Journal of Neurology

, Volume 261, Issue 1, pp 213–223

Increased gait variability is associated with the history of falls in patients with cerebellar ataxia

Authors

    • Department of NeurologyUniversity of Munich
    • German Center for Vertigo and Balance Disorders (DSGZ), University of Munich
  • Max Wuehr
    • German Center for Vertigo and Balance Disorders (DSGZ), University of Munich
  • Cornelia Schlick
    • Department of NeurologyUniversity of Munich
    • German Center for Vertigo and Balance Disorders (DSGZ), University of Munich
  • Sabrina Huth
    • German Center for Vertigo and Balance Disorders (DSGZ), University of Munich
  • Cauchy Pradhan
    • German Center for Vertigo and Balance Disorders (DSGZ), University of Munich
  • Marianne Dieterich
    • Department of NeurologyUniversity of Munich
    • German Center for Vertigo and Balance Disorders (DSGZ), University of Munich
  • Thomas Brandt
    • German Center for Vertigo and Balance Disorders (DSGZ), University of Munich
    • Institute for Clinical NeurosciencesUniversity of Munich
  • Klaus Jahn
    • Department of NeurologyUniversity of Munich
    • German Center for Vertigo and Balance Disorders (DSGZ), University of Munich
Original Communication

DOI: 10.1007/s00415-013-7189-3

Cite this article as:
Schniepp, R., Wuehr, M., Schlick, C. et al. J Neurol (2014) 261: 213. doi:10.1007/s00415-013-7189-3
  • 461 Views

Abstract

Falls are common in patients with cerebellar ataxia (CA). Identification of gait variables associated with a higher risk of falls allows us to detect fallers and initiate protective procedures early. Gait variability, which is increased in CA patients, is a good predictor of falls in elderly subjects and patients with neurodegenerative diseases. The relationship between gait variability and fall risk in patients with different cerebellar disorders was systematically investigated. A total of 48 patients with different cerebellar ataxia entities [adult-onset cerebellar atrophy (SAOA) (n = 23), unknown entity (n = 7), vascular (n = 5), post-cerebellitis (n = 6), congenital (n = 2), Louis–Bar syndrome (n = 2), ethyltoxic (n = 2) posttraumatic (n = 1)] were examined using a GAITRite® sensor mat. Spatial and temporal variability parameters were used for ANOVA testing and logistic regression models with categorized fall events as dependent variables. Gait variability in the fore–aft direction showed significant differences between the fall groups (p < 0.05–0.01). Model effects were highest for walking with slow speed (correct prediction 0.50–0.72). The speed-dependent integral of gait variability markers showed a higher discriminatory power (correct prediction 0.74–0.94). Gait variability is linked to the fall risk of patients with CA, slow walking and temporal gait variability being most relevant. The use of speed-dependent integrals of gait variability improves the accuracy of fall prediction. To predict fall risks in cerebellar ataxia, gait variability measurements made during slow walking should be included in a gait analysis procedure. The effects of speed-adjusted physiotherapeutic interventions have to be further investigated.

Keywords

Gait variabilityFallsFall-related injuriesCerebellar ataxiaGait analysis

Introduction

Gait disorders with balance problems are common in patients with cerebellar ataxia (CA). The gait of these patients is characterized by a broadened base of support, irregularity of foot trajectories and increased variability of temporal and spatial gait parameters [24, 25]. Gait alterations result in an increased risk of falls for patients with both hereditary and sporadic entities of cerebellar dysfunctions [11, 12, 46]. The implications for mobility are profound; more than 50 % of patients with a hereditary ataxia report fall-related injuries, causing morbidity and reduced quality of life [12]. Furthermore, falls might initiate a secondary cycle of intense fear of falling with a consecutive loss of independence and thus a deterioration of physical activity, which in turn affects the stance and gait control of patients [1, 45]. The supportive treatment of CA patients should therefore include the exploration and documentation of fall events as well as fall prevention procedures in order to ameliorate secondary morbidity. Tools for assessing typical gait alterations of CA patients might be useful for the differentiation of potential fallers from non-fallers. However, a systematic investigation of fall risk and quantitatively measureable alterations in gait parameters has not been conducted so far. Increased levels of gait variability were shown to be related to a higher fall risk in elderly community-living subjects [3, 19, 21, 28] and in patients with Parkinson’s disease [5, 22], Huntington’s disease [17, 20] and different types of dementia [27]. The gait patterns of CA patients also show increased gait variability, which is dependent on the actual walking speed [3739]. The highest deviations of gait variability occur during walking at slow and fast speeds; walking at preferred speed is less affected.

