Age-related prevalence of sacroiliac joint variations and their association with structural damage in axial spondyloarthritis

Abstract

Objectives

To investigate the age-associated prevalence patterns of sacroiliac joint (SIJ) variations, evaluate their association with structural damage in axial spondyloarthritis (axSpA) patients, and compare their prevalence and morphological spectrum with European data.

Methods

This retrospective study analyzed high-resolution CT scans from 806 adults. Six predefined SIJ morphotypes were evaluated. Age-associated prevalence was modeled using generalized additive models (GAM) and Bayesian additive regression trees (BART) to capture non-linear dependencies. The association between SIJ variations and structural damage severity (Innsbruck CT grade) in axSpA patients was assessed with multivariable cumulative-link mixed models. European data were derived from a random-effects meta-analysis.

Results

SIJ variation prevalence showed a strong, monotonic increase with age (OR = 1.26/year, 95% CI 1.15–1.38), accelerating after 60 years and being more common in women (73.5% vs. 25.2%, P < 0.001). Critically, the presence of any SIJ variant was significantly associated with higher odds of severe structural damage in axSpA patients (OR = 2.23, 95% CI: 1.56–3.19, P < 0.001). BART modeling provided superior net benefit for risk prediction versus GAM, facilitating exploratory risk stratification. While overall prevalence was similar to European data (44.4% vs. 41.0%, P = 0.236), morphological distributions differed significantly: semicircular defects (11.2% vs. 4.8%) and crescent-shaped plates (12.4% vs. 3.6%) were more prevalent in our cohort.

Conclusions

SIJ variations are strongly associated with age, suggesting a degenerative component, and constitute a relevant risk marker for severe structural damage in axSpA. The BART model effectively supports exploratory risk assessment. Our cohort shares a similar prevalence but demonstrates a distinct morphological spectrum compared to European populations.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12891-026-09629-9.

Introduction

The sacroiliac joint (SIJ) is a critical synovial–ligamentous articulation connecting the sacrum and ilium [1, 2]. It transmits axial loads from the spine to the lower limbs and maintains pelvic stability[3], thereby serving an essential role in posture and locomotion. Its unique architecture consists of an anterior cartilaginous compartment and a posterior ligamentous compartment, enabling it to withstand substantial mechanical stress while permitting only limited motion [4, 5]. Multiple SIJ anatomical variants—including accessory SIJ, iliosacral complex, semicircular defects, crescent-shaped ilium, bipartite iliac plate, and isolated ossification centers—have been documented via computed tomography (CT) [6, 7], magnetic resonance imaging (MRI) [8], and synthetic CT (sCT) [9], with prevalence rates varying considerably across populations [10–13]. These variants may alter joint load distribution and biomechanical properties [14], potentially contributing directly to the pathophysiology of axial spondyloarthritis (axSpA) [15, 16]. Mechanical stress resulting from atypical joint morphology is hypothesized to promote enthesitis and stimulate new bone formation [17, 18]. It was demonstrated that specific variants, such as the iliosacral complex and crescent-shaped ilium, were significantly more common in patients with symptomatic SIJ disorders—including axSpA—than in controls [19]. Previous studies also reported higher rates of erosions and bone marrow edema in axSpA patients with these variants [20], supporting a “biomechanical–inflammatory” interaction model wherein anatomical anomalies induce localized stress concentration, triggering or sustaining inflammation in genetically susceptible individuals. However, current evidence on SIJ variants predominantly originates from European populations [10–13], creating a significant knowledge gap regarding their epidemiology and phenotypic distribution in Asian cohorts. Furthermore, the nature of the relationship between SIJ variations and age remains incompletely understood, particularly with respect to non-linear associations. Importantly, while previous research in axSpA has primarily emphasized inflammatory pathologies—such as bone marrow edema detected by MRI [20]—the association of joint morphology with structural damage severity of the SIJ, as detectable with high-resolution CT (HRCT), has not been fully elucidated.

The present study systematically characterizes the prevalence and subtype spectrum of SIJ variants in an Asian adult population, compares these findings with meta-analytic European data to examine ethnic variations, and applies—for the first time—generalized additive models (GAM) and Bayesian additive regression trees (BART) to model age-associated prevalence patterns. Furthermore, using multivariable models, we evaluate the independent association of joint morphology with structural damage severity of the SIJ in axSpA patients, clarifying its role in SIJ pathology.

