Early diagnosis of axial spondyloarthritis among patients with chronic back pain using a clinical screening tool and referral strategy

Abstract

Objective

Lack of timely referral of suspected axial spondyloarthritis (axSpA) patients to rheumatologists is an important modifiable reason for diagnostic delay of axSpA. We assessed the usefulness of a self-referral strategy using a clinical feature-based screening questionnaire (SQ) (A-tool).

Methods

Finding axSpA (FaxSpA) was single-centre prospective study involving patients with chronic back pain (CBP). The A-tool, consisting of a three-question prescreen and eight-question SQ, was distributed to patients via the patient communication portal and university Facebook page. Patients with affirmative responses on all three prescreen questions, and three or more questions on SQ were eligible for study visit. Enrolled patients underwent history, physical examination, labs (CRP and HLA-B27) and imaging studies (X-ray and MRI of the pelvis). The clinician’s judgement was considered the gold standard for diagnosing axSpA.

Results

Eighty-six of the 100 enrolled patients completed all the study procedures, and 29 (34%) were diagnosed with axSpA. Seven patients had AS, and 22 had non-radiographic axSpA. Sensitivity and specificity of the individual A-tool questions for diagnosing axSpA ranged from 0.03 to 0.86 and 0.14 to 0.96, respectively. Positive likelihood ratios (+LR) of the individual items in the A-tool ranged from 0.84 to 1.34. There was low to moderate agreement between the patient responses on the online A-tool and the corresponding physician-confirmed responses.

Conclusion

A tool-based strategy for self-referral of CBP patients is a simple, practical and feasible approach for early diagnosis of axSpA. We need a larger prospective study to validate our findings.

Key messages.

• Lack of rheumatology referral is an important modifiable reason for delayed diagnosis of axial spondyloarthritis.

• We developed a simple, inexpensive and practical screening tool (A-tool) for early diagnosis of axSpA.

• A-tool based self-referral or clinician administered strategy will help to improve early diagnosis of axSpA.

Introduction

Axial spondyloarthritis (axSpA), an inflammatory rheumatic disease that predominantly involves the spine and sacroiliac joints, is an important but underrecognized cause of chronic back pain (CBP). This condition primarily affects young individuals in their productive age and can result in significant loss of function, work disability and impaired quality of life [1]. Although the population prevalence of axSpA and AS in the USA is 1.4% and 0.55% [2], the diagnostic prevalence is as low as 0.2% and 0.1%, respectively [3]. Despite the advances in our understanding of the disease and the availability of newer imaging techniques such as MRI, delayed diagnosis in axSpA remains a considerable challenge worldwide, with an average delay between symptom onset and diagnosis ranging from 8 to 14 years [4, 5]. Lack of timely referral of suspected axSpA patients to rheumatologists is one of the important and modifiable reasons for the delay [6, 7]. Previous studies conducted using different referral strategies for referring CBP patients to rheumatologists by non-rheumatology clinicians have demonstrated that approximately 35–45% of the enrolled patients had a final diagnosis of axSpA [8]. However, the majority of these studies were performed in Europe and other countries where the structure of healthcare, referral patterns and resources are different from those in the USA. Also, most of these screening strategies involve a blood test (HLA-B27) and/or imaging (X-ray or MRI), which may not serve as cost-effective screening tools for large-scale use. Also, it may not be feasible for busy primary care providers (PCP) to perform and interpret these tests which have inherent nuances. We propose that a referral strategy in the form of questionnaire created using clinical information about individual spondyloarthritis (SpA) features alone without laboratory or imaging tests, may help to provide timely and efficient self-referral of patients with a high likelihood of axSpA to rheumatologists. In this study, we prospectively assessed the usefulness of a self-referral strategy using a clinical feature-based screening questionnaire (SQ) (A-tool) in patients with CBP. The primary objective of this study was to evaluate the effectiveness and feasibility of A-tool-based referral strategy in a prospective manner.

Patients and methods

Finding axSpA (FaxSpA) was a single-centre prospective pilot study involving patients with CBP. This study was approved by the Yale University Institutional Review Board (IRB protocol 2000024619).