This study investigated the relationship between gait variability and fall events in patients with CA. Since gait variability is speed-dependent in CA patients, the assessment of gait and the data analysis were performed for slow, intermediate (preferred) and fast walking speed sections.

Methods

Patients

The patient group included 48 patients with symptoms of a cerebellar ataxia. They were recruited from the Department of Neurology and our Dizziness Clinic (German Center for Vertigo and Balance Disorders). All patients underwent a complete neurological and physical examination prior to the experimental procedures. The scale for the assessment and rating of ataxia (SARA) [36] was used for all patients; inclusion criteria were a sub-score >0 and <4 in the item “gait”, implying a gait impairment with the ability to ambulate short distances without a walking aid. Exclusion criteria were other comorbidities that affect locomotion (except sensory neuropathy). Only patients with concomitant neuropathy who did not show motor weakness or severe somatosensory dysfunctions due to neuropathy were included in the study. All subjects gave their informed written consent prior to the experiments. The study protocol was approved by the local Ethics Committee and is in accordance with the standards of the Declaration of Helsinki.

Fall assessment

All subjects underwent a standardized interview with one of the authors (R.S. or C.S.). The following information was recorded: duration of symptoms, ambulatory status, functional status, medication, and falls in the previous 6 months. All subjects completed the Falls Efficacy Scale–International (FES-I) and the Activity-specific Balance Confidence Scale (ABC) as described in [13, 33].

Gait assessment

Gait analysis was performed using a 6.7-m-long pressure-sensitive carpet (GAITRite®, CIR System, Havertown, USA) with a sampling rate of 120 Hz. All patients had to walk over the carpet at three different speeds (preferred, slow, and maximally fast). The patients were measured without using ambulatory aids. Each walk was started 1.5 m in front of the mat and continued for 1.5 m beyond it in order to provide steady-state locomotion. Each condition was tested twice. At the start and stop sections, a protective rail was available for the subjects to hold. When necessary, the examiner walked beside the subject (approximately 0.3 m behind the patient) in order to prevent falls. Measurements with near-falls were discarded and the measurement was repeated. Clinical gait assessment was performed via the Functional Gait Assessment (FGA), a ten-item gait test developed for patients with balance deficits and vestibular disorders [47]. Moreover, the Berg Balance Scale was assessed for all enrolled subjects [4].

Data analysis

Matlab® and SPSS® were used for data analysis. Leg side was used as independent factor in the ANOVA and regression models. After testing of possible side asymmetries (Wilcoxon, Mann–Whitney) for each walking condition, data of both limbs were pooled in order to increase the number of step events. Pooling the data yielded on average 19.3 ± 1.2 steps during walking at preferred speed, 28.5 ± 4.3 steps during slow speed, 14.8 ± 0.6 steps during fast walking. A Matlab® routine was written to calculate the individual CV values for the parameters stride time, stride length and base of support by using the formula:
$${\text{CV }}\left[ \% \right] \, = \frac{\text{standard deviation of parameter}}{\text{mean of parameter}} \times 100$$
Polynomial functions for all gait variability parameters were estimated based on the six measurements (three speed conditions, each condition twice). Analysis of residues showed that second-order polynomial functions are sufficient to describe the speed-dependency of the majority of these gait variables (data not shown). The fit functions were normalized to the individual’s preferred walking speed (PWS) and the function was divided into three different speed sections: (1) “slow”: 0.25–0.75 of PWS; (2) “medium”: 0.75–1.25 of PWS; and (3) “fast”: 1.25–1.75 of PWS. Subjects whose measured gait conditions did not exceed 0.25 of PWS (slow walking), respectively 1.75 of PWS (maximally fast walking), were excluded from this statistical procedure in order to avoid extrapolation of the fit function. For the three speed sections, the area under the curve was calculated using the formulas:
$${\text{AUC}}_{\text{slow section}} = \frac{{\int\nolimits_{{0.25 \times {\text{PWS}}}}^{{0.75 \times {\text{PWS}}}} {\text{CV(s)d}}s}}{{(0.75 \times {\text{PWS}} - 0.25 \times {\text{PWS}})}}$$
$${\text{AUC}}_{\text{medium section}} = \frac{{\int \nolimits_{{0.75 \times {\text{PWS}}}}^{{1.25 \times {\text{PWS}}}} {\text{CV(s)d}}s}}{{\left( {1.25 \times {\text{PWS}} - 0.75 \times {\text{PWS}}} \right)}}$$
$${\text{AUC}}_{\text{fast section}} = \frac{{\int \nolimits_{{1.25 \times {\text{PWS}}}}^{{1.75 \times {\text{PWS}}}} {\text{CV(s)d}}s}}{{\left( {1.75 \times {\text{PWS}} - 1.25 \times {\text{PWS}}} \right)}}$$
$${\text{AUC}}_{\text{overall}} = \frac{{\int \nolimits_{{0.25 \times {\text{PWS}}}}^{{1.75 \times {\text{PWS}}}} {\text{CV(s)d}}s}}{{\left( {1.75 \times {\text{PWS}} - 0.25 \times {\text{PWS}}} \right)}}$$

