This study found no change in the rate of first-time hospital admissions for JIA over 22 years in WA but a significant increase in overall rates for JIA admission and joint injections since 2003, following significant uptake of biological therapy with 1 in 2700 children in WA on TNFi therapy for JIA at the study end date. We also report a hospital-based prevalence of JIA in WA of 72.3/100.000 with levels twice as high for girls as for boys.

The demographics of this Australian group of JIA patients were similar to studies of JIA patients in other countries, where female patients outnumber male patients and average age at diagnosis was 7 years [11,12,13]. The proportion of children with JIA identifying as Aboriginal was similar to that seen in the general WA community (6.1% versus 5.9%, respectively) [14], suggesting Aboriginal children were not at an increased risk of hospitalisation with JIA. The stable rates for a first-time admission for JIA indicates that the availability of bDMARD for JIA in this century has not impacted on the perceived need for hospital-based management of JIA in WA. As therapeutic guidelines for JIA were unavailable prior to expert based recommendations in 2011 [15], JIA management in the study period was dependent on the training/usual practice of the consulted specialists with usual practice to escalate to DMARD therapy if no improvements were seen after 2–3 months of NSAID monotherapy or earlier with severe disease activity. The stable hospitalisation rate could reflect a local preference for intraarticular joint injections and/or a slow rate of uptake of DMARD initiation and limitations in early access to biologicals as first line therapy [3]. As rates for subsequent admissions were paralleled by an increase in the number of joint aspirations/injections during the disease course since 2003, this points towards an increased reliance on intra-articular corticosteroid injections as stand-alone or adjunct to initiating DMARD therapy. Alternatively, it may relate to the fact that access subsidised TNFi therapy requires active disease in the context of DMARD inefficacy/intolerance and failed corticosteroid administration and /or a lack of resources for outpatient management [3, 16, 17]. While we are unable to tease out the details behind this, the presented data indicate that the costs associated with the increasing TNFi usage in WA have not been offset by reduced hospitalisation rates.

In agreement with other studies, lower limb joints were most frequently affected in this cohort [18, 19]. Based on an admittedly crude approximation of ILAR classification through ICD codes, most patients in this study had polyarticular or unspecified JIA with only 9% recorded to have oligoarticular disease. Oligoarticular presentation is considered the most common subtype of JIA and, although this is not the case in all countries [20,21,22], patients with oligoarticular JIA in this study may have been (mis)classified as unspecified or are truly underrepresented in this study. As treatment recommendations for oligoarticular but not polyarticular JIA suggest intraarticular corticosteroid injections as first line [3, 15], the number of oligoarticular JIA patients in this study was likely higher than the 9% coded as oligoarticular JIA. Finally, the distribution of affected joints has been the basis for the classification of JIA subtypes [2, 3]. However, as this distribution changes frequently over time [23] and recent studies demonstrate that a large proportion of JIA patients meet classification criteria for idiopathic arthritis in adults, the usefulness for separate phenotypes for childhood arthritis is being questioned [8,9,10].

The global prevalence of JIA shows considerable variation and while our hospital based data likely underestimate the true population prevalence [24, 25], the 2012 point-prevalence is at the higher end of prevalence estimates reported from other hospital-based studies [12, 26]. However, it remains well below the estimated prevalence reported from a previous cross-sectional Australian study of 12 year-old metropolitan students where approximately 4/1000 were found to have JIA [25], while similar studies from Belgium (in children aged 12–18 years) and a recent study from Brazil (children aged 1–16 years) found community-based prevalence estimates of 1.7 and 2 per 1000 children, respectively [27, 28]. Together these data illustrate that there is a number of patients with JIA that have milder and/or spontaneously remitting JIA, that will not be captured in hospital-based data [29].

Key strengths of this study included the availability of state-wide population-based data in WA, the large number of patients and the longitudinal analysis allowing examination of trends over time. Limitations of this study include the reliance on hospital data, likely resulting in an underestimation of the community prevalence of JIA. Also, validation of administrative data for identifying JIA in Australia has not been carried out, but studies from North America have reported ≥ 90% positive predictive value of at least one rheumatology-based code diagnostic in children [12, 26]. Together with the high capture between 80–90% for other rheumatic diseases (RA, AS and SLE) in WARDER [8], this suggests robustness of the current data. The Australian health care system relies on a combination of private and public health care providers, but all can refer children to the Children’s hospital in their state. There is no restriction on access to biologicals other than synthetic DMARD resistant disease activity, while for political more than medical reasons, important valuable outpatient data are not collected/registered in Australia and thus not available for study. Our data are drawn from deidentified administrative health data, that represent the clinician’s main discharge diagnoses, but do not have details of clinical, laboratory or imaging findings. The TNFi dispensing data were not individualized, making it impossible to correct for potential confounders and did not allow direct linkage to the deidentified patient data in the WARDER dataset. Thus, we could not determine the direct impact of drug type/usage/switch on admission rates. We used a 10-year look back period to ascertain new cases but cannot exclude not capturing a small group of young children < 6 years diagnosed prior to 1980, possibly contributing to an underestimation during the early 1990s.