Systematic review and meta-analysis of diagnostic delay in axial spondyloarthritis

Background. Delay to diagnosis in axial spondyloarthritis (axSpA) is longer than many other rheumatic diseases. Prolonged delay has been shown to associate with poorer outcomes including functional impairment and quality of life. Our aims were to describe 1) global variation in delay to diagnosis, 2) factors associated with delay, and 3) differences in diagnostic delay between axSpA and psoriatic arthritis (PsA). Methods. We searched Medline, PubMed, EMBASE and Web of Science using a predened protocol in accordance with PRISMA guidelines. Delay to diagnosis was dened as years between age at symptom onset and age at diagnosis. We pooled mean diagnostic delay using random-effects inverse variance meta-analysis. We examined variations in pooled estimates using pre-specied subgroup analyses and sources of heterogeneity using meta-regression. Results. A total of 64 studies reported mean diagnostic delay in axSpA patients. The pooled mean delay was 6.7 years (95% condence interval 6.2 to 7.2) with high levels of heterogeneity. Delay to diagnosis did not improve over time when stratifying results by year of publication. Studies from high-income countries (dened by the World Bank) reported longer delay than those from middle-income countries. Factors consistently reported to be associated with longer delay were: lower education levels, younger age at symptom onset and absence of extra-articular manifestations. Pooled estimate for diagnostic delay from 8 PsA studies was signicantly shorter, at 2.6 years (95%CI 1.6 to 3.6). Conclusion. For axSpA patients, delay to diagnosis remains unacceptably prolonged in many parts of the world, although some countries have reported remarkable improvements. Patient factors (education) and disease presentation (age at onset and extra-articular manifestations) should inform awareness campaigns to improve delay. Targets for improvement should aim to resemble delays in other spondyloarthritis patients.


Introduction
Axial spondyloarthritis (axSpA) is a chronic in ammatory disease characterised by signi cant in ammatory pain, stiffness and functional impairment [1]. Symptoms typically begin in early adulthood, which is a critical time for education, career, social networks and development of personal identity in general. Consequently, axSpA can signi cantly impact on mental health, quality of life and work productivity over the life course, at costs to the individual and the economy [2,3].
The disease impact is often compounded by a prolonged diagnostic delay, that is, time from onset of symptoms to getting a diagnosis. This is may be explained by the lack of awareness of axSpA, the high prevalence of other causes of back pain, a perception that musculoskeletal symptoms are self-limiting in young adults, or limited access to rheumatology services. Duration of delay is reported to range from 8 to Page 3/15 study to study. Some studies have also found no improvement in diagnostic delay over recent decades [4], despite improved understanding of the disease and access to imaging.
There is abundant evidence that diagnostic delay is associated with worse functional impairment, greater radiographic progression, poorer quality of life and reduced response to treatment [5,6]. Those with longer delays to diagnosis also report greater work disability, unemployment and healthcare costs [5]. Although the impact of delay is well described, potential causes of delay (i.e., how delay can be improved) are not.
Examining how delay durations vary across parts of the world and factors associated with delay will help inform targets for improvement.
The aim of this systematic review was to 1) describe global variation in diagnostic delay and 2) describe patient and disease factors that have been reportedly associated with delay to diagnosis. We also sought to 3) formally compare delay duration in axSpA with other SpA (e.g., psoriatic arthritis) to highlight the need and target for improvement.

Methods
We performed a systematic review in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [11]. The protocol for this review was pre-registered in advance (PROSPERO: CRD42020161887). We searched Medline, PubMed, EMBASE and Web of Science for relevant literature in September 2019 using the following search terms: (ankylosing OR spondyloarthritis OR psoriatic) AND ((delay AND diagnosis) OR (symptom AND (onset OR duration))).
This search strategy was designed to include studies that describe diagnostic delay, even if delay was not their research objective. In response to peer review, bibliographies of all eligible studies and a prior systematic review [7] were also manually searched to identify additional titles.
Studies were included if they reported mean delay to diagnosis (i.e., the mean difference between age at symptom onset and diagnosis) or if they reported both mean age at onset and at diagnosis. Studies that de ned eligibility using these variables were not eligible since estimates would not be representative. We excluded studies using the same (or very similar) cohort to studies already included. We also excluded studies reporting medians only. Letters and published conference abstracts were considered, as some prevalence studies may not be published as full articles but may have su ciently detailed methodology and results. Reviews, comments and editorials were excluded.
Two independent reviewers screened titles and abstracts, assessed full-texts for eligibility and extracted data from qualifying studies (BP, NH). Any discrepancy at each stage was resolved through discussion moderated by a third reviewer. Information from included studies was extracted into prede ned tabulated summaries (Supplementary Table S1). Studies were assessed for risk of bias using adapted versions of the Newcastle Ottowa Scale (Supplementary Table S2).

