Defining and validating a high specificity MRI cut-off for the radiographic part of the modified New York criteria in patients with axial spondyloarthritis

Abstract

Background

According to the modified New York criteria (mNYc), moderate-severe sacroiliac joint (SIJ) structural damage on radiography is necessary for classifying radiographic axial spondyloarthritis (r-axSpA). In contrast to radiography, MRI provides no ionizing radiation, higher sensitivity for structural damage, better inter-reader reliability, and is gradually replacing radiography. Therefore, we aimed to develop and validate a high specificity MRI cut-off to the radiographic component of the mNYc (rad-mNYc), defined as MRI findings that correspond to an mNY-positive radiograph.

Methods

Patients diagnosed with axSpA with available SIJ MRI and radiograph from five European clinical registries in the EuroSpA Collaboration were included. MRIs were read according to the SPARCC structural score (based on 5 predefined or all slices) and radiographs the rad-mNYc. Cut-offs with 95% specificity for rad-mNYc-positivity for MRI sum scores for individual lesions and for an MRI SIJ hierarchical Composite Structural Damage Score (CSDS₅/CSDS_all based on erosion, backfill and ankylosis in 5/all slices) and ROC curves were calculated. Internal validation by bootstrapping was performed.

Results

Overall, 640 patients were included; 296 (46%) were rad-mNYc positive. The 95%-specificity cut-off was ≥ 30 for the composite score CSDS₅ predicting rad-mNYc (specificity 0.95, sensitivity 0.39, area-under-curve 0.83, accuracy with mNYc 69.2%. Internal validation showed moderate accuracy and high positive predictive value. Cut-offs for sclerosis, erosion, backfill and ankylosis were ≥ 9, ≥12, ≥ 5, and ≥ 1, respectively, with sensitivities all ≤ 0.21 except ankylosis 0.34 and accuracy ~ 60%, except ankylosis 67%. Cut-offs using all MRI-slices performed slightly worse or similarly.

Conclusions

A composite MRI-score based on assessing erosion, backfill and ankylosis for five predefined semicoronal SIJ slices (CSDS₅) was the optimal high specificity MRI cut-off to rad-mNYc positivity, being potentially usable in both research and clinical settings, with different thresholds depending on whether sensitivity or specificity is prioritized.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13075-026-03865-z.

Introduction

Axial spondyloarthritis (axSpA) is an inflammatory disease affecting the spine and the sacroiliac joints (SIJs). According to the Assessment of Spondyloarthritis Society (ASAS) working group, axSpA can be divided into radiographic axSpA (r-axSpA) which is largely similar to ankylosing spondylitis (AS) defined by the modified New York (mNY) criteria, and non-radiographic axSpA (nr-axSpA) [1–4].

Fulfillment of the radiographic part of the mNY criteria (rad-mNYc) is required in the definitions of both AS and r-axSpA and is defined as either at least unilateral grade 3 changes (unequivocal abnormality) or bilateral grade 2 changes (minimum abnormality) [1]. SIJ magnetic resonance imaging (MRI) allows assessment of both early inflammatory changes and structural changes in the SIJs in one examination and without ionizing radiation and has therefore become an important part of the ASAS classification criteria for axSpA. However, for classification of axSpA according to the ASAS classification criteria’s “imaging arm”, active sacroiliitis on MRI defined as “inflammation of the SIJs highly suggestive of SpA“ is required [5]. Therefore, structural changes alone cannot fulfil the definition of a positive MRI according to the ASAS criteria, despite that it has been documented that MRI can provide a more sensitive and equally specific assessment of structural lesions in the SIJs, as compared to radiography, when computed tomography (CT) is used as the standard reference [6].

Despite the high performance of MRI, radiography of the SIJs is still frequently used in axSpA partially due to availability issues and that it is still recommended as the first imaging modality for diagnosing axSpA by the ASAS and EULAR groups [7, 8]. Disadvantages of radiography are (1) it does not visualize inflammation, (2) it causes ionizing radiation, and (3) the inter-reader agreement for scoring SIJs is at best moderate [6, 9–13]. Therefore, the detection of unambiguous structural SpA findings on MRI, regardless of findings of active inflammation, could be a useful substitute for radiographically detectable damage both for diagnosis and classification of AS/r-axSpA. Thus, as further described below, it would be relevant to gather knowledge about what lesions are seen on MRI when an SIJ radiograph is rad-mNYc-positive, i.e. to define a high specificity MRI cut-off to the radiographic part of the mNY criteria.

Defining a high specificity MRI cut-off to the rad-mNYc is relevant for academic, diagnostic, and therapeutic reasons. Firstly, it will provide an understanding of what the different grades of the rad-mNYc represent, when assessed by means of a more sensitive and accurate imaging modality. Secondly, the distinction between nr- and r-axSpA is important for patients because in some countries certain treatments, such as physical therapy and/or some biological therapies, are only offered to/approved for patients with AS/r-axSpA but not with nr-axSpA [14]. Thirdly, the definition of a high specificity MRI cut-off would make the use of standard radiography more of a choice due to availability and cost and would therefore be unnecessary in some countries and clinics where MRI is readily available, thereby minimizing patients’ radiation exposure. Finally, patients with AS/r-axSpA represent an extensively studied and well-defined population [15, 16] and therefore having AS/r-axSpA has considerable prognostic value including long-term outcomes and the development of comorbidities. For future routine care patients, where MRI but not radiography may be used, a high specificity MRI cut-off could allow the clinician to use knowledge from previous AS/r-axSpA cohorts to guide clinical decisions.

Therefore, the aim of this study was, in routine care patients with a clinical diagnosis of axSpA, to develop and validate a high specificity MRI cut-off to fulfilment of the rad-mNYc, defined as MRI findings with ≥ 95% specificity corresponding to a rad-mNYc-positive radiograph.

Methods

Study design

The study involved collection and central reading of corresponding MRIs and radiographs of the SIJs, acquired as part of routine clinical practice in patients with axSpA, in five European countries participating in the European Spondyloarthritis Research Collaboration Network (EuroSpA) of clinical registries in SpA (Czech Republic, ATTRA registry; Denmark, DANBIO registry; Iceland, ICEBIO registry; Slovenia, biorx.si registry, Switzerland, SCQM registry). Detailed information on the registries can be found in Supplementary Table S1. The imaging part of the EuroSpA collaboration is explained in detail by Hadsbjerg et al. [17]. Corresponding clinical data were extracted from the clinical registries, pseudonymized and linked at the individual patient level.

Image collection

The patients in the study were selected based on image availability. For inclusion in the study, the following was required: a registered diagnosis of SpA (i.e. both axSpA and PsA) in one of the clinical registries, age ≥ 18 years at initial diagnosis and availability of: MRI of the SIJs (including semi-coronal T1-weighted images and semi-coronal short tau inversion recovery (STIR) or T2-weighted fat-saturated images), and radiograph of the SIJs within ± 6 months of the MRI. The MRIs and radiographs were pseudonymized in each country before transfer to the EuroSpA Coordinating Centre in Copenhagen via a secure server.

