Bone edema scored on magnetic resonance imaging scans of the dominant carpus at presentation predicts radiographic joint damage of the hands and feet six years later in patients with rheumatoid arthritis




Magnetic resonance imaging (MRI) is capable of revealing synovitis and tendinitis in early rheumatoid arthritis (RA), as well as bone edema and erosion. These features are visible before radiographic joint damage occurs. We sought to examine whether MRI of one body region (the wrist) can be used to predict whole-body radiography scores reflecting joint damage at 6 years.


We conducted a 6-year prospective study of a cohort of patients who fulfilled the criteria for RA at presentation, using clinical parameters, radiographs, and MRI scans of the dominant wrist. Of the 42 patients enrolled at baseline, full MRI, radiographic, and clinical data were available for 31 at 6-year followup. MRI scans were scored by 2 radiologists, using a validated scoring system. Radiographs of the hands and feet were graded using the modified Sharp scoring method. MRI and radiography scores obtained at baseline and 6 years were compared, and baseline MRI scores were examined for their ability to predict radiographic outcome at 6 years.


At 6 years, the total Sharp score correlated significantly with the total MRI score and the MRI erosion score (r = 0.81, P < 0.0001 and r = 0.79, P < 0.0001, respectively). The 6-year Sharp score also correlated with the baseline total MRI and MRI erosion scores (r = 0.56, P < 0.0001 and r = 0.33, P = 0.03, respectively). MRI synovitis and bone edema scores remained constant for the group as a whole over 6 years, but bone erosion scores progressed (P = 0.0001), consistent with radiographic deterioration. Erosions on 6-year MRI scans were frequently preceded by MRI bone edema at baseline (odds ratio 6.5, 95% confidence interval 2.78–18.1). Regression models indicated that the baseline MRI bone edema score was predictive of the 6-year total Sharp score (P = 0.01), as was the C-reactive protein (CRP) level (P = 0.0002). Neither shared epitope status nor swollen or tender joint counts predicted radiographic outcome in this cohort. A model incorporating baseline MRI scores for erosion, bone edema, synovitis, and tendinitis plus the CRP level and the erythrocyte sedimentation rate explained 59% of the variance in the 6-year total Sharp score (R2 = 0.59, adjusted R2 = 0.44).


MRI scans performed at the first presentation of RA can be used to help predict future radiographic damage, allowing disease-modifying therapy to be targeted to patients with aggressive disease.

Rheumatoid arthritis (RA) is the most common form of inflammatory arthritis, affecting ∼1% of the population (1). It is a heterogeneous disorder in terms of both clinical presentation and outcome. In some patients a mild, nonerosive form of RA associated with little disability develops. Other patients have persistent and aggressive disease that produces severe articular damage after only a few years (2), often requiring joint replacement surgery. The ability to predict aggressive disease when a patient first presents is an important clinical goal, because this would allow potent and potentially toxic disease-suppressing medication to be targeted to patients who are most in need. This predictive ability has become even more desirable from a medico-economic viewpoint with the advent of anti–tumor necrosis factor α therapies, which have powerful antierosive effects (3) but are costly, are sometimes associated with adverse effects, and should not be prescribed for patients in whom the potential to develop erosions is absent.

Clinical and demographic factors that have been identified as predictors of poor outcome include female sex, rheumatoid factor (RF) positivity, evidence of active inflammation (including high tender and swollen joint counts, an elevated erythrocyte sedimentation rate [ESR] and C-reactive protein [CRP] level [4, 5]), and the presence of HLA–DRB1*04/01 alleles bearing the shared epitope (6). Unfortunately, despite the development of multifactorial predictive models using these parameters (7), it is still difficult to identify individual patients in whom severe disease will develop, because the majority of patients presenting with early RA will be female, RF-positive (within the first year), and will have active joint inflammation. Early detection of radiographic erosions is powerfully associated with poor outcome (4), but in many patients radiographic erosions are not apparent until at least 12 months after symptom onset.

We and other investigators have studied MRI as a potential tool for prognostication in RA (8–13). For the past 6 years, we prospectively studied a cohort of patients with early RA, monitoring their clinical and radiographic progression as well as the MRI appearance of their dominant wrists. Baseline MRI scores for synovitis, tendinitis, bone edema, and bone erosion, when combined as a total MRI score, were found to predict MRI and radiographic erosion scores at 1 and 2 years (9, 11). However, these outcome erosion scores were confined to the area of the MRI examination (the dominant wrist) and did not reflect the degree of whole-body erosion. We now report results of our data analysis at 6 years, investigating whether MRI of this single area at presentation can be a predictor of long-term overall radiographic joint damage, as measured by the modified Sharp score (14, 15).


Patient population and clinical assessments.

An inception cohort of 42 patients with early RA has been studied since symptom onset. Details of recruitment, baseline demographics, and clinical assessments have been previously described (8, 9). Briefly, all patients fulfilled the 1987 American College of Rheumatology (formerly, the American Rheumatism Association) criteria for RA (16) and had symptoms for 6 months or less (median 4 months) at the time of entry into the study. All patients were assessed clinically for disease activity, using the tender joint count (the temperomandibular, acromioclavicular, and sternoclavicular joints, shoulders, elbows, wrists, metacarpophalangeal [MCP] and proximal interphalangeal joints, hips, knees, ankles, and the midtarsal, metatarsophalangeal [MTP], and proximal interphalangeal joints of the foot; maximum possible number of joints = 60), the swollen joint count (maximum possible number of joints = 58 [hips excluded]), the Ritchie Articular Index (17), the pain score (as measured on a 10-cm visual analog scale), the Health Assessment Questionnaire score (18), the ESR and CRP levels, and the disease activity (DAS) scores at baseline and at 1, 2, and 6 years. The DAS was calculated using 3 variables, according to the method described by van der Heijde et al (19). Radiographs of the hands and feet were obtained at baseline and at 1, 2, and 6 years. MRI scans of the dominant wrist were obtained at baseline and at 1 year and 6 years. Demographic and current medication details for the 6-year cohort are summarized in Table 1. Functional outcome data will be described in a future report.

