Objective To determine candidate lesion-based criteria for a positive sacroiliac joint (SIJ) MRI based on bone marrow oedema (BMO) and/or erosion in non-radiographic axial spondyloarthritis (nr-axSpA); to compare the performance of lesion-based criteria with global evaluation by expert readers.
Methods Two independent cohorts A/B of 69/88 consecutive patients with back pain aged ≤50 years, with median symptom duration 1.3/10.0 years, were referred for suspected SpA (A) or acute anterior uveitis plus back pain (B). Patients were classified according to rheumatologist expert opinion based on clinical examination, pelvic radiography and laboratory values as having nr-axSpA (n=51), ankylosing spondylitis (n=34) or non-specific back pain (n=72). Four blinded readers assessed SIJ MRI, recording the presence/absence of SpA by concomitant global evaluation of T1-weighted spin echo (T1SE) and short τ inversion recovery (STIR) scans and, thereafter, whether BMO and/or erosion were present/absent in each SIJ quadrant of each MRI slice. We derived candidate lesion-based criteria based on the number of SIJ quadrants with BMO and/or erosion and calculated mean sensitivity and specificity for SpA.
Results For both cohorts A/B, global assessment showed high specificity (0.95/0.83) compared with the Assessment in SpondyloArthritis international Society (ASAS) definition (0.76/0.74). BMO ≥3 (0.89/0.84) or ≥4 (0.92/0.87) showed comparably high specificity to global assessment. Erosion ≥2 and/or BMO ≥3 or ≥4 were associated with comparably high sensitivity to global assessment without affecting specificity. These combined criteria showed both higher sensitivity and specificity than the ASAS definition.
Conclusions Lesion-based criteria for a positive SIJ MRI based on both BMO and/or erosion performed best for classification of axial SpA, reflecting the contextual information provided by T1SE and STIR sequences.
- Magnetic Resonance Imaging
- Ankylosing Spondylitis
Statistics from Altmetric.com
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.
MRI of the sacroiliac joints (SIJs) is the preferred imaging examination in patients with back pain clinically suspected to have axial spondyloarthritis (SpA) when pelvic radiographs show normal or equivocal findings.1 Global assessment of SIJ MRI is the common diagnostic approach with simultaneous evaluation of T1-weighted spin echo (T1SE) and short τ inversion recovery (STIR) sequences globally, not restricted to a single lesion type. Low confidence in a subtle lesion seen on the STIR sequence may be enhanced by additional information on the corresponding T1SE sequence when viewed simultaneously.
Classification of SpA by the Assessment in SpondyloArthritis international Society (ASAS) has incorporated an imaging criterion which can include either radiographic or MRI evidence of sacroiliitis.2 In contrast to the global assessment of SIJ MRI used to diagnose sacroiliitis, the definition of a positive MRI for classification is lesion-based, requiring subchondral or periarticular bone marrow oedema (BMO) on STIR sequence, or osteitis on T1-weighted gadolinium-enhanced sequence.3 To provide guidance on what might be considered highly suggestive for SpA, the ASAS definition includes a quantitative dimension with ≥2 lesions on one slice or one lesion on ≥2 consecutive slices.
The ASAS definition of a positive SIJ MRI was developed by consensus. Since then, data have been reported demonstrating that BMO meeting the ASAS definition is evident in 23.1% of patients with non-specific back pain (NSBP) and 6.8% of healthy controls,4 raising questions as to whether this definition has sufficient specificity for a classification criterion. We, and others, have also reported that SIJ erosion has higher specificity than BMO and that including erosion in a lesion-based definition for a positive SIJ MRI improved sensitivity without worsening specificity compared with the ASAS definition in patients with non-radiographic axial spondyloarthritis (nr-axSpA) having short symptom duration (mean 29 months).4 However, it has since become clear that symptom duration in patients with nr-axSpA may extend over several years before a diagnosis is made despite active symptoms.5 Further study of an appropriate lesion-based definition therefore needs to examine a wider spectrum of patients seen in clinical practice (including patients with longer symptom duration), and its performance should preferably be comparable to global assessment of MRI scans by expert readers.