Assessment of MRIs

MRIs of the SIJs were read centrally by three readers: an experienced musculoskeletal radiologist, an experienced reader (radiologist or rheumatologist) with more than 5 years’ experience in assessment of MRI of the SIJs and a centrally trained less experienced reader (novice). All readers calibrated before reading (for details, see [18]).

The MRIs were read blinded to other imaging and clinical information (except age and sex). The assessment consisted of two parts: a detailed scoring and a questionnaire. The detailed scoring was done according to the Spondyloarthritis Research Consortium of Canada MRI Sacroiliac Joint Structural Score (SPARCC SSS) [19, 20], however, only for erosion, backfill and ankylosis, and in addition sclerosis > 1 cm in depth was scored. We added sclerosis because it is visualized on radiography, and we considered it relevant to investigate if sclerosis on MRI would improve/be of importance in the MRI model. Similarly, we did not use fat lesions, since they cannot be visualized by radiography. The questionnaire contained questions on whether (1) the overall findings on the MRI of the SIJs were indicative of SpA, (2) the MRI fulfilled the ASAS definition of a positive MRI of the SIJs [2, 5, 21], (3) there were typical active inflammatory lesions indicative of axSpA, and (4) there were typical structural changes indicative of SpA. In relation to each of these questions, the reader indicated the level of confidence on a scale from − 5 (definitely no) to + 5 (definitely yes). In addition, lesions defined by the ASAS imaging working group [2, 5, 21] were assessed per left/right ilium/sacrum. Alternative diagnoses were also registered, with a fixed vocabulary for the most common differential diagnoses.

The SPARCC scorings were done from slices with at least 1 cm visible cartilaginous joint anteriorly to slices with at least 1 cm visible joint posteriorly for both inflammatory and structural lesions.

The three tasks were distributed as follows: two readers (one musculoskeletal radiologist and one experienced reader) answered the questionnaire and two readers (one experienced reader (the same who answered the questionnaire) and one novice) did the detailed SPARCC scoring.

Assessment of radiographs

Radiographs were assessed by two readers blinded to other imaging and clinical information (except age and sex). All readers were experienced in scoring according to the rad-mNYc and calibrated before scoring (for details, see [22]).

Radiographs of the SIJs were assessed by applying the rad-mNYc (grade 0–4 for each SIJ) [4]. The changes on the radiographs were graded from 0 to 4: grade 0 is a normal SIJ; grade 1 is suspicious changes; grade 2 is minimum abnormality (small localized areas with erosion or sclerosis, without alteration in the joint width); grade 3 is unequivocal abnormality (moderate or advanced sacroiliitis with erosions, evidence of sclerosis, widening, narrowing or partial ankylosis); grade 4 is severe abnormality (total ankylosis) [1]. Moreover, the specific features in the criteria (erosion, sclerosis, joint space widening (JSW), joint space narrowing (JSN) and ankylosis) were assessed. The readers judged whether the overall changes on the radiograph were indicative of SpA or not, and the level of confidence was registered in the same way as for MRIs, as described above.

Adjudication

MRIs were adjudicated if the readers disagreed on (1) whether the MRI was indicative of SpA or not, (2) fulfilment of the ASAS MRI criteria, and/or (3) if at least one reader had registered either fracture, neoplasia, or infection as a differential diagnosis.

Radiographs were adjudicated if the readers disagreed on whether the radiograph was positive/negative for the rad-mNYc.

The adjudicator was an expert musculoskeletal radiologist (member of the ASAS/EULAR MRI group) who was blinded to assessments by previous readers.

MRI sum scores

Sum scores for the lesions on MRI (sclerosis, erosion, backfill and ankylosis) were calculated for both “five slices” and “all slices”. “Five slices” were the five slices defined in the SPARCC MRI SSS, i.e. the transitional slice and four consecutive slices anteriorly [20]. “All slices” were the available slices with ≥ 1 cm of the cartilaginous part of the joint visible anteriorly to ≥ 1 cm visible joint posteriorly. The sum scores were calculated as the sum of the lesion in each slice and these sums were added for the five/all slices. Sclerosis and erosion were assessed per quadrant per joint, i.e., four quadrants (upper iliac, upper sacral, lower iliac and lower sacral) in an SIJ equals a maximum score of four per joint per slice. Backfill and ankylosis were scored per joint half (upper and lower) per slice giving a maximum score of two per joint per slice. Total score ranges per patient for “five slices” were 0–40 for sclerosis and erosion and 0–20 for backfill and ankylosis.

A composite sum score defined by Wetterslev et al. [23] was also calculated for “five slices” and “all slices”. The composite score used was the hierarchical Composite Structural Damage Score (CSDS_hierarchical), in which later stages overweigh the earlier stages in each individual joint half with the following formula: CSDS_hierarchical = erosion*1 < backfill*4 < ankylosis*6, where the sign “<” indicates a hierarchical order, meaning that erosion was not included in the calculation if backfill was present in the same joint half, and erosion and backfill were not included in the calculation if ankylosis was present in the joint half. For erosion to be part of the score, neither backfill nor ankylosis could be present. The score is calculated per joint half (i.e., upper and lower) and then summed for the entire joint and slices (five and all slices). The CSDS_hierarchical was calculated for “five slices” (CSDS₅) and “all slices” CSDS_all). The total sum score range for CSDS₅ was 0-120.

Statistics

For the MRI SPARCC scores, the experienced readers’ scores were used for the calculations described below. For the radiographs, only one reader’s scoring was used since readers were considered equally competent, the scores used were based on random selection unless the images were adjudicated. If the images were adjudicated, the adjudicator’s score was always used.

Descriptive statistics were used to assess the structural lesions on MRI and radiography, rad-mNYc positivity on radiographs and whether the MRI was indicative of SpA as well as clinical data.

Agreement between readers of radiographs was assessed with Cohen’s Kappa, quadratically weighted. Agreement between novices and experts on reading MRIs according to the SPARCC scoring system was assessed for sclerosis, erosion, backfill and ankylosis using intraclass correlation coefficient (ICC), two-way random model.

Cut-offs for the composite scores, CSDS₅ and CSDS_all, were calculated against rad-mNYc positivity (cut-off on patient level), rad-mNYc grade ≥ 3 (cut-off on right/left joint), and bilateral rad-mNYc grade ≥ 2 according to the 95% specificity method which maximizes sensitivity given at least 95% specificity. By applying this method, we aimed to avoid patients who are radiographically mNY negative being misclassified as positive by MRI-based estimated rad-mNYc grading. The cut-offs were calculated for both five slices and for all slices, as defined above. For each cut-off sensitivity, specificity, area under curve (AUC), accuracy (i.e. the proportion of correctly classified patients among the total number of patients), positive predictive value (PPV), and negative predictive value (NPV) were calculated.