Table 1. Patient medications and disease activity*
 Baseline (n = 42)1 year (n = 42)2 years (n = 40)6 years (n = 34)
  • *

    NSAID = nonsteroidal antiinflammatory drug; DMARD = disease-modifying antirheumatic drug; HAQ = Health Assessment Questionnaire; DAS = Disease Activity Score; ESR = erythrocyte sedimentation rate; CRP = C-reactive protein.

 NSAID, no. (%)38 (90)25 (60)24 (60)20 (59)
 DMARD, no. (%)21 (50)30 (71)29 (73)21 (62)
  Sulfasalazine, 1–3 gm/day1816139
  Methotrexate, 7.5–17.5 mg/week1101314
  Hydroxychloroquine, 400 mg/day2372
  Penicillamine, 375 mg/day110
  Azathioprine, 100 mg/day11
  Leflunomide, 20 mg/day2
  Combination DMARD6
 Prednisone, 1–10 mg/day, no. (%)5 (12)10 (24)8 (20)9 (26)
Disease activity, median (range)    
 Ritchie index13.5 (0–37)5.5 (0–25)6 (0–26)5 (0–15)
 Swollen joint count14.5 (0–38)2 (0.21)2 (0–18)3.5 (0–21)
 Pain score3.3 (0.7–9.3)2.3 (0–7.4)2.2 (0–6.5)2.1 (0–8.6)
 HAQ score0.6 (0–1.8)0.1 (2–4)0.2 (0–1.5)0.3 (0–1.6)
 DAS4.2 (1.3–7.6)1.3 (0.3–2.8)2.4 (0.9–4.4)2.7 (0.9–4.6)
 ESR, mm/hour29 (10–131)16 (1–113)19 (3–113)26 (5–129)
 CRP, mg/liter18 (0–150)5 (0–147)9 (0–86)7 (3–69)

Of the original 42 patients, 34 participated in the followup study at 6 years. Among the 8 nonparticipants, 4 declined to be involved or withdrew prior to their clinical assessment, and 4 either had left the country or were untraceable. In 3 patients, MRI scanning of the dominant wrist was not performed at 6 years, because they had undergone orthopedic surgery involving placement of pins across the carpus. Thus, at 6 years a total of 34 patients underwent clinical and radiographic assessments, and 31 patients underwent clinical review, radiography, and MRI scanning (Figure 1).

Figure 1.

Details of patient followup over 6 years. XR = x-ray (radiography); MRI = magnetic resonance imaging. ∗ = C-reactive protein values were available for 41 patients at baseline and for 29 patients at 6 years.

MRI scans.

An MRI scan of the dominant wrist was obtained using the GE Signa Horizon 1.5T MRI scanner (General Electric Signa, Milwaukee, WI) with a dedicated wrist coil (Medical Devices, Waukesha, WI), which was the same scanner used at baseline and at 1 year for this group of patients. The hand was placed in the wrist coil, where it fitted snugly by the patient's side with the palm facing the body and the thumb in the anterior position. Each sequence of the 6-year followup study was planned using localizing sequences to match the first study sequence as closely as possible. All parameters for both studies were identical. The 8-cm field of view included the distal radioulnar, radiocarpal, and midcarpal joints as well as the metacarpal bases. The small field of view was chosen to optimize resolution and did not include the MCP joints. Coronal and axial T1-weighted sequences were performed, followed by axial T2-weighted fat-suppressed fast spin-echo sequences. Then coronal T1-weighted fat-suppressed sequences were performed after injection of gadolinium diethylenetriaminepentaacetic acid (Gd-DTPA) (Omniscan; Amersham Health, Lindesnes, Norway), which acts as a contrast agent. An axial fat-suppressed T1-weighted post–Gd-DTPA sequence was also included in the 6-year study.

The system used to score MRI scans has been described previously (8). Briefly, erosions were defined as focal areas of loss of low signal cortex with sharply defined margins, identified on both T1- and T2-weighted sequences. The cortex was replaced by well-circumscribed intermediate signal tissue on T1-weighted images, which was intermediate to bright on T2-weighted images and was enhanced with Gd-DTPA. Erosions were scored only if they were visible in 2 planes, with a cortical break seen in at least 1 plane. Erosions were differentiated from intraosseous cysts, which appeared as well-circumscribed, rounded lesions within bone without any associated cortical break, as described previously (20). MRI erosions were scored at 15 sites within the carpus. A total MRI score for the carpus was derived from the sum of scores for erosions, bone marrow edema, synovitis, and tendinitis (8). MRI scans were scored independently by the same 2 radiologists with musculoskeletal expertise (NS and JC) who scored the baseline and 1-year scans. The scorers were blinded to clinical, genetic, and radiographic data. Scans at 6 years were scored without reference to the baseline or 1-year scans.

Bone marrow edema was identified as a poorly defined area of low signal within bone on T1-weighted images, which had high signal on T2-weighted fat-suppressed images. Edema was scored, at the same sites as erosions, as follows: 0 = no edema, 1 = minor edema involving ≤50% of the bone, and 2 = gross edema involving >50% of the bone marrow. The total bone marrow edema score was derived from the sum of scores at all 15 sites (maximum possible score = 30).

