The concept of non-radiographic axial spondyloarthritis (nr-axSpA) gained prominence from the understanding that definite plain radiographic features of sacroiliitis evolve over many years, and diagnosis is often delayed, while treatment is particularly effective in the early stages of disease. Many patients presenting with features suggestive of SpA but without radiographic sacroiliitis were indiscriminately labeled as “undifferentiated SpA” but then denied highly effective tumor necrosis factor inhibitor (TNFi) therapies because the drug label confined treatment to those with radiographic sacroiliitis. It became necessary to capture these patients with early disease in new classification criteria that would include patients within a broader spectrum of axSpA, which became possible with the advent of MRI for early detection of sacroiliitis. This was accomplished using a two-pronged approach in the 2009 Assessments in SpondyloArthritis international Society (ASAS) classification criteria [1]: (1) An imaging arm allows patients to be classified as having axSpA if they have MRI evidence of sacroiliitis and at least one SpA feature. (2) A clinical arm permits classification of axSpA in the absence of MRI inflammation if the patient is positive for HLA B27 and has at least two SpA features. A positive MRI for the purposes of classification was defined by a 2009 consensus of ASAS experts as bone marrow edema (BME) on fat-suppressed scans or osteitis on T1-weighted contrast-enhanced scans in a typical subchondral location [2]. This definition required the presence of at least two BME lesions on a single semicoronal slice through the SIJ or a single lesion on two consecutive slices. The lesion also had to be considered “highly suggestive” of axSpA although what characteristics of the lesion would define it as “highly suggestive” were not elaborated. A 2016 consensus update of the 2009 ASAS definition further elaborated that the concomitant presence of structural lesions, especially erosion, could help determine whether the BME lesion was “highly suggestive” of axSpA [3].

Soon after publication of the criteria, several studies examined the characteristics of patients classified as nr-axSpA and demonstrated more females and a lower prevalence of HLA-B27 as compared to studies that had classified patients using the modified New York criteria [4,5,6]. This led some to question the accuracy of the criteria, especially the clinical arm, by referencing studies which demonstrated that patients fulfilling only the clinical arm did not demonstrate progression to radiographic sacroiliitis and did not respond to TNFi in placebo-controlled randomized controlled trials (RCT) unless objective features of inflammation in the form of an elevated C-reactive protein (CRP) or MRI inflammation were evident [6, 7]. Proponents of the criteria pointed to the observation that the clinical arm had similar predictive validity to the imaging arm for rheumatologist follow-up diagnosis of axSpA in the ASAS classification cohort [8]. In addition, patients fulfilling the “clinical arm” in the ASAS classification cohort actually had a mean of 3.4 SpA features and meta-analyses of international cohorts demonstrated similar sensitivity/specificity performance of the clinical and imaging arms for rheumatologist diagnosis of axSpA [9].

The authors of the review entitled “Understanding the Paradigm of Non-Radiographic Axial Spondyloarthritis (Benavent and Navarro-Compán)” [10] fall firmly into the camp of the proponents in arguing that the data supports the view that nr-axSpA resembles radiographic-axSpA (r-axSpA) in terms of the clinical manifestations, disease burden, and treatment response. Certainly, it is now undeniable that patient self-reported symptomatology and impact on quality of life are comparable between these two categories. In comparing patient characteristics and manifestations of disease, it is important to clarify that the diagnosis of axSpA is challenging in its early stages and disease manifestations may vary according to the duration of disease. Consequently, even though the purpose of classification criteria is to select patients with homogeneous clinical characteristics across different cohorts, some differences between published cohorts might still be expected for frequency of HLA-B27 and certain clinical manifestations that appear with increasing duration and/or severity of disease as well as factors such as patterns of referral to the rheumatologist and inclusion criteria for the cohort. Benavent and Navarro-Compán provide evidence in their review of the literature that disease characteristics do indeed appear to be similar among different cohorts and between r-axSpA and nr-axSpA when the ASAS classification criteria are applied with the exception of a consistently higher percentage of females and a lower frequency of acute anterior uveitis (AAU) in nr-axSpA. The latter is to be expected as the frequency of patients with AAU increases with disease duration [11]. A higher percentage of women with nr-axSpA is consistent with data demonstrating lower severity of disease in this gender [12]. Consequently, it can indeed be argued that this data is consistent with nr-axSpA and r-axSpA being part of the spectrum of a single disease entity that is embraced by the ASAS criteria. However, assessment of treatment responses, and especially data from recent RCTs of interleukin-17 targeted therapies in nr-axSpA do indeed raise potential concerns as to the capacity of the criteria to capture a homogeneous disease entity under the category of nr-axSpA.

