A novel radiographic scoring system for growth abnormalities and structural change in children with juvenile idiopathic arthritis of the hip

Juvenile idiopathic arthritis (JIA) is an autoimmune disease of unknown origin, with an onset before the age of 16 years and a reported incidence of 1–2:1,000 children [1, 2]. It is characterised by synovial inflammation, with soft-tissue oedema and effusion of the involved joints, followed by destructive changes of cartilage and bone if treatment fails. Although early treatment with biologics and methotrexate does reduce the morbidity, reports addressing the long-term side effects are concerning [3, 4]. This highlights the need for accurate tools to monitor treatment response.

Approximately 20–50% of children with JIA, particularly those with systemic-onset disease, have hip involvement within 1–6 years of disease onset [5, 6]. Hip involvement is often a predictor of a severe disease course and carries a high risk of disability. Typically both hips are affected, but unilateral involvement sometimes occurs. Traditionally, conventional radiography has been used to detect, quantify and monitor osteochondral changes such as growth disturbances and destructive change (narrowed joint space from thinning and loss of cartilage, erosions, and avascular necrosis as a complication to treatment) [7‐9]. Radiographic findings vary according to mode of onset and age; in younger children the initial findings might be developmental rather than destructive, i.e. overgrowth/flattening, with subsequent development of coxa magna and varus deformity, whereas in children with clinical hip involvement, destructive changes might supervene. Children with later-onset JIA sometimes have destruction/narrowed joint space as the first feature, often followed by development of protrusio acetabuli [5, 8, 10]. Premature closure with subsequent growth arrest might also occur [11].

Several scoring systems for radiographic JIA changes have been proposed; however most are based on hand and wrist radiographs [12‐17]. Ideally a radiographic score for structural change should reflect the total burden of disease damage; however no generally accepted score, be it based on an “index” joint or a composite score, has been identified [12, 18].

Scoring systems for radiographic hip joint changes in children with JIA are lacking. One exception is a system by Bertamino et al. [18] in 2010, the Childhood Arthritis Radiographic Score of the Hip (CARSH), devised by a panel of five paediatric rheumatologists. Based on an evaluation of joint space narrowing, erosions, growth abnormalities, subchondral cysts, malalignment, sclerosis of the acetabulum and avascular necrosis of the femoral head, the hips were scored as 0 (normal) to 32 (extensive, destructive disease in both hips); based on high intra-class correlation, the authors concluded that the system was reliable [18]. However no information was given as to differences between or within scorers.

In the present study, our goal was to examine the repeatability of radiographic findings used in the CARSH scoring system: erosions, flattening of the femoral head, joint space, fovea assessment and sclerosis. Second, we aimed to validate additional markers for both growth abnormalities and structural damage, for the purpose of informing future scoring systems for structural hip damage.

This study is part of a large longitudinal multi-centre project (Health-e-Child) to establish imaging-based scoring systems for children with JIA with wrist or hip involvement. The project was approved by the institutional research ethics committees at Great Ormond Street Hospital, London, UK, and Ospedale Bambino Gesu, Rome, Italy, and written informed consent was obtained from all the patients or their caregivers.

For this particular study, we reviewed a subset of 75 hip radiographs from 75 children with JIA with hip involvement (59 seen at the outpatient clinic at Great Ormond Street Hospital during 2006–2016, and 16 seen at Bambino Gesu Hospital). This subset was chosen based on clinical demographic information to reflect a wide range of disease duration, JIA subset and severity in order to robustly test the variables within our radiographic scoring system.

All hip radiographs were scored by two sets of radiologists, blinded to each patient’s length of disease, prior imaging, clinical symptoms and subtype of JIA. The images were all scored once in consensus by an experienced paediatric radiologist (LTdH, 10 years of paediatric radiology experience) and general radiology consultant (PLdP, 5 years of paediatric radiology experience), and twice by consensus between an experienced paediatric radiologist (K.R., 25 years of paediatric radiology experience) and a paediatric radiology research fellow (S.C.S., 3 years of paediatric radiology experience). Two weekend calibration sessions were conducted prior to scoring, using imaging examples not included for analysis in this study.

We included one set of radiographs from a total of 75 children (39 females), mean age 13 years 1 month (range 6–21 years) with hip JIA. Twenty-eight had the polyarticular form of JIA, while 23 had oligo JIA, 11 had enthesitis-related arthritis, 8 had systemic-onset JIA, 1 had psoriatic JIA and 4 had non-differentiated JIA. Mean duration of disease at the time the radiographs were obtained was 5.1 years (range 1 year to 15 years). The distribution of changes seen for right and left hips separately is shown in Fig. 7.

We have identified a set of radiographic markers suggestive of chronic disease in children with hip JIA. The markers include the assessment of destruction, such as erosions of the femoral head and neck, flattening of the femoral head and joint space narrowing, as well as measures of growth abnormalities such as length and width of the femoral neck, projected centrum–collum–diaphysis angle and trochanteric–femoral head length. Their clinical validity remains to be determined.

Although subjective assessment of femoral head flattening performed well for the same observer, using Mose circle seemed to be a more robust method between assessors. This technique has been validated for use in other paediatric hip diseases such as Perthes disease and avascular necrosis [22]. The agreement for scoring erosions of the femoral head and neck using 0–4 scale and 0–1 scales, respectively, was moderate to good within readers, while the agreement for acetabular erosions performed poorer. Agreement between observers was only fair, underscoring the importance of reader calibration workshops and standardising radiographs when performing clinical trials.

