Abstract
Objective Ankylosing Spondylitis Disease Activity Score based on C-reactive protein (ASDAS-CRP) is recommended over ASDAS based on erythrocyte sedimentation rate (ASDAS-ESR) to assess disease activity in axial spondyloarthritis (axSpA). Although ASDAS-CRP and ASDAS-ESR are not interchangeable, the same disease activity cut-offs are used for both. We aimed to estimate optimal ASDAS-ESR values corresponding to the established ASDAS-CRP cut-offs (1.3, 2.1, and 3.5) and investigate the potential improvement of level of agreement between ASDAS-ESR and ASDAS-CRP disease activity states when applying these estimated cut-offs.
Methods We used data from patients with axSpA from 9 European registries initiating a tumor necrosis factor inhibitor. ASDAS-ESR cut-offs were estimated using the Youden index. The level of agreement between ASDAS-ESR and ASDAS-CRP disease activity states was compared against each other.
Results In 3664 patients, mean ASDAS-CRP was higher than ASDAS-ESR at both baseline (3.6 and 3.4, respectively) and aggregated follow-up at 6, 12, or 24 months (1.9 and 1.8, respectively). The estimated ASDAS-ESR values corresponding to the established ASDAS-CRP cut-offs were 1.4, 1.9, and 3.3. By applying these cut-offs, the proportion of discordance between disease activity states according to ASDAS-ESR and ASDAS-CRP decreased from 22.93% to 19.81% in baseline data but increased from 27.17% to 28.94% in follow-up data.
Conclusion We estimated the optimal ASDAS-ESR values corresponding to the established ASDAS-CRP cut-off values. However, applying the estimated cut-offs did not increase the level of agreement between ASDAS-ESR and ASDAS-CRP disease activity states to a relevant degree. Our findings did not provide evidence to reject the established cut-off values for ASDAS-ESR.
The Ankylosing Spondylitis Disease Activity Score (ASDAS) is the recommended instrument for assessing disease activity in patients with axial spondyloarthritis (axSpA).1 ASDAS is a composite index with continuous measurement properties developed by the Assessment of SpondyloArthritis international Society (ASAS) for assessing ankylosing spondylitis.2 ASDAS was developed in 2 versions that include the same variables, except for the acute-phase reactants C-reactive protein (CRP) or erythrocyte sedimentation rate (ESR), which have slightly different weighting.2 ASDAS-CRP is the recommended version of ASDAS, with ASDAS-ESR as an alternative in case CRP is not available.3 Although both ASDAS versions demonstrated a high level of agreement when they were developed, it has been suggested that they might not be interchangeable.3
The ASAS group has established cut-off values (1.3, 2.1, and 3.5) to classify patients into 4 distinct disease activity states: inactive disease (ID; < 1.3), low disease activity (LDA; 1.3 ≤ ASDAS-CRP < 2.1), high disease activity (HDA; (2.1 ≤ ASDAS-CRP ≤ 3.5) and very high disease activity (VHDA; > 3.5).4,5 These cut-offs were first developed for ASDAS-CRP and then cross-validated for ASDAS-ESR, leading to a recommendation to use the same cut-off values for both ASDAS versions. However, ASDAS-ESR showed a trend to classify higher numbers of patients in lower disease activity states compared to ASDAS-CRP and lower numbers of patients in higher disease activity states.4 Thereafter, some studies reported equal mean ASDAS scores,6-8 whereas others have observed differences.9-12 Hence, there may be clinically relevant discrepancies between the 2 versions and, consequently, between their cut-off values for disease activity states, which could affect treatment decisions in a treat-to-target regimen.
For this study based on a large, multinational cohort of patients with axSpA, we aimed to (1) assess the level of agreement between ASDAS-CRP and ASDAS-ESR disease activity states using the established formulas and cut-off values; (2) identify optimal ASDAS-ESR values corresponding to the established ASDAS-CRP cut-off values; and (3) assess the level of agreement between ASDAS-CRP and ASDAS-ESR disease activity states when applying the newly estimated cut-off values.
