In this validation exercise, MRI assessment of JSN in the hands and wrists of patients with RA provided the same results as did XR assessment of the same joints. MRI was both highly sensitive and specific for JSN that was demonstrable with XR. This suggests that MRI-JSN scoring may offer a viable alternative to XR-JSN scoring in multicenter clinical trials of RA. This is important because clinical trials with XR have become more costly and time consuming over the past decade. RA patients appropriate for clinical trials are increasingly difficult to recruit, and the switch to active comparator study designs has increased the number of patients and the observation time required to discriminate reliably the differences in change between treatment arms. The increased sensitivity of MRI for detecting bone erosions [2
] should allow it to demonstrate structure-modifying treatment more quickly and with fewer patients than XR could. However, the MRI scoring method that has been used in most clinical trials thus far, the Outcome Measures in Rheumatology Clinical Trials RA MRI Score (OMERACT-RAMRIS) [13
], did not include cartilage loss or JSN, and thus lacked the content validity of XR for structural joint damage. This was a significant limitation because bone erosion and cartilage loss do not always show the same pattern of response to therapy, as illustrated in the randomized, controlled trial of the inhibitor of the receptor for activated nuclear factor-κB (RANKL), denosumab, reported by Cohen et al.
]. In that study, 227 patients with established RA were treated with either placebo plus methotrexate or one of two doses of denosumab plus methotrexate and followed up longitudinally with XR (erosion and JSN) and MRI (erosion only). Although both MRI and XR showed denosumab to have strong erosion-suppressing effects, XR showed denosumab to have no effect on preventing JSN. Had XR been excluded from the study, the lack of efficacy on this structural end point would not have been noticed.
Fortunately, MRI is well suited for imaging articular cartilage [14
]. It depicts joint anatomy tomographically and therefore without projectional distortions that can mimic JSN on conventional XR, and MRI also is able to visualize the articular cartilage tissue directly, rather than only as a space between opposing articular cortices, as with XR. Moreover, MRI shows the same distribution of involvement of joints in the hands and wrists of patients with RA as does XR [19
]. Fat-suppressed, T1
-weighted 3D gradient-echo scans, as were used in this study, have been shown to delineate articular cartilage accurately in various joints, including the MCPs [15
], and is commercially available on all clinical MRI systems operating at a magnetic field strength of 1.0 T or higher. Systems operating at lower field strengths currently have difficulty with this technique because of limitations in spectral fat suppression or selective water excitation. Selective fat suppression or water excitation are important for increasing T1
contrast between cartilage and adjacent joint fluid or subchondral bone (marrow fat) and for eliminating chemical-shift effects [16
], which distort cartilage-bone interfaces and can simulate cartilage thinning and JSN. Increasing receiver bandwidth can reduce chemical shift, but this reduces the signal-to-noise ratio of the images and does not completely eliminate the problem. Fat-suppressed, T1
-weighted, 3D gradient-echo is also the most commonly used pulse sequence for evaluating bone erosion in multicenter randomized controlled trials of RA [9
]. Thus, MRI protocols do not require expansion to add JSN to assessments of joint damage.
One technical challenge in MRI-JSN assessment is achieving adequate coverage of all PIP joints, as these joints are at the distal limit of the field of view used in most clinical-trial protocols. In this study, 35 (19%) of 188 of PIP joints were not adequately covered and therefore excluded from analysis. Despite this limitation and the fact that images from two different multicenter clinical trials were pooled for this investigation, MRI-JSN correlated strongly with XR-JSN, attesting to the robustness of the MRI-JSN method.
Because of the longer imaging time required for MRI than for XR, MRI protocols in most clinical trials include only one or two hand(s)/wrist(s) per patient, whereas XR protocols typically include both hands/wrists and feet. Despite this greater anatomic coverage of XR, however, adding XR data from the other hand/wrist and both feet did not increase the number of JSN-positive patients in this study. Conversely, adding data from the eight additional joints included by MRI increased the number of JSN-positive joints by 36 and the number of JSN-positive patients by one. Thus, MRI of one hand/wrist seems to offer at least the same sensitivity for detecting RA patients with JSN as does XR of both hands and feet.
The MRI-JSN score used this study was modeled after the Genant-modified Sharp XR-JSN score [11
], which has been used in multiple clinical trials to gain regulatory approval of structure-modifying therapies, including abatacept [24
], rituximab [25
], and tocilizumab [26
]. In contrast to the van der Heijde-Sharp JSN-XR score, which uses a 5-point scale, ranging from 0 to 4 in increments of 1 [27
], the Genant-modified Sharp XR-JSN score uses a 9-point scale, also ranging from 0 to 4, but in increments of 0.5. Both methods examine PIP 2, PIP 3, PIP 4, PIP 5, MCP 1, MCP 2, MCP 3, MCP 4, MCP 5, CMC 3, CMC 4, CMC 5, and the capitate-scaphoid and radius-scaphoid joints, but the Genant-modified Sharp method combines CMC 3 to 5 into a single space and adds the IP 1, capitate-lunate, and radius-lunate joints. Direct comparisons of these two XR scoring methods by using images from multicenter clinical trials, found that despite the differences in scales and specific joints examined by each, both methods showed the same discriminative power for XR-JSN change over time and between treatment arms [28
The advantage of a 9-point scale over a 5-point scale is that including smaller increments across a similar range of structural damage may improve sensitivity to change, although comparative longitudinal studies are needed to evaluate this directly. Additionally, all images in this study were evaluated by a single, expert radiologist with 20 years of experience reading XR and MR images in RA clinical trials. As such, the results represent a best-case scenario and may not generalize to analyses performed by less-experienced readers. Interreader variability and smallest detectable change were not evaluated. Time required to perform MRI-JSN scoring also was not evaluated; however, because the technique is extremely detailed, it may be time consuming for inexperienced readers to perform. In a previous investigation [7
], we demonstrated the 9-point MRI-JSN scale to be more sensitive to change than was the 9-point XR-JSN scale. We and others similarly found high interreader agreement and strong XR correlation of a 5-point MRI-JSN score that was similar to the van der Heijde XR-JSN score [4
]. McQueen et al.
] recently compared MRI with XR by using such a 5-point scale in the wrists of 38 patients with RA and 22 control subjects by using 2D spin-echo and 3D gradient-echo MRI at 3 T but without fat suppression, except on postcontrast 3D scans. Although chemical-shift artifacts are greater at 3 T than at 1.5 T, and the joints evaluated by MRI and XR in their study were not exactly the same as those included in either modified-Sharp XR method, correlations were high between the total MRI cartilage score in the wrist and total XR-JSN score in the same wrist (0.61 to 0.74) or in both hands, wrists, and feet (0.68 to 0.78). A subsequent study by Ostergaard et al.
] similarly found strong correlation between a similar 5-point MRI-JSN scale and the van der Heijde-modified Sharp XR-JSN scale in the same hand and wrist.