Patients were categorized according to their individual frequency of fall events within the preceding 6 months. The statistical analysis comprises two branches with different categorical systems: binomial categories with the groups “no falls” and “falls” and multinomial categories with group 0: no falls; group 1: one fall, group 2: ≥2 falls.

As a first step, analysis was performed by multivariate analysis of variance (ANOVA) testing for significant differences of mean gait parameters between the different categorical fall groups. “Walking speed” was used as a cofactor in a two-way ANOVA model.

Second, a stepwise forward logistic regression model (controlled for age and gender) was performed to identify variables that can predict the individual fall status. Therefore, categorized fall frequencies (binomial and multinomial) were used as dependent variables. Different gait variability markers (CV of stride time, CV of stride length, CV of base of support) and their speed-dependent integrals (AUC) were used as covariates.

The results were considered significant if p < 0.05.

Results

Characteristics of the patients

The CA cohort had a mean age of 64.3 ± 17.8 years with a mean duration of symptoms of 55 ± 32 months. Ataxic symptoms were mild to moderate, with a median SARA score of 10 [range 3; 20].

The entities of CA were heterogeneous (details Table 1). A total of 34 patients performed balance training in a physiotherapeutic setup, 1–3 times a week. Nine patients had clinical signs of a peripheral neuropathy (reduced vibrotactile thresholds, reduced ankle jerk reflexes) with sensory deficits of the legs. The presence of peripheral neuropathy did not show any significant effects on the fall status of patients. A moderate but not significant trend towards reduced walking speeds and increased gait variability could be observed in CA patients with peripheral neuropathy.
Table 1

Demographic and clinical data of the enrolled subjects

Demographic characteristics

 N

48 (22 women)

 Mean age (years)

64.3 ± 17.8

 Height (m)

1.71 ± 0.06

 Weight (kg)

76.3 ± 8.6

 Duration of symptoms in months

54.8 ± 32.3

 Median SARA score [range]

10 [3; 20]

 Median Berg Balance Scale [range]

4 [2; 8]

Etiology of ataxia

 SAOA

23

 Idiopathic (without atrophy)

7

 Vascular

4

 Astrocytoma (post-surgery)

1

 Ethyltoxic atrophy

2

 Posttraumatic lesions

1

 Cerebellitis

Unknown entity 2, paraneoplastic 4

 Congenital

Unknown entity 1, Arnold-Chiari 1

 Louis–Bar Syndrome

2

Walking performance

 Median FGA score [range]

17 [6; 30]

Ambulatory aids

 None

24 (50 %)

 Intermediate, personal help

3 (6 %)

 Walking stick(s)

10 (21 %)

 Walking frame

11 (23 %)

Fall frequency in preceding 6 months

 No falls

21 (44 %)

 1 fall

10 (21 %)

 ≥2 falls

17 (35 %)

SARA scale for the assessment and rating of ataxia, SAOA sporadic adult onset cerebellar atrophy, FGA functional gait assessment

Ambulatory status and fall history

The Functional Gait Assessment (FGA) scores of the enrolled subjects showed a moderate to severe impairment of gait function with a median of 17 [range 6; 30]. The Berg Balance Scale (BBS) and the GAITRite® specific functional ambulatory profile (FAP) [30] also showed moderate to severe impairments of gait performance (Table 1 and Table 5, supplemental data). None of these parameters showed a significant correlation with the duration of symptoms.

