Validation of the Chinese version of the 12-Item multiple sclerosis walking scale in individuals with multiple sclerosis: a psychometric and item response theory analysis

Abstract

Background

The 12-item Multiple Sclerosis Walking Scale (MSWS-12) is widely used to assess patient-perceived walking disability in individuals with multiple sclerosis (MS). Despite its international application, a validated Mandarin version has been lacking. This study aimed to translate and validate the Chinese version of the MSWS-12 and examine its psychometric properties using both classical test theory (CTT) and item response theory (IRT) approaches.

Methods

The MSWS-12 was translated into Chinese following international guidelines, including forward and backward translation, expert panel review, and cognitive debriefing. Cognitive debriefing was conducted with 20 individuals with MS (10 urban, 10 rural; 40% with a high school education or lower), confirming that all items were clear, culturally relevant, and conceptually equivalent. Participants were recruited from neurology clinics and included adults with clinically confirmed MS who had sufficient cognitive and physical function to complete the self-report instrument. Measurement properties were evaluated in accordance with the COnsensus-based Standards for the selection of health Measurement INstruments (COSMIN) guideline, including internal consistency, test–retest reliability, structural validity, measurement invariance, and item-level analysis using the Graded Response Model (GRM).

Results

Participants (N = 292) had a mean age of 41.72 years (SD = 5.23), with 60.6% identifying as female, and represented a diverse educational backgrounds. The Chinese version of the MSWS-12 demonstrated excellent internal consistency (Cronbach’s α = 0.92; McDonald’s ω = 0.91) and strong test–retest reliability (intraclass correlation coefficient [ICC] = 0.91, 95% CI [0.86, 0.94]), indicating high temporal stability. CFA supported a unidimensional structure with excellent model fit (CFI = 0.987, TLI = 0.984, IFI = 0.987, SRMR = 0.032, RMSEA = 0.038. 90% CI [0.015, 0.056]). Measurement invariance was confirmed across sex, residency, and disease duration at configural, metric, scalar, and strict levels (ΔCFI < 0.010, ΔRMSEA < 0.015). IRT analysis showed high item discrimination parameters (α = 2.01 - 2.95) and well-ordered respnse thresholds. No evidence of differential item functioning (DIF) was found (Δpseudo-R² < 0.02). Criterion-related validity was supported by strong correlations with the T25FW (r = .64, 95% CI [0.57, 0.70]) and the 6MWT (r = –.61, 95% CI [–0.67, –0.53]). Convergent validity was indicated by moderate associations with the EDSS (r = .47, 95% CI [0.37, 0.55]) and the MFIS (r = .45, 95% CI [0.36, 0.54]), while discriminant validity was confirmed by a weak association with the PHQ-9 (r = .28, 95% CI [0.17, 0.38]).

Conclusion

The Chinese version of the MSWS-12 is a psychometrically sound, culturally valid tool for assessing ambulatory disability in Mandarin-speaking individuals with MS. Its demonstrated reliability, validity, and measurement invariance support its application in clinical practice and cross-cultural research.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40359-025-03607-x.

Introduction

Walking dysfunction represents a core clinical manifestation of multiple sclerosis (MS) [1], an immune-mediated demyelinating disorder characterized by progressive neurological deterioration [2]. The disease pathophysiology significantly compromises ambulatory function, resulting in measurable deficits in gait velocity, postural stability, physical endurance, and overall locomotor performance [3, 4]. These impairments lead to progressive loss of functional independence [5] while substantially compromising psychosocial well-being [6] and health-related quality of life [7]. Given the complex and multidimensional nature of walking disability in MS, there remains a critical need for assessment tools that not only objectively quantify motor decline but also capture the patient’s subjective experience of disease progression [8].

Developed as a patient-reported outcome (PRO) measure, the 12-item Multiple Sclerosis Walking Scale (MSWS-12) provides a standardized method for evaluating walking ability from the patient perspective [9]. This validated instrument comprehensively assesses self-reported limitations across critical functional domains, including gait parameters, balance control, endurance capacity, and the broader psychosocial impact of mobility impairment [10]. The MSWS-12 has demonstrated robust cross-cultural adaptability through rigorous linguistic validation studies conducted in diverse populations, including Brazil [11], Iran [12], Italy [13], South Korea [14], Germany [15], Japan [16], and Croatia [17]. These multinational studies have consistently confirmed the scale’s excellent reliability and stable factorial structure across different cultural contexts.

While performance-based measures like the Six-Minute Walk Test (6MWT) [18] and Timed 25-Foot Walk (T25FW) [19] provide objective and quantifiable indices of ambulatory function, they lack sensitivity to the psychosocial ramifications of walking disability, including its effects on social engagement, occupational performance, and overall life satisfaction [20]. The MSWS-12 addresses this critical gap by effectively capturing the subjective, emotional, and social consequences of walking impairment, thereby serving as an indispensable complement to performance-based assessments [21]. This integrated evaluation approach enables a more holistic, patient-centered understanding of mobility limitations in MS [22]. Notably, comparative studies have established the MSWS-12’s superior responsiveness in detecting clinically meaningful improvements following rehabilitation interventions when compared to brief walking tests [23]. This enhanced sensitivity underscores the instrument’s unique capacity to identify treatment-responsive aspects of walking disability that often elude detection by conventional performance measures alone.

