Objective To determine which of two referral strategies, when used by referring physicians for patients with chronic back pain (CBP), is superior for diagnosing axial spondyloarthritis (SpA) by rheumatologists across several countries.
Methods Primary care referral sites in 16 countries were randomised (1 : 1) to refer patients with CBP lasting >3 months and onset before age 45 years to a rheumatologist using either strategy 1 (any of inflammatory back pain (IBP), HLA-B27 or sacroiliitis on imaging) or strategy 2 (two of the following: IBP, HLA-B27, sacroiliitis, family history of axial SpA, good response to non-steroidal anti-inflammatory drugs, extra-articular manifestations). The rheumatologist established the diagnosis. The primary analysis compared the proportion of patients diagnosed with definite axial SpA by referral strategy.
Results Patients (N=1072) were referred by 278 sites to 64 rheumatologists: 504 patients by strategy 1 and 568 patients by strategy 2. Axial SpA was diagnosed in 35.6% and 39.8% of patients referred by these respective strategies (between-group difference 4.40%; 95% CI −7.09% to 15.89%; p=0.447). IBP was the most frequently used referral criterion (94.7% of cases), showing high concordance (85.4%) with rheumatologists' assessments, and having sensitivity and a negative predictive value of >85% but a positive predictive value and specificity of <50%. Combining IBP with other criteria (eg, sacroiliitis, HLA-B27) increased the likelihood for diagnosing axial SpA.
Conclusions A referral strategy based on three criteria leads to a diagnosis of axial SpA in approximately 35% of patients with CBP and is applicable across countries and geographical locales with presumably different levels of expertise in axial SpA.
- Ankylosing Spondylitis
Statistics from Altmetric.com
Axial spondyloarthritis (SpA) is an inflammatory disease of the spinal column with a worldwide prevalence, including its early stages, of 0.5–1%.1–3 Surveys show that a final diagnosis of axial SpA is delayed 5–10 years from the onset of initial symptoms.4–7 During this period, undiagnosed patients may experience continuous pain, stiffness and fatigue; undergo fruitless, costly and sometimes harmful diagnostic testing; and have potentially progressive loss of spinal mobility and function, each contributing to decreased health-related quality of life.8 Early diagnosis of ankylosing spondylitis (AS), the prototypical form of axial SpA, may allow initiation of treatment with a tumour necrosis factor (TNF)α blocker at a time when greater benefits may be realised.9–13
The Assessment of SpondyloArthritis International Society recently developed and validated classification criteria to identify whether patients with either radiographic or non-radiographic sacroiliitis have axial SpA.14 ,15 Although this new classification system may help specialists to diagnose axial SpA at an earlier stage, it does not obviate the need for improved methods of patient referral from primary care settings. Delays in axial SpA diagnosis may stem from difficulty in distinguishing patients with axial SpA (back pain caused by inflammation) from the large group of patients with chronic back pain (CBP), which is a relatively common complaint in primary care that is frequently related to diseases other than SpA.8
Opportunities for patient referral and diagnosis of axial SpA and AS are frequently missed in primary care.16 ,17 Current referral methods have low sensitivity owing to the use of multiple combined criteria. At the same time, it may not be practical for primary care physicians to investigate all features typical of axial SpA because not all may be needed for adequate referral. These factors underscore the need for improved screening strategies,8 which in turn should improve referral from primary care, shorten the time to diagnosis and ultimately lead to better treatment and improved prognosis. Previous studies from Germany and Austria using inflammatory back pain (IBP) alone18 ,19 or one out of three suggested parameters20 as a referral strategy have reported a diagnosis of axial SpA in 33–50% of the referred patients. The Recognising and Diagnosing Ankylosing Spondylitis Reliably (RADAR) study was designed to determine which of two referral strategies, when used by referring physicians for patients with CBP in 16 different countries, is superior for diagnosing definite axial SpA by rheumatologists. Secondary objectives included usage of the referral criteria in primary care, effectiveness of different combinations of referral criteria for diagnosis of axial SpA, and determining whether patient-driven data compared with physician data are sufficient for effective screening of axial SpA.
