Utilizing the ISway to Identify and Compare Balance Domain Deficits in People with Multiple Sclerosis

Abstract

Objective.

To develop a multiple sclerosis (MS)-specific model of balance and examine differences between 1) MS and neurotypical controls and 2) people with MS (PwMS) with (MS-F) and without a fall history (MS-NF).

Design and Setting.

A cross-sectional study was conducted at the Gait and Balance Laboratory at the University of Kansas Medical Center. Balance was measured from the instrumented sway system (ISway) assessment.

Participants.

118 people with relapsing-remitting MS (MS-F = 39; MS-NF = 79) and 46 age-matched neurotypical controls.

Intervention.

Not applicable.

Outcome Measures.

22 sway measures obtained from the ISway were entered into an exploratory factor analysis (EFA) to identify underlying balance domains. The model-derived balance domains were compared between 1) PwMS and age-matched, neurotypical controls and 2) MS-F and MS-NF.

Results.

3 distinct balance domains were identified: 1) sway amplitude and velocity, 2) sway frequency and jerk mediolateral (ML), and 3) sway frequency and jerk anteroposterior (AP), explaining 81.66% of balance variance. PwMS exhibited worse performance (i.e., greater amplitude and velocity of sway) in the sway velocity and amplitude domain compared to age-matched neurotypical controls (p = .003). MS-F also exhibited worse performance in the sway velocity and amplitude domain compared to MS-NF (p = 0.046). The AP and ML sway frequency and jerk domains were not different between PwMS and neurotypical controls nor between MS-F and MS-NF.

Conclusion.

This study identified a 3-factor, MS-specific balance model, demonstrating that PwMS, particularly those with a fall history, exhibit disproportionate impairments in sway amplitude and velocity. Identifying postural stability outcomes and domains that are altered in PwMS and clinically relevant (e.g., related to falls) would help isolate potential treatment targets.

Multiple sclerosis (MS) is a neurodegenerative disease resulting in axon demyelination throughout the central nervous system ¹. Balance impairments are a common symptom of MS ^2,3 and are related to falls ⁴. Research shows that 56% of people with MS (PwMS) will experience a fall in any three-month period, while 37% of PwMS are classified as frequent fallers ⁵. The physical, psychological, and social consequences of falls for PwMS are severe, diminishing daily function and quality of life ^4,6,7. Therefore, identifying aspects of balance that may distinguish PwMS most at risk for falling is warranted.

Quantitative posturography via body-worn inertial measurement units has become a convenient, mobile, and low-cost alternative to force platform-based postural sway assessments for clinical populations such as PwMS. For example, the instrumented sway system (ISway) is a valid and reliable clinical balance assessment that uses body-worn accelerometers to measure postural sway ⁸. However, these approaches yield a vast range of outcome measures. In a recent meta-analysis of 43 studies investigating balance in PwMS, 105 postural stability-related outcome measures were reported ⁹. The heterogeneity and redundancy across outcomes impede clinicians’ ability to interpret data and provide evidence-based care. Reduction techniques such as principal component analysis (PCA) or exploratory function analysis (EFA) offer data-driven ways to group many similar outcomes into functionally distinct domains ^10–17. Although most of this work has focused on neurotypical older adults, work has begun to identify domains in neurological populations ^10,11,17,18. For example, studies in Parkinson’s disease (PD) established distinct mobility domains during instrumented assessments of standing and walking ^10,11. Regarding balance, Morris et al. identified 4 independent domains: 1) sway area/jerkiness, 2) sway velocity, 3) sway frequency anteroposterior (AP), and 4) sway frequency mediolateral (ML) ¹¹. In PwMS, we recently conducted a similar approach in gait, establishing a 6-domain model in PwMS¹⁸. To date, no work has applied such methods to balance in PwMS.