Ongoing concerns regarding the cross-cultural applicability of the MSWS-12 have been substantiated by several scholars who have demonstrated that specific items may not maintain invariant psychometric properties across diverse demographic groups [24–26]. These findings suggest potential variability in item interpretation and response behavior, thereby jeopardizing the comparability of scores across populations. Establishing measurement invariance is essential to ensure the validity, reliability, and interpretive robustness of the MSWS-12 within multicultural and heterogeneous clinical populations [27, 28]. Such concerns underscore the imperative of conducting item-level psychometric evaluations—most notably differential item functioning (DIF) analyses—to determine whether the scale performs equivalently across subgroups [29, 30].

The present study

Validating the MSWS-12 for Mandarin-speaking populations is essential to ensure psychometric robustness, minimize cultural bias in item interpretation, and establish its utility as a reliable and equitable tool for clinicians and researchers working in diverse sociocultural settings. Although the MSWS-12 has been extensively applied across cultures, its psychometric performance has not yet been systematically examined in Mandarin-speaking populations. Given the instrument’s potential to inform both clinical decision-making and research in China, a culturally adapted validation is critical to ensure interpretive accuracy and measurement equity.

The methodological standards established by the COnsensus-based Standards for the selection of health Measurement INstruments (COSMIN) initiative emphasize that rigorous psychometric evaluation must integrate both classical test theory (CTT) and modern frameworks such as item response theory (IRT), as these approaches provide complementary insights into dimensionality, reliability, and differential item functioning [31]. Within IRT, Graded Response Model (GRM) is particularly suitable for instruments with ordered categorical responses, such as the Likert-type items in the MSWS-12, as it estimates both item discrimination and threshold parameters [32].

Accordingly, the present study sought to translate and psychometrically validate the MSWS-12 for Mandarin-speaking MS, employing a combined CTT–IRT approach. We hypothesized that the MSWS-12 would demonstrate a unidimensional factor structure, confirmed through CFA, and would yield high internal consistency alongside strong test–retest reliability over a two-week interval. With respect to construct validity, we anticipated strong correlations with objective walking speed and endurance measures, moderate correlations with overall disability and fatigue indices, and weak correlations with depressive symptoms, thereby supporting discriminant validity. Furthermore, we predicted measurement invariance across gender, residency status, and disease duration, ensuring that the scale’s latent structure, loadings, and intercepts would remain stable across these subgroups.

At the item level, GRM-based analyses were expected to provide evidence of excellent model fit, with items discriminating effectively across levels of walking disability and showing minimal local dependence. Finally, DIF examined through ordinal logistic regression, was anticipated to be non-significant across gender, residency status, and disease duration, confirming that the MSWS-12 items functioned equivalently across these demographic groups.

Method

Study design and participants

This cross-sectional psychometric validation study investigated the measurement properties of the Chinese version of the MSWS-12 in individuals diagnosed with MS. This study was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki. Ethical approval was granted by the Zhejiang Wanli University Student Mental Health Education and Counseling Center (Approval No. SC1032440880380), and written informed consent was obtained from all participants. All procedures were implemented in compliance with China’s Personal Information Protection Law (PIPL) and the psychometric evaluation standards established by the COSMIN initiative [33], which provides internationally recognized guidelines for evaluating the methodological quality of health-related measurement instruments (see Supplementary File 1).

Participants and procedure

A total of 292 participants (177 women and 115 men) were recruited through convenience sampling from four tertiary neurology clinics and two regional MS registries across China. This strategy was designed to ensure demographic and clinical representation across geographic regions, healthcare settings, and sociocultural contexts.

Participants were eligible if they had a clinically confirmed diagnosis of MS according to the McDonald criteria [34], were aged 18 years or older, had the ability to ambulate with minimal assistance as indicated by an Expanded Disability Status Scale (EDSS) score of ≤ 6.5 [35], and were fluent in Mandarin. Exclusion criteria included severe cognitive impairment, mobility restrictions not attributable to MS, a relapse within the preceding 30 days, or the use of medications that could affect motor functioning. Additionally, participants were excluded if they submitted surveys with incomplete responses to the MSWS-12 items.

All participants were informed of their right to withdraw at any stage without penalty. No financial or material incentives were provided. A priori power analysis for CFA based on a unidimensional model with 12 items and expected loadings ≥ 0.50 indicated that a sample of at least 250 participants was needed [36]. To ensure sufficient power and accommodate potential attrition, the final sample size was increased to 292.

Cross-Cultural adaptation

The translation and cultural adaptation of the MSWS-12 followed internationally endorsed guidelines [37]. Two bilingual experts in neurology and psychometrics independently translated the instrument into Mandarin, and discrepancies were resolved through consensus discussions. This preliminary version was subsequently back-translated into English by two independent translators who were blinded to the original instrument.