The goals of this study were (1) to determine candidate lesion-based criteria for a positive SIJ MRI based on BMO alone, erosion alone, or on combined BMO and/or erosion in patients with a broad clinical spectrum of nr-axSpA; (2) to compare the performance of candidate lesion-based criteria for a positive SIJ MRI, including the ASAS definition,3 with global evaluation by expert readers. Rheumatologist expert opinion, which was used as the traditional ‘gold standard’ for three SpA classification criteria sets over the past two decades (Amor,6 European Spondylarthropathy Study Group (ESSG),7 ASAS2), served as the gold standard for classification of axial SpA.
Patients and methods
The demographic and clinical characteristics of the study sample have been reported previously.8 ,9 Two independent inception cohorts of consecutive patients aged ≤50 years with back pain newly referred to two university outpatient clinics were recruited by two different approaches. Cohort A (n=69) comprised patients with back pain referred by rheumatologists and primary care physicians for further evaluation of clinically suspected SpA. Twenty age-matched healthy controls, defined by the Nordic questionnaire10 and by absence of clinical features indicative of SpA, were concomitantly enrolled from hospital staff of the same university clinic. Patients of cohort B (n=88) presented with acute anterior uveitis to a university ophthalmology department and were referred to the rheumatology department for assessment of SpA if they indicated past or present back pain for ≥3 months on a structured questionnaire.
In both inception cohorts, clinician expert opinion by one local rheumatologist (UW for cohort A and WPM for cohort B) based on clinical examination inclusive of structured questionnaires on SpA-related features (modified Outcome in Ankylosing Spondylitis International Study protocol11 for cohort A and Spondyloarthritis Research Consortium of Canada protocol12 for cohort B), pelvic radiographs and laboratory values (HLA-B27, C-reactive protein) served to classify the patients with back pain as having nr-axSpA (n=20/31, for cohorts A/B, respectively), ankylosing spondylitis (AS) (n=10/24) and NSBP (n=39/33). Two readers at each site independently categorised pelvic radiographs according to the modified New York criteria.13 SIJ MRIs were not used for classification of the study subjects to avoid circular reasoning14 since the diagnostic utility of various features on the scans was the primary objective of this study. Patients with continuing or previous treatment with biological agents were excluded from both cohorts. The local ethics review boards approved the study protocols, and the study participants gave written informed consent.
Evaluation of MRI
The technical parameters for STIR and T1SE sequences of semicoronal MRI SIJ scans have been described previously.4 The MRI scans (1.5T Avanto, Siemens Medical Solutions, Erlangen, Germany, for both cohorts A and B) were read and scored independently by four blinded readers (one radiologist: VZ; three rheumatologists: SJP/UW/WPM). The images from each cohort were evaluated separately in random order on electronic work stations, and MRI scores were entered into a customised online data entry module.
The evaluation of SIJ MRI followed a previously published standardised module4 ,15 (online supplementary text S1) with two sections: first a global assessment indicating the presence/absence of SpA according to all MRI findings by viewing T1SE and STIR sequences simultaneously; then a detailed assessment recording BMO and erosion according to standardised lesion definitions and a reference SIJ MR image set developed by consensus among study investigators.16 The presence/absence of BMO and erosion in the cartilaginous joint compartment was recorded as a binary variable in each quadrant (upper and lower ilium, upper and lower sacrum) of each SIJ, based on evaluation of each MRI slice.4 The ASAS definition was considered to be met if BMO was present in ≥2 SIJ quadrants on the same slice or in the same SIJ quadrant on ≥2 consecutive slices. We also assessed the MORPHO definition, which was fulfilled if either BMO or erosion was present in ≥2 SIJ quadrants or if both were present in ≥1 SIJ quadrant.4
Differences in demographic and clinical characteristics between cohorts A and B were assessed by Fisher's exact test for nominal and Wilcoxon test for continuous variables. Mean intraclass correlation coefficients (ICC(3, 1); range) among the six reader pairs were used to calculate the reproducibility of BMO and erosion. ICC values >0.4, >0.6, >0.8 and >0.9 were regarded as representing moderate, good, very good and excellent reproducibility.17 ,18 The frequency of MRI lesions at the group level was analysed descriptively as the median number (IQR) of SIJ quadrants affected by BMO and erosion in patients and controls. At the individual level, we calculated the mean number (percentage) of patients and controls showing ≥1/≥2/≥3/≥4/≥5 SIJ quadrants with BMO or erosion for four readers. For the four readers we also calculated in a binary mode (present/absent) the mean number (percentage) of patients and controls showing ≥1 SIJ quadrant with BMO and ≥1 SIJ quadrant with erosion, ≥1 SIJ quadrant with BMO only, ≥1 SIJ quadrant with erosion only, or showing neither of the two lesions.
