Real-world exercise and disease activity in axial spondyloarthritis: does the association vary by age, BMI, or socioeconomic status?

Abstract

Background

Exercise is recommended in axial spondyloarthritis (axSpA), yet evidence on habitual exercise in routine care and whether its association with disease activity varies by age, body mass index (BMI), or socioeconomic position remains limited. We assessed the association between self-reported regular exercise during the previous year and disease activity (ASDAS) at the study visit.

Methods

We performed a cross-sectional analysis from two multicentre registries (REGISPONSER and RESPONDIA), including 1233 patients with axSpA. The exposure was self-reported regular exercise in the preceding year. Missing data were handled using multiple imputation. We estimated the association between exercise and ASDAS using stabilised inverse probability of treatment weighting (IPTW) based on a propensity score model including age, sex, BMI, HLA-B27 status, and country (Spain vs. Latin America). Doubly robust augmented IPTW (AIPW) and sensitivity analyses using a clinically expanded confounder specification and sex-stratified models were also performed. We then explored whether the exercise–ASDAS association varied across age, BMI, and socioeconomic position (Graffar scale) by fitting weighted interaction models, using spline terms for age and BMI, and deriving marginal effects across the observed ranges and by Graffar levels.

Results

Regular exercise was associated with lower disease activity. In IPTW analyses, exercise was associated with a reduction in ASDAS of − 0.153 points (95% CI − 0.276 to − 0.031; p = 0.014). Results were consistent in doubly robust AIPW analyses (− 0.155, 95% CI − 0.277 to − 0.034; p = 0.012) and in a clinically expanded sensitivity analysis (IPTW: −0.157, 95% CI − 0.279 to − 0.035; AIPW: −0.158, 95% CI − 0.280 to − 0.036). Exploratory analyses suggested possible variation in the association across age, BMI, socioeconomic position, and sex, but global interaction tests were not statistically significant, and sex-stratified estimates in women were imprecise.

Conclusions

Self-reported regular exercise during the previous year was associated with modestly lower ASDAS at the clinical visit, consistently across IPTW and doubly robust estimators. These findings are consistent with current recommendations supporting regular physical activity as a core component of axial spondyloarthritis management; however, the cross-sectional design, retrospective self-report of exercise, and the potential for residual confounding and reverse causation preclude causal inference.

Clinical trial number

Not applicable

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s41927-026-00672-6.

Introduction

Axial spondyloarthritis (axSpA) is a chronic immune-mediated inflammatory disease that is characterised by persistent pain, functional limitation, and substantial impairment in quality of life. Management strategies increasingly follow a treat-to-target framework aimed at suppressing inflammation, improving symptoms and function, and limiting long-term disability through a combination of pharmacological and non-pharmacological care. Across international recommendations, exercise and physiotherapy are not ancillary but foundational components of comprehensive axSpA management [11, 12, 17].

Most evidence on exercise in axSpA comes from structured, often supervised programmes. These studies show that exercise can improve symptoms and function and may also modestly reduce disease activity in some settings [1, 5, 6, 15]. Less is known about whether habitual exercise performed in the year before a routine clinical visit is linked to lower ASDAS in real-world settings, and whether this association is consistent across patient subgroups. In particular, age, body mass index, and socioeconomic position may shape both the ability to exercise and the inflammatory and functional response to physical activity [7, 14]. We therefore examined whether the association between regular exercise in the previous year and ASDAS at the study visit varies by age, BMI, and social class (Graffar scale).

Patients and methods

Study design

This cross-sectional analysis drew on merged data from two Ibero-American multicenter registries—REGISPONSER (Spain, 2004–2007, 31 centres) and RESPONDIA (Latin America, 2006–2007, 33 centres across eight countries)—both enrolling consecutive spondyloarthritis patients meeting ESSG criteria with identical case-report forms. Full details of registry design, sampling, and recruitment have been published previously [2, 10, 11]. Data from both registries were pooled into a single multinational cohort of 1233 patients with axial spondyloarthritis for the present analysis.