Identifying balance outcomes and domains that are both altered in PwMS and clinically relevant (e.g., related to falls) would help isolate potential treatment targets. Recently, Swanson et al. performed a comprehensive comparison of mobility measures between PwMS and middle-and old-aged adults ¹⁹. Although several postural sway measures were significantly impaired in PwMS compared to neurotypical controls, they had a variable ability to discriminate between PwMS and older adults ¹⁹. While this work provides important insights into balance deficits in MS, the sample size of 30 hindered analyses testing the clinical relevance of these outcomes, such as their ability to discriminate PwMS with and without a fall history. Further, factor analyses to identify domains of balance that are impacted by MS or fall status (thus reducing redundancy and heterogeneity of outcome measures) have not been carried out.

Therefore, the purpose of this study was to conduct a factor analysis to establish specific domains of balance in PwMS and determine which of these outcomes are 1) altered in PwMS and compared to neurotypical controls and 2) altered across PwMS with and without a fall history. Identifying clinically-relevant balance domains impacted in PwMS may help limit the redundancy when reporting balance outcomes and inform more targeted balance rehabilitation to help reduce fall risk in PwMS.

Methods

Participants

118 people with relapsing-remitting MS and 46 neurotypical controls were recruited via the MS clinic at the University of Kansas Medical Center (Table 1). MS participants were excluded based on the following criteria: 1) a Kurtzke Expanded Disability Status Scale (EDSS) Score >5.5 (could not walk 100 meters without assistance), 2) an MS symptom exacerbation in the previous sixty days that required treatment, 3) an inability to give consent, 4) any musculoskeletal or orthopedic impairments that would affect balance or mobility and, 4) any neurological disorder other than MS. Neurotypical participants were matched for gender, age, and body mass index with the overall MS cohort. They also had to be free of any neurological or musculoskeletal deficits that could affect balance or gait. PwMS were classified into two groups; no fall history (MS-NF) and a fall history (one or more falls; MS-F) based on their self-reported fall history in the previous six months. Falls were operationally defined as “An unexpected event in which the participants come to rest on the ground, floor, or lower-level” ²⁰. Informed consent was acquired from all participants before study participation. The University of Kansas Medical Center ethics board approved the study protocol, which was conducted in accordance with the Declaration of Helsinki.

Table 1.

Participant Characteristics.

	Control (n = 46)	MS (n = 118)		MS vs. CON (p-value)	MS-F vs. MS-NF (p-value)
		MS-NF (n = 79)	MS-F (n = 39)
Age	43.06 (10.23)	44.60 (8.61)	47.31 (9.84)	0.23	0.13
Sex (Female/Male)	38/8	65/14	27/12	0.51	0.11
Height (m)	1.66 (0.11)	1.66 (0.10)	1.68 (0.10)	0.47	0.33
Weight (kg)	72.56 (15.80)	80.43 (19.16)	77.69 (23.78)	0.02	0.50
Body Mass Index (kg/m2)	26.25 (4.61)	29.07 (6.79)	27.11 (6.77)	0.02	0.14
EDSS (min - max)	- -	2.00 (0.00 – 4.00)	2.50 (0.00 – 5.50)	- -	<0.001
Physical Functioning (SF-36)	96.15 (5.28)	79.68 (21.60)	60.13 (25.51)	<0.001	<0.001
Energy and Fatigue (SF-36)	70.42 (15.19)	50.96 (22.71)	37.69 (19,09)	<0.001	0.002
Berg Balance Scale	- -	54.29 (2.68)	51.13 (6.74)		0.01
Falls Efficacy Scale	18.08 (2.12)	23.00 (6.72)	31.18 (9.24)	<0.001	<0.001
Stroop Word Test (CW Score)	26.19 (4.71)	24.05 (4.54)	21.50 (5.40)	<0.001	0.01

Reported as mean and standard deviation; EDSS is reported a median (min - max). Legend: MS- multiple sclerosis; MS-NF- People with MS with no fall history; MS-F- People with MS with a fall history; EDSS- Expanded Disability Status Scale. Fall status was determined if one or more fall was reported in the previous six months. Berg Balance data was collected in the MS group only.

