Objectives To examine the construct validity of MRI in the detection of structural hand osteoarthritis features with conventional radiography (CR) as reference and explore the association between radiographic severity and MRI-defined pathology.
Methods 106 hand osteoarthritis patients (97 women, mean age 68.9 years (SD 5.6)) had 1.0T contrast-enhanced MRI and CR of the dominant hand. The 2nd–5th interphalangeal joints were scored according to the preliminary Oslo hand osteoarthritis MRI score and Kellgren–Lawrence (KL) scale and Osteoarthritis Research Society International atlas for radiographs. The authors compared the number of joints with structural features by MRI and CR (Wilcoxon signed-rank test) and examined concordance at the individual joint level. The OR of MRI features in joints with doubtful (KL grade 1), mild (2) and moderate/severe (≥3) radiographic osteoarthritis was estimated by generalised estimating equations (KL grade 0 as reference).
Results MRI detected approximately twice as many joints with erosions and osteophytes compared with CR (p<0.001), but identification of joint space narrowing, cysts and malalignment was similar. The sensitivity of MRI was very high for osteophytes (1.00) and erosions (0.95), while specificity was lower (0.22 and 0.63). The prevalence of most MRI features increased with radiographic severity, but synovitis was more frequent in joints with mild osteoarthritis (OR2.1, 95% CI 1.4 to 3.2) than in moderate/severe osteoarthritis (OR1.4, 95% CI 1.0 to 2.2).
Conclusion MRI detected more osteophytes and erosions than CR, suggesting that erosive osteoarthritis may be more common than indicated by CR. Synovitis was most common in mild osteoarthritis. Whether this is due to burn-out of inflammation in late disease must be investigated further.
Statistics from Altmetric.com
Conventional radiography (CR) is currently considered as the gold standard for the assessment of structural hand osteoarthritis features.1 2 CR provides a two-dimensional image of the bony changes and joint space narrowing (JSN) as an indirect measure of cartilage loss. However, osteoarthritis is currently recognised as a failure of the whole joint,3 and Magnetic Resonance Imaging (MRI) has a unique advantage with multiplanar visualisation of all joint components.
MRI is accepted as an outcome measure in both knee osteoarthritis and in inflammatory joint diseases. Features that are not visible on CR such as bone marrow lesions (BML) and synovitis are proven as important markers for prognosis and symptomatology in both knee osteoarthritis and rheumatoid arthritis,4,–,7 and numerous studies in rheumatoid arthritis have shown that MRI is more sensitive than CR in the detection of structural damage in the wrist and finger joints.8 9 However, only a few MRI studies have been performed in hand osteoarthritis.10,–,12 Wittoek et al.12 recently compared the agreement between CR, MRI and ultrasonography in patients with hand osteoarthritis. However, the study only included 14 patients, and the MRI reading was performed without any standardised scoring system. We have recently published the first proposed MRI scoring system for hand osteoarthritis, which showed very good intra-reader and inter-reader reliability.13
In this study of 106 hand osteoarthritis patients, our first objective was to examine the construct validity of MRI in comparison with CR and to examine the concordance between the two modalities in the detection of structural features (osteophytes, JSN, erosions, cysts and malalignment). Our second aim was to explore the frequency of MRI features in joints with different levels of radiographic osteoarthritis severity.
The Oslo hand osteoarthritis cohort was established in after 2001, and 209 patients (190 women) underwent an extensive examination in 2001–3.14 Of those, 128 patients (117 women) met for a follow-up examination (including MRI) in 2008–9. In this study, we included 106 (97 women) patients who had available MRI scans with T1-weighted and short TI inversion recovery (STIR) sequences and CR of the dominant hand. Ten of those had no postgadolinium T1-weighted images. The examinations in each individual were performed on the same day, except for two patients with MRI 13 and 18 days after CR.
