Validity of ankylosing spondylitis and undifferentiated spondyloarthritis diagnoses in the Swedish National Patient Register.

OBJECTIVES
Epidemiological studies of spondyloarthritis (SpA), using ICD codes from the Swedish National Patient Register (NPR), offer unique possibilities but hinge upon an understanding of the validity of the codes. The aim of this study was to validate the ICD codes for ankylosing spondylitis (AS) and undifferentiated SpA (uSpA) in the NPR against the established classification criteria [modified New York (mNY), Assessment of SpondyloArthritis international Society (ASAS), Amor, and European Spondyloarthropathy Study Group (ESSG) criteria].


METHOD
All patients with an ICD-8/9/10 code of AS or uSpA in the NPR 1966-2009 at a visit to a specialist in rheumatology or internal medicine or corresponding hospitalization, alive and living in Sweden 2009, were identified (n=20,089). Following a structured procedure to achieve geographical representativeness, 500 random patients with a diagnosis of AS or uSpA in 2007-2009 were selected. Based on a structured review of clinical records, positive predictive values (PPVs) for fulfilling the criteria sets were calculated.


RESULTS
For those having received an ICD code for AS, the PPVs for fulfilling the mNY criteria or any set of SpA criteria were 70% and 89%, respectively. For those with an uSpA diagnosis (and never an AS diagnosis), the corresponding PPVs were 20% and 79%. The subset with both AS and uSpA diagnoses (overlap=12%) were as likely to fulfil the mNY criteria as the group that had been coded as AS only.


CONCLUSIONS
The diagnosis codes for AS or uSpA had high PPVs, suggesting that our case identification in the Swedish NPR can be used for nationwide, population-based, epidemiological studies of these diseases.

Patients with a diagnosis of spondyloarthritis (SpA) represent a heterogeneous spectrum of phenotypes, where patients at one end have a predominantly axial disease with syndesmophyte formation (ankylosing spondylitis, AS), and at the other end have a predominantly peripheral disease with enthesial and synovial involvement. Most frequently these manifestations occur together and often in combination with other related inflammatory manifestations such as anterior uveitis, psoriasis, and inflammatory bowel disease (IBD).
Attempts that have been made to classify patients with SpA include the Rome criteria (1), the modified New York (mNY) criteria for AS (2,3), the European Spondylarthropathy Study Group (ESSG) (4) and Amor (5) criteria for SpA, and most recently the Assessment of SpondyloArthritis international Society (ASAS) criteria for axial SpA (6), as well as peripheral SpA (7). In the Swedish healthcare system, these subjects are diagnostically coded according to the World Health Organization (WHO) International Classification of Disease (ICD) codes (8) and subsequently registered in the National Patient Register (NPR). The ICD coding system has not changed in parallel with the classification criteria. In the ICD-10 (and previous ICD versions) there are specific codes for some specific entities of SpA, such as AS (M45.9), psoriatic arthritis (M07.0-M07.3), reactive arthritis (M01.3-M02.9, M03.6), and arthritis associated with IBD (M07.4-M07.5). The remaining group, not being categorized as either AS or any of the other specific forms of SpA, may only be given the non-specific codes for undifferentiated SpA (uSpA) (M46.0, M46.1, M46.8, M46.9; also including sacroiliitis and enthesiopathy of the spine). The latter group is likely to include a majority of the non-radiographic axial SpA not related to psoriasis, IBD, or reactive arthritis.
Previous epidemiological studies have focused primarily on AS, while studies on other forms of SpA are relatively scarce. Approaches in studies of prevalence of AS include identification of cases through telephone surveys (9,10), blood donor registers (11)(12)(13)(14), relatives of patients (15), hospital records (16,17), or small community-based studies (18)(19)(20). Identification of individuals with SpA or AS on a national level, based on healthcare registers, has not previously been described, but offers unique possibilities to study the occurrence, risk factors, and outcomes for this group of diseases. However, the Swedish NPR is used extensively for research for a number of other diseases, where the positive predictive value (PPV) for diagnoses in the In-Patient Register (IPR) has varied between 85% and 95% (21). For rheumatoid arthritis (RA) the PPV has been shown to be 90% (22), but for AS and the other SpA subtypes it has not previously been investigated at a national level. Before using healthcare data from the NPR, the validity of the ICD codes used by the healthcare system to classify these patients needs to be also established for AS and the other SpA subtypes, The aims of this study were: first, to validate the ICD codes in the NPR, for AS (M45) and uSpA (M46.0, M46.1, M46.8, M46.9) against the mNY criteria and the various classification criteria for SpA; and second, to assess the possibility of identifing incident cases based on the first date of a recorded diagnosis.