Methods

Participants

Patients who underwent sacroiliac joint (SIJ) high-resolution computed tomography (HRCT) at the First Medical Center of the Chinese People’s Liberation Army General Hospital between January 1 and June 30, 2024, were included. Exclusion criteria were: (1) repeated examinations within six months, in which case only the first scan was retained; (2) age < 18 years; (3) non-Asian ethnicity; and (4) The anatomical architecture of the sacroiliac joint could not be identified due to complete ankylosis [7]. A total of 806 patients were ultimately included (Fig. 1). The axSpA diagnosis was defined by a rheumatologist and fulfilled ASAS classification criteria [21].

Fig. 1.

Patient Selection and Classification Flowchart. Flowchart illustrating patient selection and classification process.From 842 initially screened patients undergoing CT imaging, 806 eligible patients were included after applying exclusion criteria. Patients were classified by sacroiliac joint variant type and grouped into axSpA (n = 310) and non-axSpA (n = 496) cohorts according to ASAS criteria, with axSpA patients further classified using Innsbruck criteria. SIJ, sacroiliac joint; CT, computed tomography; axSpA, axial spondyloarthritis; ASAS, Assessment of SpondyloArthritis international Society

For the epidemiological analysis in European populations, PubMed, the Cochrane Library, Embase, and Web of Science were systematically searched for relevant studies published up to 30 June 2024. The detailed search strategy is provided in Table S1.

CT assessment

All scans were performed using a Revolution CT scanner (GE Healthcare, United States) at 120 kVp with automatic tube current modulation, and reconstructed in a bone window with a slice thickness of 1.0–1.2 mm (window width: 2600 HU; window level: 800 HU). Two rheumatologists, after conducting a calibration exercise by jointly scoring 30 cases outside the study cohort [7, 9], independently documented six predefined sacroiliac joint variants according to the Prassopoulos [6] classification—accessory SIJ, iliosacral complex, semicircular defects, crescent-shaped ilium, bipartite iliac plate, and isolated ossification centers (Figure S1)—and graded structural changes using the Innsbruck classification [22, 23]. This methodology of image assessment by trained rheumatologists aligns with established practices in research on sacroiliac joint variations [7, 9]. Discrepancies were resolved by a third senior rheumatologist. Inter-observer agreement for sacroiliac joint variant subtypes was assessed using Cohen’s κ [24]. Given the ordinal nature of the grading system, agreement for Innsbruck grades was evaluated with a weighted κ (quadratic weights) [25].

SIJ structural changes were graded according to the Innsbruck classification [22, 23], which is specific to HRCT: Grade I(A): joint space > 4 mm; Grade I(B): joint space < 2 mm; Grade II(A): articular surface irregularity; Grade II(B): bone erosion; Grade III(A): subchondral sclerosis; Grade III(B): osteophyte formation; Grade IV(A): bony bridging; Grade IV(B): complete ankylosis.

Data collection

Demographic and clinical data, including age, sex, and relevant clinical characteristics, were extracted from medical records. Imaging data included SIJ variant type, laterality of involvement, and Innsbruck grade.

Statistical analysis

All analyses were conducted using R (v4.3.2). Data are presented as mean ± SD for continuous variables or frequency (%) for categorical variables. Normality was assessed with the Shapiro-Wilk test. Group comparisons used t-tests for normal data or Mann-Whitney U tests for non-normal data (reported as median [IQR]). Categorical variables were compared with chi-square or Fisher’s exact test (expected counts < 5). Multivariable logistic regression was further applied for categorical outcomes. Age-related risk curves were fitted using Bayesian Additive Regression Trees (BART) and penalized spline generalized additive models (GAM). Convergence of the BART Markov chain Monte Carlo algorithm was assessed using the Gelman–Rubin statistic (≤ 1.01). The association between variations and the Innsbruck grade was analyzed using a cumulative link mixed model (CLMM), generalized random forests (GRF), a Bayesian CLMM, and average treatment effect (ATE) analysis, with patient ID included as a random effect and with adjustments for age and sex.