As additional analyses, cut-offs of the sum score for the different structural lesions (i.e. sclerosis, erosion, backfill and ankylosis) were calculated. Moreover, cut-offs for the composite score were developed according to alternative ROC methods, i.e. 90% specificity, Youden index and the point closest-to-(0,1) (or ROC01). This Youden method defines the optimal cut-off as the point maximizing the difference between sensitivity and 1-specificity while the point closest-to-(0,1) defines the optimal cut-off as the point minimizing the distance between the ROC curve and the (0,1) point.

Internal validation based on nonparametric bootstrapping was performed to assess the performance of the selected cut-offs. Given that data include N patients, a bootstrap sample was created by drawing N patients with replacement from the original data. The optimal cut-offs were estimated in the bootstrap sample and validated in the rest of the patients from the original data. This process was repeated 1,000 times and finally the accuracy, PPV, and NPV of the selected optimal cut-offs were estimated across all bootstrap samples.

As additional validation of all cut-offs described above, accuracy, PPV, and NPV were calculated using the novices’ MRI SPARCC scores.

Results

Overall, 640 patients with corresponding MRI and radiograph of the SIJs were included; mean age was 38 years (± 11 years) and 60% were male (see Supplementary Figure S1 for flowchart of patient inclusion). For the MRIs, 69% were overall indicative of SpA and for the radiographs, 46% were rad-mNYc positive (Table 1). Descriptive statistics on the lesions on radiographs and MRIs can be found in Table 1. The median time between MRI and radiograph was 33 days (mean 48 days) with a minimum of 0 days, and a maximum of 183 days. A sensitivity analysis using only images with less than three months between MRI and radiograph (n = 525) showed no difference in cut-offs or sensitivity (ROC curves are presented in Supplementary Figure S2). The interreader agreement as measured by Kappa was 0.55 for rad-mNYc grading (grade 0–4) for right SIJ and 0.56 for left SIJ. For MRI, the interreader agreement for five slices, as measured by ICC, was 0.55 for sclerosis, 0.30 for erosion, 0.67 for backfill and 0.82 for ankylosis; ICCs for all slices were similar (see Supplementary Table S2).

Table 1.

Descriptive statistics on clinical and imaging data

Clinical data from registries
	No. available	Mean (SD); n (%)
Age	640	38 (11)
Sex, male	640	386 (60%)
Diagnosis, axSpA	640	627 (98%)
Disease duration	577	3.52 (6.86)
BASDAI	534	4.66 (2.45)
HAQ	213	0.56 (0.57)
ASDAS-CRP	475	2.91 (1.12)
Tender joints, 28	80	0.95 (2.26)
Swollen joints, 28	238	0.32 (1.37)
Tender joints, 68	39	4.72 (5.84)
Swollen joints, 66	23	1.74 (2.20)
		Positive		Negative
Fulfilment of mNY criteria, n (%)	264	191 (72%)		73 (28%)
Fulfilment of ASAS criteria (in registries), n (%)	503	492 (98%)		11 (2.2%)
Fulfilment of CASPAR criteria (in registries), n (%)	40	40 (100%)		0 (0%)
HLA-B27, n (%)	603	432 (72%)		171 (28%)
Radiographs, central reading, n (%)
	No. available	Positive		Negative
Radiographic part of mNY criteria	640	296 (46%)		344 (54%)
Sacroiliitis grade	640	Right		Left
0		164 (26%)		180 (28%)
1		151 (24%)		143 (22%)
2		134 (21%)		129 (20%)
3		130 (20%)		143 (22%)
4		57 (8.9%)		44 (6.9%)
		Present		Not present
		Right	Left	Right	Left
Sclerosis, ilium	639	403 (63%)	391 (61%)	236 (37%)	248 (39%)
Sclerosis, sacrum	639	274 (43%)	273 (43%)	365 (57%)	366 (57%)
Erosion, ilium	639	258 (40%)	381 (60%)	251 (39%)	388 (61%)
Erosion, sacrum	639	136 (21%)	147 (23%)	503 (79%)	492 (77%)
Joint space widening	639	83 (13%)	94 (15%)	556 (87%)	545 (85%)
Joint space narrowing	639	102 (16%)	107 (17%)	537 (84%)	532 (83%)
Ankylosis	639	95 (15%)	91 (14%)	544 (85%)	548 (86%)
MRI, central reading
MRI overall indicative of SpA, n (%)	440 (69%)
Patient level sum scores, mean (min-max)	No. available	Five slices		All slices
Sclerosis	640	1.8 (0–24)		2.8 (0–41)
Erosion	640	3.8 (0–33)		5.0 (0–45)
Backfill	640	1.3 (0–17)		1.6 (0–24)
Ankylosis	640	1.7 (0–20)		2.0 (0–34)
Composite score	640	18 (0-120)		25 (0-204)
Joint level sum score, mean (min-max)		Right	Left	Right	Left
Sclerosis, sacrum	640	0.06 (0–7)	0.04 (0–4)	0.12 (0–10)	0.08 (0–5)
Sclerosis, ilium	640	0.87 (0–10)	0.84 (0–10)	1.29 (0–21)	1.27 (0–20)
Erosion, sacrum	640	0.52 (0–9)	0.42 (0–7)	0.68 (0–11)	0.55 (0–9)
Erosion, ilium	640	1.44 (0–10)	1.47 (0–10)	1.94 (0–15)	2.0 (0–15)
Backfill	640	0.66 (0–10)	0.65 (0–10)	0.85 (0–14)	0.78 (0–14)
Ankylosis	640	0.82 (0–10)	0.84 (0–10)	1.2 (0–18)	1.2 (0–18)
Composite score	640	9 (0–60)	9 (0–60)	13 (0-108)	12 (0-108)

Five slices are defined in the SPARCC structural score as the transitional slice and 4 consecutive slices anteriorly. All slices are defined as all available slices from 1 cm visible cartilaginous joint posteriorly to 1 cm anteriorly. The composite score used was the CSDS_hierarchical, in which later stag-es overweigh the earlier stages with the following formula: CSDS_hierarchical = erosion*1 < backfill*4 < ankylosis*6, where the “<” indicates a hierarchical order, meaning that erosion was not included in the calculation if backfill was present in the same joint half, and erosion and backfill were not included in the calculation if ankylosis was present in the joint half

Abbreviations: SD standard deviation, axSpA axial spondyloarthritis, ASAS the Assessment of SpondyloArthritis international Society, HLA-B27 human leukocyte antigen B27, mNY modified New York, MRI magnetic resonance imaging. CSDS composite structural damage scores, ASDAS-CRP Ankylosing spondylitis disease activity score with C-reactive protein, BASDAI Bath Ankylos-ing Spondylitis Disease Activity Index, HAQ Health Assessment Questionnaire

The mean number of slices needed to cover the entire cartilaginous part of the joint was 9 (± 1.65).