Study setting
The NPR is composed of the IPR and the Out-Patient Register (OPR). The IPR was initiated in 1964 and has since then (with an increasing number of counties enrolled) collected data (medical and administrative) from all patients discharged from hospitals in Sweden. From 1987 the coverage is considered to be close to 100% (23). The OPR for specialist care was initiated in 2001 and contains data on specialized out-patient care (such as a visit to a rheumatologist). The coverage is 80%, which is lower (24) than for the IPR, mainly because of missing data from private caregivers. Furthermore, primary care is not included in the OPR.
The medical data collected include primary and secondary (up to eight) diagnoses, injuries, procedures, age, and sex. The administrative data include dates of admission and discharge, hospital, department, and county. The data are linked to a personal identity number (PIN), allowing for identification and cross-linking with other national Swedish registers. The use of PINs is central to the Swedish healthcare system, and allows for deterministic record linkage. Under certain circumstances PINs can be reused or changed but this accounts for less than 0.7% of all PINs assigned since 1969 (25).
Diagnoses are registered according to the Swedish version of the ICD: ICD-8 for the period 1968-1986, ICD-9 for the 1987-1996, and ICD-10 for 1997 to the present day (26).

Study population
All patients registered with an AS-or uSpA-related diagnosis in the NPR, 1966-2009, who were alive and residing in Sweden in 2009, were identified. The year 1966 was chosen as the first year because the New York classification criteria for AS were published in that year (27). A subgroup of the AS and uSpA cases identified was selected for the validation exercise, based on receiving an AS or uSpA diagnosis at least once (in-patient or outpatient care) in a department of rheumatology or internal medicine 2007-2009. The rationale for this was to ensure that the medical records would be accessible, sometimes dating back to 1966. AS and uSpA patients in Sweden are usually treated in out-patient clinics, not necessarily in specialized rheumatology or internal medicine departments, but also in primary care. The vast majority, however, are likely to be diagnosed at least once in specialized care (28), most often at a department of rheumatology (29).

Validation cohort
From the subset of register-identified patients, diagnosed with AS or uSpA in a department of rheumatology or internal medicine 2007-2009, 500 patients were randomly selected in a stratified mode. Conditions for selection were set up to maximize the probability of retrieving the data necessary for validation, with a nationwide geographical distribution ( Figure 1). The selected patients had to have at least one out-patient visit in 2007-2009, at one of five rheumatology departments (100 patients each), in different regions of Sweden (those surrounding Stockholm, Gothenburg, Malmö/Lund, Umeå, and Falun). Hospitals were selected to represent a gradient from north to south as well as both county and university clinics. From each department 25 patients were randomly selected based on having been given an AS diagnosis before 2007; 25 based on a first AS diagnosis 2007-2009; 25 on having been given an uSpA diagnosis before 2007; and 25 on a first uSpA diagnosis 2007-2009. A randomly generated number was applied to every identified patient at each location, and the 25 lowest numbers from each group were selected. The stratified selection enabled validation separately of AS and uSpA, as well as for cases that, from a register identification point of view, appeared to be incident (first occurrence of the relevant ICD codes during 2007-2009) and prevalent, respectively.
To retrieve data on the time-point of disease onset, time-point of diagnosis, and the information needed to assess fulfilment of the various classification criteria, information from the 500 records was extracted following a structured procedure. Classification criteria, for which information was sought, were mNY (3), ASASperipheral (7), ASAS-axial (30), ESSG (4), and Amor (5) criteria. For the criteria features related to imaging, reports from radiologists were used to judge whether the patient fulfilled radiographic sacroiliitis and sacroiliitis on magnetic resonance imaging (MRI), according to the mNY or ASAS criteria, respectively. For the features corresponding to biomarkers [C-reactive protein (CRP) and human leucocyte antigen (HLA)-B27], information was retrieved from both laboratory records and the physicians' notes in the medical records. For the remaining features information was sought among the physicians' notes in the medical records: inflammatory back pain (IBP), alternating buttock pain, uveitis, enthesitis, dactylitis, synovitis, psoriasis, previous urethritis, cervicitis, or diarrhoea, impaired spine mobility, family history for SpA (and SpA-related disease manifestations), and good response to non-steroidal anti-inflammatory drugs (NSAIDs). Only information about enthesitis in the heel was recorded, not counting enthesitis in other locations (contrary to the instructions for the ASAS criteria for peripheral disease).
In addition to validating the diagnoses against the different criteria, we attempted to record how confident the physicians were regarding the diagnoses (the gold standard for the diagnoses), according to the given ICD codes, based on notes in the medical records. The review of the medical records was performed, at the respective hospitals, by five different reviewers, trained in rheumatology, using a structured Case Report Form (CRF) discussed and agreed upon in advance by the reviewers.
PPVs for the different sets of criteria were determined separately for the cases with a diagnosis of AS, those with a diagnosis of both AS and uSpA, and those with only a diagnosis of uSpA. The PPV was calculated by dividing the number of cases fulfilling the different criteria sets by the number of patients in the respective diagnosis group, based on ICD codes. Missing data were treated as negative, but separate subset analyses were performed for only the cases with the relevant data available (see subset analyses in next section). Sensitivity and specificity was not calculated, as this would have required a second population to determine the proportions of false and true negatives.

