Introduction

Radiographic damage is one of the core outcomes in axial spondyloarthritis (axial SpA) (including both non-radiographic axial SpA and ankylosing spondylitis (AS)) as recommended by the Assessment of SpondyloArthritis international Society (ASAS) [1]. Cross-sectionally, it is associated with impairment in spinal mobility [2, 3] and longitudinally with functional disability [4], emphasizing the importance of assessment. ASAS recommends routine radiography of the lateral cervical and lumbar spine for assessing damage over time, but radiographs should not be repeated more frequently than every 2 years, unless indicated in individual cases, who might show faster progression [5, 6].

Different scoring methods have been developed to quantify structural damage in axial spondyloarthritis: the Bath AS Radiology Index (BASRI) [7], the Stoke AS Spine Score (SASSS) [8] and a modification of the SASSS, the mSASSS [9]. In a formal comparison, the mSASSS has shown best reliability and sensitivity to change [10]. Consequently, it is the preferred scoring method for assessing structural damage in the spine for use in clinical trials, as endorsed by ASAS and Outcome Measures in Rheumatology Clinical Trials (OMERACT) [11]. The mSASSS assesses the presence of erosions, sclerosis, squaring, syndesmophytes and bridges at the anterior vertebral corners (VCs) of both the cervical and lumbar spine [9].

More recently, a new scoring method, the Radiographic AS Spinal Score (RASSS), has been proposed that includes the lower thoracic vertebrae, under the hypothesis that most progression is found in these segments [12]. Four thoracic VCs are added and the same features are scored as for the mSASSS, though with slightly modified scoring rules.

The usefulness of the RASSS has not been further evaluated thus far. Hence, it is important to compare both scoring methods, in order to establish the preferred method for the assessment of structural damage as an outcome measure.

Outcome measures should be valid in all their aspects. To standardize the nomenclature of validity, the OMERACT filter has been proposed and this includes three aspects: discrimination, truth and feasibility [13]. The main objective of the present study was to compare the mSASSS and RASSS with regard to performance, taking the aspects of the OMERACT filter into account.

Materials and methods

Patients and radiographs

Radiographs from patients included in the Outcome in Ankylosing Spondylitis International Study (OASIS) were used [14, 15]. The OASIS study is a prevalence cohort including 217 consecutive patients with AS from the Netherlands, Belgium and France that started in 1996. According to protocol, cervical and lumbar spine radiographs were taken biannually for 12 years, with a total of seven possible time points per patient. For the present study, patients were included if they had at least one time point in which at least one of the radiographic damage scores could be calculated.

Scoring methods

The two scoring methods used were the mSASSS [9] and the RASSS [12] (Table 1). In the mSASSS the anterior VCs of the cervical (lower border of C2 to upper border of T1) and lumbar (lower border of T12 to upper border of S1) segments (a total of 24 VCs) are scored at a lateral view, for the presence of erosion and/or sclerosis and/or squaring (1 point), syndesmophyte (2 points) and bridging syndesmophyte (3 points). The total score ranges from 0 to 72 [9]. The RASSS is similarly scored as the mSASSS with 3 modifications: 1) inclusion of the lower thoracic spine (lower border of T10 to upper border of T12; total of 28 VCs); 2) erosions are not scored; 3) squaring is not scored in the cervical spine. The RASSS ranges from 0 to 84 [12].

Table 1 Description of the mSASSS and RASSS scoring systems.

The radiographs were independently scored according to both scoring methods by two trained experts (SR and CS) who were blinded to demographic and clinical data. Both readers registered all the changes identified in each VC (for example, erosions, sclerosis and squaring) separately so that afterwards both scores could be computed. Because radiographs were taken in different formats during the 12 years of follow-up, enabling the readers to identify the points in time, they were scored with known chronology. All the available films per patient were scored at the same time.