When scoring 6-year MRI scans, in some instances the erosive damage was so severe that individual erosions could not be counted or scored. In these situations, the carpal site in question was allocated an arbitrary high score of 6 (equivalent to 6 small erosions or 3 large erosions). When a site could not be scored for bone edema, synovitis, or tendinitis because of surgical alteration of anatomy, that site was allocated the same score as the median for the rest of the carpal sites scored (15 sites scored for bone edema, 7 for synovitis, and 9 for tendinitis) (8).

Dynamic MRI studies.

Details of the dynamic MRI scans performed at baseline in this cohort have been described previously (10). Briefly, dynamic scans used a coronal fat-suppressed fast multiplanar spoiled GRASS technique. Scans were localized using an initial axial spin-echo sequence. Slices were centered on the mid lunate. The following imaging parameters were used: repetition time 150 msec, time to echo 9.1 msec, flip angle 60°, matrix 256 × 256 pixels, and slice thickness 3 mm, with a 2-mm gap between slices. Each sequence consisted of 6 slices and was obtained in 42 seconds. A precontrast scan was performed, then 10 ml of Omniscan gadodiamide (0.5 mmol/ml) (Amersham Health) was injected intravenously through a 23-gauge cannula into the opposite arm over a period of 20–30 seconds, with a subsequent flush of 10 ml of normal saline. Nine sequences were then obtained (total imaging time 6 minutes, 30 seconds). The last sequence was examined, and the slice with the greatest degree of synovial enhancement (usually from the midcoronal position) was chosen for measurements of pixel intensity.

Analysis of data was performed on an Advantage Windows 2.1 work station (General Electric Medical Systems, Milwaukee, WI). A region of interest circle (8–12 mm2) was placed over the region of maximal synovial enhancement identified by a radiologist (NS). Following injection of Gd-DTPA, a curve was obtained, plotting pixel intensity against time. The rate of increase in pixel intensity over the steep linear region of the curve (E-rate) was calculated as follows: E-rate = SIt − SI0/▵t, where SI0 was the signal intensity at the takeoff point and SIt was the signal intensity reached after time t. E-rate data were used as a measure of the intensity of synovitis.

Scoring radiographs.

Plain radiographs of the hands and feet (anteroposterior views) were obtained at baseline and at 1, 2, and 6 years. Films obtained at baseline and 6 years were scored separately by 2 observers (NB and DP) using the modified Sharp/van der Heijde method (14), and the mean score of the 2 readers for each patient was used in data analysis. Observers were blinded to clinical and MRI data. They did not have other radiographs of the sequence to use as references when scoring films but were aware of whether films were obtained at baseline or 6 years.

Genetic studies.

The methodology used for HLA–DRB1 typing has been previously described (8). Briefly, DNA was extracted from anticoagulated blood that was obtained from each patient at the time of recruitment into the study. Low resolution typing was performed using sequence-specific oligonucleotide polymerase chain reaction (PCR) with a standardized panel of 24 oligonucleotide primer pairs. In subjects with alleles of the DRB1*01, 04, 10, or 14 groups, the sequence of the subtype-determining region of exon 2 of the DRB1 gene(s) was obtained by direct sequencing of PCR products.

Statistical analysis.

Intraclass correlation coefficients (ICCs) were calculated to investigate the interobserver reliability of MRI and radiography scores (21). Each person was rated once by each observer, and the observers were considered to be a random sample of all possible observers. Pearson's correlations were used to investigate the relationships between MRI and clinical scores at baseline and at 6 years. Changes over time in scores for edema, erosion, and synovitis were investigated with a repeated-measures analysis that used the generalized estimating equations (GEE) approach (22) for correlated data. The scores obtained by the 2 observers were averaged and analyzed as ordinal variables, using a multinomial model (23).

The relationship between the presence of MRI bone edema at baseline (as scored by one or both observers) and the presence of MRI erosion at the same site at 6 years (as scored by one or both observers) was investigated initially by combining data from all patients over all the sites and performing a chi-square test. Only those baseline sites with edema but no erosions were chosen. Then, selecting the 10 most common sites of bone edema at baseline, a logistic regression analysis (with site as a repeated measure) was used to determine whether the presence of MRI bone edema at baseline at a particular site was predictive of MRI erosion at the same site 6 years later. The GEE approach (22) was used to account for any possible correlation between measurements at different sites within the same individual.

The same procedures were used to investigate whether synovitis at baseline was associated with the development of erosions at adjacent carpal bones at 6 years. The analysis included only the 10 most common bony sites adjacent to regions of synovitis at baseline (8), and only those sites that were not involved with erosion at baseline.

Linear regressions were used to investigate the ability of baseline MRI and clinical and radiography scores to predict 6-year radiography scores. A square root transformation was used on the 6-year total Sharp score, because the data were positively skewed. Six-year Sharp score data for the dominant wrist were grouped into quartiles (0, 0–2, 2–15, and >15) because of the large number of patients with zero scores, and ordinal logistic regression was used. Initially, regressions were performed using each baseline measure singly as a predictor of 6-year radiography measures. Then the best subset of all the baseline measures was found using the R2 value in linear regression and the likelihood score statistic in ordinal logistic regression.


Interobserver reliability for radiography and MRI scoring at 6 years.

