Non-steroidal anti-inflammatory drug use is determined by disease activity in axSpA and decreased by biologicals: a longitudinal analysis


 Objective

To evaluate non-steroidal anti-inflammatory drug (NSAID) use and Assessment in Spondyloarthritis International Society (ASAS)-NSAID scores in patients with axial spondyloarhritis (axSpA) in a longitudinal study.
Methods

In total, 429 patients with axSpA (59% male; 63.6% with AS) were included in this study. Data about disease activity, C-reactive protein (CRP) levels, and NSAID use and dosage were collected at 0, 12, 24, and 52 weeks retrospectively. The relationship with NSAID use /NSAID scores and other factors was tested using generalized estimating equations (GEE).
Results

At baseline (0 weeks), 92.8% of patients started biologic disease-modifying anti-rheumatic drugs (bDMARDs) and 82.1% were conventionally treated with NSAIDs. At baseline, the proportion (p=0.03) and the median (IQR) ASAS-NSAID scores were higher in biologic treatment group [100 (50) vs 50 (83.4); p<0.001]. During follow-up, NSAID use and ASAS-NSAID scores decreased significantly in patients treated with bDMARDs (p<0.001) and the reduction remained stable throughout the follow-up However, neither NSAID use (p=0.06) nor ASAS-NSAID scores changed in conventionally treated patients (p=0.15). In bDMARD-treated patients, ASDAS-CRP and BASFI scores were independent determinants for NSAID use, and BASDAI and patient global assessment (PGA) were determinants for NSAID dosage. There was no independent significant predictor for ASAS-NSAID scores; PGA was the only significant predictor for NSAID use in the conventional treatment group.
Conclusion

Concurrent biologic treatment was associated with low NSAID intake in patients with axSpA, and NSAID use was determined mainly by disease activity and partly by function during bDMARD treatment.


Introduction
Axial spondyloarthritis (axSpA) is a chronic, in ammatory rheumatic disease affecting the axial skeleton, which typically leads to in ammatory back pain. Axial spondyloarthritis is subdivided into two formsradiographic (r-axSpA) and non-radiographic axSpA (nr-axSpA) -based on the presence or absence of unequivocal radiographic changes in sacroiliac joints (1). The disease also causes peripheral arthritis, enthesitis, and dactylitis.
The treatment goals of axSpA are reducing symptoms, improving and maintaining spinal exibility and normal posture, reducing functional limitations, maintaining the ability to work, and decreasing the extramusculoskeletal manifestations and complications associated with the disease (2).
The clinical e cacy of non-steroidal anti-in ammatory drugs (NSAIDs) on axSpA has been shown in numerous clinical trials (3)(4)(5)(6), and NSAIDs are the rst treatment option in patients with axial pain and/or stiffness (2,7). NSAIDs improve back pain, spinal function, morning stiffness and also peripheral joint and entheseal pain. Moreover, there is also evidence suggesting that continuous or high NSAID intake may have an impact on slowing radiographic damage of the spine (8,9). As a consequence, continuous and maximum tolerated doses of NSAIDs are recommended for active, symptomatic disease (9). However, NSAIDs may be associated with an increased risk of cardiovascular events, and gastrointestinal and renal toxicity (10)(11)(12)(13).
The Assessment in SpondyloArthritis international Society (ASAS)/European League Against Rheumatism (EULAR) management recommendations for axSpA advised biologic disease-modifying antirheumatic drugs (bDMARDs) for active axial disease despite at least two maximum doses of NSAIDs (2).
In daily clinical practice, tapering or withdrawing NSAIDs may be expected in patients who have begun taking bDMARDs. However, there are limited data concerning NSAIDs dosage and the determinants of NSAID use in patients with axSpA who have been treated with bDMARDs. To standardize the quantitation of NSAID use in patients with axSpA, the ASAS working group suggested using NSAID scores (14).
In a randomized placebo-controlled trial, baseline disease activity and ASAS-NSAID scores were similar between the groups; however, a signi cantly higher proportion of patients treated with etanercept vs. placebo reduced or stopped NSAID at week 8 (15). It was also shown in an observational study that treatment with tumor necrosis factor inhibitor (TNFi) was associated with a decrease in the proportion as well as the dosage of the NSAIDs in patients with axSpA in a real-life setting (16). Recently, another observational study revealed that longitudinal NSAID use was associated with disease activity, particularly in patients treated with TNFi (17).
Therefore, in the present study, we aimed to evaluate NSAID use longitudinally during a 52-week period in an observational cohort of patients with axSpA and to determine factors associated with NSAID intake in a real-life setting.