Variation rates from European studies were logit-transformed and pooled using a random-effects GLMM-logit model, providing estimates with 95% confidence and prediction intervals adjusted via the Hartung–Knapp–Sidik–Jonkman (HKSJ) method. Heterogeneity was assessed using τ²(REML), I², and the Q test. Subgroup analyses were conducted by sex and subtype, and a mixed-effects meta-regression was performed with the “Europe–Asia” variable as a covariate.

Ethics statement

The study protocol was approved by the Ethics Committee of the Chinese PLA General Hospital (approval No. 2025-863-1). As all participants had provided written informed consent prior to their previous HRCT examinations, and given the retrospective nature of this study, the requirement for additional informed consent was waived. All procedures were conducted in accordance with the Declaration of Helsinki.

Results

Sacroiliac joint variations

The baseline demographic and clinical characteristics of the study cohort are summarized in Table 1. The study included 806 participants, consisting of 485 males (60.2%) and 321 females (39.8%). Male participants exhibited a significantly lower mean age than females (32.0 [26.0–40.0] years vs. 39.0 [32.0–49.0] years). Based on diagnostic classification, 310 individuals were diagnosed with axSpA and 496 were classified as non-axSpA. The diagnoses in the non-axSpA group comprised lumbar disc herniation, spinal stenosis, lumbar muscle strain, fasciitis, and osteitis condensans. SIJ variations were identified in 44.4% of the participants, with a significantly higher prevalence among females compared to males (73.5% vs. 25.2%, P < 0.001). Among specific morphological variants, bipartite iliac bony plate was observed in 20.3% of the cohort and demonstrated a pronounced sex-based disparity, being substantially more prevalent in females (48.0% vs. 2.1% in males, P < 0.001). Similarly, semicircular defects were more common in females (15.3% vs. 8.5%, P = 0.004), as were crescent-like iliac bony plates (19.0% vs. 8.0%, P < 0.001). Accessory joints, present in 8.8% of participants, also showed higher prevalence in females (14.3% vs. 5.2%, P < 0.001). In contrast, no significant sex differences were observed in the prevalence of iliosacral complex, ossification centers, or other variants.

Table 1.

Demographic characteristics and prevalence of sacroiliac joint variants/related lesions

	Total (n = 806)	Bilaterally involved, %	Sex			Diagnosis
			Male n (%)	Female n (%)	P (Sex)	axSpA n (%)	Non-axSpA n (%)	P (Diagnosis)
General Characteristics
Number of subjects	806	–	485 (60.2)	321 (39.8)		310	496
Age, years (median [IQR])	35.0 [28.0–44.8]	–	32.0 [26.0–40.0]	39.0 [32.0–49.0]	< 0.001
SIJ Variants and Lesions
SIJ Variation	358 (44.4)	–	122 (25.2)	236 (73.5)	< 0.001	130(41.9)	228 (46.0)	0.295
Accessory Joints	71 (8.8)	54.9	25 (5.2)	46 (14.3)	< 0.001	24 (7.7)	47 (9.5)	0.473
Iliosacral Complex	36 (4.5)	72.2	18 (3.7)	18 (5.6)	0.271	9 (2.9)	27 (5.4)	0.128
Bipartite Iliac Bony Plate	164 (20.3)	54.3	10 (2.1)	154 (48.0)	< 0.001	47 (15.2)	117 (23.6)	0.005
Semicircular Defects	90 (11.2)	62.2	41 (8.5)	49 (15.3)	0.004	41 (13.2)	49 (9.9)	0.176
Crescent-like Iliac Bony Plate	100 (12.4)	63	39 (8.0)	61 (19.0)	< 0.001	47 (15.2)	53 (10.7)	0.077
Ossification Centers	4 (0.5)	0	4 (0.8)	0 (0.0)	0.155	1 (0.3)	3 (0.6)	1

Baseline characteristics and prevalence of sacroiliac joint (SIJ) variations/lesions (n = 806). Continuous and categorical data are presented as mean ± SD and n (%), respectively. Bilaterally involved: Percentage among those with lesions. SIJ sacroiliac joint, SD standard deviation, axSpA axial spondyloarthritis