Development of A-tool and referral strategy

Based on a comprehensive literature review, SpA features with reasonable sensitivity, specificity and positive likelihood ratios were used to develop the A-tool [8]. We excluded SpA features involving lab tests such as CRP, HLA B-27 or imaging tests (X-ray or MRI). Also, we included individual items in inflammatory back pain (IBP) criteria that had a stronger association with the diagnosis of axSpA rather than using one single construct of IBP. The aim was to create a simple and practical screening tool that could be used by patients themselves as well as administered by healthcare providers.

A-tool consists of a prescreen and a SQ. Subjects could access the SQ only if they answered all three questions in the prescreen affirmatively (age of onset <45 years, CBP > 3 months and gradual onset of back pain). The SQ has eight questions that address IBP, enthesitis, peripheral arthritis, psoriasis, IBD and uveitis, and family history of spondyloarthritides. The subject must pass the prescreen and answer three or more of the eight questions in the SQ affirmatively to be eligible for enrolment in the study. The decision to have three or more affirmative answers in the SQ was made based on a review of existing studies and editorials [9].

Patient selection

Patients were recruited using the MyChart patient portal of EPIC electronic medical records (EMRs) and Facebook. Participants were recruited via MyChart by the Yale Joint Data Analytics Team, which identified patients 18 to 60 years old with CBP reported in their problem list. These patients were invited to complete the A-tool via an email, and participants were also recruited by using a social media-based (Facebook) recruitment strategy. We created a study page on the Facebook account of Yale Center of Clinical Investigation, and flyer was distributed to people in New Haven County. Interested patients filled out the A-tool on a secure web-based Qualtrics survey. Eligible patients were contacted by the research coordinator to ensure eligibility. Written informed consent was obtained by the study coordinator or rheumatologist on the day of the visit. Patients with preexisting diagnosis of axSpA were excluded.

Rheumatology assessments and diagnosis

The rheumatologist (A.D.) performed a detailed history and physical examination including enthesitis assessment and spinal metrology. All patients underwent lab testing (CRP [(mg/l] and HLA-B27 [flow cytometry]) and imaging studies (X-ray of the pelvis [AP and Ferguson views] and MRI of the pelvis without contrast). One patient with a contraindication to MRI did a CT pelvis to look for sacroiliitis. Imaging studies were reviewed by an expert musculoskeletal radiologist (A.H.).

We collected data about validated disease activity measures and functional indices for axSpA. The clinician’s judgement was considered the gold standard for the diagnosis of axSpA. Patients were diagnosed to have axSpA or no axSpA. Those with axSpA were further classified as or non-radiographic axial spondyloarthritis (nr-axSpA). Diagnostic assessment was made at the stage of history and physical and after the complete evaluation (at the end of labs and imaging). Assessment in Ankylosing Spondylitis (ASAS) classification criteria were applied for patients diagnosed with axSpA and ASAS IBP criteria were applied to those diagnosed with IBP by clinician.

Patients had the option to contribute additional blood for future immunophenotyping and cytokine analysis for biomarker research.

Statistical analysis

Demographics were summarized with mean and SD for continuous variables and frequency and column percent for categorical variables. Differences in demographics were computed using a t-test for continuous variables and a chi-squared test or Fisher’s exact test for categorical variables, as appropriate.

The association of each A-tool question to the outcome was evaluated using univariate logistic regression. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio (+LR) and negative likelihood ratio (−LR) were computed for each A-tool question.

A multivariable logistic model and receiver operating characteristic (ROC) were constructed to predict the probability of axSpA given a patient’s answers on the A-tool. The outcome was axSpA status (yes or no), and the dependent variables were all A-tool questions; those with composite scores were combined into a single variable score.

Agreement between A-tool questions and corresponding physician-confirmed responses was calculated using the kappa statistic and 95% CIs.