Interobserver reliability was high for scoring radiographs at 6 years. The ICC for the dominant wrist Sharp score was 0.92 (95% CI 0.85–0.95) and for the total modified Sharp score was 0.97 (95% CI 0.94–0.98) (Table 2). Reliability of radiographic scoring was significantly less at baseline (ICC 0.73 [95% CI 0.50–0.86] for total Sharp score), reflecting difficulty in defining early erosive change on plain radiographs. Scoring was also consistent between readers of MRI scans at 6 years, with ICCs of 0.84 (95% CI 0.69–0.92) for MRI erosion score and ICCs of 0.87, 0.86, and 0.85, respectively, for synovitis, bone edema, and tendinitis scores. Baseline and 1-year reliability data for scoring MRI scans have been previously published (8) but are included for comparison in Table 2.

Table 2. Reliability of MRI scoring system at baseline, 1 year, and 6 years and of total modified Sharp score at baseline and 6 years*
 Baseline1 year6 years
  • *

    Values are the intraclass correlation coefficients (95% confidence intervals). MRI = magnetic resonance imaging.

MRI score   
 Interobserver reliability   
  Synovitis0.74 (0.56–0.85)0.90 (0.82–0.95)0.87 (0.75–0.93)
  Tendinitis0.77 (0.60–0.87)0.73 (0.54–0.84)0.85 (0.71–0.92)
  Bone edema0.84 (0.72–0.91)0.83 (0.71–0.91)0.86 (0.73–0.93)
  Erosion0.77 (0.60–0.87)0.75 (0.53–0.87)0.84 (0.69–0.92)
  Total0.81 (0.64–0.89)0.87 (0.77–0.93)0.90 (0.82–0.95)
 Intraobserver reliability   
  Observer 1   
   Total0.94 (0.66–0.99)
   Erosion0.98 (0.93–0.99)
  Observer 2   
   Total0.81 (0.05–0.95)
   Erosion0.92 (0.48–0.98)
Radiography score   
 Interobserver reliability   
  Total Sharp score0.73 (0.50–0.86)0.97 (0.94–0.98)

Correlations between MRI and radiography scores.

Of the 34 patients who underwent radiography at 6 years, 33 had baseline films for comparison (baseline radiographs were missing for 1 patient). Nineteen of these 33 patients had a total modified Sharp score >0 (as confirmed by both observers) at baseline, and 15 of these patients had erosions. However, baseline average Sharp scores were low (median 4, range 0–22) compared with 6-year average Sharp scores (median 25, range 1–178) (Figure 2A).

Figure 2.

A, Total modified Sharp scores (average of 2 observers) in 33 patients at baseline and 34 patients at 6 years. B, Correlation between total magnetic resonance imaging (MRI) score and total modified Sharp score (r = 0.81, P < 0.0001). C, Correlation between MRI erosion score and total modified Sharp score (r = 0.79, P < 0.0001).

The total modified Sharp score correlated strongly with the total MRI score (Figure 2B) and the MRI erosion score (Figure 2C) at 6 years (r = 0.81, P < 0.0001 and r = 0.79, P < 0.0001, respectively), as did the dominant wrist Sharp score (r = 0.84, P < 0.0001 and r = 0.83, P < 0.0001, respectively). Correlations between radiography and MRI scores were weaker at baseline, when the total Sharp score correlated better with the total MRI score (r = 0.56, P < 0.0001) than with the MRI erosion score (r = 0.33, P = 0.03).

Correlations between MRI and clinical scores.

Correlations between MRI scores and clinical measures of disease activity were examined at baseline and at 6 years (Table 3). At baseline, all MRI scores except those for tendinitis correlated with the CRP level. The strongest correlations were between the MRI synovitis score and the CRP level (r = 0.49, P = 0.001) and the total MRI score and the CRP level (r = 0.50, P = 0.001). Interestingly, the DAS was most strongly correlated at baseline with the MRI bone edema and MRI bone erosion scores (r = 0.44, P = 0.004 and r = 0.45, P = 0.003, respectively).

Table 3. Correlations between clinical and MRI parameters at baseline and 6 years*
Clinical scoresMRI score
SynovitisTendinitisBone edemaBone erosionTotal
  • *

    Values are the correlation coefficients (P values). MRI = magnetic resonance imaging; ESR = erythrocyte sedimentation rate; CRP = C-reactive protein; DAS = Disease Activity Score.

Tender joint count     
 Baseline−0.08 (0.6)0.10 (0.5)0.23 (0.1)0.14 (0.4)0.10 (0.5)
 6-year0.15 (0.4)0.35 (0.05)0.13 (0.5)0.32 (0.08)0.33 (0.08)
Swollen joint count     
 Baseline0.03 (0.8)0.43 (0.005)0.10 (0.5)0.24 (0.1)0.23 (0.1)
 6-year0.43 (0.02)0.45 (0.01)0.48 (0.007)0.39 (0.04)0.42 (0.02)
Pain score     
 Baseline0.32 (0.04)0.10 (0.5)0.42 (0.006)0.42 (0.006)0.40 (0.009)
 6-year0.13 (0.5)0.32 (0.08)0.25 (0.2)0.31 (0.09)0.25 (0.2)
 Baseline0.30 (0.05)0.28 (0.07)0.42 (0.006)0.47 (0.002)0.42 (0.005)
 6-year0.19 (0.3)0.38 (0.04)0.26 (0.02)0.53 (0.002)0.53 (0.002)
CRP level     
 Baseline0.49 (0.001)0.23 (0.2)0.35 (0.03)0.48 (0.002)0.50 (0.001)
 6-year0.49 (0.007)0.13 (0.5)0.59 (0.0007)0.59 (0.0007)0.59 (0.0007)
 Baseline0.12 (0.4)0.38 (0.01)0.44 (0.004)0.45 (0.003)0.36 (0.02)
 6-year0.39 (0.03)0.46 (0.009)0.38 (0.04)0.53 (0.002)0.56 (0.001)

At 6 years, correlations between MRI scores and disease activity measures were slightly stronger, particularly between the CRP level and the total MRI score (r = 0.59, P = 0.0007) as well as its component MRI scores (once more excluding the MRI tendinitis score). The 6-year DAS correlated most strongly with the 6-year total MRI score (r = 0.56, P = 0.001) and the 6-year MRI erosion score (r = 0.53, P = 0.002).