When analyzed by diagnostic category, bipartite iliac bony plate prevalence was significantly higher in non-axSpA patients than in axSpA patients (23.6% vs. 15.2%, P = 0.005). No statistically significant differences between diagnostic groups were found for accessory joints, semicircular defects, crescent-like iliac bony plates, or iliosacral complex. Regarding laterality patterns, certain variants exhibited substantial bilateral involvement. Specifically, iliosacral complex and semicircular defects demonstrated bilateral presence in over 60% of affected cases, while bipartite iliac bony plate showed bilateral involvement in more than half of identified cases.

Association between sacroiliac joint variation and age

Age-stratified modeling using Bayesian Additive Regression Trees (BART) and generalized additive models (GAM) demonstrated a continuous increase in the probability of detecting sacroiliac joint (SIJ) anatomical variants with advancing age (Fig. 2). Both models exhibited monotonically increasing curves, with BART consistently predicting higher probabilities of detection than GAM, indicating a more conservative estimation approach. The smooth term of the GAM was highly significant (P < 0.001). The 95% confidence intervals for both models widened substantially beyond 60 years of age, reflecting increased estimation uncertainty in older populations.

Fig. 2.

Age-Dependent Probability of Sacroiliac Joint Variation Derived from Generalized Additive Modeling. Fitted curve (solid line) represents the smoothed probability of sacroiliac joint (SIJ) variation as a function of age, generated by a generalized additive model (GAM). The shaded band denotes the 95% confidence interval. The smooth term for age was statistically significant (P < 0.001). Sample size: n = 806

Decision curve analysis (Fig. 3, upper panel) demonstrated a significant influence of model selection on clinical utility, with the BART model yielding a superior net benefit across the clinically decisive threshold probability range of 0.25 to 0.65. To translate these predictions into stratifiable groups, we defined exploratory risk strata based on this decisive range: low (< 30%), intermediate (30–60%), and high (> 60%) probability [26]. Consequently, risk stratification analysis (Fig. 3, lower panel) visualized how these models classify individuals across age groups. Notably, the BART model demonstrated a more distinct and monotonic progression towards higher probability categories with advancing age, clearly classifying the oldest individuals (e.g., > 75 years) into the high-probability (> 60%) group. In contrast, GAM resulted in a broader, less decisive intermediate-probability (30–60%) population, particularly in the 40–70 age range.

Fig. 3.

Decision Curve Analysis and Clinical Risk Stratification for Sacroiliac Joint Variation. A Decision curve analysis comparing the net benefit of Bayesian additive regression trees (BART; red solid line), generalized additive models (GAM; blue solid line), “Treat All” (black dashed line), and “Treat None” (gray dashed line) strategies across threshold probabilities (0–1.0). B Exploratory risk stratification distribution showing low (< 30%), intermediate (30–60%), and high (> 60%) risk categories across age groups. The BART model demonstrated superior net benefit at threshold probabilities ≤ 0.65. n = 806

Model reliability was confirmed through comprehensive convergence diagnostics (Figure S2). Markov chain Monte Carlo (MCMC) traces, posterior density plots, and autocorrelation functions all indicated excellent convergence (Gelman-Rubin statistic = 1.001), with a modal tree depth of 2 consistent with default prior settings.

Association between sacroiliac joint variation and Innsbruck grading

A total of 620 sacroiliac joints from 310 patients with axSPA were analyzed, among which 219 joints exhibited sacroiliac joint variations (Var n = 219), and 401 joints were normal (NoVar n = 401) (Fig. 4). A significant association between SIJ morphological variation and the severity of structural damage was confirmed. In a multivariable CLMM (Fig. 4) adjusted for age and sex, SIJ variation was independently associated with higher Innsbruck grades (OR = 2.23, 95% CI: 1.56–3.19, P < 0.001); per one-year increase in age, odds rose by 5% (OR = 1.05, 95% CI 1.04–1.06, P < 0.001); female sex showed a marked protective association (OR = 0.32, 95% CI 0.22–0.48, P < 0.001). Consistent with the primary model, Fig. 5 provides distributional corroboration: in the no-variation group, nearly half of cases fell within Innsbruck grades 1–2, whereas in the variation group the distribution shifted rightward, with more than 70% of cases at grade ≥ 3. The CLMM effect size displayed in Fig. 5 (OR = 2.23) closely matches Fig. 4, underscoring internal coherence between the model-based estimates and the descriptive distribution.