An expert panel (neurologists, rehabilitation specialists, psychometricians) reviewed translations for equivalence. Cultural appropriateness was a priority, and certain phrases (e.g., “balance while walking”) were revised to align with Mandarin linguistic norms while avoiding stigmatizing connotations in collectivist settings. To ensure comprehensibility and relevance, cognitive debriefing was conducted with 20 individuals with MS (10 urban, 10 rural; 40% with a high school education or lower). Participants engaged in think-aloud protocols and provided item clarity and relevance ratings using Likert scales. Minor modifications were implemented based on participant feedback to enhance clarity and cultural alignment while preserving the original conceptual intent.

Measures

Walking disability

Perceived walking disability was assessed using the Mandarin version of the MSWS-12 [9]. This instrument contains 12 items rated on a 5-point Likert scale (1 = “Not at all” to 5 = “Extremely”) and captures walking-related limitations such as gait speed, balance, and endurance. Raw scores (12–60) were linearly transformed to a 0–100 scale, with higher scores indicating greater walking disability.

Overall disability

Overall neurological impairment was evaluated using the EDSS [35], a clinician-administered tool that quantifies functional disability across multiple systems including pyramidal, cerebellar, brainstem, and sensory domains, as well as ambulation. Scores range from 0 (normal neurological function) to 10 (death due to MS), with 0.5-point increments.

Walking endurance

Walking endurance was assessed using the 6MWT [18], a widely used and validated performance-based measure of functional exercise capacity. Participants were instructed to walk at their own pace along a flat, straight course for six minutes, and the total distance covered during this time was recorded in meters. Greater walking distances reflected higher levels of endurance.

Walking speed

Walking speed was assessed using the T25FW [19]. Participants were instructed to walk a distance of 25 feet as quickly and safely as possible, and the time (in seconds) was recorded. Higher scores indicate slower walking speed and greater ambulatory limitation.

Fatigue

Fatigue was evaluated using the Modified Fatigue Impact Scale (MFIS) [38], a 21-item self-report measure encompassing physical, cognitive, and psychosocial dimensions. Items are rated on a 5-point Likert scale (0 = “Never” to 4 = “Almost always”), with higher total scores indicating greater fatigue-related impairment.

Depressive symptoms

Depressive symptoms were measured using the Patient Health Questionnaire-9 (PHQ-9) [39], a 9-item scale assessing the frequency of depressive symptoms over the preceding two weeks. Each item is scored from 0 (“Not at all”) to 3 (“Nearly every day”), yielding a total score ranging from 0 to 27. Higher scores indicate more severe depressive symptomatology.

Statistical analysis

Descriptive and group comparisons

All statistical analyses were conducted using R (version 4.3.2) in RStudio (version 2023.09.1, Build 494) and AMOS. Descriptive statistics (means, standard deviations, frequencies) were computed for all variables. Normality assumptions were assessed using skewness (< |2.0|) and kurtosis (< |7.0|) [40]. Group comparisons were conducted using chi-square tests for categorical variables and independent t-tests for continuous variables. Cohen’s d was calculated for significant effects, using standardized mean differences appropriate for group comparisons [41].

Reliability analysis

Internal consistency was assessed using Cronbach’s alpha and McDonald’s omega, with coefficients ≥ 0.70 indicating acceptable reliability [42]. Composite reliability (CR) and average variance extracted (AVE) were also calculated to assess convergent validity, using thresholds of CR > 0.70 and AVE > 0.50.

[43]. Test-retest reliability was examined in a subset of 70 participants who completed the MSWS-12 twice over a two-week interval. Intraclass correlation coefficients (ICCs) were calculated using a two-way mixed-effects model, and ICC values ≥ 0.70 indicated adequate temporal stability [44].

Validity analysis

Pearson correlation coefficients were used to assess convergent and discriminant validity. Criterion-related validity was examined using performance-based mobility benchmarks, specifically the T25FW and 6MWT. Strong positive correlations with T25FW and strong negative correlations with 6MWT were hypothesized, reflecting the MSWS-12’s sensitivity to walking capacity. Convergent validity was evaluated through associations with the EDSS and MFIS, with moderate-to-strong correlations expected due to their conceptual alignment with disability and fatigue. Discriminant validity was tested against the PHQ-9, where weak correlations were anticipated to confirm the scale’s specificity to mobility rather than mood. Correlation strength was interpreted according to Cohen’s guidelines (small: r =.10–0.29; moderate: r =.30–0.49; large: r ≥.50) [41].

A CFA was conducted using the Maximum Likelihood (ML) estimator to examine the latent structure of the MSWS-12, as the data approximated normality and displayed adequate category distribution. Model fit was evaluated using the Comparative Fit Index (CFI) and Tucker–Lewis Index (TLI) (≥ 0.95), Root Mean Square Error of Approximation (RMSEA ≤ 0.08), and Standardized Root Mean Square Residual (SRMR ≤ 0.06) [45]. To test factorial stability, multi-group CFA (MG-CFA) with ML estimation and robust corrections was performed across sex (male vs. female), residency (urban vs. rural), and disease duration (≤ 5 vs. >5 years). Invariance was examined sequentially through configural, metric, scalar, and strict models, each imposing increasingly restrictive equality constraints on parameters. Model comparisons were assessed using Δχ² (p >.05) as the primary criterion and changes in fit indices as supplementary evidence; invariance was accepted when ΔCFI ≤ 0.010, ΔTLI ≤ 0.010, and ΔRMSEA ≤ 0.015 [46].