Progression of MRI erosions over 6 years.

Median scores for MRI erosions, bone edema, synovitis, and tendinitis were examined at baseline and at 1 and 6 years (Figure 3). Average levels between the 2 observers for bone edema, synovitis, and tendinitis did not change over time (P = 0.1, 0.6, and 0.1, respectively), but the MRI bone erosion score progressed, with the median score increasing from 0.5 at baseline to 2.0 at 1 year to 12.0 at 6 years (P = 0.0001). This compares with progression of the median total Sharp score from 3.5 at baseline to 24.5 at 6 years.

Figure 3.

Box plots showing range of magnetic resonance imaging (MRI) data (whiskers) plus upper and lower quartiles (boxes) and medians (lines within boxes) for observers 1 and 2 at baseline, 1 year, and 6 years. A, MRI synovitis scores did not change over the 6-year period (P = 0.6). B, MRI bone edema scores did not change over the 6-year period (P = 0.1). C, MRI erosion scores progressed from baseline to 1 year to 6 years (P = 0.0001).

Role of bone edema as a preerosive lesion.

Site-specific analysis examining bone edema and erosion was performed at the 15 sites of the carpus scored on MRI (11). At baseline, 41 sites had edema but no erosion. Of these sites, 34 developed erosions by 6 years, while 7 did not. In comparison, of 366 sites without edema or erosion at baseline, 155 were involved with erosion at 6 years, while 211 were not. Thus, erosion was more likely to be detected at 6 years if bone edema had been scored at baseline (OR 6.5, 95% CI 2.78–18.1). A more stringent repeated-measures analysis (22), which allowed for correlation between sites in the same person, still found baseline bone edema to be significantly predictive of erosion at 6 years (P = 0.02). Examples of baseline bone edema proceeding to erosion at 6 years are shown in Figure 4.

Figure 4.

Axial T1-weighted magnetic resonance images of the carpus showing 2 examples of baseline bone edema (A1 and B1) progressing to erosion at 6 years (A2 and B2).

Role of synovitis as a preerosive lesion.

An analysis similar to that used to examine bone edema and erosion was used to examine the role of baseline synovitis as a predictor for erosion at an adjacent carpal bone at 6 years. When sites were treated independently, no association between baseline synovitis and 6-year erosion was demonstrated, and this was confirmed on the repeated-measures analysis (P = 0.5).

Baseline MRI bone edema predicts 6-year radiography scores.

When best subsets regression models were used to determine which baseline MRI parameters were individually predictive of 6-year radiographic measures, baseline MRI bone edema was found to be predictive of the total Sharp score (R2 = 0.20, P = 0.01) and the dominant wrist Sharp score (P = 0.05) (Table 4). Interestingly, MRI bone edema at baseline predicted the joint space narrowing component of the total modified Sharp score at 6 years (P = 0.02) as well as the erosion score component (P = 0.003). The combined baseline MRI edema plus MRI erosion score was also predictive of the total Sharp score at 6 years (R2 = 0.17, P = 0.01), but the baseline MRI erosion score alone was not (R2 = 0.06, P = 0.1).

Table 4. Baseline predictors of radiographic damage at 6 years*
 Total modified Sharp score at 6 yearsDominant wrist Sharp score at 6 years
F1,32 testPR2χ2 (ldf)PR2
  • *

    A square root transformation was used on the Sharp score at 6 years. Six-year Sharp score data for the dominant wrist were grouped into the following quartiles: 0, 0–2, 2–15, and >15. MRI = magnetic resonance imaging.

  • Value was determined using regression models with only 1 independent variable.

  • Value was determined using generalized coefficient of determination.

MRI features at baseline      
 Synovitis score1.350.
 E-rate (dynamic enhanced scans)0.940.30.030.310.60.01
 Erosion score2.
 Bone edema score2.570.010.203.700.050.12
 Bone edema + erosion score2.550.010.174.640.030.15
 Tendinitis score0.010.90.00050.010.90.0005
 Total MRI score3.920.060.122.910.090.10
Clinical feature at baseline      
 Swollen joint count0.0050.90.00020.040.90.0002
 Tender joint count0.070.80.0030.050.80.002
 Ritchie index2.560.10.073.560.060.12
 Pain score4.500.
 C-reactive protein level17.330.00020.366.210.010.23
 Erythrocyte sedimentation rate7.760.0090.193.680.050.13
 Disease Activity Score1.910.20.060.770.40.03
 Rheumatoid factor0.270.60.0091.040.30.03
Genetic measures      
 Shared epitope–positive0.720.
Radiographic measure at baseline      
 Total modified Sharp score4.860.040.135.540.020.20

Baseline MRI scores in patients requiring orthopedic intervention by 6 years.

At 6 years, 6 patients had undergone orthopedic surgery for management of rheumatoid joint damage, which involved the dominant carpus in 3 patients. All patients had MRI erosions at baseline, and many had high scores for synovitis and bone edema. Table 5 details their clinical and MRI features at baseline and Sharp scores at 6 years. Figure 5 shows an example of florid bone edema on baseline carpal MRI followed by the development of severe erosive radiographic changes at 6 years in one of these patients (patient 33).