Analysis was performed using R Statistical Software (v4.2.1; R Core Team 2022). R packages utilized include gtsummary (Sjoberg DD et al., 2021), caret (Kuhn M, 2022), Hmisc (Harrell Jr F, 2022), psych (Revelle W, 2022) and pROC (Robin X et al., 2011). A two-sided P-value of 0.05 was considered statistically significant. Given that this is an exploratory analysis, no multiple testing correction was done.

Results

A total of 1274 patients completed the A-tool between April 2019 to February 2022, 50% via Facebook and 50% via MyChart. Among the responders, 507 (40%) had a positive A-tool. We excluded individuals with confirmed diagnosis of SpA and those with an obvious alternative aetiology for back pain. One hundred patients were enrolled in the study: 39 patients via Facebook and 61 patients via MyChart-based recruitment. Of the enrolled patients, 86 completed all the study procedures, including the study visit, labs and imaging, and of those, 29 (34%) were diagnosed with axSpA. There were seven patients with AS and 22 with nr-axSpA. The remaining 54 patients were diagnosed as having no axSpA, and 3 were deemed uncertain. The level of confidence of investigator in those diagnosed as axSpA was graded as ‘definite axSpA’ for 19 patients and ‘probably axSpA’ for 10 patients.

The demographic and clinical characteristics of patients in axSpA and those in no SpA groups are shown in Table 1. Most of the study population were female (n = 58; percentage = 67%) and white (57; 87%). The average age was 44 (SD 10) years. Patients in the axSpA group were older than those in the no axSpA group (47 [8] vs 42 [11]). The average age of onset of symptoms for those with axSpA was 32 (7) years vs 26 (8) in the no axSpA group. More patients in the axSpA group (26; 90%) were diagnosed with IBP clinically than those in the no axSpA group (35; 64%).

Table 1.