More than half of the patient cohort (n = 27; 17 patients with SAOA, four patients with unknown entities of cerebellar ataxia, two patients with ethyltoxic atrophy, three patients with cerebellitis, one patient with Louis-Bar syndrome) reported at least one fall in the preceding 6 months. Seventeen of the 27 “fallers” had more than one fall event in the preceding 6 months (for details, see Table 1). Fall categories were independent of the clinical scores (SARA, BBS, FGA), the diagnosis groups, the duration of symptoms and the presence of a concomitant neuropathy. Ambulatory aids, walking sticks or a walking frame were used by 21 patients, but mainly the patients received personal help. All had a positive history of falls. Only three patients reported that the fall event had occurred when using the aid, whereas the majority (24 of 27 patients) reported that falls usually occurred without the aid.

Gait characteristics of CA patients

Although not specifically compared to a healthy subject control group, the characteristics of the gait disorders in our cohort were similar to those of our previous studies (details in Table 5, supplementary data). A one-way ANOVA for walking speed revealed significant changes of nearly all gait variables, except for the coefficients of variation of base of support. As in previous research, we identified a U-shaped curve of speed-dependent changes of the CV of stride length and the CV of stride time. For both parameters, the highest values were found under the condition of walking at slow speed (Table 5, supplementary data).

Significant correlations with the duration of symptoms (in months) were found for the base of support during walking at preferred speed (R = 0.37, p < 0.05) and during walking at fast speed (R = 0.32, p < 0.05). No significant effects were found when testing for the factor “diagnosis group”.

The area under the curve (AUC) could be calculated for 40 of the 48 patients. Eight patients had to be excluded since walking speeds did not exceed the external boundaries of the speed sectors (0.25 of PWS, 1.75 of PWS). The AUCs for the CV of stride length and for the CV of stride time were comparable. Highest individual AUC values were found for the slow speed AUC (0.25 of PWS to 0.75 of PWS). ANOVA testing revealed a significant speed effect for both variability markers (p < 0.01, respectively p < 0.05) (Table 6, supplemental data).

Fall frequency and speed-dependent gait variability

Across the different fall categories, significant differences for the means of CV of stride length and the CV of stride time were found (in model 1 and 2) with post hoc analysis revealing significance during walking at slow speed (model 1) (Figs. 1I, 2I, Table 2a, b) and during walking at slow and preferred speeds (model 2). The F value of the CV of stride time was higher than the F-value of the CV of stride length.
https://static-content.springer.com/image/art%3A10.1007%2Fs00415-013-7189-3/MediaObjects/415_2013_7189_Fig1_HTML.gif
Fig. 1

Speed-dependent gait variability of CA patients (two-group comparison). The patients are categorized according to their reported fall status (binomial categories: no falls vs. falls). Boxplots (with median, upper and lower quartile and 0.95 boundaries) represent the raw values of the coefficient of variation of stride time in section I and the areas under the curve (AUC) of the coefficient of variation of stride time in section II. In both sections, a indicates walking at slow speed (respectively AUC 0.27–0.75 of PWS); b indicates walking at preferred speed (respectively AUC 0.75–1.25 of PWS); c indicates walking at maximally fast speed (respectively AUC 1.25–1.75 of PWS); d indicates all speed sections, only possible for the AUC procedure (AUC 0.25–1.75). ANOVA with post hoc Scheffé tests revealed significant differences between the fall groups (*p < 0.05; **p < 0.01; ***p < 0.001). Note that the CV and AUC values are significantly different between the groups only for slow walking, whereas fast walking speeds does not reveal any significant difference. AUC values also showed differences for preferred walking and the overall speed spectrum. AUC area under the curve, CV coefficient of variation, PWS preferred walking speed

https://static-content.springer.com/image/art%3A10.1007%2Fs00415-013-7189-3/MediaObjects/415_2013_7189_Fig2_HTML.gif
Fig. 2

Speed-dependent gait variability of CA patients (three-group comparison). The patients are categorized according to their reported fall frequency (multinomial categories: no falls vs. one fall vs. ≥ two falls). Boxplots (with median, upper and lower quartile and 0.95 boundaries) represent the raw values of the coefficient of variation of stride time in sectionI and the areas under the curve (AUC) of the coefficient of variation of stride time in section II. In both sections, a indicates walking at slow speed (respectively AUC 0.27–0.75 of PWS); b indicates walking at preferred speed (respectively AUC 0.75–1.25 of PWS); c indicates walking at maximally fast speed (respectively AUC 1.25–1.75 of PWS), and d indicates all speed sections, only possible for the AUC procedure (AUC 0.25–1.75). ANOVA with post hoc Scheffé tests revealed significant differences between the fall groups (*p < 0.05, **p < 0.01; ***p < 0.001). Note that the CV and AUC values are significantly different between the groups only for slow and preferred walking conditions, whereas fast walking speeds does not reveal any significant difference. CV values showed significant differences between the group of non-fallers and the fallers. AUC values also showed differences between occasional and frequent fallers. AUC area under the curve, CV coefficient of variation, PWS preferred walking speed

