Background
Juvenile spondyloarthritis (SpA) is a term that encompasses a group of conditions characterized by inflammatory arthritis, enthesitis, HLA-B27 positivity, acute anterior uveitis, inflammatory bowel disease, and psoriasis. The arthritis of juvenile SpA (JSpA) can be peripheral or axial (sacroiliac joints or spine). While the diagnosis of peripheral arthritis can typically be made by clinical examination, confirmation of sacroiliitis often requires imaging. Prior studies have shown that tenderness to palpation and physical examination maneuvers such as the flexion abduction external rotation (FABER) hip test have low sensitivity and specificity for sacroiliitis using magnetic resonance imaging (MRI) as the reference standard [
1]. For historical reasons, radiographs are currently the gold standard for making the diagnosis of ankylosing spondylitis and are frequently a prerequisite for obtaining an MRI study under many insurance plans in the United States. Radiographs, however, only show bony damage and are not sensitive enough to detect early disease or incremental changes over short periods of time [
2]. Given the relatively short disease duration and rare occurrence of ankylosis in children [
1,
3,
4], the value of radiographs at the time of diagnosis or in evaluation of suspected early inflammatory sacroiliitis in children is unclear. The practice of routinely obtaining radiographs may cause unnecessary radiation exposure and result in early cases of sacroiliitis going undetected and untreated if MRI is not subsequently performed [
1]. Misdiagnosis may also result in inappropriate therapy.
MRI has become increasingly utilized to detect inflammation in the sacroiliac joints (SIJs) before changes are apparent on radiographs. The Assessment of SpondyloArthritis International Society (ASAS) classification criteria for axial spondyloarthritis specifically include MRI evidence of inflammation in the sacroiliac joints in the criteria for adult SpA [
5]. Several studies have shown the value of MRI in evaluation of JSpA [
1,
6‐
8], but only one small study has directly evaluated the diagnostic utility of radiography versus MRI for children with SpA [
7]. Concordant with other studies which show unequivocal superiority of MRI over radiographs for detection of active disease [
9‐
11], the positive likelihood ratio (LR+) for a clinical diagnosis of SpA was much higher for MRI findings than radiographic findings in that small study, especially for erosions (LR+ = 6.7 vs 3.5) and global impression (LR+ = 9.4 vs 4.4) [7]. However, that study had a small sample size and a high frequency of abnormalities reported in controls; 55% and 20% had sclerosis and erosions by radiograph, respectively. This may have been due to use of oversensitive criteria for these radiographic findings [
12].
The objective of this project was to more fully evaluate the accuracy of radiographs to detect sacroiliitis in children using global impression of the MRI study as the reference standard. Our working hypotheses were that radiographs do not add incremental value to the MRI examination of the sacroiliac joint and that the test properties of radiographs are sufficiently low that follow-up MRI is needed in most cases.
Methods
Human subject protection
The protocol for the conduct of this study was reviewed and approved by the Children’s Hospital of Philadelphia Committee for the Protection of Human Subjects (IRB 16-013013).
Study population
This was a retrospective cross-sectional study of all children with suspected or confirmed JSpA who underwent both pelvic radiograph and MRI separated by no more than 6 months between January 2012 and May 2016. Eligible children were ages 6–18 years at the time of clinical care and had the following imaging protocols performed at our institution: anterioposterior (AP) view of the pelvis or dedicated radiographs and MRI of the sacroiliac joints that included coronal oblique T1 and STIR sequences performed at either 1.5 or 3 Tesla. All MRI assessments were made using noncontrast sequences. Demographic characteristics and indications for imaging were abstracted from the electronic medical record and the imaging studies were obtained from the picture archiving and communication system (PACS).
Evaluation of imaging studies
All scoring exercises were completed within a web-based environment (
CaREArthritis.com) or Research Electronic Data Capture (REDCap) [
13]. REDCap is a secure, web-based application designed to support data capture for research studies. Four raters were musculoskeletal radiologists (DMB, RGL, JLJ, NAC) and one rater was an adult rheumatologist with SIJ imaging expertise (WPM). All images were reviewed in random order and blinded to clinical details. All raters have had extensive training in the interpretation of pelvic radiographs and MRI.