METHODS
Study population. We used data from 9 European registries with available ASDAS-ESR and ASDAS-CRP values that participated in the EuroSpA Research Collaboration Network: ARC (the Netherlands), ATTRA (Czech Republic), Biorx.si (Slovenia), BSRBR-AS (UK), NOR-DMARD (Norway), Reuma.pt (Portugal), ROB-FIN (Finland), RRBR (Romania), and SRQ (Sweden). We included patients with a clinical diagnosis of axSpA at age ≥ 18 years, who initiated a tumor necrosis factor inhibitor (TNFi) as a first biologic treatment between January 1, 2010, and December 31, 2018. Data at baseline and 3 follow-up visits (6, 12, and 24 months) during the first TNFi treatment course were collected within specified time windows.13
To have the best representation of the disease activity states, follow-up data (6, 12, or 24 months) were used to assess the ASDAS-CRP cut-offs for ID and between LDA and HDA, whereas baseline data were used to assess the cut-off for VHDA. Therefore, only patients with a complete registration of ASDAS scores at baseline and at least 1 complete follow-up registration of ASDAS scores were included.
Statistical methods. The relationship between ASDAS-CRP and ASDAS-ESR was analyzed using Spearman correlation coefficient (ρ) and the agreement between the 2 ASDAS versions was additionally visualized using Bland-Altman plots. ASDAS-CRP and ASDAS-ESR were compared according to their cut-off values using proportion of discordance, sensitivity, specificity, balanced accuracy and Cohen kappa coefficient (
). In addition, the level of agreement between disease activity states based on the ASDAS-CRP and ASDAS-ESR cut-off values was measured using proportion of discordance and weighted 
coefficient. These analyses were carried out in data from up to 3 different follow-up timepoints for each patient included in the study.
We performed receiver-operating characteristic (ROC) analyses to determine the optimal ASDAS-ESR values corresponding to ASDAS-CRP cut-offs using the Youden index (= sensitivity + specificity − 1). The stability of the selected optimal cut-off values was assessed by nonparametric bootstrap sampling, based on 500 bootstrap samples.14 Multiple samples were drawn randomly with replacement from the original data, and the optimal cut-off values were estimated in each sample. For the selection of ASDAS-CRP cut-offs for ID and between LDA and HDA (follow-up data), the earliest timepoint was prioritized.
Ethics. All participating registries obtained the necessary approvals in accordance with legal, compliance, and regulatory requirements from national data protection agencies and/or research ethics boards prior to the data transfer to the EuroSpA coordinating center.
RESULTS
We analyzed data from 3664 patients with a mean (SD) age of 42.0 (12.3) years and disease duration of 6.7 (8.4) years at initiation of their first TNFi, of whom 2244 (61.24%) were men. The mean (SD) values of ASDAS scores at all available baseline (N = 3664) and follow-up (N = 7824) observations were 3.6 (1.0) and 1.9 (0.9) for ASDAS-CRP, and 3.4 (1.0) and 1.8 (0.9) for ASDAS-ESR, respectively (see Supplementary Table S1 for corresponding values of ASDAS components and disease activity states per timepoint, available with the online version of this article).
ASDAS-CRP and ASDAS-ESR were very strongly correlated at all timepoints (ρ ranged from 0.87 to 0.92, P < 0.001). Bland-Altman plots showed that ASDAS-CRP was higher than ASDAS-ESR; however, discrepancies seemed to be evenly distributed (Supplementary Figure S1, available with the online version of this article). Fewer patients met the ID and LDA cut-offs of < 1.3 and < 2.1, respectively, by ASDAS-CRP than by ASDAS-ESR, whereas the percentage of patients with VHDA by the ASDAS-CRP cut-off of > 3.5 was higher than by ASDAS-ESR (Table 1). We observed good agreement between the established ASDAS-ESR and ASDAS-CRP cut-offs, but slightly worse agreement for the cut-off for VHDA compared to the other 2 cut-offs. The proportions of discordance between disease activity states based on the ASDAS-ESR and ASDAS-CRP cut-off values were 22.93% in baseline data and 27.17% in follow-up data, with weighted 
values of 0.59 (P < 0.001) and 0.54 (P < 0.001), respectively, indicating a moderate level of agreement (Table 2).
Statistical measures comparing ASDAS-ESR and ASDAS-CRP cut-off values for classifying patients in different disease activity states.
Distribution of observations in different disease activity states by established ASDAS-ESR and ASDAS-CRP cut-off values and level of agreement between disease activity states in baseline data (N = 3664).