Study design and patient population
This was a retrosprective observational study. All patients with axSpA were aged 18 years and over, according to the ASAS classi cation criteria for axSpA, patients from the SpA cohort of a tertiary referral hospital were included in the analysis (1).
Patients who had any contraindication for NSAID use such as pregnancy, lactation, and renal failure were excluded. This study was approved by the ethics committee of Izmir Katip Celebi University (approval number of 2020-0559).
In the present cohort, bDMARD treatment was started according to the ASAS/EULAR and American College of Rheumatology/Spondylitis Association of America/Spondyloarthritis Research and Treatment Network (ACR/SAA/SPARTAN) management recommendations and local guidelines (2,7,18). NSAIDs were used based on ASAS/EULAR recommendations (7).

Collected Data
Baseline information regarding socio-demographic (including age, sex, education level, smoking status, body mass index) and disease-related characteristics were recorded. During follow-up (at 0, 12, 24, and 52 weeks), data about axSpA-related features [including the presence or the development of skeletal (peripheral arthritis, dactylitis, heel enthesitis)], extra-musculoskeletal manifestations [uveitis, psoriasis, in ammatory bowel disease (IBD)] in addition to disease activity and function measurements were collected.
The Bath Ankylosing Spondylitis Disease Activity Index (BASDAI), Bath Ankylosing Spondylitis Functional Disease Index (BASFI), serum C-reactive protein (CRP) levels, Ankylosing Spondylitis Disease Activity Score (ASDAS-CRP), and Patient Global Assessment (PGA) of disease activity were collected using a structured form. In each visit, NSAID intake, type, and the dosage and frequency of NSAIDs were also recorded to determine the ASAS-NSAID score.
The ASAS-NSAID score was calculated by using the following formula: (equivalent NSAID score) × (days of intake during the period of interest) × (days per week)/ (period of interest in days) as suggested by the ASAS working group (14).

Statistical analysis
All statistical analyses were performed using the Statistical Package of the Social Science (SPSS) software package version 13.0 (Chicago, IL). The distribution of continuous variables was investigated using visual (histograms, probability plots) and analytical methods (Kolmogorov-Smirnov). Most of the continuous variables were not distributed normally; therefore, values were presented as medians and interquartile range (IQR) for continuous and as percentages for categorical variables. The Mann-Whitney U test was used to compare non-normally distributed variables between the groups. The Chi-square test or Fisher's exact test was used for the comparison of categorical data.
Friedman tests were conducted to test whether there was a signi cant change in the ASAS-NSAID scores and disease activity measurements due to violations of parametric test assumptions (non-normal distribution). The Wilcoxon test was performed to test the signi cance of pair-wise differences using Bonferroni correction to adjust for multiple comparisons.
NSAIDs use and scores over time were investigated using generalized estimating equations (GEE). The GEE procedure allows the analysis of repeated measurements in longitudinal studies. Traditional statistical techniques such as linear or logistic regression cannot be used in longitudinal studies because observations are not independent of each other and special techniques such as the GEE were developed to this end (19,20). GEE is suitable for the evaluation of the longitudinal relationship with continuous and dichotomous outcome variables (dependent), and several time-dependent and independent covariates allow the use of unequal numbers of repeated measurement data, even if they deviate from normality (19).
In the rst step of the present study, individual longitudinal interactions between ASAS-NSAID scores, demographic and clinical characteristics, and NSAID use were assessed and variables with p-values of <0.10 were included in regression models. Different multivariate longitudinal models were run, mainly differing in the disease activity measures.
All statistical tests were two-tailed and p-values of <0.05 were considered to be statistically signi cant.