Demographic and clinical characteristics of the study population

Characteristic	axSpA (N = 291)	No axSpA (N = 571)	Overall (N = 861)	P-value
Age on study (years), Mean (SD)	47.21 (8.19)	41.89 (10.59)	43.69 (10.11)	0.0122
Female gender, n (%)	22 (76%)	36 (63%)	58 (67%)	0.23
Race				>0.9
Black or African American, n (%)	1 (3.4%)	3 (5.3%)	4 (4.7%)
Other, n (%)	2 (6.9%)	5 (8.8%)	7 (8.1%)
White, n (%)	26 (90%)	49 (86%)	75 (87%)
Ethnicity				0.94
Hispanic or Latino, n (%)	2 (6.9%)	7 (12%)	9 (10%)
Non-Hispanic, n (%)	26 (90%)	48 (84%)	74 (86%)
Unknown, n (%)	1 (3.4%)	2 (3.5%)	3 (3.5%)
Age of onset of initial back pain/hip pain (years), Mean (SD)	32.41 (6.97)	26.35 (7.82)	28.40 (8.04)	<0.001
Clinically IBP, n (%)	26 (90%)	35 (64%)	61 (73%)	0.0113
ASAS IBP criteria fulfilled?, n (%)	28 (97%)	45 (82%)	73 (87%)	0.0884
Peripheral inflammatory arthritis, n (%)	2 (6.9%)	3 (5.3%)	5 (5.8%)	>0.9
Enthesitis on exam, n (%)	12 (43%)	29 (55%)	41 (51%)	0.33
Dactylitis, n (%)	3 (11%)	1 (1.8%)	4 (4.8%)	0.114
Acute anterior uveitis, n (%)	2 (6.9%)	2 (3.5%)	4 (4.7%)	0.64
Psoriasis of skin and/or nails, n (%)	2 (6.9%)	5 (8.8%)	7 (8.1%)	>0.9
Ulcerative colitis, n (%)	2 (6.9%)	2 (3.5%)	4 (4.7%)	0.64
Crohn’s disease, n (%)	0 (0%)	2 (3.5%)	2 (2.3%)	0.54
Family history (first or second degree relatives)
Psoriasis, n (%)	12 (41%)	11 (20%)	23 (27%)	0.0323
Crohn’s disease, n (%)	5 (17%)	7 (12%)	12 (14%)	0.54
Ulcerative colitis, n (%)	7 (24%)	5 (8.9%)	12 (14%)	0.104
Ankylosing spondylitis, n (%)	1 (3.4%)	1 (1.8%)	2 (2.4%)	>0.9
Modified Schober test (cm), Mean (SD)	4.67 (0.87)	5.05 (1.87)	4.92 (1.61)	0.22
BASDAI (cm), Mean (SD)	5.25 (2.19)	6.11 (2.10)	5.82 (2.16)	0.0882
ASDAS (cm), Mean (SD)	2.65 (0.84)	3.01 (0.91)	2.89 (0.90)	0.0792
BASFI, Mean (SD)	3.59 (2.24)	4.48 (1.93)	4.18 (2.07)	0.0772
BASMI, Mean (SD)	2.33 (0.83)	2.22 (0.78)	2.26 (0.80)	0.52
RAPID3 score (0–30), Mean (SD)	12.41 (5.32)	14.32 (5.29)	13.68 (5.35)	0.122
P4, Mean (SD)	17 (6)	18 (6)	18 (6)	0.22
Physician global, Mean (SD)	4.10 (1.14)	4.17 (1.19)	4.15 (1.16)	0.82
CRP, Mean (SD)	4.27 (5.39)	3.91 (4.21)	4.03 (4.61)	0.82
Positive HLA B27, n (%)	7 (24%)	2 (3.5%)	9 (10%)	0.0064
Radiographic sacroiliitis (mNY grading), n (%)	7 (24%)	0 (0%)	7 (8.9%)	<0.001
MRI sacroiliitis, n (%)	15 (52%)	4 (7.3%)	19 (23%)	<0.001
Diagnosis of axSpA suspected based on history and physical exam, n (%)	23 (79%)	24 (42%)	47 (55%)	0.0013
Final diagnosis of axSpA?				<0.001
No, n (%)	0 (0%)	54 (96%)	54 (64%)
Uncertain, n (%)	0 (0%)	2 (3.6%)	2 (2.4%)
Yes, n (%)	29 (100%)	0 (0%)	29 (34%)
ASAS classification met?, n (%)	26 (90%)	0 (0%)	26 (62%)	<0.001
How did the ASAS classification met (clinical vs imaging)				>0.9
Clinical arm, n (%)	5 (19%)	0 (NA%)	5 (19%)
Imaging arm, n (%)	21 (81%)	0 (NA%)	21 (81%)
Fulfilled mNY criteria?, n (%)	7 (26%)	0 (0%)	7 (18%)	0.0844

axSpA: axial spondyloarthritis; BASDAI: Bath Ankylosing Spondylitis Disease Activity Index; ASDAS: Ankylosing Spondylitis Disease Activity Score; BASFI: Bath Ankylosing Spondylitis Functional Index; BASMI: Bath Ankylosing Spondylitis Metrology Index; RAPID3: Routine Assessment of Patient Index Data 3; P4: Pain, Physical function, Patient global, and Physician (MD) global; mNY: modified New York; ASAS: Assessment of Spondyloarthritis International Society.

While extra-articular features were not different between the two groups, axSpA patients had an increased prevalence of psoriasis in first- or second-degree family members (41% vs 20% P = 0.07). Patients suspected to have axSpA at the stage of clinical evaluation (history and physical) were more likely to have axSpA diagnosis as the outcome (P = 0.0013). HLA B27 was more prevalent in the axSpA group (24%) than in the no axSpA group (3.5%) (P = 0.0064). Elevated CRP was distributed similarly among the two groups. Levels of disease activity as measured by Bath Ankylosing Spondylitis Disease Activity Index (BASDAI), Ankylosing Spondylitis Disease Activity Score (ASDAS) and Routine Assessment of Patient Index Data 3 (RAPID3) were not significantly different among the two groups, though there was a trend towards higher scores in axSpA groups. Physical function, mobility limitation and physician global were comparable among the two groups.