Distribution of observations in different disease activity states by established ASDAS-ESR and ASDAS-CRP cut-off values and level of agreement between disease activity states in follow-up data (N = 7066).
The estimated ASDAS-ESR values for ID, between LDA and HDA, and for VHDA corresponding to ASDAS-CRP cut-offs 1.3, 2.1, and 3.5 were 1.4, 1.9, and 3.3, respectively (Table 3). The stability of the optimal cut-off values was assessed by bootstrapping. Out of the 500 bootstrap samples, the triplet of the estimated cut-off values (1.4, 1.9, and 3.3) was selected 114 (22.8%) times, whereas (1.4, 1.9, and 3.2) was selected 72 (14.4%) times. Frequencies of selection for individual values are shown in Supplementary Figure S2 (available with the online version of this article).
ASDAS-ESR cut-off estimation that corresponds to established ASDAS-CRP cut-off values.
The proportion of discordance comparing the estimated ASDAS-ESR (< 1.4, < 1.9, and > 3.3) and ASDAS-CRP (< 1.3, < 2.1, and > 3.5) cut-off values was better overall in baseline data as compared to applying the same cut-offs for both ASDAS versions, whereas the proportion of discordance in follow-up data increased slightly with the estimated cut-off values (Supplementary Table S2, available with the online version of this article). Proportions of discordance between disease activity states based on the estimated ASDAS-ESR and the established ASDAS-CRP cut-off values decreased slightly from 22.93% to 19.81% in baseline data, but it increased from 27.17% to 28.94% in follow-up data, with weighted 
values of 0.64 (P < 0.001) and 0.53, respectively (P < 0.001; see Supplementary Table S3). Comparing the estimated and the established ASDAS-CRP cut-off values, we observed that 11.35% and 13.38% of observations changed disease activity states at baseline and follow-up, respectively (Supplementary Table S4).
DISCUSSION
In our present study, using large European observational datasets, fewer patients were classified as having ID and LDA when using the ASDAS-CRP than ASDAS-ESR, whereas more patients were classified as having VHDA. The ASDAS-ESR cut-off values identified through our analyses were only slightly different from the established ASDAS-CRP cut-offs. Applying the newly estimated cut-offs did not yield a clinically relevant decrease in discordance in classification. Our findings are important for clinicians managing patients with axSpA, as they indicate that the established ASDAS-CRP cut-offs can be reliably used for assessing disease activity and guiding treatment decisions, despite the slight differences observed between ASDAS-CRP and ASDAS-ESR.
Our cohort revealed higher mean ASDAS-CRP than mean ASDAS-ESR at both baseline and follow-up, which is consistent with a previous smaller study,12 although other studies have demonstrated lower ASDAS-CRP than ASDAS-ESR.9-11 In previous studies, the absolute difference between mean ASDAS-CRP and mean ASDAS-ESR ranged from zero6-8 to 1 unit or higher,9,10 whereas we found corresponding differences close to zero. This discrepancy suggests that findings from small single-country cohorts may not be generalizable to the overall axSpA population. Considering the disease activity states according to the established ASDAS-CRP and ASDAS-ESR cut-off values (1.3, 2.1, and 3.5), the observed proportions of discordance were in line with a German cross-sectional study.8 However, the reported weighted 
coefficient was higher in the German study than in our data.
We used ROC analyses to identify ASDAS-ESR cut-off values corresponding to established ASDAS-CRP cut-off values for disease activity states. An alternative would have been to redevelop cut-offs for both ASDAS-ESR and ASDAS-CRP against external clinical criteria. When we performed bootstrapping analyses, a high stability of the identified values was shown, whereas a trend to even lower estimated cut-offs between LDA and HDA and for VHDA than the established ones was illustrated.
Although Machado et al4 developed the original ASDAS cut-offs based on the first published formula for ASDAS-CRP,2 we applied the updated ASDAS-CRP formula, where the constant value of 2 mg/L was used for CRP values < 2 mg/L, which showed better agreement with ASDAS using high-sensitivity CRP.15 An additional strength of our analyses is the use of a large dataset from 9 countries, limiting the effect of single-center factors (eg, genetic composition of the included population). In addition, our study used real-world data, enhancing the generalizability of our findings. The lack of strict inclusion criteria is evident, since a minority of patients were already in LDA or inactive at baseline. In clinical practice, such patients may have initiated a TNFi due to extramusculoskeletal manifestations such as psoriasis or inflammatory bowel disease. A potential limitation for our analyses is the varying limits of detection for laboratory instruments measuring CRP across centers in these countries, which may have influenced the estimation of the optimal cut-off values, especially the cut-off for ASDAS-ESR ID.