Composite MRI scores versus radiographic assessment

The distribution of the composite MRI scores according to rad-mNYc positivity and the different rad-mNYc grades (0-4) is shown in Fig. 1.

Fig. 1.

Distribution of MRI composite sum score divided on radiographic mNY positivity and radiographic mNY grade for five (CSDS₅) and all slices (CSDS_all) on patient level (A), for the right (B) and for the left joint (C). “Five slices” are defined in the SPARCC structural score as the transitional slice and 4 consecutive slices anteriorly. “All slices” are defined as all available slices from 1 cm of visible cartilaginous joint posteriorly to 1 cm anteriorly. The composite score used was the CSDS_hierarchical, in which later stages overweigh the earlier stages with the following formula: CSDS_hierarchical = erosion*1 < backfill*4 < ankylosis*6, where the “<” indicates a hierarchical order, meaning that erosion was not included in the calculation if backfill was present in the same joint half, and erosion and backfill were not included in the calculation if ankylosis was present in the joint half. NAs in mNY grade are due to some sacroiliac joints not being scoreable. However, for all patients at least one joint was scoreable and there were no patients with unilateral grade 0–2 and NA in the other joint, therefore assessing mNY positivity was possible in all patients. Abbreviations: mNY: modified New York. CSDS: composite structural damage scores

Cut-offs for CSDS5

For the composite sum score per patient for five slices (CSDS₅) versus the rad-mNYc, the MRI cut-off to achieve 95% specificity for fulfillment of the rad-mNYc was a CSDS₅ ≥30, which gave a sensitivity of 0.39. When this cut-off was applied, the accuracy between MRI-estimated and radiography-observed presence/absence of rad-mNYc positivity was 69.2%, whereas AUC was 0.83 (Table 2). ROC curves for five MRI slices are shown in Fig. 2.

Table 2.

Cut-off values for MRI composite sum scores for fulfilment of the radiographic part of the mNY criteria (per patient), mNY grade ≥ 3 and bilateral mNY grade ≥ 2 (per patient) with sensitivity, specificity and validation analyses

	Cut-off, ≥	Specificity	Sensitivity	AUC	Accuracyexperts	PPVexperts	NPVexperts	Accuracynovices	PPVnovices	NPVnovices	Accuracyboot (%)	PPVboot (%)	NPVboot (%)
Cut-offs with ≥ 95% specificity
Against mNY positivity, per patient
CSDS5	30	0.95	0.39	0.83 (0.80–0.86)	443/640 (69.2%)	116/133 (87.2%)	327/507 (64.5%)	455/640 (71.1%)	118/125 (94.4%)	337/515 (65.4%)	69.7 (63.0-73.4)	87.6 (83.2–89.5)	65.0 (58.3–69.2)
CSDSall	38	0.95	0.39	0.83 (0.80–0.86)	443/640 (69.2%)	116/133 (87.2%)	327/507 (64.5%)	456/640 (71.2%)	124/136 (91.2%)	332/504 (65.9%)	69.3 (63.2–73.5)	87.4 (83.5–89.8)	64.6 (58.6–69.4)
Against mNY grade ≥ 3, per SIJ
CSDS5, right SIJ	21	0.95	0.40	0.82 (0.78–0.85)	507/640 (79.2%)	80/102 (78.4%)	427/538 (79.4%)	509/640 (79.5%)	75/90 (83.3%)	434/550 (78.9%)	79.0 (75.9–82.3)	78.4 (72.2–82.7)	79.3 (75.7–83.1)
CSDS5, left SIJ	20	0.95	0.39	0.81 (0.78–0.85)	505/640 (78.9%)	75/97 (77.3%)	430/543 (79.2%)	508/640 (79.4%)	76/96 (79.2%)	432/544 (79.4%)	78.9 (75.8–81.7)	77.4 (70.9–81.9)	79.2 (75.6–82.8)
CSDSall, right SIJ	30	0.95	0.40	0.82 (0.78–0.85)	501/640 (78.3%)	74/96 (77.1%)	427/544 (78.5%)	505/640 (78.9%)	69/82 (84.1%)	436/558 (78.1%)	78.2 (74.8–81.2)	77.4 (70.9–81.6)	78.5 (74.8–82.2)
CSDSall, left SIJ	25	0.95	0.40	0.82 (0.78–0.85)	514/640 (80.3%)	83/104 (79.8%)	431/536 (80.4%)	512/640 (80.0%)	84/108 (77.8%)	428/532 (80.5%)	80.1 (77.2–84.1)	79.4 (73.7–85.5)	80.3 (77.1–84.6)
Against bilateral mNY grade ≥ 2, per patient
CSDS5	32	0.95	0.40	0.83 (0.79–0.86)	455/640 (71.1%)	110/127 (86.6%)	345/513 (67.3%)	458/640 (71.6%)	105/114 (92.1%)	353/526 (67.1%)	71.1 (65.8–76.2)	86.4 (82.7–90.2)	67.4 (61.8–72.3)
CSDSall	43	0.95	0.37	0.81 (0.79–0.86)	449/640 (70.2%)	104/121 (86.0%)	345/519 (66.5%)	457/640 (71.4%)	106/117 (90.6%)	351/523 (67.1%)	70.2 (64.7–74.2)	85.8 (81.8–88.7)	66.6 (61.2–71.1)
Alternative cut-offs
90% specificity
CSDS5	16	0.90	0.57	0.83 (0.80–0.86)	480/640 (75.0%)	169/202 (83.7%)	311/438 (71.0%)	494/640 (77.2%)	194/238 (81.5%)	300/402 (74.6%)	75.1 (69.6–79.8)	83.6 (80.3–86.9)	71.3 (64.6–76.9)
CSDSall	21	0.90	0.56	0.83 (0.80–0.86)	475/640 (74.2%)	165/199 (82.9%)	310/441 (70.3%)	494/640 (77.2%)	192/234 (82.1%)	302/406 (74.4%)	74.1 (68.6–78.3)	74.1 (68.6–78.3)	70.2 (64.0-75.3)
Youden
CSDS5	8	0.81	0.74	0.83 (0.80–0.86)	497/640 (77.7%)	219/285 (76.8%)	278/355 (78.3%)	485/640 (75.8%)	239/337 (70.9%)	246/303 (81.2%)	77.0 (73.3–79.7)	76.5 (67.9–82.8)	77.8 (71.6–82.6)
CSDSall	12	0.85	0.78	0.83 (0.80–0.86)	499/640 (78.0%)	208/261 (79.7%)	291/379 (76.8%)	488/640 (76.2%)	229/314 (72.9%)	259/326 (79.4%)	77.2 (73.9–80.1)	77.3 (72.5–87.2)	77.4 (69.5–80.8)
ROC01
CSDS5	7	0.78	0.77	0.83 (0.80–0.86)	495/640 (77.3%)	228/305 (74.8%)	267/335 (79.7%)	475/640 (74.2%)	240/349 (68.8%)	235/291 (80.8%)	77.0 (73.2–79.8)	74.7 (67.9–79.3)	79.2 (74.5–83.7)
CSDSall	7	0.75	0.80	0.83 (0.80–0.86)	493/640 (77.0%)	236/323 (73.1%)	257/317 (81.1%)	466/640 (72.8%)	246/370 (66.5%)	220/270 (81.5%)	76.7 (72.8–79.4)	73.8 (64.1–77.7)	79.8 (76.2–86.0)