Fig. 4.

Adjusted Odds Ratios of Key Predictors for Higher Innsbruck CT Grade. Forest plot displaying adjusted odds ratios (log scale) and 95% confidence intervals for three predictors associated with higher Innsbruck CT grade severity: sacroiliac joint (SIJ) variation, sex (female), and age (per year). Estimates were derived from a multivariable cumulative-link mixed model adjusting for potential confounders. SIJ variation (OR = 2.23; 95% CI: 1.56–3.19; P < 0.001) and increasing age (OR = 1.05 per year; 95% CI: 1.04–1.06; P < 0.001) were significantly associated with higher grade severity, while female sex was associated with significantly lower odds (OR = 0.32; 95% CI: 0.22–0.48; P < 0.001). The horizontal axis uses a logarithmic scale to facilitate visualization of effect sizes

Fig. 5.

Distribution of Innsbruck CT Grades in Subjects With and Without Sacroiliac Joint Variations. Stacked bar chart showing the proportional distribution of Innsbruck CT-based structural damage grades stratified by the presence of sacroiliac joint (SIJ) variations. The “NoVar” group (n = 401) represents subjects without anatomical variations, while the “Var” group (n = 219) represents subjects with at least one SIJ variation. Grade 4 A indicates the most severe structural damage. A cumulative link mixed model (CLMM) analysis revealed that the presence of any SIJ variation was associated with significantly higher odds of more severe Innsbruck grades (adjusted OR = 2.23; 95% CI: 1.56–3.19; P < 0.001)

Robustness checks upheld these findings. Figure S3 contrasts three strategies—generalized random forest (GRF) ATE, frequentist CLMM, and Bayesian CLMM—showing effect sizes on the log-OR scale clustered between 1 and 2, with substantial overlap of the associated intervals, thereby reinforcing the consistency of the association. Figure S4 juxtaposes posterior probabilities from the Bayesian cumulative link model with posterior-assigned Innsbruck grades, demonstrating a monotonic increase in risk across grades; the variant group’s average treatment effect (ATE) was 0.55 (95% CI 0.28–0.82), indicating a consistent trend toward greater damage. The posterior distribution from the Bayesian CLMM (Figure S5) yielded a median OR = 2.0 (95% CI 1.2–3.1), nearly overlapping the frequentist estimate and corroborating the reliability of the conclusions. Figure S6 displays patient-specific random intercepts with an approximately symmetric scatter, suggesting that between-patient heterogeneity was adequately captured.

Association between sacroiliac joint variation and population groups

Four studies were included in the final analysis (Table S2) [6, 7, 10, 27]. The methodological quality of these included studies was high (Table S3). A total of 2,060 individuals from four European studies were included as the control cohort (Table S4). Pooled data from European populations (Table 2) demonstrate that 41% of adults exhibit at least one sacroiliac joint (SIJ) variant, with a pronounced sex-based disparity: 60.8% in females compared to 29.4% in males. Variant-specific prevalence across European cohorts was heterogeneous: accessory joints (16.0%), bipartite iliac bony plate (16.3%), iliosacral complex (8.1%), semicircular defects (4.8%), crescent-like iliac plate (3.6%), and ossification centers (1.5%). Between-study heterogeneity was substantial (I² = 0.71–0.97), reflecting methodological variations among the included European studies. Nonetheless, the random-effects model confirms that atypical SIJ morphology represents a common anatomical variant in this population.

Table 2.