Only scores of radiographs with ≤ 3 missing VCs per segment (either cervical or lumbar) were used. For the RASSS, the same rule applied and the four additional thoracic VCs were considered part of the lumbar segment [12]. Reliability between the two readers was explored using Bland and Altman analysis [16] on the progression intervals. All radiographs from patients with at least one score being beyond the 95% level of agreement were independently scored by an adjudicator (AvT). Averaged scores per VC of the two primary readers were used. In adjudicated cases, the score of the primary reader closest to the adjudicator was used. Missing VCs were imputed using an adaptation of the last-observation-carried-forward methodology. First, a missing value for a VC was replaced with the value of the previous observation. Then, the mean spinal segment's progression score (either cervical or lumbar) per patient was calculated. This was added to the imputed value, in an attempt to more accurately reproduce the true progression. This rule was applied assuring that the score achieved per VC never exceeded a score of 3. Similarly, in case of a score missing in a patient with a score of 0 in the same VC at a subsequent time point, the score of 0 for the previous time point(s) was assumed. If the baseline score of a VC was missing, the same procedure was applied, subtracting the mean segment progression from the score of year 2 for a particular patient. If a value of this VC was also missing at year 2, then an average of the other available VCs from this spinal segment at baseline was used to replace the missing VC(s).

Status and progression scores were calculated for both scoring methods. Status scores refer to the score in each of the available time points (at baseline and every 2 years thereafter). Progression scores were calculated as the difference between the status scores of two time points. Two-year progression scores refer to the progression occurring within 2 years, that is, status score of one time point minus the status score of the immediately previous time point. Twelve-year progression scores were computed as the score at year 12 minus the score at baseline.

Use of the OMERACT filter to compare the scoring methods

The mSASSS and the RASSS were judged with respect to the different aspects of the OMERACT filter: truth, discrimination, and feasibility [13].

Feasibility

The feasibility aspect of the OMERACT filter addresses the question: can the measure be applied easily, given constraints of time, money and interpretability? The feasibility of both methods (mSASSS and RASSS) was assessed. Because the RASSS requires a further four additional thoracic VCs to be present in the radiograph of the lumbar spine, the assessment of the ability to obtain both scores is important. The availability of the VCs and the ability to assess the status and the 2-year progression scores of both scoring methods was compared and the number of available VCs out of the four additional VCs included in the RASSS was also investigated. Comparisons were performed calculating a ratio of the available cases for the RASSS over the mSASSS, taking all radiographs into account, but also restricted to 1) patients with a RASSS available at year 12 in order to assess whether the RASSS would perform differently in the subset of patients with a complete follow-up and 2) patients with the first interval between years 0 and 2 available to compare with other results available in the literature for the RASSS [12].

Discrimination

The discrimination aspect focuses on the question: does the measure discriminate between situations of interest? This aspect of the OMERACT filter pertains to sensitivity to change and reliability.

Inter-observer reliability was assessed for both status and progression scores for both mSASSS and RASSS, by means of Bland and Altman plots [16] and by calculation of the smallest detectable change (SDC) for each method. The SDC is the smallest change that can be detected beyond measurement error to determine change in an individual and was calculated as follows [17]: SDC = 1 . 96 * SD diff k * 2 . SD diff is the standard deviation (SD) of the set of differences in change scores obtained by two readers; k is the number of readers whose change scores are used (here: k = 2).

To obtain insight into sensitivity to change of the methods, the means and SDs of baseline, 2-year and 12-year status scores were assessed. Effect sizes (for all 2-year progression scores) were calculated for both mSASSS and RASSS dividing the mean value of the progression scores by the corresponding standard deviation.

