicians in their considerations of which patients need early and aggressive treatment to avoid subsequent joint damage. MRI bone marrow oedema should be recognised as an important prognostic factor in early RA.


We thank research nurse Margareth Sveinsson for collecting clinical data, research coordinator Tone Omreng for organising the data collection, Halvor Gilboe for practical assistance with case record forms, Marianne Ytrelid for technical assistance, Dr Inge Olsen for statistical advice and Drs Annelies Boonen and Desirée van der Heijde for advice and assistance with radiographic scoring methodology.


  • Funding: This study was supported in part by grants from the Research Council of Norway, the Norwegian Rheumatism Association, the Norwegian Women Public Health Association, Grethe Harbitz Legacy and Marie and Else Mustad's Legacy.

  • Competing interests: None declared.

  • Ethics approval: The regional ethics committee evaluated the study, the storage of data was licensed from the data inspectorate, approval for the collection of biological material was obtained from the Department of Health.


Free sample This recent issue is free to all users to allow everyone the opportunity to see the full scope and typical content of ARD.
View free sample issue >>

Don't forget to sign up for content alerts so you keep up to date with all the articles as they are published.