Compared to the other radiographic hip scoring system in the literature for JIA in children (the CARSH score), of the seven features that were included [18] — joint space narrowing, erosions, growth abnormalities, subchondral cysts, malalignment, sclerosis of the acetabulum and avascular necrosis of the femoral head — we found moderate agreement for assessment of erosions and moderate to good agreement for assessment of joint space narrowing on a 0–2 scale, while scoring of acetabular sclerosis performed rather poorly. On the other hand, direct measurement of joint space height turned out to be rather inaccurate, with significant variation for the same observer. The variation was even higher for the medial joint space width, reflecting difficulties in identifying precise measurement points. Although we were aware of the large specter of normal, acetabular appearances, such as an arched roof, a flat roof or an angular roof, the different shapes might have biased the measurements. Whether joint space varies according to gender, age or weight has been addressed in a few studies, with no conclusions [23]. Although reference ranges by age are lacking, joint space narrowing is still considered an important radiographic marker of destructive disease in an arthritic joint [24].

Given our results, the high inter-observer agreement reported for the CARSH score, with intra-class correlation of 0.98 for baseline scores, 0.76 for changes in scores from baseline to time 1, and 0.96 for scores obtained for the whole dataset, are surprising [18]. These results might, however, be flawed because the assumption of independent measurements was violated, and, further, the hip radiographs were analysed chronologically by readers not blinded for previous findings/scores. Moreover, the dataset did not include normal cases. The joint space narrowing, believed to be the most characteristic radiographic finding in JIA, was assessed subjectively by comparing the two hips for each of the patients. If both sides were judged to be involved, consensus was made by pediatric radiologists. In sum, all these factors might have influenced the results.

Further, in the CARSH study the authors performed a comparison between clinical and radiographic disease markers [18]. Radiographic changes from baseline to the first follow-up were moderately correlated to clinical disease markers at the final follow-up, and to the radiographic long-term outcome. Interestingly, there was a poor association between the CARSH-score and clinical disease markers such as the Childhood Health Assessment Questionnaire (CHAQ) and index wrist radiographic assessments using Sharp van der Heijde and Poznanski scores [18]. The lack of correlation with wrist radiographic scores suggests that damage scores in the hand and wrist joints do not reliably reflect damage of the hip, underscoring the need for a hip-specific radiographic score.

Our findings compare well with what has been shown for the immature wrist joint, in that the appearances of chronic change differ from those in adult rheumatoid arthritis. In rheumatoid arthritis, chronic change manifests with bone erosions, while in immature bones, increased growth might succeed bony angulation and squaring, followed by carpal crowding and bone loss — with or without erosions. The different routes of bone destruction are caused by the large amount of cartilage still existent in immature bones.

In our image analysis, we included markers of growth abnormality to reflect the chronic changes specific to a paediatric population. The clinical validity of these features must still be assessed, particularly the centrum–collum–diaphysis angle given that it relies on patient positioning (e.g., an external rotation of the leg or marked antetorsion of the femur projects the angle incorrectly). It is also known that there is a high variance reported in the medical literature for measuring this angle, with inconsistency in how radiologists and orthopaedic surgeons account for this [25]. In addition, it is also doubtful whether coxa valga is a true, primary feature of JIA because it is more likely secondary to prolonged immobilisation from disuse secondary to non-weight-bearing [26].

Our study also had several limitations. First, there is the subjective nature of any radiographic scoring and measurements, with inherent biases in readers’ experiences and understanding of the factors required to score. We endeavoured to overcome this by hosting a calibration session among all readers prior to scoring and analysis, and conducting our scoring using consensus between two groups of two readers, although we understand that consensus reads with two readers might be less representative of clinical practice. Despite our efforts, challenges still exist in measurement of joint spaces by radiograph in skeletally immature patients whose epiphyseal cartilage is not well distinguished and also in potential differences in calibration and magnification when using electronic measurement tools on differing image viewing systems.

Our study also does not address the clinical validity of our scoring variables; however this was not our intention. Our intention was primarily to understand which imaging factors radiologists agree upon as robust markers in JIA with the desire to then assess these for clinical validity, based on the understanding that these features could be more reliably scored. The strengths of this study include the use of cases from two European centres and the multireader aspect of our data analysis. We include a breadth of pathology and severity of imaging features, which were assessed by a scoring system that tested a wide variety of radiographic features of JIA.

Final important considerations for further work in providing a robust radiographic hip-JIA scoring system, particularly pertaining to clinical validity, could include the discovery that differing radiographic features might not be equally representative of significant long-term damage or clinical deterioration, and that perhaps only specific reproducible measurements are the key to prognosticating outcomes on paediatric radiographs rather than a score. Nevertheless, we are unable to define these without further assessment and long-term data, and in this paper we demonstrate specific markers that can be assessed with some degree of reproducibility to take forward. The combination and utilisation of combined imaging features from differing modalities (e.g., ultrasound, MRI and osteodenistometry) might provide an opportunity to refine and stratify clinical outcomes; however this has not been established and is beyond the remit of this study. We hope that this information can be incorporated into future scoring systems to test clinical validity and applicability to other patient populations.

Despite efforts at robust standardisation of measurements for different paediatric hip radiograph variables associated with JIA, there remains poor inter-observer variability, although reasonable intra-observer variability. We have identified several features that are more reliable for reproducible measurements and hope that these features can be helpful for use in studies that assess clinical validity and long-term patient outcomes.