The composite MRI score was against whether the radiographic part of the mNY criteria was positive or not. For right and left it was done against whether the radiograph was mNY grade ≥ 3 or not. The composite score used was the CSDS_hierarchical, in which later stages overweigh the earlier stages with the following formula: CSDS_hierarchical = erosion*1 < backfill*4 < ankylosis*6, where the “<” indicated a hierarchical order, meaning that erosion was not included in the calculation if backfill was present in the same joint half, and erosion and backfill were not included in the calculation if ankylosis was present in the joint half. All slices were defined as all available slices from 1 cm of visible cartilaginous joint posteriorly to 1 cm anteriorly, five slices were defined in the SPARCC structural score as the transitional slice and 4 consecutive slices anteriorly. Values in the six right columns are accuracy and precision for the sample (expert readings) used to calculate the cut-offs (“experts”), for the novices using the calculated cut-off (“novices”) and for the boot-strap validation (“boot”). For the boot-strap (“boot”) accuracy and precision values are the mean (confidence interval) of the 1000 repetitions

Abbreviations: MRI magnetic resonance imaging, AUC area under the curve, mNY modified New York, CSDS Composite Structural Damage Score, SIJ sacroiliac joint.

Fig. 2.

ROC curves for MRI composite scores against radiographic mNY positivity with marking of the cut-off at a specificity of 95% and AUC. The composite score used was the CSDS_hierarchical, in which later stages overweigh the earlier stages with the following formula: CSDS_hierarchical = erosion*1 < backfill*4 < ankylosis*6, where the “<” indicates a hierarchical order, meaning that erosion was not included in the calculation if backfill was present in the same joint half, and erosion and backfill were not included in the calculation if ankylosis was present in the joint half. All slices are defined as all available slices from 1 cm of visible cartilaginous joint posteriorly to 1 cm anteriorly, five slices are defined in the SPARCC structural score as the transitional slice and 4 consecutive slices anteriorly. Abbreviations: MRI: magnetic resonance imaging. mNY: modified New York. SIJ: sacroiliac joint. AUC: area under the curve. CSDS: composite structural damage scores. SPARCC: Spondyloarthritis Research Consortium of Canada

CSDS₅ cut-offs for right/left SIJ, i.e. per joint, for achieving 95% specificity against rad-mNYc grade ≥ 3 were 21/20. For both left and right sides, the predefined specificity of 0.95 was met, and the sensitivity was 0.40/0.39 for right and left, respectively. Accuracy was 79% for both sides, and AUC was 0.81–0.82 (Table 2). ROCs are shown in Fig. 2.

Using the CSDS₅ cut-offs for having a rad-mNYc grade ≥ 3 (i.e. a composite sum score of at least 21/20 for either right and/or left SIJ in five slices) the accuracy regarding predicting a positive radiograph (rad-mNYc grade ≥ 3) was 443/640 (69.2%). If we use the cut-offs for having a rad-mNYc grade ≥ 3 or bilateral ≥ 2a (i.e. a composite sum score of at least 21/20 for either right and/or left SIJ in five slices or a composite sum score of at least 32 for the entire patient) for predicting rad-mNYc positivity, the accuracy was 70.2%.

A confusion matrix for the main thresholds and alternative cut-offs can be found in Supplementary Table S3.

Examples of MRIs with corresponding radiographs showing different CSDS₅ cut-offs are presented in Fig. 3.

Fig. 3.

Examples of patients with different CSDS₅ cut-offs showing both radiographs and all five slices from the MRI. Abbreviations: MRI: magnetic resonance imaging. mNY: modified New York. CSDS₅: composite structural damage score based on five slices

Cut-offs for CSDS_all

For all slices, the CSDS_all cut-off for rad-mNYc positivity achieving a 95% specificity was ≥ 38 with a sensitivity of 0.39 and an accuracy of 69.2% (Table 2).

CSDS_all cut-offs for right/left for achieving 95% specificity SIJ against rad-mNYc grade ≥ 3 were 30/25 for all slices with the predefined specificity of 0.95 being met and a sensitivity of 0.40 for both sides. Accuracy was 78% and 80% for right and left and AUC was 0.82 for both sides (Table 2). ROC curves can be found in Fig. 2.

Using the CSDS_all cut-offs for having a rad-mNYc grade ≥ 3 (i.e. a composite sum score of at least 28/24 for either right and/or left SIJ in five slices) the accuracy regarding predicting a positive radiograph (rad-mNYc grade ≥ 3) was 441/640 (68.9%). If we use the cut-offs for having a rad-mNYc grade ≥ 3 or bilateral ≥2a (i.e. a composite sum score of at least 30/25 for either right and/or left SIJ in five slices or a composite sum score of at least 43 for the entire patient) for predicting rad-mNYc positivity the accuracy was 69.2%.

Cut-offs for the individual lesions

The distribution of the structural MRI lesions according to rad-mNYc positivity and the different rad-mNYc grades (0-4) is shown in Fig. 4 (for five slices) and Supplementary Figure S3 (for all slices).

Fig. 4.

Distribution of the individual lesions (sclerosis, erosion, backfill and ankylosis) according to mNY positivity (patient level) and mNY grade (joint level) for five slices. Five slices are defined in the SPARCC structural score as the transitional slice and 4 consecutive slices anteriorly. NAs in mNY grade are due to some sacroiliac joints not being scoreable. However, for all patients at least one joint was scoreable and there were no patients with unilateral grade 0-2 and NA in the other joint. Therefore assessing mNY positivity was possible in all patients. Abbreviations: mNY: modified New York, MRI: magnetic resonance imaging. SPARCC: Spondyloar-thritis Research Consortium of Canada. NA: not applicable

Cut-offs for the different lesion scores using 5 slices, including specificity and sensitivity, are described in Table 3. For erosion, the cut-offs were 9 (joint level) and 12 (patient level) with a specificity from 0.95 to 0.97, sensitivity from 0.06 to 0.17 and AUC from 0.58 to 0.64. For sclerosis, the cut-offs were 6 (joint level) and 9 (patient level), with specificity from 0.95 to 0.96, sensitivity from 0.06 to 0.10, and AUC from 0.52 to 0.57. For backfill, the cut-off was 5 (both joint and patient level), specificity from 0.95 to 0.97, sensitivity from 0.12 to 0.21, and AUC from 0.59 to 0.64. For ankylosis, the cut-off was 1 (both joint and patient level). Specificity for the ankylosis cut-off ranged from 0.95 to 0.97, sensitivity from 0.34 to 0.42 and AUC from 0.66 to 0.69. Similar values for all slices are provided in Supplementary Table S2.