MRI of the hand
The 2nd–5th distal interphalangeal (DIP) and proximal interphalangeal (PIP) joints of the dominant hand (n=848 joints) were examined using an extremity 1.0T MRI unit (ONI; GE Healthcare, Waukesha, Wisconsin, USA), as detailed previously.13 Coronal, sagittal and axial T1-weighted fat-suppressed pre- and post-intravenous (0.1 mmol gadolinium/kg body weight; Magnevist; Bayer Schering Pharma AG, Leverkusen, Germany) images were acquired from a three-dimensional dual-echo Dixon technique15 (repetition time 20 ms, echo time 5 ms, 1 mm slice thickness) for assessment of structural features as well as synovitis and flexor tenosynovitis (gadolinium enhancement). Coronal and axial STIR images (repetition time 2850 and 3150 ms, echo time 16.3 and 21 ms, 2 and 3 mm slice thickness, respectively) were acquired for assessment of BML. Total acquisition time was approximately 30 min.
Two readers (IKH, PB) in consensus scored the 10 first MRIs while one reader (IKH) scored the remaining MRIs according to the preliminary Oslo hand osteoarthritis MRI score;13 osteophytes (grade 0–3; distal/proximal part of the joint separately), JSN (grade 0–3), cysts (absence/presence; distal/proximal), malalignment (absence/presence; frontal (≥15°)/sagittal plane), synovitis (grade 0–3), flexor tenosynovitis (grade 0–3), BMLs (grade 0–3; distal/proximal), collateral ligament absence/discontinuity (absence/presence; radial/ulnar) and BMLs at collateral ligament insertions (absence/presence; distal/proximal and ulnar/radial). Erosions (grade 0–3; distal/proximal) and attrition of the joint plate (absence/presence; distal/proximal) were scored separately, in contrast to the proposed combined definition and grading. A detailed description of the features is shown in supplementary table S1, available online only. IKH re-scored 10 randomly selected MRI after a period of at least 7 weeks and the intra-reader reliability assessed by κ and intraclass correlation coefficients was classified as ‘good’ to ‘very good’ for all features except flexor tenosynovitis, which had ‘fair’/‘moderate’ reliability (see supplementary table S2a, available online only).16 The scoring of MRI was performed blinded for radiographic status scores and clinical determinants.
Conventional hand radiography
The patients underwent posteroanterior CR of both hands and one reader (IKH) scored the bilateral 2nd–5th DIP, 2nd–5th PIP, thumb interphalangeal, 1st–5th metacarpophalangeal and 1st carpometacarpal joints according to the Kellgren–Lawrence (KL) scale (grade 0–4).17 The presence/severity of individual radiographic features were scored according to the Osteoarthritis Research Society International (OARSI) atlas18; osteophytes (grade 0–3), JSN (grade 0–3), central erosions (absent/present), cysts (absent/present) and malalignment (absent/present). In order not to underestimate the frequency of radiographic erosions, one reader (IKH) also assessed the presence of marginal erosions (no atlas/scoring system available) in joints that had no central erosions according to the OARSI atlas.18
Thirty randomly selected radiographs were scored twice by the same reader (IKH) and by a second reader (JB). Intra-reader and inter-reader reliability for KL and individual radiographic features were classified as ‘very good’ for most features (see supplementary table S2b, available online only).16
For joints that were scored for MRI pathology in both the proximal and distal part of the joint, we used the highest score for that joint (ie, a joint with an osteophyte grade 3 distally and grade 1 proximally was interpreted as grade 3). We dichotomised the collateral ligament variables into ‘present ligaments versus one–two absent/non-continuous ligaments’ and ‘no BML versus 1–4 BMLs at the insertion sites’. Flexor tenosynovitis grade 3 was present in one joint only and we therefore collapsed grades 2 and 3.
We had a few missing values for the MRI features due to limited field of view (structural features: eight or fewer missing of 848 joints; synovitis: 13 of 768; flexor tenosynovitis: 40 of 768) and the numbers of affected joints were calculated using imputation of the mean score. We compared the number of affected (≥grade 1 pathology) interphalangeal joints of the 2nd–5th finger in the dominant hand by CR and MRI with the Wilcoxon signed-rank test and examined the concordance at individual joint level for features that were assessed by both MRI and CR; ie, osteophytes, JSN, erosions/attrition, cysts and malalignment. Osteophytes and JSN were assessed on 0-3 scales by both modalities and we calculated the percentage of exact agreement (PEA) and percentage of close agreement (PCA; difference ≤1). We calculated the PEA between presence of radiographic central erosions and MRI-defined bone attrition. Similarly, we calculated the PEA between the presence of radiographic erosions (either central or marginal) and the presence of MRI-defined bone attrition and/or erosions (grade 0, absent; grades 1–3, present). Cysts and malalignment were scored as absent/present by both modalities and we calculated the PEA. The construct validity of MRI was further assessed by calculation of κ (weighted κ for osteophytes and JSN) and sensitivity/specificity of MRI with CR used as reference.