Variables

A total of 1233 patients with axial spondyloarthritis were included. Sixty-six variables spanning four domains—personal, social, treatment, and clinical/disease activity—were collected using the standardised case-report forms common to both registries; the complete variable definitions and data-collection procedures have been described elsewhere [2]. Descriptive statistics, including missing-data proportions, are presented in Tables 1 and 2. Key variables for the present analysis are summarised below.

Table 1.

Descriptive analysis of categorical variables (n = 1233)

Variable	Category	Count	Percent_%	Missing_%
Country	ESPAÑA	1055	85,56	0
	PORTUGAL	48	3,89
	ARG	44	3,57
	CHIL	39	3,16
	VEN	18	1,46
	MEX	10	0,81
	URU	9	0,73
	PERU	7	0,57
	CR	3	0,24
Gender	man	929	75,34	0
	woman	304	24,66
Education	Elementary school	337	27,33	34,55
	High school	323	26,2
	University	137	11,11
	Illiterate	10	0,81
Marital Status	Married	586	47,53	34,47
	Single	184	14,92
	Divorced/Separated	26	2,11
	Widowed	12	0,97
Profession	Skilled worker	218	17,68	34,63
	Unskilled worker	193	15,65
	Employee	178	14,44
	Technician	121	9,81
	Universitary	96	7,79
Source Of Income	Salary	565	45,82	34,71
	Donations	126	10,22
	Rents	60	4,87
	Profits	54	4,38
Housing Conditions	Good	400	32,44	34,63
	Excelent without luxury	343	27,82
	Deficients	40	3,24
	Excelent with luxury	15	1,22
	Inadequate	8	0,65
Graffar Scale	Middle-lower class	337	27,33	34,71
	Middle class	255	20,68
	Upper-middle class	150	12,17
	Lower class	48	3,89
	Upper class	15	1,22
Race	White	724	58,72	34,47
	White-Indigenous	59	4,79
	Black	8	0,65
	Other	6	0,49
	White-Black	6	0,49
	Indigenous	4	0,32
	Black-Indigenous	1	0,08
Family History	NO	903	73,24	6,16
	YES	254	20,6
Employment Status	Employed	601	48,74	0,97
	At home	422	34,23
	Unemployed	146	11,84
	Retired	42	3,41
	student	10	0,81
Disability	NO	878	71,21	2,92
	Permanent	271	21,98
	Temporary	48	3,89
Exercise Habit	NO	702	56,93	0
	YES	531	43,07
Smoker	NO	271	21,98	68,78
	YES	114	9,25
Iritis/Uveitis	NO	962	78,02	0,81
	YES	261	21,17
Dactylitis	NO	1210	98,13	0,32
	YES	19	1,54
Palmoplantar Pustulosis	NO	1222	99,11	0,16
	YES	9	0,73
Acne Conglobata	NO	1220	98,95	0,49
	YES	7	0,57
Balanitis	NO	1223	99,19	0,24
	YES	7	0,57
Prostatitis	NO	1218	98,78	0,32
	YES	11	0,89
Heart Condition	NO	1205	97,73	0,24
	YES	25	2,03
Kidney Condition	NO	1204	97,65	0,41
	YES	24	1,95
Neurological Condition	NO	1224	99,27	0,16
	YES	7	0,57
Lung Condition	NO	1212	98,3	0,24
	YES	18	1,46
Nail Condition	NO	1199	97,24	0,73
	YES	25	2,03
Tarsitis	NO	1051	85,24	13,38
	YES	17	1,38
Nsaids	Regularly	768	62,29	0,97
	NO	266	21,57
	PRN	187	15,17
Corticosteroids	NO	1129	91,57	1,95
	YES	80	6,49
Methotrexate	NO	1138	92,3	2,27
	YES	67	5,43
Sulfasalazine	NO	1049	85,08	1,95
	YES	160	12,98
Leflunomide	NO	1197	97,08	2,27
	YES	8	0,65
Infliximab	NO	1095	88,81	2,51
	YES	107	8,68
Etanercept	NO	1172	95,05	2,35
	YES	32	2,6
Adalimumab	NO	1185	96,11	2,43
	YES	18	1,46
Enthesitis	NO	927	75,18	0,97
	YES	294	23,84
B27	Positive	860	69,75	14,36
	Negative	196	15,9
ASDAS_cat	High	471	38,2	8,03
	Moderate	268	21,74
	Very high	218	17,68
	Inactive	177	14,36
Sf12-Mental	Good	653	52,96	4,06
	Moderate	474	38,44
	Excellent	29	2,35
	Low	27	2,19
Sf12-Physical	Low	727	58,96	4,06
	Moderate	422	34,23
	Good	34	2,76
ASQoL_cat	Excellent	448	36,33	2,43
	Impaired	268	21,74
	Good	268	21,74
	Poor	219	17,76