We calculated sensitivity, specificity and positive/negative likelihood ratios as mean values of four readers for global assessment and for ASAS and MORPHO definitions with rheumatologist expert opinion as the gold standard. We derived additional candidate lesion-based criteria based on the number of SIJ quadrants with BMO alone (≥2/≥3/≥4 SIJ quadrants with BMO), erosion alone (≥2 SIJ quadrants with erosion), and combined BMO and/or erosion (≥2/≥3/≥4 SIJ quadrants with BMO and/or ≥1/≥2 SIJ quadrants with erosion) to determine which might have optimal mean sensitivity and specificity in both cohorts. Comparison groups were nr-axSpA versus NSBP and AS versus NSBP; in cohort A, we additionally compared nr-axSpA with combined data from NSBP and healthy controls. To account for four correlated observations for each study subject, 95% CIs for sensitivity/specificity were computed by the generalised estimating equations approach described by Smith and Hadgu.19–21
Characteristics of the two SpA inception cohorts, which span a wide spectrum of SpA phenotypes, are shown in table 1. The two different recruitment strategies to identify patients with early SpA resulted in substantial demographic and clinical differences between the two cohorts with cohort B (acute anterior uveitis plus back pain) having much longer symptom duration and less active disease. This disparity in symptom duration, age of study participants and disease activity led us to analyse the two cohorts separately.
Inter-reader agreement for BMO and erosion
For cohorts A/B, mean ICC (range) over six reader pairs for the number of affected SIJ quadrants was 0.88 (0.84–0.93)/0.75 (0.64–0.88) for BMO, and 0.82 (0.79–0.85)/0.71 (0.62–0.78) for erosion.
Frequency of BMO and erosion
The median number of SIJ quadrants with BMO or erosion per group and the mean number (percentage) of patients and controls having BMO or erosion (stratified from ≥1 to ≥5 affected SIJ quadrants) is shown in table 2. In patients with nr-axSpA, BMO and erosion were recorded more often in cohort A. Erosion in the absence of BMO was recorded in 7.5%/11.3% of patients with nr-axSpA in cohorts A/B, and BMO in the absence of erosion in 25.0%/24.2%. In controls, BMO in the absence of erosion was common (26.9%/32.5% in NSBP/healthy controls of cohort A, and 17.4% of NSBP controls in cohort B), whereas erosion in the absence of BMO was virtually absent (3.8%/1.3% in NSBP/healthy controls of cohort A, and 2.3% of NSBP controls in cohort B).
Sensitivity and specificity of candidate lesion-based criteria
Patients with nr-axSpA versus NSBP controls
For both cohorts, global assessment (0.95/0.83) and a candidate criterion based on erosion alone (≥2 SIJ quadrants with erosion) showed the highest specificity (0.97/0.90); however, sensitivity for the latter was much lower (0.58/0.30) than for global assessment (0.74/0.44) (table 3). Candidate criteria based on BMO in ≥3 or ≥4 SIJ quadrants had substantially enhanced specificity over both ASAS and MORPHO definitions. In addition, candidate criteria that included both BMO and erosion (≥3 or ≥4 SIJ quadrants with BMO and/or ≥2 SIJ quadrants with erosion) showed higher sensitivity and specificity in both cohorts than with either the ASAS or MORPHO definitions and were comparable in performance to global assessment (table 3). For both groups the MORPHO definition had increased sensitivity in comparison with the ASAS definition (0.88/0.54 vs 0.80/0.42) with only a slight drop in specificity (0.72/0.72 vs 0.76/0.74). In cohorts A/B, sensitivity and specificity of the ASAS definition (requiring ≥2 BMO quadrants on the same slice or in the same quadrant on ≥2 consecutive slices) was almost identical to ≥2 SIJ quadrants with BMO in any location.