Demographic and clinical characteristics of AS and nr-axSpA patients are shown in Supplementary Table S1, available at Rheumatology Advances in Practice online. Though not significantly different, patients with nr-axSpA were younger and more likely to be women as compared with those with AS. Dactylitis and ulcerative colitis were more prevalent in the AS group than in nr-axSpA. All seven patients with AS fulfilled mNY criteria, but only one had positive HLA-B27. Of 22 patients with nr-axSpA, 14 met the imaging arm of ASAS criteria, 5 met the clinical arm, and 3 patients did not meet the ASAS criteria.

The sensitivity and specificity of the individual A-tool questions for diagnosis of axSpA ranged from 0.03 to 0.86 and 0.14 to 0.96, respectively (Table 2). Positive likelihood ratios (+LR) of the A-tool items range from 0.84 to 1.34. A-tool items with >60% sensitivity included improvement with exercise, nocturnal back pain, plantar fasciitis and peripheral arthritis; items with >60% specificity included a history of PsO and IBD, uveitis and a family history of SpA. Items with positive LR >1 in descending order include a family history of SpA, improvement with exercise and good response to NSAIDs. No individual A-tool question showed a significant association with the outcome of axSpA in the logistic models.

Table 2.

Predictive values of the individual items in A-tool for axSpA diagnosis

	Sensitivity	Specificity	PPD	NPD	LR +	LR −
Improvement of backpain with activity	0.76	0.42	0.40	0.77	1.31	0.57
Backpain waking up at night	0.86	0.14	0.34	0.67	1.00	0.98
Improvement of backpain with NSAID	0.55	0.58	0.40	0.72	1.31	0.77
History of plantar fasciitis	0.62	0.32	0.32	0.62	0.91	1.20
History of peripheral arthritis	0.62	0.25	0.30	0.56	0.82	1.54
History of psoriasis, CD, UC	0.17	0.82	0.33	0.66	0.98	1.00
History of psoriasis	0.10	0.89	0.33	0.66	0.98	1.00
History of Crohn’s disease	0.03	0.93	0.20	0.65	0.49	1.04
History of ulcerative colitis	0.10	0.96	0.60	0.68	2.95	0.93
History of uveitis/iritis	0.10	0.88	0.30	0.66	0.84	1.02
Family history of psoriasis, Crohn’s disease, ulcerative colitis, ankylosing spondylitis	0.52	0.61	0.41	0.71	1.34	0.79
Family history of psoriasis	0.38	0.81	0.50	0.72	1.97	0.77
Family history of Crohn’s disease	0.21	0.93	0.60	0.70	2.95	0.85
Family history of ulcerative colitis	0.10	0.88	0.30	0.66	0.84	1.02
Family history of ankylosing spondylitis	0.10	0.84	0.25	0.65	0.66	1.06

When assessing the A-tool using a multivariable logistic model, odds ratios ranged from 0.76 (95% CI 0.14, 3.36) to 2.41 (0.87, 7.29) (Supplementary Table S2, available at Rheumatology Advances in Practice online). The area under the curve (AUC) for the ROC curve (Fig. 1) was 0.66.

Figure 1.

ROC curve for multivariable model using A-tool responses. Receiver operating characteristic (ROC) curve generated from a multivariable logistic regression model using patient responses to A-tool questionnaire items. The model included all individual A-tool items as predictors. The area under the curve (AUC) was 0.66, indicating modest discriminative ability of the tool to differentiate patients with axial spondyloarthritis (axSpA) from those without

Subsequently, we assessed the agreement between the patient responses on the online A-tool, and the corresponding physician-confirmed responses to the same questions at the research visit (Table 3). Kappa statistic for agreement between A-tool and analogous clinical questions ranged from 0.19 to 0.88. Most values were < 0.6, signifying low to moderate agreement.

Table 3.