Table 2.

Descriptive analysis of numerical variables (n = 1233)

Variable	Missing_%	Mean	SD	Median	Q1	Q3	Min	Max
Age	0,49	47,65	13,26	48	37,5	57	12	100
Delay	3,89	7,972	9,24	4	1	11	0	52
Disease Duration	2,6	20,34	13,01	19	10	29	0	65
Duration	3	12,19	10,29	10	4	19	0	56
Onset Age	1,87	27,39	10,52	25	20	32	0	99
Diagnosis Age	3,08	35,42	12,10	33	26	43	0	99
Chest Expansion (Cm)	1,38	3,655	2,09	3,5	2	5	0	15
Schober (Cm)	1,78	2,845	1,80	3	1,5	4	0	12,5
Finger-Floor Distance (Cm)	2,68	19,34	14,56	18	8	30	0	70
Occiput-Wall Distance (Cm)	0,97	4,915	6,24	2	0	9	0	35
Weight	7,3	73,76	13,80	73	64,25	81	40	170
Height	7,3	166,58	9,27	167	160	173	125	196
BMI	8,68	26,54	4,35	26,08	23,51	28,89	17,09	57,46
Number Of Swollen Joints	0,32	0,21	1,55	0	0	0	0	32
VAS	1,14	4,57	2,72	5	2	7	0	10
Asqol	2,43	6,28	5,01	6	2	10	0	18
Sf12-Physical Component	4,06	37,37	7,35	37	31	43	20	57
Sf12-Mental Component	4,06	50,46	5,44	51	46	54	32	64
ASDAS	8,03	2,47	1,09	2,45	1,64	3,31	0,06	5,80

Personal and lifestyle variables included age, sex, BMI, country, HLA-B27 status, family history, exercise habit (binary self-report of regular exercise during the preceding year), and smoking status. Social variables encompassed educational attainment, employment status, and socioeconomic stratification via the Graffar scale. Treatment variables documented NSAID use, corticosteroids, conventional synthetic DMARDs, and biologic therapies. Disease activity was assessed with the ASDAS; pain was recorded by VAS; and additional clinical features (enthesitis, extra-articular manifestations, comorbidities) were also captured (see Tables 1 and 2 for full details).

All covariates, including ASDAS and exercise habit, were measured at the same clinical visit. Exercise habit was assessed by self-report and referred to regular exercise performed during the previous year, resulting in a cross-sectional analysis with retrospective exposure assessment.

Data analysis

The primary exposure of interest was regular exercise habit (yes vs. no), defined as self-reported regular physical activity during the previous year. Patients reporting regular exercise were classified as the exposed group.

To evaluate the association between exercise habit and disease activity, we prespecified a parsimonious set of potential confounders for the primary propensity score model, focusing on patient characteristics unlikely to be consequences of exercise and measured at the same clinical visit. These included age, sex, body mass index (BMI), HLA-B27 status, and country (Spain vs. Latin America). These variables were selected because they represent pre-exposure characteristics that may influence both exercise propensity and disease activity. Variables such as current pharmacological treatment, disease duration, and related clinical measures were not included in the primary propensity score model because, in this cross-sectional setting with concurrent measurement, their inclusion could introduce overadjustment, mediator, or collider bias. Nevertheless, these variables were incorporated into the multiple imputation models and into prespecified sensitivity analyses addressing residual confounding. Specifically, we repeated the analyses using a clinically expanded specification that additionally included disease duration, diagnostic delay, smoking status, current treatment variables (biologic therapy, corticosteroid use, and regular NSAID use), and one proxy of functional burden (SF-12 physical component, ASQoL, or disability, depending on availability). In addition, because sex may influence both exercise patterns and clinical presentation, we performed sex-stratified sensitivity analyses applying the same analytical framework separately in men and women. Socioeconomic indicators (Graffar scale and employment status) and selected clinical variables were also retained for descriptive purposes and to strengthen the multiple imputation models.