Results
In total, 429 patients with axSpA (273 with r-and 156 with nr-axSpA) were included in the study. The median age was 40 (IQR: 18) years, 59% of patients were male and 61% were HLA-B27 positive. The median disease duration was 12 (IQR: 12) years. The baseline demographic and disease-related characteristics of the patients are summarized in Table 1 In the present study, 139 patients with axSpA were started on bDMARDs and 290 were treated conventionally. There was a higher proportion of males and patients with r-axSpA among the biologic treatment group. Furthermore, symptom duration and time from diagnosis were longer and, as expected, disease activity was higher in patients started on bDMARDs. During the follow-up period, the proportion of NSAID use ( gure 1) and, in particular, ASAS-NSAID scores (supplementary gure 1) decreased rapidly (as early as 12 weeks) and signi cantly in patients treated with biologic drugs; the reduction remained stable throughout follow-up (p<0.001). However, neither NSAID use (p=0.06) nor ASAS-NSAID scores changed in conventionally treated patients (p=0.15).
All disease activity measurements (BASDAI, BASFI, ASDAS-CRP, serum CRP levels, PGA of disease activity) decreased signi cantly in bDMARD-started patients in week 12 (p<0.001) and the measurements were found to be stable (supplementary gure 2-6) throughout follow-up. However, disease activity did not change in conventionally treated patients during the follow-up period (supplementary gure 2-6).
NSAID use as a continuous variable estimated using ASAS-NSAID scores [0.046-0.453]; p=0.016). Therefore, we established two multivariable models to assess the associated factors/covariates with ASAS-NSAID scores over time (one with ASDAS and the other BASDAI+CRP as disease activity measures) and showed that BASDAI and PGA of disease activity were independent determinants of ASAS-NSAID scores in patients who were biologically treated ( Table 2).  . Therefore, we again established two multivariable models to assess the associated factors/covariates with NSAID use over time (one with BASDAI and the other ASDAS-CRP as disease activity measures) and showed that ASDAS-CRP and BASFI were independent determinants of NSAID use in patients with axSpA treated with bDMARDs (Table 4). analysis showed that PGA of disease activity was the only signi cant predictor for NSAID use in this group of patients with axSpA (Table 5).