Subset analyses
Because of the retrospective design of the study, spanning more than four decades, a fair amount of missing data for the different criteria features was expected. As imaging results and/or HLA-B27 are essential to fulfilling the mNY and ASAS criteria, a separate analysis of PPV was performed, including only the portion of cases where this information was available. Separate subset analyses were also performed on the AS cases without available radiographic data, and for men and women.

Ethical approval
The study was approved by the regional ethical review board at the Karolinska Institute in Stockholm. Consent from individual patients was not required.

Statistics
Between-group comparisons for fulfilling different criteria sets were performed using χ 2 tests, by the five selected hospitals, by age at the time of review (above vs. below the median age), and by sex. Data were analysed using SPSS version 21 (IBM Corporation, NY, USA).

Results
In total, 29 673 individuals were identified in the NPR with an AS or uSpA diagnosis, still alive and living in Sweden 2009. Of these, 20 089 individuals met the study criteria ( 1 AS or uSpA diagnosis from a physician at a department of rheumatology or internal medicine 1966-2009), 11 472 with AS (34% women) and 8617 with uSpA (and never AS) (56% women). The proportion who had been given both types of diagnoses was 12% (n ¼ 2387). The subgroup of the cohort, with a diagnosis 2007-2009, from which the validated cases were selected, consisted of 5918 individuals, 3310 with AS and 2608 with uSpA (and never AS).

Validation cohort: characteristics and criteria fulfilment
Of the 500 medical records examined, 499 clinical records were recovered and reviewed. Five cases were identified as incorrectly coded (four RA and one Cogan's syndrome) and three cases had altered PINs (reused or changed). These five were not excluded, as this probably reflects the frequency of incorrect coding in the whole cohort.
The basic demographics and frequencies of criteria features are presented (Table 1) for three groups based on ICD codes in the NPR: (i) those having ever received an AS diagnosis (n ¼ 250), (ii) those having received only a diagnosis of uSpA and never a diagnosis of AS (n ¼ 186), and (iii) those having received both types of diagnoses (n ¼ 137), representing an overlap group. In the AS patients compared to uSpA, there was a lower age for symptom onset and a higher frequency of men, radiographic sacroiliitis, IBP, and impaired spine mobility (according to the medical records) ( Table 1). For all these features, the overlap group had similar frequencies to the group with AS. Of the 250 AS cases, 88, 56, and 14% had available information in the clinical records on radiographic status of the sacroiliac (SI) joints, HLA-B27, and MRI of the SI joints, respectively. For the uSpA cases, this information was available for 76, 65, and 15%, respectively.
In the AS group, 70% fulfilled the mNY criteria and 79% the ASAS criteria for axial SpA (Table 2). For uSpA, 79% fulfilled at least one of the established classification criteria for SpA, 73% either of the ASAS criteria for peripheral or axial SpA, but only 20% fulfilled the mNY criteria (Table 2). In the whole validated cohort (n ¼ 500), 98% had at least one of the SpA criteria features listed in Table 1. Of the 21 cases with only one feature, nine had synovitis, IBD, uveitis, or radiographic sacroiliitis. The remaining cases had only a family history of SpA, good response to NSAIDs, HLA-B27, impaired spine mobility, history of urethritis, cervicitis, or diarrhoea.
When applying the classification criteria to the overlap group, again the frequencies were overall more similar to the group with AS than to the uSpA group.
In the AS group there was a considerable overlap in fulfilling the axial ASAS, Amor, ESSG, and mNY criteria, where the latter almost completely constituted a subgroup of those fulfilling the axial ASAS criteria (Figure 2A). For uSpA there was a considerable overlap between those fulfilling the ASAS (peripheral or axial), Amor, and ESSG criteria. For uSpA, those fulfilling the mNY criteria represented a minority completely covered by the other criteria ( Figure 2B).
We also attempted to validate the ICD codes against the physicians' expert diagnosis. However, it turned out to be difficult to, retrospectively, judge the physicians' confidence in the diagnoses, based on the medical records. This resulted in a missing data rate of 40-60%, depending on diagnosis.
In Table 3 the frequencies of anterior uveitis, psoriasis, and IBD for the validated AS and uSpA cohorts (based on review of medical records) are compared to the frequencies in the whole cohorts identified from the NPR (based on ICD codes). The frequencies of uveitis were higher in the validated cohorts, both in AS and uSpA, while the frequencies for psoriasis and IBD were similar.