Missing data were addressed using multiple imputation by chained equations (MICE) under a missing-at-random assumption. Twenty imputed datasets were generated using 20 iterations per dataset, including all variables subsequently used in the analyses together with auxiliary variables predictive of missingness. Continuous and categorical variables were jointly imputed; categorical predictors were handled through appropriate ordinal or multinomial encoding. All downstream analyses were conducted separately within each imputed dataset and combined using Rubin’s rules to account for within- and between-imputation variability.

The primary analysis estimated the association between regular exercise and continuous ASDAS using inverse probability weighting. Within each imputed dataset, propensity scores were estimated via logistic regression including age, BMI, sex, HLA-B27, and country (Spain vs. Latin America). Stabilized inverse probability of treatment weights (IPTW) were computed, and extreme weights were truncated at the 1st and 99th percentiles to improve stability. The IPTW-adjusted effect was estimated using weighted linear regression, and covariate balance was evaluated using standardized mean differences (SMD) and effective sample size (ESS). As a robustness approach, doubly robust augmented IPTW (AIPW) estimators were also computed by combining the propensity score model with outcome regression models fitted separately within exercise groups. As prespecified sensitivity analyses, we repeated the IPTW and AIPW analyses using the clinically expanded covariate specification described above to assess the robustness of the findings to broader clinical adjustment. We also repeated the same analytical framework separately in men and women to explore whether the exercise–ASDAS association was consistent across sex strata.

Effect modification by continuous baseline covariates was examined using spline-based interactions. Age and BMI were modelled flexibly using B-splines (4 degrees of freedom), and interaction terms between exercise and the spline functions were added to the weighted outcome models to allow the exercise–ASDAS association to vary smoothly across the covariate range. Standardised marginal effects (ΔASDAS) were estimated and plotted across the 5th–95th percentile range to avoid extrapolation. Global evidence of heterogeneity was assessed using pooled Wald χ² tests for the joint significance of the exercise-by-spline interaction parameters.

As a complementary analysis, heterogeneity by socioeconomic position was assessed using the ordinal Graffar scale (five levels). An exercise × Graffar interaction term was included in the weighted models, level-specific marginal exercise effects were derived, and a pooled Wald χ² test was used to evaluate the overall socioeconomic gradient.

Results

Across the 20 imputed datasets, regular exercise was associated with lower disease activity. In the primary IPTW analysis, exercise was associated with a reduction in ASDAS of − 0.153 points (95% CI − 0.276 to − 0.031; p = 0.014). Results were consistent in the doubly robust AIPW analysis (− 0.155, 95% CI − 0.277 to − 0.034; p = 0.012). Covariate balance improved substantially after weighting, with the maximum pooled absolute standardised mean difference decreasing from 0.145 before weighting to 0.012 after weighting, indicating excellent balance (Fig. 1). Detailed covariate-specific balance metrics are provided in Supplementary file S1.

Fig. 1.

Love plot showing pooled standardized mean differences before and after IPTW

In a sensitivity analysis to address residual confounding, we repeated the analyses using a clinically expanded specification that additionally incorporated disease duration, diagnostic delay, smoking status, current treatment variables, and one proxy for functional burden. Results were virtually unchanged. In the expanded model, exercise remained associated with lower ASDAS both in IPTW analyses (ΔASDAS − 0.157, 95% CI − 0.279 to − 0.035; p = 0.012) and in AIPW analyses (ΔASDAS − 0.158, 95% CI − 0.280 to − 0.036; p = 0.011), closely matching the primary estimates (Supplementary file S2).