Table 5. Baseline clinical and MRI features and 6-year Sharp scores in patients requiring orthopedic surgery by 6 years*
Patient no./age at entry/sexClinical feature at baselineMRI scores at baselineRadiography (Sharp) score at 6 yearsOrthopedic intervention by 6 years
Swollen/tender joint countCRP, mg/literSynovitisTendinitisBone edemaErosionTotal
  • *

    Median (range) MRI scores at baseline are as follows: for synovitis, 8 (0–16); for tendinitis, 1 (0–10); for bone edema, 1 (0–11); for erosion, 1 (0–15); for total, 10 (0–35). The median (range) radiography score at 6 years is 25 (1–178). MRI = magnetic resonance imaging; CRP = C-reactive protein (normal level <1 mg/liter); THR = total hip replacement.

14/40/F18/261983221414Yttrium synovectomy (knee)
22/49/F14/16396101863Fusion (wrist)
30/42/F38/4423738321132Bilateral THR, fusion (wrist) + synovectomy, tendon rupture
33/54/M12/25601519934113Fusion (wrist), synovectomy (wrist)
40/63/F22/18661332926118Synovectomy (wrist)
Figure 5.

Magnetic resonance images (MRIs) and radiographs from patient 33, a 54-year-old man with aggressive arthritis (currently managed with methotrexate, hydroxychloroquine, cyclosporine, and low-dose prednisone). A, T1-weighted coronal MRI obtained at baseline showing extensive bone marrow edema within the lunate, triquetrum, and hamate (white arrows) with synovial thickening indicating florid synovitis at the radiocarpal joint (black arrow). B, T1-weighted coronal post–gadolinium diethylenetriaminepentaacetic acid image showing increased intensity (post-contrast) in areas of bone marrow edema and synovitis as above. C, Plain radiograph of the hands and wrists at baseline, showing minor narrowing and sclerosis of the radiocarpal joint but no erosions. D, Plain radiograph of the hands and wrists at 6-year followup, showing that extensive destructive erosive changes have occurred at the carpus bilaterally. Resection of the ulnar head has been performed bilaterally, and the carpus in the right hand has been fused.

Baseline MRI synovitis not predictive of 6-year radiography scores.

Data describing MRI synovitis were available from 2 sources. First, baseline MRI scans had been scored for synovitis by grading synovial thickening and the degree of enhancement following intravenous (IV) administration of contrast (8). This baseline “static” measure of synovitis was not a predictor of the 6-year total Sharp score (R2 = 0.05, P = 0.2). Data from dynamic enhanced MRI in these patients were also available and used the E-rate parameter (initial rate of increase of enhancement following IV injection of Gd-DTPA) as another way to measure synovitis on baseline MRI scans (10). This “dynamic” measure of synovitis also was not a predictor of 6-year radiography scores (R2 = 0.03, P = 0.3).

Clinical predictors of 6-year radiography scores.

The strongest individual clinical predictors of the 6-year total Sharp score were the baseline CRP level (R2 = 0.36, P = 0.0002) and the baseline ESR (R2 = 0.19, P = 0.009) (Table 4). However, baseline clinical scores describing peripheral synovitis (tender and swollen joint score) were not predictive. The baseline total Sharp score was a weak predictor of the 6-year score (R2 = 0.13, P = 0.04), but the presence of rheumatoid factor was not predictive in this cohort, in which 90% of patients were RF-positive.

Combined clinical and MRI scores predictive of 6-year radiography scores.

The best subsets selection procedure was used to derive a model combining baseline variables for prediction of the 6-year total Sharp score. A combination of baseline MRI scores for erosion, bone edema, synovitis, and tendinitis as well as the baseline ESR, CRP level, and Sharp score was found to explain 59% of the variance of the total Sharp score at 6 years (R2 = 0.59, adjusted R2 = 0.44). This compares with a model combining ESR, CRP level, and Sharp score at baseline, which explained 45% of the variance of the 6-year Sharp score (R2 = 0.45, adjusted R2 = 0.39).

Review of patients in clinical remission at 6 years.

In 3 patients from this cohort, complete clinical remission (no tender or swollen joints, no use of medication) was apparent at 6 years (Table 6). Two of these patients had previously been reclassified as having transient inflammatory polyarthritis, but in both patients the CRP level remained slightly elevated at 6 years. All 3 of these patients were RF-negative at 6 years (although one of them had been RF-positive at baseline), and all of them had no erosions on radiographs of the hands and feet at 6 years. MRI features at baseline and 6 years for these patients are shown in Table 6. In 2 patients, the MRI synovitis score was above the median level at baseline but was low at 6 years. In one patient (patient 24), MRI erosions were scored on baseline scans, but, unfortunately, this patient was unable to have an MRI scan at 6 years because a pacemaker was in situ. Radiographs from this patient at 6 years revealed lucencies at the metatarsal heads and joint space narrowing at the second MCP joint, but no carpal erosions.

Table 6. Clinical, radiographic, and MRI characteristics of patients in clinical remission (not taking medication) at 6 years*
Patient no./age at entry/sexACR criteria fulfilled at entryClinical featureRadiography featureMRI score (baseline)
Swollen/tender joint countCRP, mg/literSynovitisTendinitisBone edemaErosionsTotal
  • *

    See ref. 16 for an explanation of the American College of Rheumatology (ACR) criteria. Median (range) magnetic resonance imaging (MRI) scores at 6 years and baseline, respectively, are as follows: for synovitis, 7 (0–18) and 8 (0–16); for tendinitis, 2 (0–16) and 1 (0–10); for bone edema, 1 (0–15) and 1 (0–11); for erosions, 12 (0–60) and 1 (0–15); for total, 24 (0–76) and 10 (0–35). CRP = C-reactive protein; ND = not done.