Protocol and Tasks

First, demographic, anthropometric, and clinical measurements were recorded. Disease severity was measured via the EDSS ^21,22, the Berg Balance Scale (BBS) ²³ assessed functional mobility, the 36-Item Short-Form Health Survey (SF-36) estimated fatigue and physical health ²⁴, and the Falls Efficacy Scale International (FES-I) measured assessments of fall-risk ²⁵. Finally, participants completed the Stroop Color and Word Test to obtain measures of executive function ^26–28.

Quantitative balance measures were obtained via Mobility Lab’s ISway protocol^a and Opal inertial sensors^b (50 Hz; APDM Inc., Portland, OR.). Additional protocol details, as well as the validity and sensitivity of this approach, are described by Mancini & colleagues, 2012 ⁸. Outcomes were derived from an inertial sensor placed on the fifth lumbar vertebrae ⁸. Participants completed three 30-second ISway trials, and sway measures were averaged across the three trials. Participants stood quietly on a firm surface wearing comfortable athletic shoes while looking at a fixed image placed at head height on the wall, with arms comfortably over their chest (Supplementary Material 1). Foot placement was standardized using the APDM base of support plate to ensure consistent stance width.

The ISway algorithm automatically generates sway measures in the time and frequency domains using the tri-axial accelerometer from the lumbar sensor. Further details concerning outcome measurements are provided in Mancini et al., 2012 ⁸. Sway outcomes from the ISway assessment included in the EFA were informed by balance measures commonly reported in MS and previous factor analyses conducted in PD ^9,10,19,29. The definitions of the sway measures included in the EFA (n = 22) are provided in Supplementary Material 2.

Statistical analysis

EFA was used to examine the underlying structure of the 22 sway measures obtained from the ISway protocol within the MS cohort. Mean balance data from neurotypical participants were used for comparative and reference purposes. Bartlett’s test of sphericity ³⁰ and the Kaiser-Meyer-Olkin (KMO) measure of sampling accuracy (MSA) ³¹ were used to ensure that the correlation matrix was factorable and suitable for factor analysis. The KMO statistic was required to be above a minimum of 0.50 ^31,32. Univariate and multivariate distribution of items was determined through skewness and kurtosis values and Mardia’s multivariate test of normality ^33,34. Since many of the balance items were skewed, an iterated principal axis factoring (PAF) extraction method was estimated due to its tolerance of nonnormality ^35–37. Parallel analysis ³⁸ and the visual scree test ³⁹ (Supplemental Material 3) were used to determine the appropriate number of factors to retain. Since the underlying factors may not necessarily be orthogonal, the EFA was examined using promax rotation. Factor loadings with a minimum loading of 0.50 were considered practically significant ³²

Composite domain scores obtained from the EFA were compared between 1) MS and neurotypical controls and 2) MS-NF and MS-F. First, sway measures were log-transformed to reduce skewness. Then, scores were computed using an approach described by Morris et al. that involved summing z-scores for each sway measure for each domain and dividing by the number of parameters within that domain ¹¹. Independent t-tests and Mann-Whitney U tests compared the domain scores between MS and controls and MS-NF and MS-F. Shapiro-Wilks tests and histograms were inspected to assess normal distribution.

Next, exploratory, across-group analyses were run to compare individual metrics within each domain. Specifically, sway measures were compared between 1) MS and controls and 2) MS-NF and MS-F using independent sample t-tests for approximately normally distributed outcomes and Mann-Whitney U tests for those violating the normality assumption. Normal distribution was assessed via visual inspections of histograms and Shapiro-Wilks tests. Independent sample effect sizes are reported as Cohen’s d and interpreted as small (d=0.20), medium (d-0.50), and large (d=0.80) ⁴⁰. Effect sizes for Mann-Whitney U are reported as r values computed as: $r=\frac{\left|z\right|}{\sqrt{n}}$and interpreted as small (r=0.10), moderate (r=0.30), and large (r=0.50) ⁴⁰.

Finally, to provide additional insight into whether the identified factors were associated with clinical assessments, we ran bivariate correlations between each domain score and the following clinical or clinically relevant outcomes: EDSS, FES-I, BBS, and Stroop Color-Word Test. Additional details on the statistical analysis are described in Supplemental Material 3.