Each radiograph was assessed for erosions, sclerosis, joint space narrowing, joint space widening, and ankylosis. Each rater indicated whether the radiograph was globally representative of sacroiliitis (yes or no) and rated confidence in global impression (ordinal scale − 4 to 4 with anchors of “definitely no” and “definitely yes”). Erosion was defined as a cortical irregularity along the articular surface of the bone. Sclerosis was defined as increased subchondral bone density compared to the subchondral bone density in the hips/spine. Ankylosis was defined as complete obliteration of the joint space with contiguous bone between the sacrum and ilium. Joint space narrowing and widening were determined subjectively as decreased or increased width of the joint space. The presence of each lesion was recorded as occurring in the left or right joint, with an additional specification of quadrant location being made for erosion and sclerosis.
Each MRI study was evaluated for active inflammatory lesions (bone marrow edema, capsulitis, SIJ effusion, enthesitis outside of the SIJ) and structural lesions (erosion, sclerosis, fat metaplasia, backfill, ankylosis). Inflammation was assessed using the CareArthritis platform and the Spondyloarthritis Research Consortium of Canada (SPARCC) SIJ Inflammation Score (SIS) scoring module. Reliability of the SPARCC SIS has been demonstrated in the pediatric population [
14,
15]. Details about the platform and scoring have been published previously [
16]. All raters previously completed calibration exercises for the SPARCC SIS and SSS with acceptable reliability (intraclass correlation (ICC) ≥ 0.8) [
14,
15]. The presence or absence of marrow edema was scored for each joint quadrant (total score per slice 0–8). Marrow edema was deemed present if the intensity was the same or greater than the presacral veins and depth ≥ 1 cm, and was scored dichotomously for each sacroiliac joint. Positive bone marrow edema findings were defined in accordance with the ASAS criteria (bone marrow edema in two or more locations on a single MRI slice or bone marrow edema on two consecutive MRI slices). The ASAS MRImagine consensus-based eCRF for recording MRI data was used to capture the following: rater global impression of acute/active inflammatory lesions compatible with sacroiliitis (yes/no), rater global impression of structural lesions typical of axial SpA, confidence in that assessment (ordinal scale − 4 to 4 with anchors of “definitely no” and “definitely yes”), capsulitis (yes/no), SIJ effusion (yes/no), and enthesitis outside of the SIJ (yes/no) [
17,
18].
Structural lesions on MRI (erosion, sclerosis, fat metaplasia, backfill, ankyloses) were assessed using the CareArthritis platform and the SPARCC SIJ structural score (SSS) scoring module. Reliability of the SPARCC SSS has been demonstrated in the pediatric population [
19].
Statistical analysis
Subject demographic characteristics and raters’ assessments of lesions were summarized by frequencies and percentages or medians and interquartile ranges (IQR). In order to compare radiograph and MRI assessment for lesions, all MRI scoring was dichotomized. Interrater agreement was assessed using Fleiss’s kappa statistic with bootstrap confidence intervals [
20], with agreement interpreted as poor ≤ 0.40, fair 0.41–0.59, good 0.60–0.74, and excellent ≥ 0.75 [
21]. Sensitivity, specificity, positive predictive value, and negative predictive value were calculated to assess the performance of radiographs in identifying sacroiliitis (global impression “yes”) using MRI global impression of active sacroiliitis (“yes”) as the reference standard. We also conducted two analogous analyses in which the reference standard was altered: to assess performance of radiographs in identifying sacroiliitis using MRI global impression of structural lesions consistent with sacroiliitis (global impression “yes”) as the reference standard; and to assess performance of radiographs in identifying sacroiliitis using MRI global impression of active sacroiliitis (global impression “yes”)
or structural lesions consistent with sacroiliitis (global impression “yes”) as the reference standard. All analyses were performed using Stata 14.2 (2015, Stata Statistical Software Release 14; StataCorp. LP, College Station, TX, USA).
Discussion
This is a systematic analysis to compare the utility of radiographs to MRI in the evaluation of suspected sacroiliitis. Interrater reliability and confidence of MRI were superior to radiographs for global impression of sacroiliitis. The rates of misclassification were high, with false positive radiographs occurring more frequently than false negative radiographs. Radiographs do remain indispensable for identification of lesions that may be on the differential of pediatric lower back pain including osteomyelitis, septic arthritis, chronic recurrent multifocal osteomyelitis, tumor, and fracture. Our results, however, indicate that radiographs have limited utility in screening for inflammatory or structural lesions consistent with sacroiliitis in children and result in a significant proportion of both false negative and positive findings.