Patients with AS versus NSBP controls
The sensitivity and specificity for cohorts A/B followed the same patterns as in patients with nr-axSpA for all comparison groups, but showed much higher sensitivity for candidate criteria that included both BMO and/or erosion than for criteria based on BMO alone, including the ASAS definition, with minimal decrease in specificity when compared with global assessment (table 4).
Single reader data (see online supplementary table S1) reproduce the same pattern of sensitivity/specificity observed by mean data of all four readers showing lower specificity for ASAS criteria and increase in specificity for the more stringent criteria (BMO in ≥3 SIJ quadrants or combining erosion and/or BMO) consistently across both cohorts A/B and for the nr-axSpA and AS groups. Online supplementary table S2 shows 95% CIs for sensitivity/specificity.
This study in patients with nr-axSpA with a broad variability of symptom duration and disease activity reflecting clinical practice showed that candidate lesion-based criteria for a positive SIJ MRI that incorporate BMO as well as erosion perform best when compared with global assessment by expert readers. This is consistent with the integration of information from both T1SE and STIR sequences that leads to the diagnostic conclusions by expert readers. Although candidate criteria based on ≥3 SIJ quadrants with BMO alone also had high specificity comparable to global assessment, we observed a gain in sensitivity with only minimal loss in specificity when erosion was incorporated into the BMO-based candidate criteria. This difference in magnitude between sensitivity and specificity was consistent across both cohorts A/B, in nr-axSpA and more pronounced in patients with AS, and for single and mean reader data. The relative performance of these criteria was consistent between both cohorts, and these criteria also demonstrated a better performance than any criteria based on BMO alone in ≥2 SIJ quadrants, including ASAS and MORPHO proposals. These findings support a data-driven definition of a positive SIJ MRI incorporating both BMO and/or erosion for classification of SpA, which reflects the information provided by simultaneous evaluation of T1SE and STIR sequences (figure 1).
Global evaluation of SIJ MRI in patients with back pain clinically suspected to have SpA is contextual because it takes into account both active and structural lesions of varying confidence when analysing T1SE and STIR sequences simultaneously. This global assessment consistently showed the highest specificity in a previous study of 187 cases and controls4 and in our study with two independent SpA inception cohorts. Therefore, crucial aspects for the development of lesion-based criteria are to what degree candidate lesion-based criteria resemble global assessment and whether lesion-based criteria attain the high specificity found with global assessment.
Which lesion types should be incorporated into lesion-based criteria? Relevant factors for selecting certain MRI lesions over others are their reproducibility, specificity and redundancy of information. In a previous study of 30 patients with AS and 30 patients with NSBP or healthy controls, erosion and BMO were detected on SIJ MRI with a comparable degree of reliability after calibration of four readers. Reader agreement by κ values was substantial for erosion (0.72) and BMO (0.61), and ICC values for affected SIJ quadrants were high for both erosion (0.79) and BMO (0.93).22 Despite a lesion frequency comparable to erosion, the reproducibility of SIJ fat infiltration was lower than with erosion and BMO with κ/ICC values of 0.55/0.71. A separate analysis of the present dataset, which focused on the diagnostic utility of SIJ fat infiltration, confirmed it had only modest reliability, with ICC values among six reader pairs of 0.59/0.75 for cohorts A/B.9 However, the main reason for the limited diagnostic utility of SIJ fat infiltration was its strong association with concomitant BMO and/or erosion, which highlights the mostly redundant information provided by fat infiltration compared with BMO and/or erosion. We therefore limited the analysis in this report to the combination of BMO and/or erosion, which consistently showed the highest reliability in our reader team across several studies. It remains unclear whether other structural lesions such as sclerosis or alterations of joint space width might enhance the performance of lesion-based criteria for a positive SIJ MRI. There are minimal controlled data for diagnostic utility, redundancy compared with other lesions and reliability for these two lesion types.