The RADAR study was a multinational multicentre randomised study in which expert rheumatologists worked with local primary care physicians or other referring practitioners (eg, general practitioners, neurologists, orthopaedic surgeons). Participating rheumatologists were selected by the study sponsor. The selection of collaborating primary care physicians was at the discretion of the lead rheumatologist (or principal investigator) who sought out colleagues who were willing to invest time in the study.
Each local referral site was randomly assigned in a 1 : 1 ratio to apply one of two strategies when referring patients with CBP of unknown origin lasting >3 months and starting before 45 years of age to the associated rheumatologist for assessment and diagnosis. Referral strategy 1 required the presence of any of the following three criteria: IBP, positive HLA-B27 or sacroiliitis demonstrated by imaging. Referral strategy 2 required the presence of at least two of the following six criteria: IBP, HLA-B27, sacroiliitis on imaging, family history of axial SpA, good response of back pain to non-steroidal anti-inflammatory drugs (NSAIDs) or known extra-articular manifestations (EAMs) (ie, uveitis, iridocyclitis, psoriasis or inflammatory bowel disease). The presence of IBP was determined by referring physician opinion. Referring physicians and rheumatologists were asked to use the definitions they used in normal practice to determine the presence of the above criteria. Rheumatologists were permitted to educate referring physicians on the usefulness of the criteria selected for referral and on the best way to evaluate IBP, response to NSAIDs, etc. Rheumatologists were not educated specifically on how to diagnose axial SpA. Because diagnostic information such as MRI takes time to collect, rheumatologists were given approximately 3 months to collect diagnostic information for each subject. The study protocol and written informed consent form were approved by an institutional review board or independent ethics committee at each study site.
Referral centres sequentially identified patients presenting with CBP lasting >3 months, which was of unclear origin and had an onset before age 45 years. Patients with an established diagnosis of SpA at the time of referral were excluded as were any patients with other clinically significant conditions that, in the opinion of the referring physician or rheumatologist, would interfere with the study evaluations. All participating patients provided written informed consent.
Diagnostic assessment by rheumatologists
Following the initial primary care visit, eligible patients were referred to the corresponding rheumatologist for evaluation. The rheumatologist confirmed or excluded the diagnosis of axial SpA based on information captured in patient and referring physician questionnaires, diagnostic imaging, laboratory determinations (HLA-B27, erythrocyte sedimentation rate (ESR) and C-reactive protein (CRP)) and physical measurements (spinal flexion, inter-malleolar distance, occiput-to-wall distance, cervical rotation, interthorax excursion, Mennell's test and Patrick's test). No formal classification of disease was required for diagnosis; rather, the diagnosis was made on the basis of the expert opinion of the rheumatologist. As such, a definite diagnosis of axial SpA included cases with radiographic evidence of disease (ie, modified New York criteria) as well as those with non-radiographic evidence of disease (ie, non-radiographic axial SpA). Cases with only a suspicion of axial SpA were labelled as possible axial SpA. The rheumatologist also recorded the Bath Ankylosing Spondylitis Disease Activity Index and Bath Ankylosing Spondylitis Functional Index.
The primary efficacy endpoint was the proportion of patients referred according to each of the referral strategies in which the rheumatologist subsequently diagnosed definite axial SpA. The main secondary endpoints were: (1) the proportion of patients with a diagnosis of axial SpA who met each referral criterion as well as various combinations of the referral criteria based on the rheumatologist assessment; and (2) the degree of concordance of selection criteria between referring physicians and rheumatologists as well as between patient-derived data and rheumatologist data.
The primary efficacy analysis compared referral strategies for the proportion of patients with definite axial SpA as diagnosed by the rheumatologist. Weighted least-squares methods were used to adjust the analysis for the multilevel cluster design based on the assumption that the rheumatologists were a random effect and that referral sites were assigned to a referral strategy at random. The primary analysis used a binomial test of the difference in percentage of patients with definite axial SpA by referral strategy, with variance adjusted for the study design. CIs for mean or percentage differences between groups were similarly based on adjusted variances.