Data from this cohort have been published on previously ^18,41–44. However, the current analysis represents a novel analysis, focusing specifically on postural data, while previous results focused on reactive balance and/or walking outcomes.

Results

Exploratory Factor Analysis

Univariate and multivariate skewness and kurtosis values were outside the normal range ³⁴ for approximately half of the balance items (Supplemental Material 3). Mardia’s multivariate normality test also indicated that the data were not normally distributed at the multivariate level ³³ (Supplemental Material 3). Given the violation of univariate and multivariate normality, the principal axis factoring (PAF) extraction method was deemed appropriate for the EFA of the data ^34,35,37,45. Bartlett’s test of sphericity ³⁰ indicated that the correlation matrix was not random χ²(210) = 5786.883, p < .001, and the KMO statistic was 0.813, well above the acceptable threshold 0f 0.50 ³¹.

Parallel analysis ³⁸ and the visual scree test ³⁹ indicated that 3 factors should be retained. Of the 22 variables entered into the EFA, frequency dispersion AP was the only selected variable that did not meet the minimally acceptable factor loading threshold of 0.50 ³². Therefore, this item was removed, and another iteration of the EFA was performed.

The 3 factors explained 85.44% of the variance in the data set. Figure 1 illustrates the results of the EFA, and the factors were labeled based on the variables loaded with each factor: sway amplitude and velocity (47.82%), sway frequency and jerk mediolateral (ML) (25.27%), and sway frequency and jerk anterior-posterior (AP) (12.36%). Furthermore, the covariance structure of the sway variables reported in the EFA is presented in Figure 2. The correlation matrix includes a heat map that color-codes the correlations’ strengths, with blue indicating a strong positive correlation (+1.0) and red indicating a strong negative correlation (−1.0). The three orthogonal factors identified by the EFA are evident in the matrix as deep blue and red squares.

Figure 1.

Exploratory factor analysis of 21 sway measures performed during the ISway assessment in 118 people with multiple sclerosis. Three independent domains of balance were identified: Sway Amplitude and Velocity AP, Sway Frequency and Jerk ML, and Sway Frequency and Jerk AP. The percentage of variance explained by all three factors collectively and independently are provided along with the loadings of each sway measure. Factor loadings were considered relevant at >0.50 and are bolded. Factor loadings are listed in order of importance. RMS- root mean square; AP- anteroposterior; ML- mediolateral; Hz- Hertz. Jerk (m2/s2) was normalized to the range of the sway trajectory’s excursion and duration). Frequency dispersion is a dimensionless measure of variability of the frequency content of the power spectral density.

Figure 2.

The covariance structure of the ISway measures that significantly loaded on to a factor following the EFA. Blue indicates a strong positive correlation (+1.0), and red indicates a strong negative correlation (−1.0). Higher color saturation reflects a stronger correlation. RMS- root mean square; AP- anteroposterior; ML- mediolateral. Frequency Dispersion (not shown) was the only selected variable that did not significantly load on to a factor (loading for Sway Frequency and Jerk AP = −0.18). Within-parameter correlations are shown in black.

MS versus Neurotypical Controls

Comparison of domain scores between PwMS and neurotypical controls showed that PwMS had significantly greater (i.e., worse) performance of the sway amplitude and velocity domain than controls (p = 0.004, r = 0.223; Figure 3A,B). However, no differences were observed in the sway frequency and jerk ML and AP domains between groups (ML: p = 0.260, d = 0.196; AP: p = 0.165, r = 0.108; Figure 3A,B).

Figure 3.

The differences in sway domains between PwMS and controls (A & C) and MS-F and MS-NF (B & D). Higher values reflect worse balance performance. Radar plots reflect mean sway values in the MS cohort that were normalized to control data, such that the central dashed line represents control data and deviations along the radial axes indicate the number of standard deviations the MS groups differ from controls. * Indicates a statistical difference (A) between MS and controls and (B) between MS-F and MS-NF at the p <0.05 level. Box and violin plots illustrate the normalized domain scores between the MS and control (C) and MS-NF and MS-F groups. * Reflects significance at p < 0.05 level, ** represents significance at p < 0.01 level.