IRT analysis

For the IRT analysis, the GRM was selected due to its suitability for ordinal data, such as the 5-point Likert scale responses of the MSWS-12 [47, 48]. The GRM estimated item discrimination parameters (α) and threshold parameters (β), which reflect how effectively each item differentiates between levels of walking disability. Item-level model fit was assessed using the standardized chi-square statistic (S-χ²), where non-significant values (P >.05) indicate adequate fit of the model to the observed item response patterns [49]. Given the sensitivity of S-χ² to large sample sizes, additional model fit was evaluated with the RMSEA, with values below 0.08 considered acceptable [50]. Local item dependence was examined using Yen’s Q3 statistic, with Q3 values below 0.20 indicating that no pairs of items exhibited excessive residual covariance, thereby supporting the assumption of unidimensionality [51].

DIF analysis

DIF was assessed to determine whether systematic response biases existed across demographic groups among respondents with equivalent latent trait levels (θ) [52]. Both uniform and nonuniform DIF patterns were examined through ordinal logistic regression modeling. Uniform DIF, operationalized as significant differences in threshold parameters (β), was established when a demographic group exhibited consistently elevated response probabilities across the entire trait continuum. Nonuniform DIF, quantified through discrimination parameter (α) variations, was confirmed when significant group-by-trait interactions were observed in the relationship between item responses and the underlying construct [53].

The latent trait was operationalized using total scale scores in all analytical models. Three categorical grouping variables were incorporated: (1) biological sex (male vs. female), (2) geographic residency (urban vs. rural), and (3) disease duration (< 5 years vs. ≥5 years since diagnosis). The modeling framework included the latent trait as a continuous predictor, group membership as a categorical moderator, and their interaction terms to concurrently assess both uniform and nonuniform DIF effects while adjusting for underlying trait levels. DIF detection was performed using likelihood ratio χ² tests (α = 0.05), with practical significance determined through changes in McFadden’s pseudo-R² values (ΔR² ≥ 0.02 considered clinically meaningful) [54].

Results

Descriptive statistics

Descriptive statistics and group comparisons are presented in Table 1. A total of 36 surveys with incomplete MSWS-12 responses were excluded from the analysis. The final dataset had no missing values. To ensure transparency, we compared key demographic and clinical variables (e.g., overall disability measured by the EDSS and disease duration) between excluded and retained surveys using chi-square tests and independent samples t-tests, and found no evidence of systematic bias. The mean age of participants was 41.72 years (SD = 5.23). A chi-square test of independence revealed that the sex distribution was significantly uneven, with 115 males (39.4%) and 177 females (60.6%), χ²(1, N = 292) = 13.16, p <.001. The distribution by residency was not significant, χ²(1, N = 292) = 1.11, p =.292, with 137 participants living in rural areas (46.9%) and 155 in urban settings (53.1%). Finally, the proportions differed significantly across disease duration categories, χ²(1, N = 292) = 6.63, p =.010, φ = 0.15, with 124 participants (42.5%) reporting a short duration (< 5 years) and 168 (57.5%) reporting a long duration (≥ 5 years). Results from the cognitive debriefing interviews indicated high item clarity and cultural appropriateness. All participants (n = 20) rated the items as understandable and relevant. Minor linguistic adjustments were made based on participant suggestions to enhance clarity without altering the conceptual meaning.

Table 1.

Descriptive statistics and between-group comparisons (N = 292)

Variable	M (SD) or n (%)	Test Value	p-value
Age (years)	41.72 (5.23)	t(290) = −1.10	0.269
Education (≥ High School)	245 (83.9%)	χ²(1) = 134.26	< 0.001
MSWS-12	31.40 (8.84)	t(290) = −1.95	0.051
EDSS Score	2.19 (1.45)	t(290) = 0.14	0.885
T25FW (sec)	6.03 (1.70)	t(290) = 0.22	0.826
6MWT (m)	532.17 (142.04)	t(290) = 3.36	< 0.001
PHQ-9	5.95 (3.66)	t(290) = −3.14	0.002
MFIS	35.64 (9.67)	t(290) = 0.35	0.731

M Mean, SD Standard Deviation, MSWS-12 Multiple Sclerosis Walking Scale, EDSS Expanded Disability Status Scale, T25FW (sec) Timed 25-Foot Walk, measured in seconds, 6MWT (m) Six-Minute Walk Test, measured in meters, PHQ-9 Patient Health Questionnaire–9, MFIS Modified Fatigue Impact Scale. All tests were two-tailed

Independent samples t-tests revealed no significant sex differences in age, walking disability (MSWS-12), fatigue (MFIS), disability status (EDSS), or walking speed (T25FW) (P >.05). By contrast, significant differences were observed for depressive symptoms and walking endurance. Females reported higher depressive symptoms (PHQ-9), t(290) = − 3.14, p =.002, d = − 0.38, 95% CI [–0.61, − 0.14], representing a small-to-moderate effect. Males demonstrated greater endurance on the 6MWT, t(290) = 3.36, p <.001, d = 0.41, 95% CI [0.17, 0.64], corresponding to a moderate effect.