  • No MRI scans were performed at 6 years because of a pacemaker in situ.

  • Lucencies at the metatarsal heads and joint space narrowing at the second metacarpophalangeal joint of the right hand.

19/45/M1,2,3,40/02No erosions2 (9)0 (1)1 (0)0 (0)3 (9)
24/74/M1,2,3,40/0<1No erosionsND (12)ND (3)ND (5)ND (5)ND (24)
28/31/F1,3,4,60/07No erosions0 (3)2 (0)0 (0)0 (0)2 (3)

Influence of the shared epitope on MRI and radiography scores.

Shared epitope status in this cohort was determined at baseline (8). The presence of the shared epitope did not correlate with radiographic or MRI erosion scores at baseline or at 6 years and was not predictive of either the total modified Sharp score (P = 0.4) or the dominant wrist Sharp score (P = 0.7) at 6 years (Table 4).


This study is the first to show that MRI scanning is useful in predicting whole-body radiographic outcome as measured by the modified Sharp/van der Heijde score in patients with RA. Both the MRI score for bone edema and the sum score for MRI bone edema plus MRI erosion, which were derived from scans of the dominant wrist obtained when patients first presented with symptoms of arthritis, were separately predictive of the total modified Sharp score 6 years later. Patients from this cohort represented a spectrum of disease activity at baseline, with a median swollen joint count of 15 and a median CRP level of 18 mg/liter (normal <1 mg/liter) (8). Baseline total modified Sharp scores were relatively low, but almost half of the patients had radiographic erosions scored at the hands or feet, suggesting an adverse prognosis. Standard clinical and radiographic variables predicted 45% of the variance of the 6-year Sharp score, but this was improved to 59% with the addition of MRI data. This result shows that MRI, which samples a small area (the dominant wrist), may be useful in helping to predict generalized joint damage in patients presenting with RA.

An initial concern over using MRI to quantify joint inflammation and damage has been the difficulty in obtaining reproducible scores for erosions, edema, and synovitis using the various scoring systems developed (24). Scoring MRI at the wrist is a complex task, requiring a detailed knowledge of the 3-dimensional anatomy of the carpal bones and joints as well as an understanding of the different sequences used, the signal characteristics of the tissues being imaged, and potential sources of error that are intrinsic to the modality itself. The latter include phenomena such as partial voluming, where signal from 2 sharply contrasting areas is “averaged” in the intervening zone, homogeneity of fat saturation, and variations in Gd-DTPA uptake, which need to be interpreted by a trained reader. This makes MRI scoring considerably more difficult than scoring radiographs to assess joint damage. Fortunately, throughout this study, interobserver reliability for MRI scoring has been high, with ICCs ranging from 0.7 to 0.85. Our MRI scores were obtained by 2 radiologists specializing in the musculoskeletal system who have an interest in MRI interpretation. We also used a scoring system that was sensitive to change in early disease, because erosions were counted and scored for size, ensuring that even minor progression could be detected.

In this study, MRI scans and plain radiographs were read without reference to previous scans or radiographs from the same patient. This approach could decrease sensitivity for scoring abnormalities such as erosions, because a reader with access to previous images could be alerted to a particular site and therefore examine it more carefully. The advantage for our method might be improved specificity, because a reader may be less likely to score an indistinct lesion as an erosion than they would be if they knew that previous images had also shown a defect in that area. Most radiographic studies of RA progression have allowed readers access to other films in the series, and we intend to reexamine our data in that manner to further explore issues of reliability.

Are erosions seen on MRI the same lesions as those detected by plain radiography? The answer to this question seems to be yes, in the main, as illustrated by studies comparing MRI findings with those of miniarthroscopy and ultrasound (25, 26). However, cortical bone, which is so well-visualized on plain radiographs as a white line, appears on MRI as a low signal region that may be confused with focal (and potentially transient) subcortical bone marrow edema (11). Other sources of false-positive MRI erosions include degenerative bone cysts and active synovitis adjacent to irregular bony margins.

Results from this study reveal high levels of concordance between MRI and radiography erosion scores at baseline and 6 years, when radiography scores were assessed from the same area as MRI scores (the dominant wrist) and also from a larger area (both hands and feet). The modified Sharp score includes scores for both erosions and joint space narrowing and is thus influenced by preexisting degenerative change at sites such as the first MTP joint. This may account for the finding that MRI and radiography scores correlated less well at baseline, when Sharp scores reflected joint space narrowing that was not noticed using the current MRI scoring system. By 6 years, however, rheumatoid articular damage had eclipsed minor degenerative change on radiography and paralleled bony erosive change as scored on MRI. Another reason for the initial relative discordance of MRI and radiography scores is that MRI reveals erosions much earlier: 45% of this cohort had MRI erosions at the dominant wrist at baseline, and only 15% had radiographic erosions at the wrist (8).