Table 2

Between-group comparison of gait variability parameters and their integrals

A

Model 1

No falls

Falls

 

ANOVA

Slow

Mean CV of stride length [%]

7.5 ± 3.5

11.9 ± 5.3

 

F1,47 = 11.92; p < 0.01

Mean CV of stride time [%]

11.5 ± 6.7

17.9 ± 8.6

 

F1,47 = 13.88; p < 0.001

Mean CV of base of support [%]

17.7 ± 15.5

19.6 ± 12.3

 

F1,47 = 0.13; n.s.

AUC0.25–0.75 CV of stride length

13.4 ± 4.3

17.8 ± 7.2

 

F1,119 = 23.43; p < 0.001

AUC0.25–0.75 CV of stride time

17.1 ± 5.4

37.8 ± 15.5

 

F1,119 = 31.10; p < 0.001

AUC0.25–0.75 CV of base of support

30.9 ± 19.3

34.3 ± 21.3

 

F1,119 = 0.47; n.s.

Preferred

Mean CV of stride length [%]

4.9 ± 3.0

6.9 ± 3.0

 

F1,47 = 3.43; n.s.

Mean CV of stride time [%]

3.9 ± 2.5

5.8 ± 3.6

 

F1,47 = 4.04; n.s.

Mean CV of base of support [%]

19.6 ± 11.1

28.9 ± 20.9

 

F1,47 = 2.00; n.s.

AUC0.75–1.25 CV of stride length

14.4 ± 6.3

18.0 ± 4.7

 

F1,119 = 3.01; n.s.

AUC0.75–1.25 CV of stride time

14.7 ± 8.0

24.3 ± 12.2

 

F1,119 = 6.23; p < 0.05

AUC0.75–1.25 CV of base of support

34.3 ± 15.9

31.9 ± 24.8

 

F1,119 = 1.09; n.s.

Fast

Mean CV of stride length [%]

6.2 ± 4.3

9.5 ± 4.9

 

F1,47 = 3.14; n.s.

Mean CV of stride time [%]

7.3 ± 4.3

9.2 ± 5.2

 

F1,47 = 2.46; n.s.

Mean CV of base of support [%]

21.1 ± 11.4

28.0 ± 12.4

 

F1,47 = 1.35; n.s.

AUC1.25–1.75 CV of stride length

17.1 ± 4.2

21.0 ± 7.8

 

F1,119 = 1.53; n.s.

AUC1.25–1.75 CV of stride time

19.9 ± 8.4

22.5 ± 11.7

 

F1,119 = 1.36; n.s.

AUC1.25–1.75 CV of base of support

31.3 ± 9.2

28.4 ± 11.8

 

F1,119 = 0.79; n.s.

Overall

AUC0.25–1.75 CV of stride length

37.8 ± 14.2

60.8 ± 30.3

 

F1,119 = 18.62; p < 0.001

AUC0.25–1.75 CV of stride time

45.7 ± 19.1

81.5 ± 19.6

 

F1,119 = 18.31; p < 0.001

AUC0.25–1.75 CV of base of support

81.2 ± 23.7

88.0 ± 37.9

 

F1,119 = 1.43; n.s.

B

Model 2

No falls

One fall

≥two falls

 

Slow

Mean CV of stride length [%]

7.5 ± 3.5

10.8 ± 4.4

12.8 ± 4.9

F2,137 = 6.54; p < 0.05

Mean CV of stride time [%]

11.5 ± 6.7

16.0 ± 6.6

20.0 ± 6.1

F2,137 = 13.82; p < 0.01

Mean CV of base of support [%]

17.7 ± 15.5

19.3 ± 17.7

18.7 ± 17.1

F2,137 = 0.70; n.s.

AUC0.25–0.75 CV of stride length

13.4 ± 4.3

15.3 ± 6.3

19.1 ± 9.5

F1,119 = 12.39; p < 0.01

AUC0.25–0.75 CV of stride time

17.1 ± 5.4

37.6 ± 12.4

38.0 ± 15.3

F1,119 = 15.21; p < 0.001

AUC0.25–0.75 CV of base of support

30.9 ± 19.3

32.0 ± 18.8

35.7 ± 21.4

F1,119 = 1.23; n.s.