Item-level descriptive statistics and reliability indices for the MSWS-12 items are reported in Table 2. The scale demonstrated excellent internal consistency (Cronbach’s α = 0.92; McDonald’s ω = 0.91). Item-level reliability was further supported by substantial squared multiple correlations (range = 0.47–0.62). Test–retest reliability, evaluated in a subsample, was strong, with an intraclass correlation coefficient (ICC) of 0.91 (95% CI [0.86, 0.94]), supporting the temporal stability of the measure. Convergent validity indicators reinforced construct validity, as the AVE = 0.502 and CR = 0.924, exceeded conventional psychometric benchmarks.

Table 2.

Item-Level descriptive statistics and reliability indices for the MSWS-12

Item	λ	Mean	SD	SEM	SDC	Skewness	Kurtosis	SMC
i1	0.739	2.64	1.05	0.32	0.88	0.21	−0.67	0.522
i2	0.720	2.61	0.98	0.30	0.83	0.33	−0.40	0.502
i3	0.642	2.62	0.92	0.29	0.80	0.29	−0.37	0.423
i4	0.707	2.70	1.03	0.32	0.87	0.28	−0.53	0.491
i5	0.721	2.63	0.98	0.30	0.82	0.36	−0.27	0.502
i6	0.710	2.62	1.05	0.31	0.86	0.21	−0.67	0.494
i7	0.700	2.64	0.97	0.30	0.82	0.30	−0.37	0.466
i8	0.736	2.59	0.96	0.30	0.83	0.29	−0.41	0.510
i9	0.692	2.52	1.00	0.30	0.84	0.35	−0.33	0.477
i10	0.717	2.57	1.06	0.32	0.88	0.32	−0.40	0.487
i11	0.659	2.62	1.02	0.30	0.82	0.27	−0.52	0.424
i12	0.747	2.65	0.98	0.30	0.83	0.16	−0.64	0.533

λ Factor Loading, M Mean, SD Standard Deviation, SEM Standard Error of Measurement, SDC Smallest Detectable Change, SMC Square multiple correlation All values align with COSMIN thresholds (λ > 0.50; SMC > 0.30)

Bivariate correlations with 95% confidence intervals are presented in Table 3. As hypothesized, the MSWS-12 demonstrated a large positive correlation with the T25FW (r =.64, p <.001, 95% CI [0.57, 0.70]) and a large negative correlation with the 6MWT (r = –.61, P <.001, 95% CI [–0.67, –0.53]), supporting its criterion-related validity against performance-based mobility measures. Convergent validity was further supported by a moderate-to-large correlation with the EDSS (r =.47, P <.001, 95% CI [0.37, 0.55]) and a moderate correlation with the MFIS (r =.45, P <.001, 95% CI [0.36, 0.54]). In contrast, a weak correlation was observed with the PHQ-9 (r =.28, P <.001, 95% CI [0.17, 0.38]), consistent with the expectation of discriminant validity, confirming that the MSWS-12 primarily captures mobility rather than mood-related constructs.

Table 3.

Bivariate correlations among study variables with 95% confidence intervals

	Pearson Correlation	Sig. (2-tailed)	95% Confidence Intervals (2-tailed)a
			Lower	Upper
Age - MSWS12	.17	.003	.06	.28
Age - T25FW	.14	.016	.03	.25
Age - 6MWT	-.21	.000	-.31	-.09
Age - EDSS	.20	.001	.09	.31
Age - MFIS	.24	<.001	.13	.35
Age - PHQ9	-.05	.412	-.16	.07
MSWS12 - T25FW	.64	<.001	.57	.70
MSWS12 - 6MWT	-.61	<.001	-.67	-.53
MSWS12 - EDSS	.47	<.001	.37	.55
MSWS12 - MFIS	.45	<.001	.36	.54
MSWS12 - PHQ9	.28	<.001	.17	.38
T25FW - 6MWT	-.60	<.001	-.67	-.52
T25FW - EDSS	.45	<.001	.35	.54
T25FW - MFIS	.52	<.001	.43	.60
T25FW - PHQ9	.27	<.001	.16	.37
6MWT - EDSS	-.60	<.001	-.67	-.52
6MWT - MFIS	-.50	<.001	-.58	-.41
6MWT - PHQ9	-.19	.001	-.30	-.08
EDSS - MFIS	.52	.000	.43	.60
EDSS - PHQ9	.14	.017	.03	.25
MFIS - PHQ9	.25	<.001	.14	.36

Pearson correlation coefficients (two-tailed), corresponding p-values, and 95% confidence intervals are reported

Confidence intervals were estimated using Fisher’s r-to-z transformation with bias adjustment

MSWS12 12-item Multiple Sclerosis Walking Scale, T25FW Timed 25-Foot Walk, 6MWT Six- Minute Walk Test;