Truth

The truth aspect deals with the question: is the measure truthful, does it measure what is intended? Is the result unbiased and relevant? Both mSASSS and RASSS are, to a certain extent, similar, which means that they have a common part of construct validity. Therefore, we assessed the potential additional value of including the thoracic vertebrae in the RASSS, by determining the relative contribution (in %) to the 2-year total RASSS progression of each spinal segment (cervical, thoracic and lumbar) in comparison to the expected contribution. A balanced segmental progression, proportional to the number of VCs assessed in the RASSS (twelve cervical VCs, four thoracic VCs and twelve lumbar VCs) was assumed. The expected and balanced contribution assumed was 43% (12/28 VCs) for each of the cervical and lumbar segments and 14% (4/28 VCs) for the thoracic segment.

Statistical analysis

Descriptive statistics were performed, with continuous variables being presented as mean (SD) and categorical variables as frequencies. Observed and expected progression rates were compared using the chi-square test and a 5% level of significance was assumed. Stata SE version 11 was used (Statacorp, College Station, TX, USA).

Results

A total of 195 patients had at least one radiograph that could be scored (according to the mSASSS and/or RASSS), 64 had a radiograph that could be scored at year 12 and a total of 520 2-year progression intervals throughout the 12-year follow-up period were available Patients had a mean age of 42.8 (SD 12.4) years, mean disease duration since symptom onset of 20.0 (SD 11.6), mean disease duration since diagnosis of 11 (SD 8.7) years, 71% were males and 84% HLA-B27 positive. Baseline demographic, clinical and radiographic characteristics are summarized in Table 2.

Table 2 Baseline demographic, clinical and radiographic characteristics of the patients included in assessment of the radiographic progression in this study.

Feasibility

The mSASSS could be scored in a total of 809 radiographs. The RASSS could be calculated in 78% of these radiographs (n = 629) (Tables 3 and 4). In 58% of those, in which the RASSS was calculated, the score was based on one or two available thoracic VC scores and the remaining two to three needed imputation because they were missing, so were in fact non-informative. There were 520 2-year mSASSS interval progression scores available, and in 63% of them a 2-year RASSS interval progression score could be determined. If the availability of all four additional VCs was required, then the RASSS could only be calculated in 226 (36%) radiographs and in 64 (19%) progression intervals. For the subgroups of radiographs from patients with the first 2-year interval available and of radiographs from patients with a RASSS available at year 12, see Tables 3 and 4.

Table 3 Feasibility of the mSASSS vs.
Table 4 Feasibility of the RASSS - availability of the four thoracic vertebral corners added to the RASSS.

Discrimination

The first part of the discrimination aspect is reliability. Figure 1 shows the Bland and Altman plots for the progression scores of both the mSASSS and RASSS. In general, both scores could be reliably performed without clear systematic error. The SDC for the progression scores was 2.9 for the mSASSS and 3.5 for the RASSS.

Figure 1
figure 1

Bland and Altman plots: reliability of the mSASSS and RASSS progression scores. Difference against mean for mSASSS and RASSS progression scores of the two readers. The SDC for the progression scores was 2.9 for the mSASSS and 3.5 for the RASSS. mSASSS, modified Stoke Ankylosing Spondylitis Spine Score; RASSS, Radiographic Ankylosing Spondylitis Spinal Score; SDC, smallest detectable change.

Of all radiographs in which both the mSASSS and RASSS could be determined (n = 629), the mean (SD) status score was 15.5 (17.9) units for the mSASSS and 18.0 (20.9) units for the RASSS (Table 5). The mean (SD) 2-year progression score, calculated in 330 2-year intervals, was 2.0 (3.6) for the mSASSS and 2.4 (4.4) for the RASSS. The progression score per spinal segment was 1.2 (2.3) for the cervical segment, 0.8 (2.1) for the lumbar segment and 0.4 (1.2) for the thoracic segment in the RASSS. The effect size of the 2-year progression score was 0.57 (2.0/3.6) for the mSASSS and 0.55 (2.4/4.4) for the RASSS.

Table 5 Status and progression mSASSS and RASSS scores for radiographs in which RASSS was evaluable.