In sex-stratified sensitivity analyses, the association between regular exercise and lower ASDAS remained consistent among men in both the primary model (IPTW: −0.194, 95% CI − 0.336 to − 0.053; AIPW: −0.197, 95% CI − 0.335 to − 0.058) and the clinically expanded model (IPTW: −0.199, 95% CI − 0.341 to − 0.058; AIPW: −0.201, 95% CI − 0.342 to − 0.061). In women, point estimates were also negative but imprecise, with wide confidence intervals crossing the null in both the primary model (IPTW: −0.040, 95% CI − 0.289 to 0.209; AIPW: −0.030, 95% CI − 0.275 to 0.215) and the expanded model (IPTW: −0.055, 95% CI − 0.303 to 0.192; AIPW: −0.041, 95% CI − 0.282 to 0.201), precluding firm conclusions regarding sex-specific differences (Supplementary file S3).

Exploratory spline-based analyses suggested possible variation in the exercise–ASDAS association across age (Fig. 2 and supplementary file S4). Larger negative point estimates were observed in younger individuals, particularly between 30 and 42 years of age; however, confidence intervals widened outside this range, and the global Wald test for interaction was not statistically significant (χ²=6.52, df = 4, p = 0.164). Full pointwise estimates, 95% confidence intervals, and corresponding p-values are provided in Supplementary file S4.

Fig. 2.

Age-specific estimated association between regular exercise and ASDAS from spline interaction models across multiply imputed datasets (pooled estimates with 95% confidence intervals)

Similarly, exploratory BMI-varying analyses suggested possible non-linear variation in the association between exercise and ASDAS (Fig. 3 and supplementary file S5). More negative point estimates were observed around BMI values of 24–27, whereas estimates at lower and higher BMI values were smaller and less precise. The global Wald test for BMI interaction was not statistically significant (χ²=5.41, df = 4, p = 0.248). Complete pointwise results are shown in Supplementary file S5.

Fig. 3.

BMI-specific estimated association between regular exercise and ASDAS from spline interaction models across multiply imputed datasets (pooled estimates with 95% confidence intervals)

Finally, we examined whether the association between exercise and ASDAS differed across social strata defined by the Graffar scale (Fig. 4). Point estimates varied across Graffar categories, with more negative estimates in intermediate and higher strata and wider confidence intervals in the extremes. However, the global test for interaction between exercise and social class was not statistically significant (χ²=0.68, p = 0.410), indicating no clear evidence of a socioeconomic gradient in the exercise–ASDAS association (Supplementary file S6).

Fig. 4.

Estimated association between regular exercise and ASDAS across socioeconomic strata defined by the Graffar scale

Discussion

In this Ibero-American cohort of patients with axial spondyloarthritis (axSpA), self-reported regular exercise during the previous year was associated with lower disease activity at the study visit, as measured by ASDAS. Estimates were consistent across inverse probability of treatment weighting (IPTW) and doubly robust AIPW approaches, and post-weighting diagnostics indicated excellent balance across prespecified confounders. The association also remained stable in a clinically expanded sensitivity analysis that additionally accounted for disease duration, diagnostic delay, smoking, current treatment variables, and one proxy of functional burden, supporting the robustness of the findings to broader clinical adjustment. Although the pooled difference in ASDAS was modest in magnitude, the consistency of the estimates across analytic approaches supports the interpretation that regular exercise may be linked to lower disease activity in routine-care settings, consistent with international guideline frameworks that position exercise and physiotherapy as core components of axSpA management alongside pharmacological therapy [11, 17].

Our findings align with and extend the interventional literature. Randomized trials and structured exercise programmes in ankylosing spondylitis/axSpA have reported improvements across domains including disease activity, function, fatigue, and balance, with benefits observed for both high-intensity supervised programmes and aerobic training interventions [1, 6, 15]. Earlier trials also suggest that delivery format and supervision (e.g., group-based physiotherapy) can influence benefit magnitude [5], while home-based interventions have shown that structured exercise programmes can be feasible outside specialised centres [16]. Most trials, however, evaluate prescribed and often supervised regimens, whereas our exposure reflects habitual, self-reported exercise behaviour in routine care. This difference in “dose”, supervision, and standardisation may partly explain why the association observed in our registry-based analysis is smaller than effects sometimes reported under trial conditions.