Agreement analysis between patient responses on online A-tool and physician confirmed responses at the study visit

A-tool	Clinical question(s)	Kappa (95% CI)
Does your back pain improve with activities, exercises or movements?	Improvement of back pain with activities or exercises?	0.19 (0.0, 0.38)
Does the back pain wake you up at night?	Back pain at night with improvement on getting up	0.41 (0.15, 0.66)
Does the back pain resolve or get much better in 24 to 48 h after taking an NSAID such as ibuprofen (Advil, Motrin), naproxen (Aleve), meloxicam (Mobic) or diclofenac (Voltaren)?	Has there been a good response to NSAIDs?	0.26 (0.08, 0.43)
Have you had a current or past history of heel pain, especially in the morning (this is also called plantar fasciitis)?	Was there any enthesitis in past?	0.45 (0.27, 0.64)
Have you had unexplained joint pain with swelling?	i) Does patient have peripheral arthritis (synovitis) at present? ii) Does patient have history suggestive of peripheral arthritis in past?	0.22 (0.07, 0.38)
Has a physician ever diagnosed you with Psoriasis?	Did you ever have psoriasis of skin and or nails diagnosed by a physician or dermatologist?	0.86 (0.68, 1.0)
Has a physician ever diagnosed you with Crohn’s disease?	Did you ever have Crohn’s disease diagnosed by a gastroenterologist?	0.56 (0.11, 1.0)
Has a physician ever diagnosed you with ulcerative colitis?	Did you have ever have Ulcerative colitis diagnosed by a gastroenterologist?	0.88 (0.66, 1.0)
Did you ever have an inflammatory eye condition called uveitis or iritis, diagnosed by a physician?	Did you ever have diagnosis of uveitis or iritis by an ophthalmologist?	0.54 (0.23, 0.86)
Does any of your first-degree relatives (parents, full siblings or children) or second-degree relatives (grandparents, grandchildren, aunts, uncles, nephews, nieces or half-siblings) have the following?	Do you have any family history of below conditions in first degree relatives (parent, full sibling or children) in second degree relative (grandparents, grandchildren, uncles, aunts, nephews, nieces and half-siblings) ?
Psoriasis	Psoriasis	0.61 (0.42, 0.8)
Crohn’s disease	Crohn’s disease	0.69 (0.45, 0.92)
Ulcerative colitis	Ulcerative colitis	0.58 (0.32, 0.84)
Ankylosing spondylitis	Ankylosing spondylitis	0.26 (−0.03, 0.55)

Positive likelihood ratios for corresponding physician-confirmed responses ranged from 0.98 to 1.77, with most of them being just above 1 (Supplementary Table S3, available at Rheumatology Advances in Practice online). When assessing the physician-confirmed responses using a multivariable logistic model, the question related to family history of SpA had a significant association with the outcome (odds ratio 3.83 [95% CI 1.39–11.4]). The odds ratios of the physician-confirmed responses to individual A-tool questions ranged from 0.63 (95% CI 0.12–2.76) to 3.83 (1.39–11.4). Although not statistically significant, positive responses to A-tool questions about improvement with exercise, response to NSAIDs, peripheral arthritis and family history of SpA were more likely to be diagnosed with axSpA. The CIs are vast, but the sample size was small, and this was a pilot study that was not powered adequately to find statistically significant differences. The AUC for this logistic model is 0.708 (Fig. 2).

Figure 2.

ROC curve for multivariable model using physician-confirmed responses. ROC curve based on a multivariable logistic model using physician-confirmed responses corresponding to A-tool items collected during clinical visits. The model showed the strongest association with family history of spondyloarthritis. The AUC was 0.71, reflecting improved diagnostic performance compared with patient-reported responses alone

Discussion

Delayed diagnosis of axSpA is a significant problem worldwide, including in the USA [4, 5]. Early diagnosis is especially important now that we have newer effective therapies available and also, early treatment of patients with shorter disease duration is associated with better outcomes [10]. Early referral to a rheumatologist can significantly reduce the diagnostic delay [11]. It is estimated that about 5% of the patients with CBP have axSpA [12], and it is a daunting task to find these 5% patients among the vast majority of patients with chronic mechanical back pain.