Table 3.

Cut-off values for MRI lesion sum scores (five slices) for fulfilment of the radiographic part of the mNY criteria (per patient) respectively mNY grade ≥ 3 (per joint) with sensitivity, specificity and validation analyses

	Cut-off, ≥	Specificity	Sensitivity	AUC	Accuracyexperts	PPVexperts	NPVexperts	Accuracynovices	PPVnovices	NPVnovices	AccuracyBoot (%)	PPVBoot (%)	NPVBoot (%)
Against mNY positivity, per patient
Sclerosis	9	0.96	0.1	0.57	361/640 (56.4%)	31/45 (68.9%)	330/595 (55.5%)	346/640 (54.1%)	19/36 (52.8%)	327/604 (54.1%)	56.3 (52.1–60.6)	68.2 (58.9–90.6)	55.5 (51.5–59.4)
Erosion	12	0.95	0.17	0.63	378/640 (59.1%)	51/68 (75.0%)	327/572 (57.2%)	386/640 (60.3%)	74/106 (69.8%)	312/534 (58.4%)	58.4 (54.2–64.5)	73.8 (65.5–94.8)	56.8 (52.7–61.9)
Backfill	5	0.95	0.21	0.64	390/640 (60.9%)	63/80 (78.8%)	327/560 (58.4%)	395/640 (61.7%)	59/67 (88.1%)	336/573 (58.6%)	60.4 (55.4–66.2)	78.6 (70.9–88.8)	58.1 (53.6–63.0)
Ankylosis	1	0.97	0.34	0.66	436/640 (68.1%)	101/110 (91.8%)	335/530 (63.2%)	428/640 (66.9%)	98/112 (87.5%)	330/528 (62.5%)	68.1 (64.5–71.9)	91.8 (87.2–97.4)	63.2 (59.1–67.4)
Against mNY grade ≥ 3, per SIJ
Sclerosis right	6	0.96	0.07	0.55	439/640 (68.6%)	14/32 (43.8%)	431/608 (70.9%)	448/640 (70.0%)	15/31 (48.4%)	433/609 (71.1%)	69.5 (65.9–72.9)	NaN (NaN-NaN)	71.0 (67.2–74.7)
Sclerosis left	6	0.96	0.07	0.53	447/640 (69.2%)	13/31 (41.9%)	434/609 (71.3%)	440/640 (68.8%)	6/24 (25.0%)	434/616 (70.5%)	69.9 (66.4–73.3)	41.7 (24.8–60.9)	71.4 (67.8–74.9)
Erosion right	9	0.96	0.1	0.58	452/640 (70.6%)	20/37 (54.1%)	432/603 (71.6%)	439/640 (68.6%)	22/54 (40.7%)	417/586 (71.2%)	70.6 (67.1–74.0)	NaN (NaN-NaN)	71.8 (68.1–75.3)
Erosion left	9	0.97	0.06	0.58	448/640 (70%)	11/26 (42.3%)	437/614 (71.2%)	447/640 (69.8%)	21/47 (44.7%)	426/593 (71.8%)	70.0 (66.5–73.3)	42.6 (24.9–58.6)	71.4 (67.7–74.6)
Backfill right	5	0.96	0.14	0.59	459/640 (71.7%)	27/44 (61.4%)	432/596 (72.5%)	457/640 (71.4%)	18/28 (64.3%)	439/612 (71.7%)	71.6 (68.3–74.9)	60.6 (49.7–74.4)	72.6 (68.8–76.2)
Backfill left	5	0.97	0.12	0.6	463/640 (72.3%)	23/35 (65.7%)	440/605 (72.7%)	462/640 (72.2%)	19/28 (67.9%)	443/612 (72.4%)	72.0 (69.0-75.7)	61.2 (51.4–80.6)	72.8 (69.2–76.2)
Ankylosis right	1	0.95	0.39	0.68	503/640 (78.6%)	75/96 (78.1%)	428/544 (78.7%)	494/640 (77.2%)	70/95 (73.7%)	424/545 (77.8%)	78.6 (75.5–81.7)	79.0 (70.0–83.0)	78.6 (75.3–82.5)
Ankylosis left	1	0.96	0.41	0.69	509/640 (79.5%)	77/97 (79.4%)	432/543 (79.6%)	502/640 (78.4%)	72/94 (76.6%)	430/546 (78.8%)	79.3 (76.4–83.6)	79.5 (71.4–86.0)	79.4 (76.1–84.1)

The lesion sum score is against whether the radiographic part of the mNY criteria is positive. For right and left it is done against whether the radiographic mNY grade is ≥3 or not. Five slices are defined in the SPARCC structural score as the transitional slice and 4 consecutive slices anteriorly. Values in the nine right columns are accuracy and precision for the sample (expert readings) used to calculate the cut-offs (“experts"), for the novices using the calculated cut-off (“novices”) and for the boot-strap validation (“boot”). For the boot-strap (“boot”) accuracy and precision values are the mean (confidence interval) of the 1000 repetitions

Abbreviations: AUC: area under the curve, mNY: modified New York, SIJ: sacroiliac joint, SPARCC: Spondyloarthritis Research Consortium of Canada, PPV: positive predictive value, NPV: negative predictive value

For all types of lesions, the accuracy of the MRI lesion sum scores was lower when compared against rad-mNYc positivity than when compared against rad-mNYc grade ≥ 3; in general, the accuracy was around 10% points lower.

Validation

Internal

In the 1,000 bootstrap samples, the estimated accuracy and PPV of CSDS per patient for predicting rad-mNYc positivity, and their 95% CIs were 69.7% (63.0%-73.4%) and 87.6% (83.2%-89.5%) for five slices and 69.3% (63.2%-73.5%) and 87.4% (83.5%-89.8%) for all slices (Table 2).

Regarding the composite score cut-offs per joint for predicting rad-mNYc grade ≥ 3, the estimated accuracy and PPV were similar for right and left joint, and for five and all slices (Table 2). Accuracy was > 78.2% and PPV > 77.4%

For the composite score cut-offs for bilateral grade ≥2, the estimated accuracy and PPV were 71.1% and 86.4% for five slices, and 70.2% and 85.8% for all slices (Table 2).