Consistent with these results, exploratory analyses of individual sway variables demonstrated that several measures within the sway amplitude and velocity domain differed significantly between MS and controls, albeit more frequently in the AP direction (Table 2; Figure 3A). Specifically, in the AP direction, PwMS exhibited greater (i.e., worse) RMS sway (p < 0.001, r = 0.307), mean distance (p < 0.001, r = 0.317), range of acceleration (p < 0.001, r = 0.271), velocity (p = 0.02, r = 0.180), sway area (p = 0.001, r = 0.250), 95% ellipse sway area (p = 0.01, r = 0.218), and path length (p < 0.001, r = 0.262). In the ML direction, a greater path length (p = 0.02, r = 0.190) was the only significantly different variable between MS and controls within the sway velocity and amplitude domain. A reduction in the ML frequency dispersion (p = 0.04, Cohen’s d = 0.365) was the only difference between PwMS and controls within the sway frequency and jerk ML domain. These findings should be treated with caution, given the lack of correction for multiple comparisons. No sway measures within the sway frequency and jerk AP domain differed between PwMS and their neurotypical peers (Table 2; Figure 3A,B).

Table 2:

Mean and standard deviation (SD) of Sway parameters in Controls, MS (all), and MS Faller and non-fallers.

			Control vs. MS			MS-NF vs. MS-F
	Control	MS	P-value (Effect Size)	MS-NF	MS-F	P-value (Effect Size)
	Mean (SD)	Mean (SD)		Mean (SD)	Mean (SD)
Sway Amplitude and Velocity
RMS sway (AP) (m/s2)	0.066 (0.027)	0.900 (0.485)	<0.001* (0.307)	0.085 (0.045)	0.100 (0.054)	0.02* (0.213)
Mean distance (AP) (m/s2)	0.053 (0.023)	0.072 (0.038)	<0.001* (0.317)	0.068 (0.035)	0.081 (0.043)	0.02* (0.223)
Range of acceleration (AP) (m/s2)	0.338 (0.135)	0.456 (0.268)	<0.001* (0.271)	0.438 (0.272)	0.497 (0.261)	0.04* (0.195)
Mean velocity (AP) (m/s)	0.145 (0.078)	0.189 (0.128)	0.02* (0.180)	0.174 (0.110)	0.220 (0.158)	0.02* (0.211)
Total sway area (m2/s5)	0.004 (0.007)	0.006 (0.009)	0.001* (0.250)	0.006 (0.009)	0.007 (0.009)	0.06* (0.174)
95% ellipse sway area (m2/s4)	0.031 (0.046)	0.061 (0.101)	0.01* (0.218)	0.056 (0.96)	0.071 (0.114)	0.06* (0.177)
Path length (AP) (m/s2)	3.603 (1.307)	4.732 (2.457)	<0.001* (0.262)	4.528 (2.433)	5.170 (2.504)	0.12* (0.142)
Range of acceleration (ML) (m/s2)	0.138 (0.083)	0.175 (0.152)	0.19* (0.102)	0.167 (0.141)	0.194 (0.175)	0.14* (0.136)
Path length (ML) (m/s2)	2.436 (0.874)	3.111 (1.914)	0.02* (0.190)	2.966 (1.795)	3.424 (2.170)	0.34* (0.087)
RMS sway (ML) (m/s2)	0.021 (0.008)	0.029 (0.022)	0.07* (0.140)	0.028 (0.024)	0.031 (0.020)	0.15* (0.134)
Mean distance (ML) (m/s2)	0.016 (0.006)	0.022 (0.017)	0.06* (0.147)	0.022 (0.018)	0.024 (0.014)	0.15* (0.134)
Mean velocity (ML) (m/s)	0.038 (0.022)	0.050 (0.038)	0.13* (0.119)	0.050 (0.041)	0.052 (0.033)	0.30* (0.100)
Sway Frequency and Jerk ML
Centroidal frequency (ML) (Hz)	1.001 (0.220)	0.990 (0.226)	0.77 (0.050)	0.996 (0.203)	0.987 (0.269)	0.83 (0.043)
Mean frequency (ML) (Hz)	0.948 (0.250)	0.946 (0.300)	0.88* (0.011)	0.954 (0.284)	0.935 (0.338)	0.74 (0.065)
Normalized jerk (ML)	1.899 (0.412)	1.943 (0.512)	0.62* (0.039)	1.961 (0.491)	1.911 (0.567)	0.63 (0.096)
95% frequency (ML) (Hz)	2.322 (0.386)	2.255 (0.437)	0.36 (0.158)	2.285 (0.425)	2.204 (0.468)	0.34 (0.186)
Frequency dispersion (ML)	0.717 (0.069)	0.692 (0.067)	0.04 (0.365)	0.691 (0.056)	0.690 (0.086)	0.93 (0.021)
Sway Frequency and Jerk AP
Centroidal frequency (AP) (Hz)	0.569 (0.137)	0.539 (0.134)	0.25* (0.088)	0.534 (0.127)	0.552 (0.151)	0.54* (0.057)
95% frequency (AP) (Hz)	1.361 (0.422)	1.296 (0.412)	0.43* (0.062)	1.276 (0.391)	1.349 (0.460)	0.44* (0.071)
Mean frequency (AP) (Hz)	0.443 (0.109)	0.415 (0.113)	0.14* (0.116)	0.416 (0.102)	0.412 (0.136)	0.51* (0.061)
Normalized jerk (AP)	1.094 (0.214)	1.052 (0.244)	0.18* (0.105)	1.050 (0.222)	1.053 (0.291)	0.79* (0.025)