EDSS Expanded Disability Status Scale, PHQ-9 Patient Health Questionnaire-9

^aEstimation is based on Fisher's r-to-z transformation with bias adjustment. All correlations are significant at p < .05 unless otherwise noted

Structural measurement

Prior to conducting confirmatory factor analysis, the data were examined for univariate and multivariate normality. Skewness values ranged from 0.16 to 0.36, and kurtosis values ranged from − 0.67 to − 0.27 across the 12 items, all of which fall within the acceptable range of ± 2.00, indicating that the distribution of item scores approximated univariate normality. The multivariate kurtosis value was 4.22, with a critical ratio (C.R.) of 1.97, remaining well below the recommended cut-off of 5.00. These findings collectively support the assumption of multivariate normality.

The confirmatory factor analysis supported the unidimensional structure of the MSWS-12. Model fit indices indicated an excellent fit to the data (χ²(54) = 76.64, p =.023, CMIN/DF = 1.42, CFI = 0.987, TLI = 0.984, IFI = 0.987, SRMR = 0.032, RMSEA = 0.038, 90% CI [0.015, 0.056], PCLOSE = 0.848). The factor loadings ranged from 0.642 to 0.74, with the latent factor accounting for 54% of the variance. To further evaluate the assumption of local item independence, Yen’s Q3 residual correlations were computed using both the Graded Response Model (GRM) and the Rasch Partial Credit Model (PCM). For the GRM, Q3 residuals ranged from − 0.173 to 0.008 (M = − 0.082), whereas for the PCM, residuals ranged from − 0.186 to 0.026 (M = − 0.085). All values were well below the critical threshold of 0.20, indicating the absence of substantive local item dependence. Although both models met the independence criterion, the GRM demonstrated marginally superior local fit, suggesting a more precise representation of item–trait relationships. Additionally, all off-diagonal residual correlations remained below ± 0.20, further reinforcing the unidimensionality of the MSWS-12 and confirming that item responses primarily reflect the latent construct of walking disability rather than secondary dependencies (see Supplementary File 2).

The item parameter estimates obtained from the GRM are presented in Table 4. The GRM demonstrated consistently high discrimination parameters across all 12 items (α = 2.01–2.95), indicating that each item effectively differentiated respondents along the latent continuum of walking disability. The threshold parameters (β₁–β₄) increased monotonically across response options, confirming the ordered nature of the categories and supporting the conceptual coherence of the scale’s ordinal structure. Item fit statistics (Infit and Outfit mean-square values ranging from 0.866 to 0.943) fell well within the recommended range of 0.70–1.30, indicating excellent item-level fit. Additionally, all standardized chi-square (S-χ²) tests were nonsignificant (p >.05) after Bonferroni correction, and RMSEA values consistently remained below 0.04, suggesting minimal residual misfit. These findings collectively confirm the adequacy of the GRM in representing the hierarchical difficulty and discrimination structure of the MSWS-12, providing robust evidence for its construct validity and parameter stability across the measurement continuum.

Table 4.

Item parameter estimates and model fit indices for the GRM (Items i1–i12)

Item	α	β₁	β₂	β₃	β₄	Infit	Outfit	S-χ²	df	RMSEA	p-value
i1	2.56	−1.08	0.09	1.04	2.47	0.903	0.878	37.80	36	0.013	0.388
i2	2.44	−1.19	0.14	1.21	2.51	0.902	0.888	28	34	0.000	0.756
i3	2.37	−1.36	0.09	1.18	2.73	0.897	0.901	48.60	34	0.038	0.050
i4	2.71	−1.15	0.08	1.01	2.34	0.908	0.943	39.90	34	0.025	0.223
i5	2.44	−1.21	0.09	1.25	2.53	0.887	0.888	45.50	34	0.034	0.090
i6	2.35	−1.04	0.19	1.18	2.55	0.886	0.880	40	37	0.017	0.337
i7	2.26	−1.27	0.08	1.26	2.64	0.919	0.936	42	35	0.026	0.193
i8	2.46	−1.16	0.09	1.19	2.74	0.887	0.881	25.20	33	0.000	0.832
i9	2.17	−1.13	0.17	1.29	2.66	0.910	0.909	48.70	37	0.033	0.095
i10	2.01	−1.13	0.11	1.27	2.40	0.902	0.888	49.20	41	0.026	0.177
i11	2.19	−1.25	0.09	1.26	2.79	0.922	0.921	30	35	0.000	0.706
i12	2.95	−1.14	0.10	1.11	2.62	0.884	0.866	33.20	32	0.011	0.409

α = item discrimination parameter; β₁–β₄ = category thresholds; Infit/Outfit = item fit mean squares; S-χ² = item-level chi-square fit statistic; RMSEA = root mean square error of approximation; p-value corresponds to the item-level fit test. RMSEA values of 0 indicate χ² ≤ df (excellent fit)

Table 5 summarizes the results of measurement invariance testing. Full invariance was established across sex, with the configural model demonstrating excellent fit (χ²(108) = 152.07, CFI = 0.974, RMSEA = 0.038) and only minimal changes across subsequent steps (ΔCFI ≤ 0.004, ΔRMSEA ≤ 0.005). Invariance was also confirmed across disease duration groups, where the configural model showed good fit (χ²(108) = 147.80, CFI = 0.977, RMSEA = 0.036) and later models remained stable. For residency, the configural model exhibited excellent fit (χ²(108) = 149.86, CFI = 0.976, RMSEA = 0.037); although a slight decrease in CFI (ΔCFI = − 0.002) was observed at the metric level, subsequent steps improved, supporting overall factorial invariance across all subgroups.