We calculated the prevalence of MRI features in joints with no osteoarthritis (KL grade 0), doubtful (KL grade 1), mild (KL grade 2) and moderate/severe osteoarthritis (KL grade ≥3). KL grades 3 and 4 were collapsed due to few joints in each category (n=70 and n=122, respectively). Furthermore, we calculated the odds ratio (OR) for the presence of MRI features in joints with doubtful, mild and moderate/severe radiographic osteoarthritis (KL grade 0 as reference) by generalised estimating equations (unstructured correlation matrix) in order to adjust for within-subject dependency. These analyses were adjusted for age and sex.
The data collection was approved by the regional ethics committee and the data inspectorate. All patients signed informed consent.
Most patients had many osteoarthritic joints and a large proportion had radiographic erosive disease (table 1). Two patients had no hand joints with KL grade ≥2. However, both patients had one or more joint(s) with KL grade 1 and fulfilled the clinical American College of Rheumatology (ACR) criteria of hand osteoarthritis.19
The concordance between MRI and CR of the interphalangeal joints of 2nd–5th finger
MRI detected more joints with osteophytes compared with CR both on the subject level (table 2) and individual joint level (729 vs 333 joints) (table 3). The sensitivity of MRI with CR as reference was excellent; 332 of 333 joints (99.7%) with radiographic osteophytes also had MRI-defined osteophytes. The specificity of MRI was low (22.3%); MRI demonstrated osteophytes in 397 of 511 joints (77.7%) with no radiographic osteophytes. However, the majority of these osteophytes were small (grade 1) (354 of 397; 89.2%) and the PCA between MRI and CR was very good (table 3).
CR detected JSN more frequently than MRI (679 vs 578 joints) (tables 2 and 3). In total, 147 of 679 joints (21.6%) that appeared narrowed on CR were scored as normal by MRI (sensitivity of MRI 78.4%). The majority of these joints had radiographic JSN grade 1 (142 of 147; 96.6%). On the other hand, 46 of 165 joints (27.9%) that appeared normal on CR were scored as narrowed by MRI (specificity of MRI 72.1%). The majority of these joints had MRI-defined JSN grade 1 (40 of 46; 87.0%). The PCA between MRI and CR was excellent (table 3).
We compared MRI-defined bone attrition and radiographic central erosions (181 vs 170 joints) and found no significant difference in the number of affected joints and good agreement between MRI and CR (tables 2 and 3). We also compared MRI-defined erosions and/or bone attrition (combined variable) and radiographic central or marginal erosions (combined variable) (434 vs 209 joints). MRI detected significantly more joints with erosions/attrition compared with CR (table 2) and the sensitivity of MRI was excellent (table 3). Erosions/attrition were detected by both modalities in 199 joints (23.7%). The majority of these joints had MRI-defined attrition (141 of 199; 70.9%). Erosions/attrition was detected by MRI only in 235 joints (28.0%) and the majority of erosions were small (grade 1) (158 of 235; 67.2%) (figure 1). Erosions were detected by CR only in 10 joints (1.2%).
Cysts were infrequent on both MRI and CR (41 vs 38 joints) and the number of affected joints was non-significantly different between the two modalities (table 2). However, the concordance at the individual joint level was fair (table 3); cysts were detected by both modalities in six joints (0.7%), by MRI only in 35 joints (4.2%) and by CR only in 32 joints (3.8%). The majority of joints with cysts on CR only had MRI-defined erosions (27 of 32; 84.4%).
Malalignment was also infrequent on both MRI and CR (56 vs 98 joints), but the number of affected joints was significantly higher for CR (table 2). The concordance at the individual joint level was moderate, with high specificity but low sensitivity for MRI (table 3).
Similar results were found in DIP and PIP joints in stratified analyses (see supplementary tables S3 and S4, available online only).