Against novices’ scores

For the CSDS₅ cut-off, the validation using the novices’ scores showed an accuracy of 71.1%, and PPV of 94.4%. In general, a similar or higher accuracy and PPV on all different cut-offs as com-pared to the accuracy and PPV done on the sample (i.e. expert readings) (Table 2) were found.

Alternative cutoffs

Calculations of cut-offs with alternative methods (applying a 90% specificity threshold, Youden index and ROC01) are shown in Table 2. The alternative cut-offs’ specificities were notably lower than the 0.95-specificity method, and the sensitivities higher than the sensitivity for the 0.95-specificity-cut-off. The highest sensitivities were reached with Youden and ROC01, between 0.74-0.80, with specificities ranging from 0.75 to 0.85. Accuracy ranged from 74.2% to 78.0% for all alternative cut-offs. ROC curves with the alternative cut-offs annotated are shown in Figure 5.

Fig. 5.

ROC curves for MRI composite scores against radiographic mNY positivity with marking of AUC and the alternative cut-offs calculated with a specificity of 90%, Youden Index and ROC01, respectively. Abbreviations: AUC: area under curve, mNY: modified New York. CSDS: Composite Structural Damage Score

Discussion

In this study, we defined several high specificity MRI cut-offs for a rad-mNYc-positive radiograph of the SIJs, which all had a 95% specificity for rad-mNYc positivity. A composite MRI-score based on erosion, backfill, and ankylosis in five predefined semi-coronal slices (CSDS₅) had a lower cut-off, 30, compared to in all slices, 38; they performed similarly. High specificity MRI cut-offs based on MRI scores of individual lesions performed considerably worse, i.e. we do not recommend using those. Results of internal validation gave estimates of cut-offs similar to the selected cut-offs in the original data. For the composite score against rad-mNYc positivity, bootstrap analyses resulted in moderate accuracy (between 60% and 70%), but very high PPV (between 0.8 and 0.9); against rad-mNYc grade ≥ 3 both accuracy and PPV were high (between 0.7 and 0.8), for both right/left. Validation done using the novices’ scores generally showed similar or higher accuracy and PPV compared to the sample (i.e. experts’ scores).

The CSDS₅ and CSDS_all cut-offs for rad-mNYc positivity were relatively high (30 and 38 for five and all slices, respectively) indicating that there must be very substantial changes on an MRI to have a positive radiograph according to the rad-mNYc. However, it should be noted that in the composite score, backfill and ankylosis per joint-half per slice are multiplied by 4 and 6, respectively. Therefore, a score of 30 “only” requires ankylosis scores in five joint-halves within the five slices, e.g. one half in each of five slices. Erosions are not multiplied; therefore, multiple (≥ 30) erosions are required if neither backfill nor ankylosis is present. The sensitivity was relatively low for all composite score cut-offs, showing that with these cut-offs some patients, who actually are rad-mNYc positive, i.e. do have AS/r-axSpA, are missed by these MRI cut-offs. Therefore, this cut-off is a threshold for being certain that patients are rad-mNY positive that indeed, due to the low sensitivity, will miss a substantial proportion of patients who are rad-mNY positive. For this reason, alternative cut-offs for the composite sum scores against rad-mNYc positivity were investigated. Using Youden index, we found a cut-off of 8 for five slices, which provided a sensitivity of 0.74, a specificity of 0.81, and an accuracy of 83%. Similar results were seen when using the ROC01 method. This illustrates that it is important to consider the purpose of using the MRI cut-off: if a high specificity is not necessary, a lower cut-off will allow for a better sensitivity, thereby not missing that many patients that do have AS/r-axSpA. This might be useful in patients with an early disease stage, since they most likely will not be captured by a cut-off with low sensitivity and high specificity. However, using these much lower cut-offs will also result in patients being misclassified as AS/r-axSpA even though they are rad-mNYc negative. A higher sensitivity might be more desirable for diagnostic purposes, whereas a high specificity can be more desirable for situations where you need to be sure the patient does have r-axSpA e.g. when transferring knowledge from earlier radiographic studies on axSpA or deciding on treatment that requires a diagnosis of rad-mNY (i.e. for instance biological therapy in some countries). The cut-off for the MRI CSDS_all was slightly worse than the cut-off for CSDS₅ in terms of accuracy, PPV and NPV in the validation analyses. Therefore, we suggest using the cut-off for CSDS₅, also since it is more time-consuming to score all slices and therefore, five slices would for feasibility reasons be more favorable and the five slices are a more standardized approach which is not so susceptible to varying number of slices.

When creating per-joint MRI cut-offs to predict rad-mNYc grade ≥ 3, a higher accuracy was found, suggesting that these cut-offs might be valuable for determining whether an MRI is rad-mNYc positive, since unilateral grade 3 damage makes the rad-mNYc positive. Therefore, using these cut-offs fewer changes on the MRI can predict the patient being rad-mNYc positive. When comparing the MRI cut-offs that correspond to rad-mNYc ≥ grade 3 with those that correspond to rad-mNYc positivity there was no difference in accuracy, suggesting that this approach does not find more patients with AS/r-axSpA. Furthermore, applying MRI cut-offs that correspond to bilateral rad-mNYc ≥ 2 does not change the accuracy. Thus, no other cut-offs appeared better suited to predict rad-mNYc positivity than the CSDS₅ cutoff.

For the individual MRI lesions, the sensitivity for detecting rad-mNYc positivity was generally low and considerably lower than for the composite scores. An exception was ankylosis, which reached similar sensitivities as the composite scores, only by requiring a score of one (one joint half with ankylosis in one slice). This suggests that unless there is ankylosis in the joint on MRI, individual lesions on MRI cannot stand alone to fulfil the rad-mNYc criteria and thereby contribute to classifying a patient with AS/r-axSpA.

We did not use fat lesions (i.e. fat metaplasia in the bone marrow) on MRI to define the high specificity MRI cut-off. This was decided partly because fat lesions are not part of the composite scores and partly because it was decided that merely the structural lesions, which you would assume would be visible on a radiograph (i.e. changes to the bone cortex) should be used for the high specificity MRI cut-off.