MS- multiple sclerosis; MS-NF- people with MS who have reported no falls in the last 6 months; MS-F- people with MS who have reported one or more falls in the previous 6 months; AP- anteroposterior; ML- mediolateral; Hz- Hertz; Jerk (m²/s²) was normalized to the range of the sway trajectory’s excursion and duration). Frequency dispersion is a dimensionless measure of the variability of the frequency content of the power spectral density.

^*Indicates Mann-Whitney U Test. Independent sample effect sizes are reported as Cohen’s d. Effect sizes for Mann-Whitney U are reported as r values computed as:$r=\left|z\right|/\sqrt{n}$. Higher values reflect worse balance performance.

Bold indicates significance at the p < 0.05 level.

3.3. MS-NF versus MS-F

Seventy-nine PwMS had reported no falls in the last six months, while 39 PwMS had reported at least one fall. A comparison of domain scores showed that PwMS and fall history exhibited worse performance in the sway amplitude and velocity domain than PwMS and no fall history (p = 0.046, r = 0.184; Figure 3C,D). The sway frequency and jerk ML, and AP domains showed no statistically significant differences between groups (ML: p = 0.340, d = 0.188; AP: p = 0.840, d = −0.04; Figure 3C,D).

Partially with these results, exploratory analyses of individual sway variables revealed that several sway measures within the sway amplitude and velocity domain differed between MS-NF and MS-F (Table 2; Figure 3D). Specifically, PwMS with a fall history demonstrated significantly worse sway in the AP direction, specifically greater RMS sway (p = 0.02, r = 0.213), mean distance (p = 0.02, r = 0.213), range of acceleration (p = 0.04, r = 0.195), and mean velocity (p = 0.02, r = 0.211). As noted above, these findings should be treated with caution, given the lack of correction for multiple comparisons. No sway measures from the sway frequency and Jerk ML and AP domains differed between MS-NF and MS-F (Table 2; Figure 3C,D).

Finally, correlational analyses indicated that the sway and amplitude velocity factor was statistically significantly correlated to several clinically relevant balance and cognitive outcomes (EDSS, FES-I, BBS, Stroop Color-Word test, ps<0.001 for all; see supplemental material 4. Sway frequency & jerk in the ML direction were modestly correlated with Stroop (p=0.02) and EDSS (p=0.03) scores. Sway frequency & jerk in the AP direction was not statistically significantly related to any of the four clinical outcomes.