Incident cases: age of symptom onset and delay of diagnoses
For the subgroup who had their first diagnoses of AS or uSpA in 2007-2009, we evaluated how close in time this date was to the actual onset of symptoms according to the review of clinical records. The age for first reported symptoms, as documented in the medical records, of AS was lower compared to that in uSpA [mean 25.3 (SD 9.3) vs. 31.7 (SD 13.0) years; p ¼ 0.02]. Of the incident and prevalent AS cases, 95% and 97%, respectively, had symptom onset before the age of 45 years. On average, the mean time from onset of first symptom of disease to the date of first diagnosis by a specialist in rheumatology or internal medicine was 16.6 years (SD 12.3, min 0; max 53) for AS and 9.2 years (SD 11.0, min 1, max 57) for uSpA.

Subset analyses of the validation cohort
When examining only the 220 (88%) AS cases with known radiographic status of the SI joints, 80% fulfilled the mNY and 86% the ASAS criteria for axial SpA. For the portion of uSpA with available data on imaging or HLA-B27 (n ¼ 166; 89%), 89% fulfilled at least one of the SpA criteria, 82% either of the ASAS criteria (axial or peripheral), and only 26% the mNY criteria (Table 2).
For 30 (12%) cases identified as AS in the NPR, radiographic data could not be located and therefore the mNY criteria were not applicable. Of these, three were misdiagnosed (two RA, one Cogan's) and four (13%) had psoriasis. Specific analysis of these 30 cases did not reveal any significant differences compared to the 220 patients where radiographic information was available, concerning sex, sample location, or age at first symptoms. The mean age in 2007 of these 30 patients was significantly higher (54 vs. 44 years, p < 0.01), which may   explain the greater difficulty in retrieving radiographic data or other information from the time of diagnosis. In the AS group as a whole, 70% fulfilled the mNY criteria. As register validation, based on review of medical records, is limited by the information available, this should be interpreted as a worst-case scenario where all of the 30 cases (with unknown radiographic status) are assumed not to fulfil the mNY criteria. If all 30 AS cases without available radiographic data were assumed to fulfil the mNY criteria (best-case scenario), 82% of the AS cases would fulfil the mNY criteria.
In the AS group men fulfilled the mNY criteria more often than women and in the uSpA group men fulfilled the Amor, ASAS-axial and ASAS-axial/peripheral criteria more often than women (Supplementary Online  Table S1).
There was no significant difference in fulfilling the criteria between cases (both AS and uSpA) with an age above or below the median age (43 years) in 2007 (data not shown).
There were no significant geographic differences either for the AS group or for the uSpA group with regard to fulfilling the mNY, ASAS-axial, or ESSG criteria (data not shown).