This study is, to our knowledge, the first to define a high specificity MRI cut-off to the rad-mNYc. Nevertheless, our results are in accordance with earlier studies comparing radiographs, MRI and CT, showing that MRI is more sensitive for detecting structural changes than radiographs [6]; i.e. that quite extensive changes on MRI are required for radiographs to be rad-mNYc positive. Strengths of the study further include the large sample size from several European countries, the relatively equal distribution of radiographic grades in both right and left SIJ and that rad-mNYc positivity versus negativity were equally split. All these contributed to allowing robust analyses. Furthermore, we validated the cut-offs of composite sum scores for five and all slices using bootstrapping instead of splitting the cohort into a “development part” and “validation part”. We believe it is a strength in this case where the sample sizes, when divided into the five grades, would not have been that big if the patients had been split into two cohorts. Other strengths include that the images were also validated on less experienced readers’ scores and were acquired in routine clinical practice, increasing external validity and that the imaging examinations were centrally read. Using images from routine clinical practice introduces heterogeneity in how the images are obtained, since different scanners and protocols have been used. This can, of course, affect the readings, e.g., if the image quality and slice thickness vary. However, we do believe that this can also be considered a strength that images were heterogenous since it makes the MRI cut-off more generalizable in routine care, where the imaging protocols may differ from center to center. Our study also has some limitations. First, radiographs and MRIs were not necessarily obtained the same day, but within up to six months. Given the slow progression of the disease on radiographs, we considered this acceptable. The internal validation analyses showed high accuracy and high precision. This confirms the reliability of our results. Second, our aim to determine a high specificity MRI cut-off with a specificity of 0.95 limits the opportunity to get a high sensitivity. This was investigated further with calculations of alternative cut-offs with higher sensitivity and lower specificity as discussed above. Third, only internal and not external validation was done. Therefore, it still needs validation in an external cohort. Fourth, our interreader agreement on rad-mNY positivity was relatively low. In addition, it can be argued that it is well known that MRI is more sensitive and accurate than radiography for SIJ structural changes and that it therefore, may be irrelevant to investigate MRI cutoffs with radiography as the standard reference [6]. However, despite the virtues of MRI, radiography is still widely used, as it is more feasible due to both cost and availability. Thus, knowing what lesions are seen by a more sensitive method (MRI) when a radiograph is rad-mNYc positive versus negative is informative for clinicians and researchers. It is useful to anchor MRI findings to rad-mNYc positivity since many historical cohorts and trials have only used radiographs, and having the high specificity MRI cut-off for rad-mNYc positivity makes it possible to compare older and newer studies. Moreover, the cutoffs found in our study can be used to assign rad-mNYc positivity to patients based on MRI examinations with high specificity without having a radiograph available. Being able to do this allows clinicians to assign therapy based on an MRI where a diagnosis of r-axSpA is mandatory for treatment initiation (both bDMARDs and physiotherapy in certain countries), thereby saving radiation exposure to the patient. In addition, the clinician may extrapolate their knowledge of r-axSpA regarding prognosis and comorbidities to patients without a radiograph. Since the specificity is this high, it is with almost certain that the patients are radiographically positive thereby ensuring that the clinician does not treat or give information on the wrong basis.

Finally, it could be mentioned that since it may in clinical practice seem unfeasible to score the entire joint for quadrants with erosions to fulfil the cut-off, it is recommended that ankylosis and backfill should be assessed first to see if they (individually or in combination) fulfil the cut-off before potentially counting quadrants with erosion.

In conclusion, high specificity MRI cut-offs to a rad-mNYc positive radiograph were defined. A composite MRI-score based on assessing erosion, backfill and ankylosis for five predefined semicoronal SIJ slices (CSDS₅) was the high specificity MRI cut-off to rad-mNYc positivity, being potentially usable in both research and clinical settings, with different thresholds depending on whether sensitivity or specificity is prioritized.

Abbreviations

mNYc

Modified New York criteria

SIJ

Sacroiliac joint

r-axSpA

Radiographic axial spondyloarthritis

MRI

magnetic resonance imaging

rad-mNYc

Radiographic part of the mNY criteria

EuroSpA

European Spondyloarthritis Research Collaboration Network

CSDS

Composite Structural Damage Score

CSDS₅

Composite Structural Damage Score for five slices

SPARCC SSS

Spondyloarthritis Research Consortium of Canada Sacroiliac Joint Structural Score

ASAS

Assessment of Spondyloarthritis Society

AS

Ankylosing spondylitis

axSpA

Axial spondyloarthritis

CT

Computed tomography

SpA

Spondyloarthritis

STIR

Short tau inversion recovery

AUC

Area under curve

Authors’ contributions

Conceptualization: AH, SK, NV, AC, KB, MG, MN, RM, JZ, BG, BM and MØ. Software: AH, MØ, SP, and SK. Formal analysis: AH and SG. Data curation: AH, SK, KB, ZS, KP, ZR, BG, MdH, TD, IE, KG, SP, IS, RL, SW, JM, MW, KG, MD, AB, LØ, and MØ. Writing – original draft: AH and MØ. Writing – review and editing: all authors. Funding: MØ and MH. Project administration: AH, SK, LØ and MØ.

Funding

Open access funding provided by Copenhagen University. This EuroSpA study was financially supported by Novartis.

Data availability

Data generated or analyzed during the study are available from the corresponding author upon reasonable request.

Declarations

Ethical approval and consent to participate

Approvals were obtained locally in the different countries. In Denmark, the Danish Patient Safety Authority approved the project, ethics approval was waived. In Iceland, The National Bioethics Committee approved the project (18-009-V17). In Switzerland, imaging data in SCQM and EuroSpA was included in the ethics approval: Kantonale Ethikkommission, Kanton Zürich, BASEC-Nr: 2022 − 00272.

Consent for publication

Not applicable.

Competing interests

AH received grants from Novartis paid to employer. NV received grants from Novartis paid to em-ployer. SK received grants from Novartis paid to employer. MØ received research grants from Abbvie, BMS, Merck, Novartis and UCB, and speaker and/or consultancy fees from Abbvie, BMS, Boehringer-Ingelheim, Celgene, Eli-Lilly, Galapagos, Gilead, Hospira, Janssen, MEDAC, Merck, Novartis, Novo, Orion, Pfizer, Regeneron, Roche, Sandoz, Sanofi and UCB. RM received honoraria for lectures or presentations from Abbvie, Janssen and Eli Lilly. BM received travel expenditures, honoraria for lectures or presentations from Abbvie, Janssen, Novartis and Pfizer. MN received hono-raria for travel expenditures, lectures or presentations from Abbvie, Eli Lilly, Janssen, Novartis, Pfizer and UCB. SJP received grants and contracts from Innovation Fund Denmark and Nordic Bioscience A/S, consulting fees, speaking fees, and/or research support from AbbVie, Novartis, MSD, Pfizer, and UCB. KB received honoraria for lectures or presentations from Novartis, AbbVie, Eli Lilly, Pfizer, and Jansen. ZS received honoraria for presentations from BioGen. LØ received grants from Novartis and UCB paid to employer. ZR received consulting or speaker fees from Abbvie, Amgen, AstraZene-ca, Boehringer, Biogen, Eli Lilly, Janssen, Medis, Novartis, Pfizer, Sandoz Lek, SOBI. RL received consulting fees from CARE Arthritis and Image Analysis Group. MdH received honoraria for presen-tations from UCB.