Comparisons among combinations of selection criteria as judged by rheumatologists or patients were ranked for their ability to identify disease by positive likelihood ratio (PLR). All other measures (sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and negative likelihood ratio) were summarised. Comparisons among selection criteria available to referring physicians were ranked only by PPV because the other measures were not appropriate for such comparisons. Selection criteria were ranked for concordance between rheumatologists and patients using weighted averages of Cohen's kappa statistic, averaged over centres, and design-adjusted p values of differences of the statistic from zero (indicating no agreement). Concordance of selection criteria was also summarised between rheumatologists and referring physicians and between rheumatologists and patients.
A total of 1072 patients were referred by 278 referral sites to 64 rheumatologists in 16 countries (12 in Europe plus Canada, India, Israel and Venezuela) between September 2008 and May 2010 (figure 1). Twenty-three patients did not complete the evaluation process. Demographic and baseline clinical characteristics of referred patients did not differ between referral strategies (table 1). The mean age of the referred cohort was 37 years; most patients were white. Mean CRP and ESR values were comparable between referral strategy groups. SpA manifestations were seen to a similar extent in both groups.
Diagnosis of definite axial SpA
The referral strategies did not differ with respect to the proportion of patients diagnosed with definite axial SpA, regardless of whether or not the disease was radiographically identifiable. Overall, 397 of the 1049 patients (37.8%) who completed the rheumatology evaluation had a diagnosis of axial SpA: 176 patients (35.6%) referred by strategy 1 and 221 patients (39.8%) referred by strategy 2 (between-group difference 4.40%; 95% CI −7.09% to 15.89%; p=0.447). Similar results were found in between-group comparisons of radiographic SpA, non-radiographic SpA and possible axial SpA, with between-group differences of 4.46% (−6.77% to 15.69%; p=0.430), 1.56% (−5.86% to 8.98%; p=0.676) and 0.31% (−6.55% to 7.17%; p=0.928), respectively (figure 2). Interestingly, the majority of patients diagnosed with axial SpA were classified as having radiographic SpA, with relatively fewer having non-radiographic SpA or possible axial SpA (29%, 8.6% and 7.7% of the total population, respectively). The proportion of referred patients diagnosed with definite axial SpA varied by country, with the highest rates seen in Russia, Romania and India and the lowest rate in Israel (see online supplementary table S1). These differences by country were not readily explained by differences in baseline characteristics. A possible explanation could be differences in symptom duration, which was not recorded during this study.
Concordance of referral physician data with rheumatologist data
The concordance between evaluations done by referring physicians and rheumatologists was determined for each criterion. IBP was the referring criterion used most frequently by referring physicians in both strategy 1 and strategy 2 (93.1% and 96.1% of patients, respectively) and showed relatively high concordance (85.4%) with evaluations performed by rheumatologists. A good response to NSAIDs was the second most frequently used referral criterion in strategy 2, but it showed lower concordance (64.1%) with the rheumatologists’ assessments. Sacroiliitis on imaging was used for referral in 31.8% of patients, but concordance with the rheumatologists’ assessments was only 64.2%. Of note, the greatest concordance was observed for the presence of HLA-B27 (97.1%) but it was rarely used as a referral criterion (17.2% of patents). Similarly, the presence of EAMs showed relatively high concordance (76.6%) but was rarely used (19.7% of patients in strategy 2).
IBP was the most frequently used referral criterion in each of the nine countries that enrolled >50 patients and generally showed good concordance (71–100% and 51–98% when used in strategies 1 and 2, respectively) with the rheumatologists’ assessments (see online supplementary table S2). Referral based on HLA-B27 (1% of referred patients in Israel to 27% in Russia) and sacroiliitis on imaging (11% in Spain to 78% in Russia) varied widely by country.