Bone marrow edema has been recognized as being important in RA only since the advent of MRI (27). It occurs in many non-RA contexts, including trauma and degenerative joint disease; it also occurs adjacent to enthesopathy and possibly in normal subjects (28). However, is all bone edema the same? In the setting of RA it has been estimated to occur in 68–75% of patients with early disease (8, 27). In our cohort it was confirmed as a preerosive lesion at 1 year (9) and again at 6 years, when the risk of developing erosion in a bone involved by edema at baseline was 6.5-fold that of erosion occurring without preceding edema. Although the level of bone edema for the group as a whole remained static throughout the study, individuals had high scores for bone edema at various time points, and this was often a transient phenomenon (i.e., it was not observed on their subsequent MRI scans). Thus, in those cases in which bone edema was not identified at baseline but erosion had occurred at 6 years, it is possible that transient but undetected bone edema had occurred as a preceding event. More frequent MRI scans over a period of years would be needed to explore the link between bone edema and erosion and to determine the degree of causality.

The mechanism underlying the association between bone edema and erosion remains to be determined, but bone edema in this context could represent an intraosseous cellular infiltrate capable of eroding cartilage and bone from the subchondral aspect of the joint. Evidence for such an infiltrate has been described in animal models of RA, in which TNF-responsive mesenchymal cells were identified within enlarged bony canals connecting bone marrow to synovium (29). Unfortunately, there are no MRI studies of animal models of early RA, and such studies are urgently needed. The histologic features associated with bone edema in early RA in humans are yet to be described because of difficulty obtaining material to examine, but Lee et al reported a decrease in bone edema in patients with RA who are in clinical remission, supporting its link to disease activity (30). McGonagle et al suggested that synovitis might lead to bone edema and subsequent erosion (28). Consistent with this is the observation from this cohort that all patients with bone edema at baseline also had synovitis, which was present in 93% of the total group (8). However, not all patients with baseline synovitis eventually experienced joint erosion, suggesting that the proposed linear causal relationship, with synovitis leading to bone edema leading to erosion, is too simplistic.

Why, in this cohort, was baseline MRI synovitis not predictive of radiographic erosion scores at 6 years? Data obtained previously from the same patients showed that MRI synovitis at baseline was predictive of MRI erosion scores at 1 year (10) but not of radiographic erosion scores at 1 or 2 years (9, 11). Other investigators have also demonstrated a link between early MRI synovitis and subsequent MRI erosion occurring after 1 year (12, 31). Conaghan et al recently reported that the area under the curve for MRI synovitis measured over 12 months in patients with early RA was a predictor of bone damage progression at MCP joints (31).

There are several possible explanations for the lack of a predictive association between synovitis and radiographic erosions in our cohort. First, this may not have been detected because the number of patients followed for 6 years was small, and a Type II error could have occurred. Second, a single “snapshot” of synovitis at baseline is unlikely to reflect subsequent levels of joint inflammation, which is why measurements of area under the curve are more informative. Indeed, Conaghan et al also did not demonstrate that baseline synovitis, as a single measure, predicted erosive progression in their patients (31). Third, this study is the first to follow up patients for a period as long as 6 years. It is possible that other processes are combining with synovitis to drive medium-term erosive progression. Our data suggest that bone edema could be a marker of such a process. The concept of a nonsynovitis mechanism contributing to erosion in RA is not new (32, 33). Kirwan postulated separate mechanisms underlying synovial inflammation and articular damage based on results of his study of low-dose steroid use in early RA, in which groups treated with prednisolone exhibited less radiographic progression than did groups treated with other disease-modifying antirheumatic drugs (DMARDs), despite the same control of synovitis (33).

In our group of patients, results of MRI scans were available to treating physicians, which may have modified their treatment decisions. Because this was not designed as an interventional study, no attempt was made to standardize DMARD therapy. Thus, the possibility exists that early aggressive therapy could have been instituted in those patients with MRI erosions on initial scans, potentially resulting in lower Sharp scores at 6 years. This would weaken an association between initial MRI scores and final radiographic outcome, suggesting that the observed association is, if anything, an underestimate of the true effect. In this cohort, analysis of the effect of treatment on outcome was performed at 1 year (9) and revealed that patients receiving DMARDs at that point were significantly more likely to have erosive disease on MRI, but this simply reflects earlier DMARD use in those with clinically aggressive disease. None of these patients received anti-TNF therapy, which is not currently funded in New Zealand.

In summary, this is the first study to show that carpal bone edema scored on MRI scans obtained at the time of first presentation of RA can be used to help predict the total radiographic outcome at the hands and feet at 6 years, as measured by the modified Sharp score. Using a modeling approach for predicting outcome, the addition of MRI variables to baseline clinical measures of disease activity improved the prediction of variance of 6-year Sharp score data. Thus, MRI scans of the dominant carpus might be a useful addition to standard investigations of patients with RA at the first presentation, although larger studies are needed to confirm this. We did not find MRI synovitis at baseline to be a predictor of radiographic erosion at 6 years, and this could suggest that other mechanisms in addition to synovitis operate in the intermediate term to promote articular damage. Bone marrow edema, which was strongly associated with subsequent erosion at individual sites in the carpus, could represent evidence of such a process centered in the subchondral region. MRI is proving to be a useful tool with which to investigate disease processes in RA and has the potential for clinical use in determining the prognosis and targeting aggressive therapy to patients with the most destructive disease.


We wish to acknowledge the assistance of the following clinicians who referred patients for this study: Drs. Mike Butler, David Caughey, Nora Lynch, Alan Doube, Hamish Hart, Peter Gow, Raoul Stuart, Terry Macedo, Max Robertson, Roger Reynolds, and the late Bob Grigor. We are also most grateful to the technical staff at the Auckland Radiology Group, who supervised the MRI scans (in particular Ms Rika Nel), and to Ms Violetta Pokorny and Mrs. Ma Wei (Department of Molecular Medicine), who performed HLA–DRB1 genotyping.