For all radiographs from patients with a RASSS evaluable at 12-year follow-up (n = 59), the mean (SD) status score of the mSASSS was 18.0 (19.3) and of the RASSS 20.8 (22.4); and in this group the mean (SD) 2-year progression score of the mSASSS was 2.4 (4.5) and of RASSS 2.9 (5.5). Compared to the status scores of all radiographs, the status scores of the radiographs from both patients with a RASSS evaluable at 12-year follow-up and radiographs with the four additional thoracic VCs available were higher (Table 5). During the first 2-year follow-up, the progression was 1.9 (4.1) units for the mSASSS and 2.2 (5.0) units for the RASSS in patients with available intervals for both mSASSS and RASSS (n = 134).

In patients with an mSASSS evaluable at baseline and at 12-year follow-up (n = 64), the mean (SD) 12-year progression was 11.7 (11.5). In 38 out of the 64 patients, the RASSS could be assessed with a mean 12-year progression of 14.2 (14.6) and a corresponding mSASSS progression of 12.2 (12.5) in this group.

In patients with a RASSS available at year 12 (n = 59), the baseline, 2-year and 12-year RASSS status scores were 11.8 (15.3), 12.3 (13.9) and 28.5 (25.0), respectively. For the mSASSS, the corresponding status scores were 10.6 (13.6), 11.2 (12.4) and 24.5 (21.6), respectively.

Interestingly, in a few cases, the RASSS enabled the occurrence of negative progression scores. Taking all radiographs into account, in five VCs in the cervical segment, first sclerosis was considered to be present and scored as 1, but at follow-up an erosion appeared, which was still scored as a 1 in the mSASSS but was scored 0 according to the RASSS scoring rules and, consequently, the progression score for that specific VC was -1 for RASSS.

Truth

Exclusive progression in the thoracic segment, which can only be captured by the RASSS and not by the mSASSS, occurred in only 5% of the 2-year progression intervals (17 intervals out of 330). In 2% of the 2-year progression intervals, a progression of two or more units (possibly indicating new syndesmophyte formation) in the thoracic segment was found. In 25% of the intervals (81 out of 330 intervals), progression occurred exclusively in the cervical segment and in 7% (23 out of 330) in the lumbar segment. However, there were no significant differences between the observed and expected contributions of the thoracic segment to progression (16% vs. 14%, P = 0.70), whilst progression was observed more frequently than expected in the cervical spine (55% vs. 43%, P = 0.09), and less frequently in the lumbar spine (29% vs. 43%, P = 0.04) (Table 6).

Table 6 Ratio of 2-year progression in each of the spinal segments of the RASSS.

Discussion

The present study shows that the mSASSS remains the most appropriate method for scoring radiographic progression in patients with AS based on feasibility, discrimination and truth aspects of the OMERACT filter.

With regard to feasibility, the 2-year progression scores were available for the RASSS in only 63% of the cases in which mSASSS progression intervals could be calculated. In the paper describing the RASSS, the authors reported an availability of 88% of the progression scores within the first 2 years [12], while in our study only in 59% of the cases a RASSS progression score could be calculated in this first 2-year interval. Furthermore, in our study in one third of the radiographs in which the RASSS could be calculated, only one to two thoracic VCs were accessible, meaning that calculation of the RASSS was based on imputed and therefore non-informative VCs in the lumbar spine. This shows that an important number of radiographs obtained in the Netherlands, France and Belgium does not include the lower part of the thoracic spine, in contrast to what has been suggested for Germany [12]. According to the OMERACT filter, feasibility captures an essential element in the selection of measures, one that may be decisive in determining a measure's success [13]. The worse feasibility of the RASSS compared with the mSASSS jeopardizes its wide use.