Discussion

This study used EFA to identify independent domains of balance measured during the ISway assessment and compared the model-derived domains between PwMS and neurotypical controls and between PwMS with and without a history of falls. The EFA produced 3 distinct balance domains in MS: 1) sway amplitude and velocity, 2) ML sway frequency and jerk, and 3) AP sway frequency and jerk. Across these domains, the sway amplitude and velocity domain was significantly worse in PwMS than in neurotypical controls and in MS-F compared to MS-NF. Further, this domain was directly correlated to clinically relevant balance and cognitive outcomes.

Characterizing balance is complex, and often the selection of potent outcome measures can be challenging as no single measure of balance can wholly reflect performance. Further, the advent of inertial sensors to measure balance produces many outcomes that can be difficult to interpret individually. As such, methods to consolidate varying outcomes into domains through factor analysis can 1) provide a more thorough characterization of balance deficits in PwMS and 2) simplify and clarify results from inertial sensor outputs. The current study identified three orthogonal domains of postural sway in PwMS from 22 sway outcomes, accounting for 85.44% of the total sway variance. While applying dimension reduction techniques to balance performance in MS is novel, results are consistent with previous constructs in other populations ^10,11,46. Analyses in people with PD describe a four-domain balance model consisting of 1) sway area and jerk, 2) sway velocity, 3) ML sway frequency, and 4) AP sway frequency ¹¹, while an earlier study reported sway area and sway frequency as two independent domains ¹⁰. There is a fair agreement between our MS model of balance and those describing balance in PD. First, balance was partitioned into the distinct area/spatial and frequency domains, describing postural sway’s magnitude and smoothness. Second, like Morris and colleagues, independent domains were established for the directional components of sway frequency ¹¹. However, caution must be taken when comparing the MS balance model produced in this report to the models described above due to several factors. Firstly, the previous report was conducted in a PD population, while our study was centered on people with MS. Previous research investigating objective assessments of balance have noted pathology-specific balance impairments, so caution should be taken when translating findings across different pathologies. Secondly, it is imperative to consider the heterogeneity of the balance outcome measures chosen. For example, Morris and colleagues included 13 characteristics of static balance, while our study incorporated 22 measures. Despite these differences, establishing a small set of independent measures can help streamline balance assessment and potentially identify targets for rehabilitation.

We identified deficits in PwMS compared to controls primarily within the sway amplitude and velocity domain. Specifically, PwMS exhibited greater sway area and velocity than their neurotypical peers. Our findings corroborate previous work highlighting that PwMS exhibit larger sway than controls, reflected in a larger COP path length ^47–53, sway area and range ¹⁹, and a greater 95% confidence ellipse ^51,54,55. Similarly, the greater sway velocities observed within our MS cohort compared to neurotypical controls is also consistent with previous work ^50,51,56. However, while these crucial studies identified sway amplitude and velocity deficits independently, current results indicate that these two characteristics may track together, representing a single, underlying construct within the factor structure of the balance and that this factor is worse in PwMS than in controls. Further, we note that performance in this domain was directly related to several clinically relevant outcomes, including EDSS, BBS, FES-I, and Stroop Color-Word Test scores. This further underscores the potential relevance of this domain of balance clinically.

Balance deficits have been linked with fall risk in MS ^51,57–61. We extend this work by comparing balance domains (rather than individual measures) across PwMS with varied fall histories to establish their potential clinical significance. We found that the only domain that differed between MS-F and MS-NF was the sway amplitude and velocity domain. PwMS who had reported at least one fall in the previous 6 months displayed greater sway area and velocity than their peers without fall history. This result is consistent with previous work showing that MS fallers have increased sway velocity and a greater overall sway area than non-fallers ⁶¹, and that more significant sway can increase the odds of an accidental fall ⁵¹. It is possible that the greater sway amplitude and velocity in MS-F may be related to deficits in reactive balance, such that a poorer ability to sense the position of and quickly redirect the center of pressure could result in larger sway ^41,62. Regardless, we showed that these sway characteristics were common to balance performance in PwMS who had fallen in the previous six months.