Here, we studied the performance of a self-referral strategy using an online screening tool (A-tool) among patients with CBP. Among self-referred patients with CBP, 29% were diagnosed with axSpA. There were 29 patients with nr-axSpA and seven patients with AS, suggesting that an A-tool-based strategy can help identify patients in earlier stages of the disease. Our results were comparable to previous studies where approximately 35% of patients were diagnosed with axSpA [8, 13, 14]. The A-tool-based referral strategy demonstrated the qualities of an ideal screening test, including its cost-effectiveness, easily administration and applicability to large populations.

A-tool can be used to screen large population of CBP as it does not involve performing labs or imaging tests prior to referral. Incorporating digital technology, A-tool can be integrated into EMRs, made available on various websites or disseminated using social media. A-tool based strategy can be used by patients themselves (self-referral) or administered by non-rheumatology providers. The fact that 1274 patients took the survey confirms that using a self-referral strategy is a feasible approach in routine practice. We believe that patients are best advocates for themselves, and A-tool based self-referral strategy is more feasible than provider-based strategy given the limitations regarding poor awareness, knowledge of axSpA and time constraints of practices of referring non-rheumatology providers.

We found that patient portals in EMRs and social media platforms such as Facebook can be effectively leveraged to reach out directly to patients with back pain rather than depending on non-rheumatology providers for referral. Because axSpA is prevalent in young adults, the dissemination of online tools through social media is especially appealing. To our knowledge, this is the first study using social media for patient recruitment to help with the early diagnosis of axSpA.

Similar to previous studies, the presence of IBP, as recognized by rheumatologists, was associated with an increased likelihood of final diagnosis of axSpA [13]. Among the IBP items, improvement with exercise and response to NSAIDs were found to be particularly helpful. A-tool contains only high-yield items in the IBP criteria. Inclusion of the selective features is innovative concept and has not been used in the previous referral strategies. This essentially avoids one step in the evaluation of a suspected axSpA patient where a non-rheumatology clinician must first to decide whether the patient’s back pain is suggestive of IBP to determine the need for referral to a rheumatologist. Patients suspected to have axSpA based on history and physical examination at the study visit were more likely to have an axSpA diagnosis as an outcome. This underscores the importance of thorough history taking for the identification of axSpA until a specific diagnostic biomarker becomes available.

Among the physician-confirmed responses to the questions in the A-tool, the odds for an axSpA diagnosis were high for improvement in back pain with exercise, good response to NSAIDs, uveitis and a family history of SpA.

We did not identify significant differences in the individual SpA features among the axSpA and no SpA group, possibly due to the small sample size. Similar to other studies, the positive likelihood ratios of individual SpA features in the A-tool are low which underscores the need for using combination of SpA features in referral strategies to achieve good post-test probability.

ROC analysis of the pilot study data demonstrates poor discriminatory capacity between the two groups; this could be explained by the small sample size and inadequate power to detect the difference. Also, the AUC for the A-tool (0.66) and physician confirmed responses on A-tool (0.71) were modest. There was poor agreement between the patients’ responses to online A-tool questions and physicians’ confirmed responses to the corresponding questions during the study visit. Also, we found that patients were more likely to respond to online surveys, but upon the phone interview by coordinators, they are reluctant to come for the study visit. We found that some questions, particularly those about the peripheral arthritis and plantar fasciitis, were vague and did not improve the diagnostic yield. Also, we realized that the screening tool has two steps of prescreen and SQ. We have modified the A-tool to improve these deficiencies, and we plan to use modified A-tool for subsequent studies. With large-scale adoption of modified A-tool based strategy, we hope to reduce the diagnostic delay for axSpA.

The majority of the referral strategies described in previous studies have used labs (HLA-B27) and imaging tests [5, 13–15]; however, these strategies may not be feasible and cost-effective in the US healthcare system. PCPs in the USA are busy addressing comorbid medical conditions, health maintenance issues as well as challenges of administrative and other tasks [16] and may not have adequate time to perform as well as interpret the labs and imaging prior to referral.