The RASSS demonstrated a higher mean progression, but an increase in the variance of the progression scores was also observed, resulting in similar effect sizes between mSASSS and RASSS. A higher mean progression was expected because the RASSS includes four additional VCs compared with the mSASSS, so that the RASSS is by definition almost always higher than the mSASSS. Exceptions are only the cases with erosions in any segment or squaring in the cervical spine, which are scored for the mSASSS, but not for the RASSS. Nevertheless, and comparing the progression scores with the limited data available in the literature for the RASSS, our RASSS progression scores were higher. In the first 2 years, we found a progression in the RASSS of 2.2 (5.0), whereas Baraliakos et al. reported a 2-year progression of 1.6 (2.8) [12]. Also, our mSASSS progression scores were higher compared with the same study (1.9 (4.1) vs. 0.9 (2.5)) [12]. A possible explanation for the difference in progression scores can be the difference in baseline radiographic damage (baseline mSASSS of 8.1 (14.6) in German cohort vs. 10.2 (14.2) in OASIS cohort). It is well known that presence of radiographic damage is a predictor of further and faster progression of radiographic damage [1820]. Other literature on RASSS progression scores is currently lacking. However, our mSASSS progression scores can be compared with other available studies. There are reports of 2-year mSASSS progression scores of around 1 mSASSS unit [2123], 2.5 units [18] and 2.6 units (extrapolation to a 2-year period of the annual progression rate of 1.3 (2.5), and assuming linearity) [20]. The differences between scores can be attributed to differences in selection of patients, baseline radiographic damage of patients, conditions in which radiographs were read [24] or the method of imputation of missing VCs. The increase in the variance around the progression scores resulted in similar effect sizes for both methods (0.57 for the mSASSS and 0.55 for the RASSS), showing that the higher mean progression of the RASSS is offset by the increased noise. For discrimination, both scoring methods seemed to be reliable, however, the SDC for the mSASSS was slightly smaller compared to the RASSS (2.9 vs. 3.5), suggesting that the measurement error with the RASSS is somewhat higher. Reliability of the RASSS could possibly improve by having an additional and separate radiograph to score the thoracic VCs. This could reduce the parallax associated with extending the view of the lumbar radiograph to include the thoracic VCs, but would on the other hand imply higher costs and radiation for the patients. With regard to the truth aspect of the OMERACT filter, we found that most progression occurred in the cervical segment of the spine (55%), followed by the lumbar spine (29%) and only 16% was found in the thoracic vertebrae. Furthermore, the progression in the thoracic vertebrae was not significantly different from what was expected, if progression throughout the spine would occur in a balanced way. In addition, we showed that new syndesmophytes exclusively occurring in the thoracic spine occurred in a maximum of 2% of the intervals. These data should be interpreted with caution, because a progression score of 2 does not always correspond to a new syndesmophyte, but can also mean twice a score of 1 in two separate VCs, reflecting development of squaring or sclerosis. This shows that the RASSS does not capture more progression occurring in the thoracic vertebrae, as was hypothesized by Baraliakos et al. [12]. In our study, a 2-year progression in the thoracic vertebrae of 0.4 (1.4) out of a total RASSS progression of 2.2 (5.0)) was found. Baraliakos et al. reported a progression of 0.6 (3.3) out of a total RASSS progression of 1.6 (2.8), indicating a higher contribution from the thoracic VCs to the total RASSS. The reason for this discrepancy between both studies is not entirely clear, but availability of the thoracic VCs for scoring can play a role.

Some limitations of the present study should be addressed. Films were obtained throughout a 12-year follow-up, which means that some of them were old and did not have the optimal quality. Nevertheless, this limited the reading of both scoring methods similarly. Furthermore, the findings of this study may not be generalizable to countries where lumbar radiographs routinely include the low thoracic spine, which is not standard procedure in our three countries.

Conclusions

In conclusion, the calculation of RASSS for status or progression of radiographic abnormalities in the spine is frequently impossible or strongly influenced by non-contributory imputation. The effect size of both methods is similar. In comparison to the mSASSS, the contribution of thoracic VCs in the RASSS is negligible, and does not justify the additional scoring efforts. The mSASSS remains the most appropriate measure to assess radiographic damage in patients with AS.