Our candidate criteria of a positive SIJ MRI are based on SIJ quadrants obtained by subdividing each SIJ into four upper and lower iliac and sacral quadrants. The diagnostic utility of ‘≥2 SIJ quadrants with BMO’ limited to the same slice or confined to the same quadrant on ≥2 consecutive slices was virtually identical to that of ‘≥2 SIJ quadrants with BMO’ without any restrictions—that is, ≥2 single quadrants with BMO also occurring in unrelated quadrants and slices. A descriptive quadrant-based analysis (data not shown) confirmed that non-SpA-related BMO in patients with NSBP and healthy controls was nearly always clustered on the same or on two consecutive slices, as is often observed in SpA. There was no relevant difference in diagnostic utility for lesion-based criteria including BMO with either ≥1 or ≥2 SIJ quadrants with erosion. We therefore favour combinations containing ≥2 SIJ quadrants with erosion to avoid overinterpretation of lesions where erosion is considered doubtful because of its presence in only one SIJ quadrant.
Strengths of our study are consistent results across two independent SpA inception cohorts which differed substantially in clinically relevant features such as symptom duration or disease activity reflecting clinical practice. Moreover, SIJ MRI scans were assessed by rheumatologists and radiologists from three international centres. The principal limitation, inherent to imaging studies in SpA, was the selection of rheumatologist expert opinion integrating the results of clinical and radiographic examination as the gold standard with the possibility of incorrect case assignments. However, clinician expert opinion has consistently been the accepted gold standard for deriving three classification criteria over the past two decades (Amor,6 ESSG,7 ASAS2 criteria) and was used for cohorts with early SpA such as DEvenir des Spondyloarthropathies Indifférenciées Récentes (DESIR).23 Another caveat in deriving lesion-based criteria is emerging evidence, that cut-off values may vary depending on the gold standard used, on recruitment strategies, and on demographic and clinical characteristics such as age, symptom duration and disease activity.8 Studies in other cohorts with early SpA are needed to determine which of the high priority candidate criteria identified in this analysis perform best across the spectrum of SpA seen in clinical practice. Moreover, predictive validation is needed but prospective data are often limited by high attrition rates for long-term follow-up, particularly in slowly evolving conditions such as SpA.
A challenge of imaging studies focusing on detailed assessment of MRI scans based on SIJ quadrants is the transfer of this knowledge to the broad community of radiologists who spend much of their time in the evaluation of spinal disorders, where SpA constitutes a small minority of routine practice. The ASAS definition3 may be appropriate for raising diagnostic suspicion of SpA when imaging is conducted in the appropriate clinical context. Our data, however, focus on the identification of a relatively homogeneous group of patients with definite SpA that can be included in genetic studies or clinical trials.
In conclusion, our study in patients with nr-axSpA, for whom symptom duration and disease activity varied considerably, reflecting clinical practice, showed that lesion-based criteria for a positive SIJ MRI that includes both erosion and/or BMO or includes ≥3 BMO quadrants provide the best classification of axial SpA. These criteria more closely approximate global assessment by expert readers, while definitions based on only two SIJ quadrants with BMO lack specificity. This approach reflects the contextual information provided by simultaneous assessment of T1SE and STIR sequences.
The authors thank the patients and the healthy volunteers for their participation; Tracey Clare, clinical research manager, Department of Radiology, University of Alberta, and Paul Filipow, data manager, Edmonton, Canada for coordinating the web-based MRI scoring module.
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:
Handling editor Tore K Kvien
Contributors KR, MØ, RGWL, SJP, UW, WPM and ZZ drafted the study design. SMC, UW and WPM acquired the clinical data. SJP, UW, VZ, and WPM were MRI readers. UW and KR performed the statistical analysis. All authors take responsibility for interpretation of the data. UW drafted the manuscript with important contributions from all authors. All authors read and approved the final manuscript. UW had full access to all data in the study and takes responsibility for its integrity.
Funding The Canadian Arthritis Society: National Research Initiative Award. Alberta Innovates Health Solutions. Walter L and Johanna Wolf Foundation, Zurich, Switzerland.
Competing interests WPM is a scientist of Alberta Innovates Health Solutions.
Ethics approval Ethics committee of the University of Alberta, Edmonton, Canada. Kantonale Ethikkommission Zurich, Zurich, Switzerland.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement This is the first analysis of this dataset to develop and compare candidate lesion-based criteria for a positive sacroiliac joint MRI in non-radiographic axial spondyloarthritis. The dataset is available for all nine manuscript authors, but not for others. A scientific abstract from these data has been submitted to EULAR 2014. There are no further publications or public presentations of these data.