Discussion
In this study we found a PPV for AS diagnoses, given at a visit to a specialist in rheumatology or internal medicine, of 80% for fulfilling the mNY criteria for AS and 88% for fulfilling the ASAS criteria for axial SpA, if relevant imaging and/or HLA-B27 data were available. For uSpA, the PPV was 79% for fulfilling at least one of the established classification criteria for SpA, and 82% fulfilled either of the ASAS criteria, if imaging or HLA-B27 data were available. However, although there was a considerable overlap between the different sets of criteria for SpA, 75% of the uSpA cases who fulfilled at least one of the criteria set did not fulfil all three, illustrating the problem encountered when defining a disease through a set of criteria ( Figure 2B).
The frequencies of anterior uveitis, psoriasis, and IBD were in the same range in the validated cohorts compared to the total, register-based cohorts. This lends support to the notion that the validity assessed in the subgroup, given a diagnosis at a visit to a specialist in rheumatology or  internal medicine 2007-2009, could be generalized and extrapolated to larger cohorts. Unfortunately, the long lag time between onset of symptoms and first visit to a specialist in internal medicine or rheumatology precludes use of the latter date as the time-point for disease onset. Our intention was also to validate the diagnoses against the physicians' 'expert diagnosis', but this subsequently failed because of the subjective difficulty in determining the physicians' certainty of the diagnoses in the medical records. As the physicians' diagnosis is often considered the gold standard, this is unfortunate. However, the ICD codes registered are likely to reflect the physicians' opinion concerning the clinical diagnoses.
As it is inherently difficult to retrospectively retrieve information on all historical features of the disease, our results are likely to underestimate the occurrence of symptoms and signs as well as the fulfilment of classification criteria and hence the PPVs. The PPVs presented here thus suggest that using national registers for case identification by ICD codes can be a valuable tool for epidemiological research, for example for exploring risk factors and outcomes in large groups of patients with AS or uSpA. However, if the AS and uSpA cohorts identified in the NPR are to be used in further epidemiological studies, it is important to note that only subjects diagnosed at departments of rheumatology or internal medicine were included in the present validation analyses.
As shown in previous studies (28), there was an inversed gender distribution between the AS and the uSpA groups, with the latter group tending to be slightly older at onset of symptoms. The fact that men tend to develop more radiographic changes during the disease course (31) possibly explains the higher proportion of men fulfilling the mNY, Amor, and ASAS criteria, as radiographic sacroiliitis is central in these criteria sets.
The data also illustrate that there is an overlap between diagnoses, with a fraction (12%) of the patients having been given both AS and uSpA diagnoses over time. This is partly due to a transition from uSpA to AS, but also represent a difficulty in delineating the two groups clinically, a problem that is enhanced by the lack of concordance between ICD classification and classification criteria for research. The existence of multiple different sets of classification criteria, with varying degrees of overlap, further complicates this. In the overlap group, characteristics such as symptoms, signs, and fulfilment of criteria were overall more similar to those of the AS group than to those of the uSpA group, which may justify labelling them as AS for some research purposes.
The strengths of the present study include the nationwide approach, the validation of several different sets of criteria, the structured and stratified procedure when selecting cases for review to guarantee representativeness, and that almost all selected cases for record review could be traced and identified, using the PIN that identifies Swedish subjects. Some limitations should be acknowledged. First, some essential information regarding important criteria elements such as radiographs of the SI joints and HLA-B27 tests could not always be recovered or had not been performed. However, if such information had been available, it would have resulted in higher estimates for criteria fulfilment. In addition, only enthesitis in the heel was counted, which probably resulted in an underestimation of cases fulfilling the ASAS-peripheral criteria.
Second, in this study we restricted the analyses to patients with AS-or uSpA-related diagnoses, and did not explore the validity for other specific SpA diagnoses such as psoriatic arthritis, IBD-associated arthritis, or reactive arthritis. This will be the scope of future studies. Therefore, we still lack information regarding the validity for these subtypes of SpA.
Third, we did not include patients seen only in out-patient private care, which according to a previous report constitute less than 5% of the total patient group (unpublished data). Fourth, we did not include patients only registered with a diagnosis of AS or uSpA in primary care (who are likely to have a milder disease). For AS this only constitutes 3% of all patients with an AS diagnoses according to published regional data from southern Sweden (29).
To conclude, our results indicate that the Swedish NPR can be used to identify individuals with AS and uSpA, according to current classification criteria, and that further epidemiological studies can be based on this nationwide register resource.