Systematic evidence syntheses further reinforce the role of non-pharmacological care. The systematic literature review informing the ASAS–EULAR 2022 update concluded that non-pharmacological interventions—particularly structured exercise programmes—improve several disease-related outcomes in axSpA [10]. Similarly, the Cochrane review reported that exercise programmes probably improve physical function and may reduce pain, although effect sizes are generally modest and heterogeneity across programmes is substantial [13]. Our study adds to this evidence by evaluating real-world habitual exercise in a large multinational registry cohort, using ASDAS as the primary endpoint.

Interpretation of effect size should consider how ASDAS is used clinically. ASDAS cut-offs define disease activity states and improvement thresholds; clinically important improvement is typically defined as a ≥ 1.1-unit decrease (and ≥ 2.0 units as major improvement) [8]. Against these benchmarks, the average between-group difference observed here would not usually indicate a clinically important response at the individual level. At the same time, the binary and self-reported nature of the exercise variable is likely to attenuate associations through exposure misclassification, potentially masking larger differences in patients engaging in higher-intensity or more consistent activity. From a population-health perspective, even modest average reductions associated with a widely accessible, low-cost intervention may still be relevant when considered across large patient populations. In addition, the concept of a “goldilocks zone” for exercise in spondyloarthritis—where the type and intensity of physical activity may differentially influence entheseal inflammation and osteoproliferation—cannot be evaluated with a binary exposure variable, underscoring the need for future studies with more detailed exercise characterization.

Exploratory spline-based interaction analyses suggested that the exercise–ASDAS association might vary across age and BMI, with larger negative estimates in participants aged approximately 30–42 years and in those with BMI around 24–27. These patterns are biologically plausible: younger patients may have less accumulated structural damage and greater functional reserve to engage in—and benefit from—exercise-based behavioural change. BMI is also strongly implicated in axSpA outcomes; a systematic review and meta-analysis found that higher BMI is associated with higher disease activity, including ASDAS, supporting the idea that adiposity-related inflammatory and biomechanical pathways may shape disease expression and responsiveness to lifestyle factors [7]. However, the global Wald tests for the age and BMI interactions were not statistically significant, and pointwise patterns should therefore be interpreted cautiously, particularly given multiple comparisons. These findings are best viewed as exploratory and hypothesis-generating rather than confirmatory.

We additionally explored whether the exercise–ASDAS association differed by socioeconomic position using the Graffar scale. Although point estimates suggested more negative associations in higher strata, the global interaction test was not significant, providing no clear evidence of a socioeconomic gradient in the association. Nevertheless, socioeconomic context remains clinically relevant because socioeconomic disadvantage has been linked to greater disease burden and disability in axSpA even when access to biologic therapy is comparable [14]. From an implementation perspective, socioeconomic factors may influence opportunities for physical activity, health literacy, access to facilities, and sustained adherence. Consistent with this, real-world axSpA studies indicate that physical activity may be insufficiently performed and shaped by perceived barriers such as physical exertion, as well as contextual constraints [4]. More broadly, EULAR recommendations on physical activity for people with inflammatory arthritis emphasize feasibility, safety, and the need for structured advice and behaviour-change support in clinical practice [12], and an evidence-based consensus statement for ankylosing spondylitis highlights the importance of appropriate programme design and delivery to maximise benefit [9]. Collectively, these considerations support continued evaluation of contextual factors that may influence engagement with exercise and its downstream impact.

A further sensitivity analysis stratified by sex suggested that the association was more precisely estimated in men than in women. Among men, the estimates remained consistently negative and statistically significant in both the primary and clinically expanded models, whereas among women the point estimates were also negative but imprecise, with wide confidence intervals crossing the null. Given the smaller proportion of women in the cohort, these findings should not be interpreted as evidence of no association in women, but rather as indicating limited precision for sex-specific inference. This is consistent with the predominance of male patients in the cohort and supports cautious interpretation regarding generalisability across sex.