Pooled prevalence of sacroiliac joint anatomical variants in European adults

Variant	Pooled prevalence, %	95% CI	τ²	I²	Q (P)	Prediction interval, %
Overall SIJ variation	41.0	36.2–45.9	0.042	0.856	21.4 (< 0.01)	18.2–68.4
Female variation	60.8	42.4–77.0	0.21	0.969	24.9 (< 0.001)	12.7–95.7
Male variation	29.4	17.5–44.6	0.152	0.943	18.5 (< 0.001)	6.5–74.3
Accessory SIJ joints	16.0	11.7–21.5	0.021	0.876	15.0 (< 0.01)	5.7–36.4
Iliosacral complex	8.1	4.7–13.7	0.025	0.861	11.2 (< 0.01)	1.7–30.4
Bipartite iliac bony plate	16.3	9.5–26.7	0.06	0.937	13.5 (< 0.01)	2.5–59.1
Semicircular defects	4.8	2.9–7.7	0.015	0.812	8.1 (0.04)	1.0–20.9
Crescent-like iliac plate	3.6	2.0–6.4	0.009	0.715	7.0 (0.06)	0.8–15.7
Ossification centers	1.5	0.6–3.6	0.009	0.684	2.4 (0.32)	0.2–9.1

Pooled prevalence of sacroiliac joint anatomical variants from European studies (random-effects model). Results show pooled prevalence (%),

95% confidence interval (95% CI), prediction interval, and heterogeneity statistics (τ², I², Q). CI confidence interval, τ² Tau-squared (estimate of between-study heterogeneity), I² I-squared statistic, Q Cochran’s Q statistic, SIJ sacroiliac join

Comparative analysis between Asian and European cohorts (Table 3) revealed a comparable overall prevalence of SIJ variation (44.4% vs. 41.0%, OR = 0.87, P = 0.236). Marked continental disparities were observed, however, in specific variant subtypes. Asian females exhibited a significantly higher prevalence of any SIJ variant relative to European females (73.5% vs. 60.8%, OR = 0.56, P = 0.002). In contrast, accessory joints (16.0% vs. 8.8%, OR = 1.86, P = 0.005) and iliosacral complex (8.1% vs. 4.5%, OR = 1.89, P = 0.009) were significantly more prevalent in European individuals. Conversely, semicircular defects (11.2% vs. 4.8%, OR = 0.40, P = 0.001) and crescent-like iliac plates (12.4% vs. 3.6%, OR = 0.27, P < 0.001) were more common in the Asian cohort. The bipartite iliac plate showed only borderline statistical significance (20.4% vs. 16.3%, P = 0.057), and no significant difference was detected in male-specific prevalence between continents.

Table 3.

Meta-analysis comparison of sacroiliac joint (SIJ) variant prevalence between European and Asian populations using random-effects models​

Variant	Europe, %	Asia, %	OR (95% CI)	P value	τ²	I²
Overall SIJ variation	41.0	44.4	0.87 (0.70–1.09	0.236	0.042	0.810
Female variation	60.8	73.5	0.56 (0.38–0.81)	0.002	0.21	0.904
Male variation	29.4	25.2	1.24 (0.83–1.87)	0.297	0.152	0.889
Accessory SIJ joints	16.0	8.8	1.86 (1.21–2.88)	0.005	0.021	0.759
Iliosacral complex	8.1	4.5	1.89 (1.17–3.08)	0.009	0.025	0.702
Bipartite iliac plate	16.3	20.4	0.76 (0.57–1.01)	0.057	0.060	0.823
Semicircular defects	4.8	11.2	0.40 (0.24–0.67)	0.001	0.015	0.686
Crescent-like iliac plate	3.6	12.4	0.27 (0.17–0.43)	< 0.001	0.009	0.721
Ossification centers	1.5	0.5	2.97 (0.78–11.27)	0.107	0.009	0.734

Meta-analysis comparison of sacroiliac joint (SIJ) variant prevalence between European and Asian populations using random-effects models. Data are presented as prevalence percentages, odds ratios (OR), 95% confidence intervals (95% CI), P-values, and heterogeneity statistics (τ² and I²). SIJ sacroiliac joint, OR odds ratio, CI confidence interval, τ² Tau-squared (estimate of between-study heterogeneity), I² I-squared statistic (I² > 50% indicates substantial heterogeneity)

Inter-reader variability for SIJ variants

Inter-reader agreement was excellent, with Kappa values of 0.89 for accessory SIJ, 0.91 for iliosacral complex, 0.81 for semicircular defects, 0.86 for crescent-shaped ilium, 0.96 for bipartite iliac plate, and 0.93 for isolated ossification centers. Meanwhile, the weighted Cohen’s κ for the Innsbruck grading system was 0.94.