Table 5.

Measurement invariance testing across sex, residency, and disease duration

Group	Step	χ²	df	Δχ²	P	CFI	ΔCFI	RMSEA	ΔRMSEA
Sex	Configural	152.07	108	—	—	0.974	—	0.038	—
	Metric	156.15	119	4.08	0.967	0.978	+ 0.004	0.033	−0.005
	Scalar	172.73	131	16.57	0.166	0.975	−0.003	0.033	0.000
	Structural covariances	175.29	132	2.56	0.109	0.974	−0.001	0.034	+ 0.001
	Strict (residuals)	190.74	144	15.45	0.218	0.972	−0.002	0.033	−0.001
Duration	Configural	147.80	108	—	—	0.977	—	0.036	—
	Metric	166.09	119	18.29	0.075	0.973	−0.004	0.037	+ 0.001
	Scalar	174.06	131	7.97	0.787	0.975	+ 0.002	0.034	−0.003
	Structural covariances	174.99	132	0.93	0.334	0.975	0.000	0.034	0.000
	Strict (residuals)	185.03	144	10.04	0.612	0.976	+ 0.001	0.031	−0.003
Residency	Configural	149.86	108	—	—	0.976	—	0.037	—
	Metric	163.51	119	13.64	0.253	0.974	−0.002	0.036	−0.001
	Scalar	176.59	131	13.08	0.363	0.974	0.000	0.035	−0.001
	Structural covariances	176.78	132	0.19	0.659	0.974	0.000	0.034	−0.001
	Strict (residuals)	181.35	144	4.57	0.971	0.978	+ 0.004	0.030	−0.004

χ² = Chi-square value; df = degrees of freedom; CFI = Comparative Fit Index; RMSEA = Root Mean Square Error of Approximation. Δχ², ΔCFI, and ΔRMSEA represent changes in chi-square, CFI, and RMSEA respectively, from the previous step. Configural invariance tests whether the same factor structure holds across groups. Metric invariance constrains factor loadings to be equal, scalar invariance adds equality constraints on intercepts, structural covariances constrain variances and covariances, and strict invariance further constrains residual variances. Changes in CFI ≤ 0.010 and RMSEA ≤ 0.015 are generally considered evidence of invariance

The DIF was assessed across sex, residency status, and disease duration using likelihood ratio tests. Statistical significance was set at p <.05, with practical significance defined as ΔR² >0.02. A Bonferroni correction was applied (α = 0.0042) to account for multiple comparisons across 12 items. No items demonstrated DIF for sex comparisons (all p >.05). For residency status, two items showed statistically significant DIF: item 5 exhibited uniform DIF (p =.046, ΔR² = 0.008) and item 2 showed non-uniform DIF (p =.024, ΔR² = 0.009). Disease duration revealed two items with non-uniform DIF: item 3 (p =.001, ΔR² = 0.017) and item 10 (p =.045, ΔR² = 0.005). After Bonferroni correction, only item 3 remained statistically significant. Importantly, no items exceeded the threshold for practical significance (ΔR² >0.02), with the largest effect size being 0.017. These findings indicate that while some statistical differences exist in item functioning across demographic subgroups, the practical impact is minimal, supporting the MSWS-12 fairness and equivalence across the examined populations (see Supplementary 3).

Discussion

This study provides robust psychometric evidence supporting the validity and cross-cultural applicability of the Chinese MSWS-12. Grounded in both CTT and IRT, the findings affirm that the scale is appropriate for assessing self-perceived walking disability among Mandarin-speaking individuals with MS—a population that previously lacked a linguistically and culturally validated measure. These results converge with validation studies in Portuguese [11], Persian [12], Italian [13], Korean [14], German [15], Japanese [16], and Croatian [17], thereby reinforcing the cross-cultural validity and extending the generalizability of the MSWS-12.

Reliability indices strongly supported the translated scale. Both Cronbach’s α and McDonald’s ω exceeded recommended thresholds, indicating high internal consistency, while test–retest reliability, evaluated through ICC, demonstrated temporal stability. These properties are essential for outcome measures used in longitudinal clinical trials, where consistency across repeated assessments is critical. Construct validity was substantiated through strong correlations with walking speed and walking endurance, confirming convergent validity. Conversely, the modest association with depression indicated discriminant validity, underscoring the scale’s specificity in capturing mobility rather than mood-related constructs.

Unidimensionality was supported by multiple lines of evidence. The absence of local item dependence confirmed that items did not share residual variance, while CFA demonstrated excellent fit indices and strong factor loadings, reinforcing that the MSWS-12 reflects a single latent factor of mobility limitation.