Several limitations warrant emphasis. First, this was a cross-sectional analysis: ASDAS and covariates were measured at the same clinical visit, and exercise was retrospectively self-reported for the previous year. Consequently, reverse causation is possible, since patients with lower disease activity may be more able to exercise. Second, residual confounding cannot be fully excluded despite weighting, multiple imputation, and sensitivity analyses with broader clinical adjustment. Third, exercise was measured as a binary self-report without information on intensity, frequency, type, or adherence, making misclassification and recall bias likely. Fourth, interaction analyses were exploratory and were not supported by statistically significant global interaction tests. Fifth, the predominance of male patients may limit precision and generalisability for women. Sixth, the registries used ESSG classification criteria, which may introduce classification bias relative to more recent ASAS criteria, and this should be considered when comparing our findings with studies using ASAS-classified populations. Finally, the inclusion period predates the availability of newer targeted therapies, such as IL-17 and JAK inhibitors, that may modify the relationship between exercise, pain, and inflammatory burden in contemporary practice. Strengths include the large harmonised multinational cohort integrating the REGISPONSER and RESPONDIA registries, robust missing-data handling via multiple imputation, careful diagnostics of balance after weighting, and the use of doubly robust estimation. Prior analyses from these registries further support the multifactorial nature of disease expression: inflammatory activity under anti-TNF therapy is shaped by multiple clinical determinants [3], and social factors independently influence pain severity in spondyloarthritis [2].

Conclusions

In this large Ibero-American cohort of patients with axial spondyloarthritis, self-reported regular exercise during the previous year was associated with lower disease activity at the study visit, as measured by ASDAS. This association was consistent across inverse probability–weighted, doubly robust, and clinically expanded sensitivity analyses, with good post-weighting covariate balance. Exploratory analyses suggested possible variation in the association across age, BMI, and socioeconomic position, although global interaction tests were not statistically significant. In sex-stratified sensitivity analyses, the association remained consistent and more precisely estimated among men, whereas estimates in women were imprecise and should be interpreted cautiously.

These findings are consistent with current recommendations supporting regular physical activity as a core component of axial spondyloarthritis management and extend trial-based evidence to real-world habitual exercise behaviour. However, the cross-sectional design, retrospective self-report of exercise, and potential for residual confounding preclude causal inference. Future studies should prioritise prospective longitudinal designs and/or pragmatic interventions using objective physical activity measures and more detailed characterisation of exercise type, frequency, and intensity.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

David Castro Corredor: study conception, methodological validation, interpretation of clinical findings, writing and critical revision of the manuscript.Luis Ángel Calvo Pascual: study conception, quantitative analysis, methodology development, results interpretation, writing, and revision of the manuscript.Eduardo Collantes-Estévez: data collectionClementina López-Medina: data collection.

Funding

The authors disclosed receipt of the following financial supportfor the research, authorship, and/or publication of this article: The REGISPONSERregistry was funded by the Spanish Society of Rheumatology. There was no funding for this subanalysis.

Data availability

The data that support the findings of this study are available from the REGISPONSER and RESPONDIA registries (Spanish Society of Rheumatology), but restrictions apply to the availability of these data, which were used under agreement for the current study, and so are not publicly available. Data are, however, available from the corresponding author on reasonable request and with permission of the Spanish Society of Rheumatology.

Declarations

Ethical approval

The study was conducted in compliance with the recommendations of the Declaration of Helsinki and was centrally approved by the Ethics Committee of the Reina Sofia University Hospital from Córdoba (Spain) on 21 April 2006 (“Comisión de Ética e Investigación Sanitaria, REGISPON-2004 y RESPONDIA-2006”). All participants provided written informed consent before enrolment.

Informed consent/ Patient consent

All patients provided written informed consent to participate in the REGISPONSER and RESPONDIA registries.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.