Analysis
We pooled mean diagnostic delay using inverse variance weighted random-effects models (DerSimonian-Laird method). This was performed for studies of axSpA (including AS), then separately for psoriatic arthritis (PsA) and spondyloarthritis (SpA, which includes axSpA, PsA and other members of the SpA family). Where mean delay to diagnosis was not reported, it was imputed as the difference in mean age at symptom onset and mean age at diagnosis. Where the standard deviation of diagnostic delay was missing, we imputed it using methods recommended by Cochrane (in essence, based on standard deviations of age at onset, age at diagnosis and their correlation in all studies [8]) or the standard deviation of a study reporting the most similar mean delay duration. We performed sensitivity analyses without imputed values. Further sensitivity analyses requested at peer review were performed by 1) restricting to studies using classi cation criteria only, and 2) excluding studies with potentially nonrepresentative sampling (e.g., entirely male populations, or sample sizes of <30 that may give unstable population estimates). Heterogeneity of meta-analysis estimates was presented using the I 2 statistic. Funnel plots were used to assess risk of publication bias.
We used random-effects meta-regression to examine whether heterogeneity in axSpA diagnostic delay could be explained by study characteristics, i.e., year of publication (pre-2010, 2010-2015, post 2015), geography (regions de ned by the World Health Organisation [9]), economic status of the country (World Bank economic class [10]), sample sources (e.g., single centre, multicentre etc), age at symptom onset (tertile) and proportion of males (tertile). Meta-regression was not performed for PsA and SpA due the limited number of studies. Analyses were performed using R version 3.6.2 and the "meta" and "metafor" packages.

Results
A total of 3286 publications were found from the literature search. After excluding duplicates, irrelevant and ineligible studies, 86 studies remained. A further 9 studies were found through manual bibliography searches. 16 studies using the same cohorts (or subsets thereof) were excluded. The study by Rojas-Vargas et al was excluded as it only included patients with ≤2 years of symptoms. The selection owchart is shown in Supplementary Figure S1. The 78 included studies are summarised in Supplementary Table S1. 64 studies reported delays among axSpA patients, 8 PsA and 8 SpA. Feld et al [11] and Sørensen et al [12] reported delay in both axSpA and PsA. Bias scores were mostly 3 to 4 out of 6 stars (Supplementary Table S2 and Figure S2) indicating moderate bias.

Diagnostic delay in axSpA
Sample size for axSpA studies ranged from 5 to 2,887 patients. 47 studies were of AS (including 32 using modi ed New York criteria) and 17 of axSpA (including 11 using the ASAS criteria). Delay ranged from 2.8 years in a small Albanian study (of 54 cases over 6 years), to 11.1 years in a single UK centre [13,14].
Results of strati ed meta-analysis are shown in Table 2. 39 axSpA studies were from countries in the European region, 9 West Paci c, 8 Eastern Mediterranean, 5 Americas and 3 South East Asia. Across these WHO regions, the mean delay and heterogeneity were not signi cantly different. When these studies were strati ed according to World Bank economic class, the High-income group had longer mean delays than the upper-and lower-middle income countries. When mean delays were pooled according to country (with ≥3 studies), the average diagnostic delay was signi cantly shorter in Turkey and China than in the UK. Mean delay duration did not differ according to year of publication. Studies with older mean age of symptom onset showed trends for shorter delay durations.

PsA and SpA
Sample size for 8 PsA studies ranged from 69 to 1970 patients. Diagnostic delay ranged from 1.0 years in the Dutch South-West Psoriatic Arthritis to 4.6 in a Swedish population-based cohort [28,29]. The mean delay to PsA diagnosis was 2.6 years (95%CI 1.6 to 3.6, I 2 =99%) (Figure 2).
Eight SpA studies ranged from 16 to 708 participants in size and 1.6 to 7.6 years in diagnostic delay. The mean delay to SpA diagnosis was 4.9 years (95%CI 3.3 to 6.6, I 2 =96%) ( Figure 2). Sensitivity analysis restricting to 3 PsA and 6 SpA studies using classi cation criteria showed similar results (supplementary gure S5).