Discussion

In this cohort of 806 adults, SIJ anatomical variants were common—affecting approximately half of the general population—and their prevalence increased with age. Among patients with axSpA, those with SIJ variants exhibited more severe structural osseous damage than those without variants.

The observed prevalence of 44.4% closely matches the pooled European estimate of 41%, indicating that SIJ variants are a widespread phenomenon rather than one confined to specific ethnic groups. Nevertheless, clear morphological differences were evident between populations: while a bipartite iliac plate predominated in both, semicircular defects and a crescent-shaped iliac plate were 2–4 times more frequent in Asians, whereas the iliosacral complex and accessory SIJ were more common in European series. These marked differences in morphological spectra may reflect variation in pelvic biomechanics, lifestyle, or genetic background. They underscore the need to include participants from diverse regions and genetic backgrounds in epidemiologic studies of SIJ variants to enhance generalizability (external validity). In line with CT findings from a German axSpA registry [19] and MRI observations by El Rafei et al. [11], we also confirmed a higher prevalence of SIJ variants in women than in men, potentially related to hormonal influences, pregnancy, or pelvic anatomical differences. This highlights the importance of accounting for sex when analyzing SIJ variants as outcomes. Corroborating previous research, most variants in our cohort were bilateral. Consistent with prior studies, the predominantly bilateral distribution of the variants in our cohort provides a plausible explanation for the influence of systemic and symmetric mechanical stress.

We also observed a notable sex disparity: females had a significantly higher prevalence of SIJ variants compared to males. However, our multivariable analysis revealed an intriguing “gender paradox”: despite having a higher burden of anatomical variants, female sex was a strong independent protective factor against severe structural damage (OR = 0.32), implying that males are more predisposed to severe osseous progression. This discrepancy can be explained by sex-specific differences in biomechanical adaptation versus osteogenic response. As hypothesized by Ulas et al. [28], the female SIJ is anatomically adapted for higher mobility (nutation) and childbirth, with naturally laxer ligaments; consequently, females may develop adaptive morphological variants (e.g., accessory joints) to stabilize the joint under stress without progressing to fusion. In contrast, as highlighted by Ziegeler et al. [29], the male SIJ is structurally more rigid, and the male biological environment has a stronger propensity for robust bone formation (osteophytosis and sclerosis). Thus, under mechanical stress and inflammation, female joints tend to undergo “morphological adaptation,” whereas male joints are more likely to progress toward irreversible “structural ankylosis.”

Previous reports have noted a higher prevalence of SIJ variants in individuals older than 40 years [30], and another study identified age as an independent risk factor for accessory joints [31]—both implying an age-related association that had not been rigorously examined. Using BART and GAMs, we analyzed the dynamic relationship between SIJ variants and age and observed a continuously increasing, non-linear probability of SIJ variants across adulthood. Rather than purely developmental traits, these findings support the concept of “degenerative remodeling” as described by Tok Umay and Korkmaz [12]. Age likely serves as a proxy for cumulative mechanical loading. As a load-bearing articulation linking the spine and pelvis, the SIJ is subjected to lifelong compressive and shear forces. Consistent with Trentadue et al. [31], who found accessory joints to be acquired features increasing with age, our data suggest that repetitive loading contributes to cartilage wear and bony remodeling, resulting in variant formation as an adaptive response to longstanding stress. The cumulative effects of inflammation and repair also merit attention. Even low-grade, subclinical microtrauma or stress-related inflammation can, through repeated injury–repair cycles, leave a lasting imprint by promoting fibrous tissue deposition and heterotopic bone formation that alters normal joint morphology. In addition, age-related changes in the biomechanical milieu—disc desiccation and degeneration, vertebral height loss, and weakening of periarticular muscles and ligaments—perturb global spinopelvic balance. As a key node in force transmission, the SIJ may remodel its morphology to seek stability under the shifted mechanical environment.