An interesting observation is the specific directional deficits in balance. Although AP and ML sway area & velocity outcomes were included in the same domain, most of the sway differences between controls and MS and between MS-F and MS-NF occurred in the AP direction. This is somewhat surprising since ML sway is speculated to be more impaired in PwMS than older adults ^{50,51,56,63,64}, and MS fallers and non-fallers ^61,65. Moreover, ML sway amplitude has been shown to predict fall risk in MS ⁶⁶. A possible explanation for the lack of differences observed in the ML direction may be our relatively mild cohort (median EDSS 2.0), as increased sway has been associated with a higher total EDSS score ⁴⁷. Additionally, the difficulty of the balance assessment can also influence sway. For example, deficits in ML sway are more apparent as the balance challenge increases (e.g., with eyes closed) ^19,47,53,61. However, it is notable that the output of the EFA model showed that sway frequency and jerk in the ML direction accounted for almost double the variance than that in the AP direction. Further, our approach showed that the sway amplitude and velocity domain, encapsulating both AP and ML components measured using a validated and reliable clinical balance assessment, remained significantly different between MS and controls and MS-F and MS-NF.

Study Limitations

There are several limitations to this study. First, our study accounted for a retrospective fall history within the previous 6 months, and no additional information surrounding the cause of the falls was collected. Therefore, there is potential for recall bias to influence our data and analyses. Second, we dichotomized our MS cohort into those with and without a fall history based on a self-report of 1 or more falls in the previous 6 months. Analyses including single vs. multiple fallers can, in some cases, provide additional insight. However, splitting the sample of fallers into subgroups substantially reduces sample size per group and power. Further, fall history data was not collected in our neurotypical control group. Such data may have aided the interpretation of our findings and supported the association between specific balance parameters and risk for falls. Finally, the MS cohort reported in this study was restricted to the relapsing-remitting subtype of MS, and our cohort reported a relatively mild level of disability. Thus, our finding’s generalizability may be limited to these characteristics.

Conclusions

This study used the ISway protocol, a validated, quantitative clinical assessment of balance, to identify three independent balance domains in a relatively mild cohort of MS consisting of 1) sway amplitude and velocity, 2) sway frequency and jerk ML, and 3) sway frequency and jerk AP. Furthermore, we showed that PwMS exhibited deficits within the sway amplitude and velocity domain than neurotypical controls, with a similar effect observed between MS-F and MS-NF. While our findings substantiate research that balance performance is compromised in MS, we extend this work by identifying a streamlined balance model in MS consisting of independent domains. The recognition of such domains that are altered in PwMS and related to falls may be useful for efficient and strategic balance assessment and rehabilitation. While the current cross-sectional approach related a specific balance domain to falls, follow-up, longitudinal studies will be necessary to determine whether deficits in this domain (or variables associated with this domain) are causally linked to falls in people with MS.

Highlights.

• A 3-domain balance model was identified in a group of people with multiple sclerosis (MS)

• Domains included: sway amplitude & velocity and AP & ML sway frequency and jerk

• Sway amplitude was worse (i.e., increased) in the MS group than the control group

• MS fallers had worse (i.e., increased) sway amplitude and velocity than non-fallers

List of Abbreviations

MS

multiple sclerosis

PwMS

people with multiple sclerosis

ISway

instrumented sway system

PCA

principal component analysis

EFA

exploratory factor analysis

PD

Parkinson’s Disease

AP

anteroposterior

ML

mediolateral

EDSS

Expanded Disability Status Scale

MS-NF

people with MS with no reported falls in the previous 6 months

MS-F

people with MS with one or more falls in the previous 6 months

BBS

Berg Balance Scale

SF-36

36-Item Short-Form Health Survey

FES-I

Falls Efficacy Scale International