We observed good agreement between established ASDAS-CRP and ASDAS-ESR cut-offs in terms of proportion of discordance, sensitivity, specificity, balanced accuracy, and Cohen 
. When comparing the newly estimated and the established ASDAS-ESR cut-offs, an improvement in the statistical measures was demonstrated only for VHDA. Somewhat to our surprise, we did not find an overall increase in agreement regarding the disease activity state of the individual patients when assessing the newly estimated and the established ASDAS-ESR cut-offs against ASDAS-CRP.
Across European registries, we estimated the optimal ASDAS-ESR values corresponding to established ASDAS-CRP cut-off values. Regarding the agreement between disease activity state between ASDAS-CRP and ASDAS-ESR, the original cut-offs overall performed similarly as the ones we estimated. Our findings did not provide evidence to reject the established ASDAS-ESR cut-off values.
ACKNOWLEDGMENT
On behalf of the EuroSpA Scientific Committee, the authors acknowledge Novartis Pharma AG and IQVIA for supporting the EuroSpA Research Collaboration Network.
Footnotes
The EuroSpA Research Collaboration Network was financially supported by Novartis Pharma AG. Novartis had no influence on the data collection, statistical analyses, manuscript preparation, or decision to submit the manuscript.
S. Georgiadis and L.M. Ørnbjerg contributed equally as joint first authors.
M.L. Hetland and M. Østergaard contributed equally as joint senior authors.
The following conflicts of interest are reported: SG: Novartis; LMØ: Novartis; BM: Novartis; TKK: AbbVie, Amgen, BMS, Celltrion, Galapagos, Gilead, Grünenthal, Novartis, Pfizer, Sandoz, UCB; JKW: AbbVie, Amgen, Eli Lilly, Novartis, Pfizer; JZ: AbbVie, AstraZeneca, Egis, Eli Lilly, Novartis, Sandoz, Sanofi, Sobi, UCB; SAP: Boehringer Ingelheim; AMR: AbbVie, Amgen, Novartis, Pfizer; MJS: AbbVie, AstraZeneca, Janssen, Lilly, Novartis, Pfizer; ZR: AbbVie, Amgen, AstraZeneca, Biogen, Eli Lilly, Janssen, Lek, Medis, MSD, Novartis, Pfizer, Sanofi, SOBI, Swixx BioPharma; KPP: AbbVie, Boehringer Ingelheim, Eli Lilly, Janssen, Lek, Medis, MSD, Novartis, Pfizer; DN: AbbVie, BMS, Lilly, MSD, Novartis, Pfizer, Roche, UCB; GJM: GSK; GTJ: AbbVie, Amgen, GSK, Janssen, Pfizer, UCB; IvdHB: AbbVie, BMS, Lilly MSD, Novartis, Pfizer, UCB; PH: Novartis; MØ: AbbVie, BMS, Boehringer Ingelheim, Celgene, Eli Lilly, Hospira, Janssen, Merck, Novartis, Novo, Orion, Pfizer, Regeneron, Roche, Sandoz, Sanofi, UCB; MLH: AbbVie, Biogen, BMS, CellTrion, Eli Lilly, Janssen Biologics, Lundbeck Fonden, MSD, Medac, Novartis, Pfizer, Roche, Samsung Biopies, Sandoz, UCB. DDG and EKK declare no conflicts of interest relevant to this article.
- Accepted for publication March 13, 2024.
- Copyright © 2024 by the Journal of Rheumatology
This is an Open Access article, which permits use, distribution, and reproduction, without modification, provided the original article is correctly cited and is not used for commercial purposes.
REFERENCES
DATA AVAILABILITY
The data in this article were collected in the individual registries and made available for secondary use through the EuroSpA Research Collaboration Network (https://eurospa.eu/#registries).
ONLINE SUPPLEMENT
Supplementary material accompanies the online version of this article.