MRI features in interphalangeal joints of 2nd–5th finger with different levels of radiographic severity
The prevalence of most MRI features was highest in joints with moderate/severe radiographic osteoarthritis (KL grade ≥3) (table 4). However, small/mild osteophytes, JSN, erosions, synovitis and absent/non-continuous collateral ligaments were frequent even in joints without radiographic osteoarthritis (KL grade ≤1).
We found higher odds of structural MRI features such as osteophytes, JSN, erosions, attrition, malalignment and collateral ligament absence/discontinuity in joints with increasing radiographic severity after adjustment for age and sex (table 5).
Joints with mild osteoarthritis (KL grade 2) had higher odds of synovitis than joints with moderate/severe osteoarthritis (KL grade ≥3) (table 5). A similar pattern was found for the presence of moderate/severe synovitis (grades 2–3); KL grade 1: OR 3.5 (95% CI 1.7 to 7.5; p=0.001); KL grade 2: OR 7.5 (95% CI 3.7 to 15; p<0.001) and KL grade 3 or greater: OR 4.5 (95% CI 2.2 to 8.9; p<0.001).
Joints with moderate/severe osteoarthritis (KL grade ≥3) had highest odds of BMLs (table 5). A similar pattern was found for moderate/severe BMLs (grades 2–3); KL grade 1: OR 0.50 (95% CI 0.1 to 49; p=0.76); KL grade 2: OR 11 (95% CI 1.3 to 89; p=0.03) and KL grade 3 or greater: OR 14 (95% CI 2.3 to 90; p=0.005).
The odds of flexor tenosynovitis and cysts were not significantly higher in joints with radiographic osteoarthritis (table 5).
Similar patterns were found for DIP and PIP joints in stratified analyses (see supplementary table S5, available online only).
This study is the first to compare MRI and CR in a large hand osteoarthritis cohort using standardised scoring methods for both modalities. First, we found that MRI was more sensitive than CR in the detection of erosions and osteophytes. Second, the frequency of MRI-defined osteophytes, JSN, erosions, attrition, malalignment, absent/discontinuous collateral ligaments and BMLs increased with radiographic severity. Synovitis was frequent and was most common in mild radiographic osteoarthritis.
The classification of erosive hand osteoarthritis has traditionally been based on the presence of radiographic central erosions, which typically show classic ‘seagull wing’ patterns. Previous studies on radiographic hand osteoarthritis have shown that cartilage loss precedes radiographic erosive evolution,20 21 suggesting that biomechanical factors may play a role in the development of these central erosions. However, studies using high-resolution MRI have also identified periarticular/marginal erosions, which possibly have a different pathogenesis.10 11 Erosions were associated with pathological collateral ligaments and synovitis, and these associations need to be examined for causality in longitudinal studies.10 11
Based on these findings, we decided to score these two types of erosions separately in this study. MRI-defined attrition was scored as present if the joint plate was flattened/depressed or had the typical ‘seagull wing’ configuration with or without accompanying erosions (distinct cortical break seen in at least two planes). We found good agreement between MRI-defined attrition and radiographic central erosions (according to the OARSI atlas). However, the total number of joints with MRI-defined erosions and/or attrition were higher than the total number of joints with radiographic erosions (both marginal and central), which is in line with previous MRI studies in hand osteoarthritis.10 12 Erosions may be located on the dorsal/volar side of the joint and may therefore be less visible on CR. These marginal erosions have traditionally not been assessed in hand osteoarthritis and future studies should explore their role in hand osteoarthritis with regard to the pathogenesis of osteoarthritis, prognostic value and their associations with patient-reported outcomes.
We similarly found that MRI was more sensitive than CR in the detection of osteophytes, which is probably due to the multiplanar demonstration of the joint by MRI. CR with the traditional posteroanterior view can demonstrate ‘real’ intra-articular osteophytes at the joint margins. The sagittal view of MRI is also able to visualise traction spurs, which represent physiological responses at the insertion of tendons and ligaments. As reported by Tan et al,22 we frequently found osteophytes at the insertion of the extensor tendon. Although the pathogenesis of osteophytes may differ, we did not score these types of osteophytes separately for feasibility reasons. In contrast to our findings, Wittoek et al12 found that MRI was less sensitive than CR in the detection of osteophytes, and argued that this could be due to the signal void of densely packed calcium in osteophytes. However, the authors did not use a standardised scoring method for the assessment of osteophytes and it remains unclear how they defined osteophytes and which sequences and planes were used for this assessment.