Concordance of patient data with referral physician and rheumatologist data
Similar concordance occurred between referral physician data and patient data that were imputed from patient questionnaires that captured information on inflammatory back symptoms, EAMs, family history of inflammatory disease, presence of HLA-B27 and response to NSAIDs. Concordance between referral physicians and patients was highest for HLA-B27 positivity (100%) and lowest for response to NSAIDs (60.2%).
The degree of agreement between rheumatologists and patients was assessed using Cohen's kappa statistic, which ranges from −1 (perfect disagreement) to +1 (perfect agreement). Strong agreement was found for HLA-B27 positivity (κ=0.984); moderate agreement for family history of axial SpA (κ=0.620), response to NSAIDs (κ=0.545) and known EAMs (κ=0.407); and weak agreement for IBP (κ=0.283). Each of these κ values differed significantly from zero.
Referral criteria associated with axial SpA diagnosis
Of the individual referral criteria applied by the referring physician, sacroiliitis on imaging had the greatest predictive value for definite axial SpA in both strategies 1 and 2; the overall PPV was 72.5% (table 2). The next most predictive criteria were HLA-B27 (PPV 67.1%) and known EAMs (PPV 55.6%). When applied by the rheumatologist, sacroiliitis on imaging also had the greatest PPV (93.6%) as well as the best balance between sensitivity (75.6%) and specificity (96.2%) (table 3). HLA-B27 and EAMs had the next greatest predictive value, but both had lesser sensitivity and specificity than sacroiliitis. IBP had high sensitivity but low specificity, whereas a positive family history had high specificity but low sensitivity.
The predictive value of each referral criterion varied by country, with sacroiliitis on imaging consistently showing the greatest PPV and best balance between sensitivity and specificity (see online supplementary table S3). Other referral criteria with PPV and sensitivity >70% included HLA–B27 in Romania and Russia; IBP in India, Romania and Russia; and response to NSAIDs in Romania and Russia, although the specificity of the latter two criteria was low.
Predictive value of combinations of referral criteria
The rheumatologist's assessment was also used to evaluate the predictive value of combinations of referral criteria. The two-of-two combination criteria that included sacroiliitis on imaging had the highest predictive value (≥94.7%) and specificity (≥97.6%) for an axial SpA diagnosis (table 4). Of these, the combination of sacroiliitis on imaging and IBP had the highest sensitivity (72.5%) and NPV (81.5%) and, therefore, would serve best as a two-of-two referral combination. When considering criteria combinations that do not require imaging evaluations, combinations of HLA-B27 with either known EAMs, response to NSAIDs or IBP had high PPV and specificity, although none had high NPV and sensitivity.
When any two of three criteria were considered, combinations that included sacroiliitis on imaging had the best predictive value for a diagnosis of definite axial SpA (table 4). Of these, the combination of sacroiliitis on imaging, family history and known EAMs had the best PPV (92.1%) and specificity (97.0%) for use in diagnosis, but this combination did not have adequate sensitivity (45.2%) to serve as an effective referral combination. Only sacroiliitis, HLA-B27 and IBP had PPV, NPV, sensitivity and specificity >80%. When imaging evaluation is excluded, none of the combinations involving HLA-B27 had NPV and sensitivity >80%. The best criterion (HLA-B27 with family history or IBP or with EAMs and response to NSAIDs) had NPV >75% and sensitivity of 68% (table 4). Because some referral centres may not have access to imaging and HLA-B27 testing, combinations of other criteria were considered. Of these, any two of IBP, EAMs and response to NSAIDs offered the best balance of NPV (77.8%) and sensitivity (85.3%) for use in patient referral.
Similar analyses were conducted using data imputed from the patient questionnaires. These criteria were much less effective than the rheumatologist data in predicting a diagnosis of axial SpA. In the two-of-two combinations, the highest PLR from the patient data was 2.65 for HLA-B27 and response to NSAIDs, whereas the highest PLR in the any two-of-three combinations was 2.25 for family history, HLA-B27 and response to NSAIDs. Both were much lower than the best PLRs from the rheumatologist data. Moreover, none of the combinations of criteria based on imputed patient data had acceptable levels (>80%) for both sensitivity and specificity.