A further key finding is that, after adjustment for age and sex, axSpA patients with SIJ variants had 2.23-fold higher odds of being classified into a higher grade on the Innsbruck scale compared with those without variants. This indicates that SIJ variants act as significant risk markers associated with more severe structural osseous damage in axSpA, although the cross-sectional design precludes establishing strict causality. A plausible mechanism involves a vicious cycle between variant-driven biomechanical alterations and immune-mediated inflammation. Anatomical variants disrupt normal force distribution and stress buffering within the SIJ, leading to focal stress concentration [32, 33]. Persistent microinjury activates mechanosensitive cells and amplifies inflammatory signaling: increased fluid shear stress exacerbates RANKL–OPG imbalance to drive osteoclastogenesis and bone resorption [34–36], while mechanical stress synergizes with inflammatory cytokines such as TNF-α and IL-17 to promote further release of IL-6 and PGE2 [37], thereby intensifying inflammation. Furthermore, recent research by Ziegeler et al. [20] demonstrated that intra-articular variants significantly increase the risk for erosion and bone marrow edema in axSpA, supporting the view that altered joint mechanics may exacerbate the inflammatory burden. Concurrently, downregulation of inhibitors of the Wnt/β-catenin pathway lifts the brake on osteoblast activity [38, 39], accelerating aberrant new bone formation. The abnormal morphology of variant joints is also more prone to degeneration and microdamage, providing additional niches for inflammatory cell infiltration and dysregulated tissue repair, culminating in rapid progression of structural lesions, including erosion, sclerosis, and ankylosis [40, 41]. Collectively, these findings suggest that SIJ variants may serve as predictive markers of structural damage severity in axSpA. Clinically, the identification of such variants should prompt rheumatologists to consider the potential for mechanically driven progression. While our data do not yet justify altering pharmacological strategies based solely on variants, they suggest that incorporating biomechanical considerations—such as tailored physical therapy to redistribute mechanical loads—might be a valuable adjunct to standard care for patients with these specific anatomical features [42, 43].

This study has some limitations. Our cohort comprised East Asian individuals and employed a single-center, retrospective design, which constrains generalizability within Asia. To obtain a comprehensive and accurate picture of the epidemiology of SIJ variants across Asian populations, multicenter, prospective cohort studies across diverse regions are warranted. The cross-sectional design precludes causal inference, leaving it unclear whether variants precede or follow structural damage; longitudinal imaging with low-dose CT or MRI could clarify temporal sequencing. We did not capture functional measures such as pain severity, biomechanical parameters, or quality of life, leaving the clinical correlates of specific variants incompletely defined.

In summary, SIJ variants are common among Asian adults, more frequent in women, and increase with age in a non-linear fashion. Although overall prevalence is similar to that reported in European cohorts, the morphological spectrum differs, and SIJ variants are closely associated with more severe structural damage.

Abbreviations

SIJ

Sacroiliac joint

axSpA

Axial spondyloarthritis

CT

Computed tomography

HRCT

High-resolution computed tomography

BART

Bayesian additive regression trees

GAM

Generalized additive models

OR

Odds ratio

CI

Confidence interval

Authors’ contributions

Jian Zhu conceived the initial study design. Jian Zhu, Jiali Yu, Xiaojian Ji, Jiaxin Bai, Wenrui Zhang, Simin Liao, Yiwen Wang, Yufei Guo, Chao Xue and Feng Huang contributed to the discussion and refinement of the study design. Jiali Yu and Jiaxin Bai contributed to the investigation and data collection. Jian Zhu, Jiali Yu and Xiaojian Ji contributed to data analysis and interpretation. The manuscript was drafted by Jiali Yu. All authors critically reviewed the manuscript for important intellectual content and approved the final version for publication. All authors take responsibility for the integrity of the work as a whole. Jian Zhu is the guarantor. Jiali Yu, Xiaojian Ji and Jiaxin Bai contributed equally and share first authorship.

Funding

The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author, Dr. Jian Zhu (jian_jzhu@126.com), upon reasonable request.

Declarations

Ethics approval and consent to participate

This study involving human participants was approved by the Ethics Committee of the Chinese PLA General Hospital (approval No. 2025-863-1). Given the retrospective design and anonymized data processing, the requirement for informed consent was waived by the committee.

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.