Several limitations should be considered while interpreting our findings. First, our sample size was limited to 60 sets of radiographs and MRI scans. The number of children who had imaging performed at our institution during the study timeframe was much larger, but many did not have both radiograph and MRI performed within 6 months of each other. Further, MRI at our institution prior to 2012 did not routinely include coronal oblique sacral sequences, which we consider vital for adequate assessment of the sacroiliac joints. Nevertheless, even with our limited sample size we were able to identify significant shortcomings in the use of radiographs in children. Second, this was a retrospective study using sets of images that could have been obtained at any point in the child’s disease course. Perhaps the utility of radiographs is higher with longer disease duration or older age and more closely matches the utility seen in adults in these cases of prolonged disease exposure. The vast majority of children, however, have relatively short duration of symptoms and disease at the time imaging is ordered. We think our data reflect the typical use of radiographs for routine practice in the evaluation of the sacroiliac joints in children with both suspected and established JSpA. Third, some of the subjects had imaging performed shortly after spondyloarthritis diagnosis as part of a prior study. Our sensitivity analysis investigating this limitation using a sample restricted to those patients imaged specifically because of pain (no prior research subjects) demonstrated that our range of estimates of the radiograph test properties did not vary between the full and restricted samples. Fourth, since this was not a prospective study, there was no imaging protocol and there were differences in imaging sequences obtained. All children had at least a single AP view radiograph performed, by study design, but some children had dedicated films with multiple views. It is possible that the sensitivity and specificity of a single view radiograph is inferior to multiple views. However, in a study of adults with seronegative spondyloarthritis, there was excellent agreement (greater than 86% for both left and right joints) between AP views and AP plus oblique projections [
23]. All MRI sequences, by study design, included coronal oblique views on both T1 and T2 sequences to ensure adequate visualization of the sacroiliac joints. There are no strict measurements for depiction of sacroiliac joint effusions on MRI in children. The presence or absence of abnormality is, therefore, a subjective call which is likely the reason why these lesions had low agreement between the raters. Until there are sufficient normative data in children, definitive characterization of abnormal amounts of fluid for the sacroiliac joints will remain difficult. Fifth, our study results are not applicable to parts of the world in which a MRI scan is difficult to obtain. In these areas, radiographs remain the only option for screening. Sixth, unlike the other prior study [
7], this imaging-only study uses MRI findings as the gold standard for sacroiliitis, without direct use of an external reference standard. Pathologic confirmation of sacroiliitis, for example by biopsy, is rarely feasible, and clinical diagnosis of spondyloarthropathy is complex; determining the relation between MRI findings of sacroiliitis and a clinical diagnosis of spondyloarthritis or juvenile idiopathic arthritis is beyond the scope of this work.
A few findings from this study warrant additional consideration. First, the specificity of radiographs for detection of sacroiliitis using MRI as the reference standard was relatively high (60.8–92.2) with slightly higher negative predictive values (82.7–93.9). However, when we looked at the children who actually had sacroiliitis, the majority had normal radiographs. The positive and negative predictive values and rates of misclassification were very similar when the reference standard was changed to structural lesions on MRI. This means that if radiographs remain the gold standard for screening, then almost all cases of sacroiliitis, even if symptoms prompted the imaging, would be missed. If we apply the concept of an early treatment window, as has been shown in rheumatoid arthritis [
24], waiting to declare sacroiliitis as a disease manifestation until changes appear on radiographs is a missed opportunity to maximally improve long-term clinical, functional, and radiographic outcomes. Further, radiographs are not without consequences to children and their parents, including anxiety (from extra imaging procedures or false positive results), radiation exposure, and costs. Procedural costs for a single AP pelvic radiograph at our institution is approximately $97.00 for one or two views and $132.00 for dedicated sacroiliac joint views; additional fees are charged for professional interpretation.
Second, imaging studies always need to be clinically correlated, considering both the pretest probability and suspicion for disease. All of the children included in the study had suspected inflammatory sacroiliitis either because of underlying diagnosis or symptoms. The majority of positive MRI studies had normal radiographs but, given clinical suspicion, MRI studies were ordered anyway. At least half of the radiographs considered indicative of sacroiliitis by all raters were accompanied by a normal MRI scan. Further, even when we think the radiograph is truly depicting evidence of joint damage and not a false positive result, the findings do not indicate whether the disease remains active and requires treatment or whether the disease has already burnt out. Therefore, abnormal radiographs are almost always followed by an MRI study. If the pretest suspicion of sacroiliitis is high enough that we are going to order the MRI scan regardless of the radiograph findings, why not save the patient and family from anxiety, radiation exposure, and healthcare dollars and jump straight to the MRI?