There are at least three studies that used referral strategy based on clinical features only and also online tool or referral strategy directed towards patients. Our outcomes are comparable or better than these studies. Hermann et al. invited PCPs to refer patients <45 years fulfilling Calin’s criteria of IBP [17]. Two physicians made a diagnosis of axial SpA according to the criteria proposed by Rudwaleit et al. [18]. In this study, 30 (33%) of the 92 patients were diagnosed to have SpA, and 16 (53%) of them had AS.

Brandt et al. 2013 evaluated the performance of a patient-based online questionnaire in identifying patients with axSpA [19]. It contained items regarding back pain, other SpA features, elevated ESR or CRP, HLA-B27-positivity and imaging results (if available). Post-test probability (PTP) was calculated using the algorithm published by Rudwaleit et al. [20]. This is a complicated formula that may or may not be feasible in routine practice. Patients who had PTP of >50 were invited. Fourteen of the 97 patients (14%) were diagnosed to have axSpA (7 with AS and 7 with nr-axSpA).

In the SPACE cohort, patients with CBP for ≥3 months but ≤2 years with the onset before the age of 45 years were referred to the rheumatology outpatient clinic by GP and specialists. The rheumatologist diagnosed axSpA in 65/157 (41.4%) of the patients. Similar to our study, the SpA features were not significantly differently expressed by axSpA and no SpA groups [21, 22].

In the more recent OptiRef study, investigators compared the online self-referral tool (OSR) to the Berlin referral strategy. OSR was created as per the ASAS referral recommendations and was made accessible through major search engines, the website of the clinic and from posters in the public transportation in Berlin area [23]. A total of 35/180 patients (19.4%) in the self-referral group and 71/181 patients (39.2%) in the physician-referral group were finally diagnosed with axial SpA. Like our study, axSpA patients from the OSR group were more often HLAB27 negative, females and were more frequently at a non-radiographic stage.

Limitations

This is a pilot study, and a subsequent study with a larger number of patients is warranted for validating our results. In the present study, the sample size of enrolled patients was smaller than ideal sample size of 350 patients which would require a multicentre study with three to five collaborating centres. This pilot study will help to further refine the A-tool and undertake a larger study. Given the COVID-19 pandemic, our enrolment was limited. Due to the exploratory nature of this study, no multiple testing correction was performed. This may increase the risk of type 1 error. Poor performance of individual questions in the A-tool with low sensitivity (0.03 to 0.86), varying specificity (0.14 to 0.96) and the lower positive likelihood ratios (0.84 to 1.34) was a limitation of our study. It will be important to reassess their performance in a larger study.

Low referral completion was a specific problem that we faced. Even after multiple calls and follow-up emails, fewer patients responded and agreed to the study visit. This may introduce a selection bias in the study. It may also imply that patients who were having severe symptoms came for a visit, thus increasing the pretest likelihood of having the axSpA. Those with mild pain may not feel the need to spend time and money (travel, childcare arrangements, etc., as applicable) to undergo the assessments. However, one of the previous studies showed similar limitations [24], suggesting that this may be an issue with many self-referral strategies. This could indicate the possibility that patients underestimate the back pain due to the chronic nature and commonality of the back pain. Also, it may convey that patients are not as serious about online surveys distributed via social media as, to date, social media is being used more as an entertainment and communication tool than as a research tool.

Conclusion

In summary, our findings suggest that A-tool based strategy for self-referral of CBP patients is simple, practical and feasible approach for early diagnosis of axSpA. We need larger prospective study to validate our findings and confirm the effectiveness of this strategy.

Supplementary material

Supplementary material is available at Rheumatology Advances in Practice online.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.

Funding

Novartis Pharmaceuticals.

Disclosure statement: Abhijeet Danve: Research grants from Novartis and Eli Lilly; advisory boards for Janssen, AbbVie, Amgen, and Novartis. The remaining authors have declared no conflicts of interest. A-tool is the original work of Dr. Abhijeet Danve and is included here under a CC BY-NC license solely for limited academic, research, and non-commercial use. Any commercial use, reproduction, or distribution of the A-tool is prohibited without prior written permission from the author. For licensing or partnership inquiries, contact: drdanve@hotmail.com.