The results of this multinational study show that the two defined referral strategies yielded similar proportions of patients with rheumatologist-confirmed definite axial SpA, both overall and when stratified by the presence or absence of radiographic evidence of disease. Thus, the simpler strategy 1 (one of three criteria) performed as well as the more complex strategy 2 (any two of six criteria). Comparable results were reported recently from a prospective non-randomised study conducted in Germany, in which strategy 1 was as effective as a more complex referral strategy (any two of five criteria; same as strategy 2 except EAMs were not included).21 The RADAR study extends these earlier findings by (1) showing that the simpler strategy is applicable across countries and geographical locales with presumably different levels of expertise in axial SpA; (2) providing more detailed information about concordance between referring physician and rheumatologist as well as between patient and rheumatologist; and (3) identifying criteria that were important to the rheumatologist in making the SpA diagnosis. This study also includes a thorough analysis of single referral parameters and different combinations of referral parameters.
According to strategy 1, sacroiliitis on imaging, a positive HLA-B27 test or IBP is sufficient for patient referral. In RADAR, sacroiliitis on imaging and HLA-B27 were the most predictive criteria for definite SpA. However, these assessments were rarely used for referral, suggesting that access to these tests may have been limited at many primary care sites, possibly because of costs. HLA-B27 showed high concordance between referring physician and rheumatologist, reflecting the positive/negative results of the genetic test. In contrast, sacroiliitis on imaging showed low concordance, suggesting that more training in collection and interpretation of the diagnostic image may be needed. The difficulty of the interpretation of sacroiliac joint images, in addition to the costs and potential radiation exposure, suggests that imaging of the sacroiliac joints may not be an ideal screening criterion in primary care and should not be promoted as such. However, available images of the pelvis/lumbar spine that are often obtained as part of a general diagnostic approach for low back pain can result in evidence of sacroiliitis and should be used to evaluate whether referral is indicated. IBP was the most frequently used criterion for patient referral and showed high concordance between referring physicians and rheumatologists, suggesting that this straightforward assessment may be effective as a referral tool, although it goes along with low LR and PPV.
An ideal referral strategy should identify all patients with axial SpA and exclude those without it. Although it may be argued that sensitivity is more important than specificity in a referral setting, a balance between sensitivity and specificity is important for identifying as many patients as possible while limiting the cost of referring patients who ultimately are not diagnosed with axial SpA. The referral strategies used in this study identified >35% of patients with definite axial SpA but did not exclude the >60% of patients without SpA. Therefore, each individual criterion as well as two-of-two and any two-of-three combination criteria based on rheumatologist data were evaluated to identify tests that may function better in a referral setting. Although many criteria sets had strong PPV and PLR, few afforded sensitivity and specificity at the 80% level that is conventionally considered optimal. No single referral criterion had adequate sensitivity and specificity based on the rheumatologist data, with sacroiliitis on imaging having the strongest disease-predictive value and specificity but a sensitivity of 75.6%. This indicates that rheumatologists rely heavily on imaging when diagnosing axial SpA. The moderate performance of HLA-B27 as a single referral criterion may reflect its relatively low usage by referring physicians. None of the two-of-two combination criteria had 80% sensitivity and specificity for definite axial SpA, although sacroiliitis on imaging and IBP reached this level for diagnosis of radiographic SpA. In the any two-of-three combinations, only sacroiliitis on imaging, HLA-B27 and IBP had sensitivity and specificity >80% (with PPV and NPV >80% as well).