The GRM further provided a detailed evaluation of item-level performance. All items displayed high discrimination (α), effectively differentiating respondents across the continuum of walking disability. Ordered response thresholds validated the appropriateness of the Likert-type scaling, ensuring interpretability across functional levels. This sensitivity is particularly important for MS, a condition characterized by heterogeneity in onset, progression, and severity [55]. The scale’s robust cross-cultural performance additionally underscores its utility in contexts where cultural norms and health beliefs may shape self-reported disability.

A major contribution of this study lies in its rigorous evaluation of DIF across subgroups (sex, residency, disease duration). Using a dual-method approach—ordinal logistic regression and IRT-based likelihood ratio testing—no systematic item bias was detected. The absence of sex-related DIF strengthens confidence that observed score differences reflect true experiential disparities rather than measurement artifacts. Measurement invariance analyses confirmed configural, metric, scalar, and strict invariance, across gender, residency, and disease duration were supported, indicating that factor structure, loadings, and intercepts were stable across groups. This ensures that comparisons between these subgroups are meaningful and unbiased.

Practical implications

The findings provide compelling psychometric and clinical evidence supporting the MSWS-12 as a reliable and valid instrument for assessing mobility in individuals with MS. The clinical relevance of the MSWS-12 is further demonstrated through its significant associations with key functional indicators. Meaningful correlations with walking speed and endurance, as well as significant correlations with overall disability and fatigue, provide further support for its sensitivity to clinically relevant mobility-related outcomes. In contrast, the weaker correlation with depressive symptoms confirms discriminant validity, indicating that the MSWS-12 primarily measures functional mobility rather than mood-related constructs. Collectively, these findings reinforce the scale’s applicability as a sensitive and specific instrument for evaluating functional status and mobility outcomes in both clinical and research settings.

Female participants reported higher PHQ-9 scores and exhibited lower endurance on the 6MWT, findings that may reflect the complex interplay of biological, psychological, and sociocultural determinants. Sex- and gender-specific mechanisms—including neuroimmune pathways, hormonal modulation, and psychosocial stressors—have been consistently implicated in shaping disease trajectories among women with MS [56, 57]. Furthermore, variations in immune profiles and pro-inflammatory processes may exacerbate both susceptibility and disease progression [58, 59]. These findings highlight the necessity of implementing sex- and gender-sensitive assessment and intervention strategies in the clinical management of MS. Given that women demonstrated elevated depressive symptomatology alongside diminished physical endurance, clinicians should adopt an integrative, multidimensional approach that concurrently addresses psychological distress and functional limitation. The findings underscore the importance of gender-sensitive rehabilitation frameworks that integrate physiological and psychosocial considerations [60, 61]. Tailored interventions can improve engagement, enhance adherence, and optimize outcomes by addressing the multidimensional contributors to mobility impairment in MS [62].

Limitations

While this study adhered to the COSMIN framework and demonstrated acceptable psychometric properties, several limitations should be acknowledged. First, the use of a convenience sampling strategy may constrain the generalizability of the findings, particularly to socioeconomically disadvantaged or rural populations who may differ in healthcare access, illness perceptions, or symptom reporting. Second, the cross-sectional design precludes any evaluation of the instrument’s responsiveness to change over time. Given that responsiveness is a critical measurement property for outcome tools used in longitudinal research and clinical trials, future investigations should adopt prospective designs to assess sensitivity to change. Third, although the sample was clinically and demographically diverse, individuals with severe cognitive impairment or advanced ambulatory dysfunction were excluded due to the self-report nature of the instrument. This exclusion may restrict the applicability of findings to the most functionally impaired subgroups of the MS population. Additionally, the sample primarily included younger individuals with low to moderate levels of disability, which may further limit generalizability to older adults or those with more severe impairment. Finally, floor and ceiling effects were not formally assessed. This omission constrains understanding of the scale’s sensitivity, particularly among individuals at the extremes of walking ability, and should be addressed in future validation efforts.

Conclusion

Taken together, these findings affirm that the Chinese-translated MSWS-12 is a psychometrically sound, culturally attuned patient-reported outcome measure. Its ability to capture functional limitations with both clinical precision and contextual sensitivity addresses a critical need in the assessment of mobility impairment among Mandarin-speaking individuals with MS. Furthermore, its demonstrated invariance and absence of DIF strengthen its utility in large-scale epidemiological studies and multinational clinical trials, where comparability and fairness of measurement are paramount.

Authors' contributions

Conceptualization and methodology: H.Z., N.N.Data curation and formal analysis: H.Z.Writing – original draft: N.N.Writing – review and editing: All authorsAll authors have reviewed and approved the final version of the manuscript for submission.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability

The datasets generated and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki. Ethical approval for this study was obtained from the Student Mental Health Education and Counseling Center at Zhejiang Wanli University (Approval No. SC1032440880380). All procedures were conducted in accordance with the ethical standards outlined in China’s Personal Information Protection Law (PIPL).

Written informed consent was obtained from all participants prior to data collection.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.