CR demonstrated JSN more frequently than MRI in this study. Malpositioned/flexed normal joints may appear narrowed on CR and this may have contributed to the decreased sensitivity of MRI compared with CR. However, MRI is able to demonstrate intra-articular osteophytes (and thus cartilage defects) that are not visualised on CR, which may contribute to the decreased specificity of MRI.
As previously reported by Tan et al,11 we found that virtually all structures were affected in both mild and severe osteoarthritis, confirming that osteoarthritis is a ‘whole-organ disease’. All features, except synovitis, were more frequent in joints with moderate/severe radiographic osteoarthritis than in no/mild osteoarthritis. Osteoarthritis has traditionally been considered a non-inflammatory disease. However, inflammation is increasingly recognised in both early and late stages of the disease contributing to symptoms and progression of osteoarthritis.23 Recent studies using ultrasonography have shown that synovitis is frequent in patients with hand osteoarthritis.24,–,28 In this study, we found a high proportion of mild MRI-defined synovitis also in joints with no/doubtful osteoarthritis (KL grade ≤1), as previously shown for knee osteoarthritis.5 Synovitis was most frequent in joints with mild osteoarthritis, which could indicate a ‘burn-out’ of inflammation in later stages or possibly more easily detection of synovitis in joints with more anatomical space. In knee osteoarthritis, Benito et al29 showed that cell infiltration and vascular proliferation were more frequent in synovial biopsies from patients with early osteoarthritis (<1 year symptom duration) than in severe late osteoarthritis. However, other studies have shown that synovitis is most frequent in severe knee osteoarthritis.30 31
Studies using high-resolution MRI have shown that collateral ligament pathology (thickening/inflammation/disruption) is common in both early (<1 year symptom duration) and chronic hand osteoarthritis with close anatomical relation to both erosions, BMLs and bone formation.11 With the use of conventional MRI, we similarly found that collateral ligament disruption/discontinuation was most frequent in moderate/severe osteoarthritis, but was also present in normal joints and joints with doubtful/mild osteoarthritis. However, it is not yet clear whether collateral ligament pathology is age related or whether it is implicated in the pathogenesis of hand osteoarthritis.
All patients had either radiographic hand osteoarthritis and/or fulfilled the clinical ACR criteria.19 The ACR criteria do not necessarily pick up patients with isolated carpometacarpal joint osteoarthritis and six of seven patients who did not fulfil the criteria had radiographic carpometacarpal osteoarthritis and no/limited amount of osteoarthritis in the DIP/PIP joints (one patient had only one DIP joint with definite osteoarthritis).
The strengths of the study are the large study sample and the extensive joint assessment by MRI and CR. Study weaknesses include the lack of MRI of the carpometacarpal and metacarpophalangeal joints. Participants in the Oslo hand osteoarthritis cohort were recruited from the rheumatology outpatient clinic and the generalisability of the results to the general hand osteoarthritis population must be investigated further. Future studies using computer tomography and histology should explore whether MRI-defined erosions, osteophytes, synovitis and BMLs represent true findings.
It is too early to recommend MRI for the diagnosis/monitoring of hand osteoarthritis in clinical practice. Even if MRI is able to detect erosive disease earlier, we have currently no disease-modifying treatment and the prognostic value of MRI is unknown.
In conclusion, conventional MRI was more sensitive than CR in the detection of osteophytes and erosions in patients with hand osteoarthritis, indicating that erosive hand osteoarthritis is more frequent than previously indicated by CR. Joints with increasing radiographic severity had a higher occurrence of most MRI features. Synovitis was most commonly present in mild radiographic disease, which may indicate a ‘burn-out’ of inflammation in later stages. However, this hypothesis needs to be examined further in longitudinal studies.
The authors thank the patients of the Oslo hand osteoarthritis cohort for participating in this study, study nurses and technicians in the Department of Radiology for helping to collect the data and Inge C Olsen for help with the statistical analyses.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Web Only Data - This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Handling editor Dr. Johannes WJ Bijlsma
Funding The study is supported by grants from the South-Eastern Norway Regional Health Authority.
Competing interests None.
Patient consent Obtained.
Ethics approval The data collection was approved by the regional ethics committee and the data inspectorate.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.