Discussion
The results of this long-term follow-up study showed that NSAID use in patients with axSpA was determined mainly by disease activity and partly by function. Almost all of the activity measurements used in daily clinical practice, such as BASDAI, PGA of disease activity, and serum CRP levels were found to be associated with longitudinal NSAID use in the present cohort. However, the disease subtype had no impact on both NSAID use and dosage in patients with axSpA. In our study groups, biologic treatment was associated with both clinical improvement and a decrease in NSAID use and dosage. Using GEE analysis, we also showed that in biologic treated patients, NSAID use was determined mainly by disease activity, irrespective of treatment type during the follow-up. There are limited data evaluating the determinants of longitudinal NSAID use in patients with axSpA (15)(16)(17).
In our study, at the baseline visit, both the proportion of NSAID use and the median ASAS-NSAID scores were statistically higher in the biologic treated patients with axSpA. However, both NSAID use and dosage decreased signi cantly only in the advanced treatment group. Similar to our ndings, NSAID use and dosage were found to be decreased in follow-up in comparison with the baseline visit in patients with axSpA treated with TNFi in the Groningen Leeuwarden Ankylosing Spondylitis (GLAS) cohort. Nevertheless, in contrast to our ndings, the authors also reported that baseline NSAID use was similar in the TNFi and conventional treatment groups (79% vs 74%; p=0.131) (17). The reason for the high NSAID use in both treatment groups was not clear in this study because baseline disease activity was higher in the patients who were put on TNFi treatment (17).
NSAIDs are the rst choice of treatment in patients with axial pain and/or stiffness and biologic agents should be considered in patients who have high disease activity despite the maximum tolerable dose of NSAIDs. In the GLAS cohort, the proportion of NSAID use was lower in the TNFi treatment group than in our study; this result might be related to the requirement of TNFi treatment for some different involvements in patients in the GLAS cohort; the higher conventional DMARD use in the patients with TNFi treatment supports this hypothesis. In both the Devenir des Spondylarthropathies Indifférenciées Récentes (DESIR) cohort and SPARSE study, NSAID use was similar to our patients with axSpA and higher than in the GLAS cohort (15)(16)(17).
In our study, we analyzed the entire axSpA patient group as well as the biologically and conventionally treated groups separately to understand the determinants of NSAID use and dosage longitudinally. In our axSpA cohort, longitudinal multivariate analysis revealed that, besides disease activity, therapy with bDMARDs was an independent predictor of NSAID use as well as dosage, suggesting a NSAID-sparing effect of bDMARDs in patients with axSpA. In the GLAS cohort, the authors also revealed to NSAID use was signi cantly related with disease activity (17). However, they analyzed conventional and bDMARD groups separately and did not evaluate whole group of patients. Also, they only used ASDAS-CRP, BASDAI, and serum CRP levels as disease activity measurements. In our study, we additionally analyzed function and PGA of disease activity for patient-reported outcomes. Furthermore, when we assessed patients in two treatment groups (conventional and biologic), we found that NSAID use was related to BASDAI, ASDAS-CRP, BASFI, and PGA of disease activity scores in biologically treated patients. By contrast, PGA of disease activity was the only determinant of NSAID use in the conventional treatment group. In addition to ASAS-NSAID scores, we analyzed NSAID use as a nominal variable. However, we believe that ASAS-NSAID scores could be a more reliable and objective marker for NSAID use in patients with axSpA because they are calculated using equivalent NSAID scores, dosage, and the percentage of days with NSAID use in a de ned period. As shown in our study, in the SPARSE study, which evaluated the NSAID-sparing effect of TNFi agents in patients with axSpA, the primary endpoint was ASAS-NSAID scores to examine this effect and the authors showed the importance of ASAS-NSAID scores (15).
The results of the present study provide additional evidence regarding the substantial e cacy of TNFi in patients with axSpA because all disease activity measures decreased signi cantly after as little as 12 weeks of follow-up. Furthermore, in our biologic treatment group, the proportion of patients using NSAIDs and median ASAS-NSAID scores decreased in 12 weeks and remained stable at the 24 and 52-weeks follow-up visits. The authors of the GLAS cohort study showed that ASAS-NSAID scores signi cantly decreased with TNFi therapy in 6-12 weeks (17). In the SPARSE study, NSAID use was reduced or stopped at week 8 in patients treated with etanercept (15). The percentage of patients with axSpA, in whom NSAID scores decreased or NSAID treatment was discontinued, was higher among the TNFi treatment group, also in the DESIR cohort (16).
Our study has some limitations. First, it was a retrospective study, inevitably resulting in missing data in the follow-up period. Second, we did not evaluate the different treatment groups (TNFi agents and secukinumab in addition to COX-2 inhibitors and traditional NSAIDs) as a sub-analysis due to the low number of patients.
The main strengths of the study were the patient populations and sample size because patients with both r-and nr-axSpA were included in the study. This is the rst study to show that in addition to biologic treatment, disease activity might affect NSAID use and dosage in patients with both r-and nr-axSpA.
Additionally, we used GEE for longitudinal analysis, which is suitable for the evaluation of the relationship with continuous as well as dichotomous outcome variables and several time dependent and independent covariates.

Conclusion
This is the rst study to evaluate the relationship between NSAID use and disease activity of all subtypes of axSpA in a longitudinal analysis. Our results showed that NSAID use and dosage were signi cantly higher in patients with axSpA who were indicated bDMARD treatment. NSAID use decreased signi cantly over time with biologic treatment ( as soon as 12 weeks ) and independently determined by disease activity. However, it was stable in conventionally treated patients with axSpA. Calculating ASAS-NSAID scores may be a more appropriate outcome than nominal NSAID use to investigate NSAID use in patients with axSpA. Funding: No speci c funding was received from any bodies in the public, commercial or not-for-pro t sectors to carry out the work described in this article.

Con ict of interests: No con ict of interests
Authors' contributions: EDE, DS and SA contributed to the conception and design of this study. GK, SG, HC and EOA participated in data collection. EDE, SA, DS and MO contributed to the data analysis. All authors contributed to the data interpretation, reviewed the article and approved the nal draft for submission.