Discussion
The mean delay to diagnosis was 6.7 years across 64 axSpA studies worldwide. Interestingly, countries classed as high-income by the World Bank had signi cantly longer delays to diagnosis than mediumincome countries. Factors associated with delay to diagnosis varied and were often contradictory across studies; the most consistently reported factors were lower education, absence of extra-articular manifestations and younger age of onset. Diagnostic delay in axSpA was signi cantly longer than studies of PsA (2.6 years) and SpA (4.9 years).
Mean duration of delay varied signi cantly within (e.g., from 5.7 to 11 years in the UK) and between countries. This may re ect multiple factors that could not be assessed in this review, such as local healthcare infrastructure and awareness of the disease. Our nding that delay was longer in high-income countries was unexpected. It may be that research centres in these countries received referrals for the most diagnostically challenging cases or served comparably deprived areas. Conversely, it may be that only centres with good referral infrastructure are publishing research in middle-income countries.
Our meta-analysis showed no meaningful change in diagnostic delay over (publication) time. This is consistent with results from the UK [4,30], France [21] and Germany [20]. In stark contrast, delay to diagnosis improved dramatically in Japan  [32]) and Australia [27]. We could not examine the cause of this variation in detail, but diagnostic approaches likely varied from country to country. For example, the extent to which HLA-B27 and gender were associated with delay differed between countries, suggesting that these factors may have differential importance in their respective diagnostic process.
In ammatory back pain in axSpA typically has an insidious onset, with subtle signs on clinical examination. There is also a plethora of highly prevalent differential diagnoses that may be incorrectly used to explain symptoms; for example, lumbar disc disease can co-exist with axSpA and prolong delay to diagnosis [6,26]. Peripheral joint involvement is relatively more acute in presentation, with clearer signs such as swelling and erythema. This may explain the much shorter diagnostic delay in PsA than in axSpA. Among axSpA studies, the presence of peripheral joint involvement was associated with shorter delay to diagnosis in Italian [15], UK [4], French [21] and Japanese [22] studies, while these patients had longer delays in Iran [16,18]. It may be the case that these Iranian patients were given other diagnoses prior to the correct axSpA label.
To reduce delay to diagnosis, intuitive targets would be to improve awareness of axSpA as a cause of back pain; general education was inversely associated with delay. Younger age of onset was also consistently associated with prolonged delay. (Although this may be an artefact of "delay" being derived from, and being dependent on, age at onset.) Education is needed among non-rheumatologists that axSpA is a cause of back pain in young people. However, there will be cases that remain more diagnostically challenging, such as patients with few SpA features.
A key strength of this review is the large and globally representative number of studies. Our unique search strategy allowed us to include studies that described diagnostic delay, even if delay was not their research objective. Such studies are less likely to be subject to subconscious bias from a prior delay hypothesis, thus their inclusion is a strength rather than weakness. There were however limitations. Diagnostic delay is known to be right-skewed in distribution, meaning that the mean is in ated above the median by a high proportion of people with disproportionately long delays. In other words, the mean may be sensitivity to these outliers (e.g., atypical clinical features or individuals with poor access to healthcare) and remain unchanged, even if diagnostic delay generally improved for many patients. We chose mean rstly because it permits meta-analysis, but also because median would take emphasis away from those with unusually long delays -precisely the individuals needing improvement to diagnosis. Some meta-analysis estimates for delay had negative lower-bounds in the con dence interval, which is not possible by de nition. This is an artefact of the random-effects methodology; in each case, there is one study with a much shorter delay than others in the category, resulting in wide intervals required to cover the pooled estimate for this subgroup. This artefact disappears in xed-effects models, which were not used in this study due to high heterogeneity between the studies. We did not review the impact of delay to diagnosis as this was recently reviewed by Yi et al [5]. Meta-regression examines relationships between summary data and should not be interpreted as traditional hypothesis testing of individual patient data. For example, proportion of males was not associated with diagnostic delay; this does not rule out a difference in delay between the sexes. Although most studies in our review did not report a statistically signi cant difference, a prior meta-analysis of SpA (excluding PsA) did [7]. Delay over time should also be interpreted with caution. Year of publication was the only available proxy for calendar time, since the recruitment period can be over many years and was often not reported. The intervals between recruitment and publication were generally homogenous in studies that did report this data.

Conclusion
The delay from symptom onset to diagnosis is 6.7 years on average in axSpA, which is signi cantly longer than 2.6 years for PsA. Although delay has improved over time in some parts of the world, many countries such as the UK need additional efforts to improve delay to diagnosis. Lower education levels, absence of EAMs and younger age of onset were associated with longer delays; therefore, improved education for physicians and patients with back pain may help reduce diagnostic delay.  Pooled estimate of diagnostic delay in axial spondyloarthritis (including ankylosing spondylitis). Results ordered according to geography and year of publication. *Diagnostic delay calculated from summary data for age at onset and diagnosis. **Standard deviation imputed for meta-analysis.

Figure 2
Pooled estimate of diagnostic delay in psoriatic arthritis and spondyloarthritis. *Diagnostic delay calculated from summary data for age at onset and diagnosis. **Standard deviation imputed for metaanalysis.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.