It is important to recognise that diagnostic imaging and genetic testing may not always be accessible in primary care. Accordingly, local adoption of specific criteria needs to consider the types of tests that can be easily performed and the concordance of those referral criteria with secondary care specialists. Considerable variation in the use of HLA–B27 and sacroiliitis on imaging as referral criteria was evident among countries, as were concordance rates for sacroiliitis. In general, greater use of sacroiliitis as a referral criterion was associated with greater concordance with the rheumatologists’ assessment (Russia, India and Romania) whereas lower usage was associated with lower concordance (Spain, Slovak Republic and Israel). Of note, usage of individual criteria by referring physicians, their concordance with the rheumatologists’ assessments, and their predictive value for an axial SpA diagnosis varied by country.
Three limitations to this study may have introduced bias. First, voluntary participation of rheumatologists and referring sites could have affected the results. However, the large number of centres and participating countries is likely to have minimised bias. Second, education of referring physicians about this trial may have increased their awareness and expertise in dealing with axial SpA. This may have led to better identification of patient characteristics and potentially affected the proportion of patients correctly referred. Third, information about who determined the presence of sacroiliitis in primary care was not collected. It is possible that the suspected diagnosis in primary care and in rheumatologist practice relied on a radiologist's written report. Although this may have influenced our results, it would also reflect the decision-making process in actual clinical practice.
According to RADAR results, the any two-of-three combination criteria of IBP, good response to NSAIDs and EAMs was the best performing set that did not involve imaging or HLA-B27 testing. It had good sensitivity (85.3%) although specificity was rather low (47.8%). Moreover, concordance between referring physicians and rheumatologists for IBP and EAMs was good (85.4% and 76.6%, respectively), although concordance for response to NSAIDs was relatively low (64.1%). Nevertheless, this set of combination criteria is worthy of consideration for locales with limited technical accessibility.
Medical writing assistance was provided by Barry M Weichman, Ellen Stoltzfus and Angela Cimmino of JK Associates, Inc., Conshohocken, Pennsylvania, USA. This assistance was funded by Merck Sharp & Dohme Corp, a subsidiary of Merck & Co Inc, Whitehouse Station, New Jersey, USA.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online supplement
Handling editor Tore K Kvien
Contributors All authors are responsible for the work described in this paper. All authors were involved in at least one of the following: conception, design, acquisition, analysis, statistical analysis and interpretation of data. All authors were involved in drafting the manuscript and/or revising the manuscript for important intellectual content. All authors provided final approval of the version to be published.
Funding This study was supported by a grant from Schering-Plough (presently Merck & Co Inc, Whitehouse Station, New Jersey, USA).
Competing interests JS: grants/research support from Merck, Pfizer, Abbott, UCB and Roche; consultant for Abbott, Merck, Pfizer, UCB and Roche; speakers’ bureau at UCB, Roche, Merck, Pfizer and Abbott. SS: previous employee of Merck. DC: grants/research support from Amgen, BMS, Merck, Pfizer, Abbott, UCB and Roche; consultant for Abbott, BMS, Amgen, Merck, Pfizer, UCB and Roche; speakers’ bureau at Amgen, BMS, UCB, Roche, Merck, Pfizer and Abbott. KP: consultant for Abbott, MSD, Pfizer, Roche; speakers’ bureau at Abbott, MSD, Pfizer, Roche. AGL: consultant for MSD, UCB, Abbott and Pfizer; speakers’ bureau at MSD. PG: consultant for Schering-Plough; speakers’ bureau at MSD, Roche, UCB and Pfizer. TR: grants/research support from MSD. SE: speakers’ bureau at MSD. AK: research funding from Abbott; speakers’ bureau at Wyeth, Pfizer, Abbott and Schering-Plough. JG: grant/research support from Cambridge University Hospital; consultant for Schering-Plough and Abbott; speakers’ bureau at Wyeth. SP: speakers’ bureau at Abbott, MSD and Roche. NV: shareholder and employee of MSD. OZ, HM, DD, FC, CA, HR, AK: none declared.
Ethics approval The appropriate IRB/IEC at each study site reviewed and approved the study, consistent with local regulations.
Provenance and peer review Commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.