The recently reported trials of secukinumab 150 mg in axSpA reported differences in ASAS40 responses at week 16 between active drug and placebo that were 29% and 25% for r-axSpA (MEASURE-1 and 2, respectively) and 12% for nr-axSpA [13, 14]. For the ixekizumab trials, differences in ASAS40 responses at week 16 between active drug and placebo were 30% for r-axSpA and 16% for nr-axSpA [15, 16]. A similarly lower ASAS40 response was noted in patients receiving etanercept in the trial of nr-axSpA as compared to the trial of patients with ankylosing spondylitis [17, 18]. Moreover, the RCT of adalimumab in nr-axSpA demonstrated treatment group differences in ASAS responses only in patients with elevated CRP or the presence of MRI inflammation [6] and all subsequent trials of TNFi therapies have recruited patients with objective evidence of active disease rather than just a high level of symptoms despite use of non-steroidal anti-inflammatory drugs. A propensity-matched analysis of patients with nr-axSpA from the DESIR cohort did not demonstrate a significantly higher ASAS40 response in patients receiving TNFi as compared to usual care in those patients who only met the clinical arm of the criteria [19]. Additional observational cohort data has reported lower responses to TNFi agents in women as compared to men [20]. These data with TNFi agents in nr-axSpA contrast with placebo-controlled trial data in early axSpA conducted prior to the publication of the ASAS criteria where patients with short symptom duration, B27 positivity, and MRI inflammation, demonstrated substantially higher responses to TNFi than those observed in phase III trials of TNFi agents in patients with ankylosing spondylitis [21, 22]. It could therefore be argued that patients recruited to trials of TNFi and IL17-targeted therapies in nr-axSpA should have demonstrated even higher responses than those observed in phase III trials of r-axSpA since we would expect early axSpA to respond better than established, long-standing axSpA. If the paradigm of axSpA being a continuum of disease from nr-axSpA to r-axSpA is correct, we should expect to observe higher responses in nr-axSpA, especially in those patients selected for RCTs because these require objective evidence of inflammation and this has been shown to further enhance responses to bio-DMARDs [23]. Why was this not observed?

An obvious concern relevant to this question is the ASAS definition of a positive MRI. The application of this definition has identified false positive BME in 20–40% of healthy individuals and those with non-specific spinal disorders [24,25,26]. The selection of patients for an RCT of axSpA begins with an accurate diagnosis incorporating clinical, lab, and imaging data. Is it possible that clinicians are incentivized to recruit patients for an RCT once some BME meeting the quantitative aspect of the ASAS definition is evident on MRI? After all, the imaging arm requires only a single clinical SpA feature and once this is met, the patient meets trial inclusion criteria for both disease classification and active nr-axSpA. The patient may in fact have pain from a concomitant degenerating disc, could have inflammatory-type back pain as 40% of such patients meet criteria for inflammatory back pain [27], and could have tried and failed several NSAIDs, and the BASDAI score could be > 4. Such a patient is now eligible for participation in a trial of nr-axSpA. This example serves to highlight the potential for classification criteria to be misused for diagnostic and other purposes.

In conclusion, the trial data in nr-axSpA continues to seed doubts as to the robustness of the criteria and whether they might additionally include patients whose symptoms of back pain are related to mechanical spinal and/or pain-sensitization disorders and not a consequence of axSpA. While the 2009 criteria clearly represent a step in the right direction towards inclusion of patients with early axSpA, we now see that it seems likely that aspects of the criteria, especially how we define a positive MRI, require greater stringency.