Preferred

Mean CV of stride length [%]

4.9 ± 3.0

5.4 ± 2.9

7.9 ± 4.3

F2,137 = 3.96; p < 0.05

Mean CV of stride time [%]

3.9 ± 2.5

4.1 ± 1.5

7.1 ± 4.4

F2,137 = 5.36; p < 0.05

Mean CV of base of support [%]

19.6 ± 11.1

31.6 ± 30.5

25.9 ± 24.1

F2,137 = 1.21; n.s.

AUC0.75–1.25 CV of stride length

14.4 ± 6.3

16.8 ± 3.5

18.7 ± 6.8

F1,119 = 2.04; n.s.

AUC0.75–1.25 CV of stride time

14.7 ± 8.0

23.1 ± 12.1

24.7 ± 11.0

F1,119 = 3.76; p < 0.05

AUC0.75–1.25 CV of base of support

34.3 ± 15.9

37.5 ± 31.8

28.6 ± 21.2

F1,119 = 0.22; n.s.

Fast

Mean CV of stride length [%]

6.2 ± 4.3

8.5 ± 3.2

10.1 ± 6.6

F2,137 = 3.22; p < 0.05

Mean CV of stride time [%]

7.3 ± 4.3

8.3 ± 4.1

9.8 ± 6.4

F2,137 = 2.32; n.s.

Mean CV of base of support [%]

21.1 ± 11.4

22.7 ± 13.7

24.5 ± 12.2

F2,137 = 0.85; n.s.

AUC1.25–1.75 CV of stride length

17.1 ± 4.2

20.3 ± 8.3

21.4 ± 7.5

F1,119 = 1.89; n.s.

AUC1.25–1.75 CV of stride time

19.9 ± 8.4

21.6 ± 5.9

23.0 ± 14.2

F1,119 = 2.50; n.s.

AUC1.25–1.75 CV of base of support

31.3 ± 9.2

22.9 ± 12.6

31.7 ± 11.3

F1,119 = 1.12; n.s.

Overall

AUC0.25–1.75 CV of stride length

37.8 ± 14.2

60.2 ± 28.3

61.1 ± 31.2

F1,119 = 8.62; p < 0.01

AUC0.25–1.75 CV of stride time

45.7 ± 19.1

71.3 ± 22.2

87.5 ± 18.4

F1,119 = 10.51; p < 0.001

AUC0.25–1.75 CV of base of support

81.2 ± 23.7

87.3 ± 29.4

88.4 ± 41.4

F1,119 = 2.69; n.s.

Fore–aft and medio-lateral gait variability parameters and their speed-dependent integrals are listed separately according to the actual walking speed. Model 1 describes a two-group (falls vs. no falls) and model 2 describe a three-group (no falls vs. one fall vs. ≥2 falls) comparison. ANOVA results for group differences are listed in the right column

AUC area under the curve, CV coefficient of variation, PWS preferred walking speed

The AUC of the CV of stride length and the CV of stride time also showed significant differences between the different fall groups (Figs. 1II, 2II; Table 2a, b). We found significant increased AUC predominantly during slow walking and for the overall speed section (all p < 0.01) in the faller groups compared to the non-faller group.

In the logistic regression model, we found significant effects for the CV of stride length (p < 0.05 for slow walking), for the CV of stride time (p < 0.01 for slow walking) (Table 3a) and for the AUCs of both parameters (for all speed sectors; p values <0.01 to <0.05) (Table 4a) in model 1 (fallers vs. non-fallers prediction). The highest correct prediction rate was found for the AUC0.25–0.75 (slow walking) of the CV of stride length.
Table 3

Logistic models for the magnitudes of gait variability and the fall frequencies

A

Logistic regression model 1

Non-fallers vs. fallers

Slow

Preferred

Fast

CV of stride length

Chi2; p value

16.78; p < 0.05

7.00; n.s.

3.57; n.s.

 

Correct prediction

0.65

CV of stride time

Chi2; p value

22.62; p < 0.01

4.44; n.s.

2.47; n.s.

 

Correct prediction

0.82

CV of base of support

Chi2; p value

3.21; n.s.

0.17

1.37; n.s.

 

Correct prediction

B

Logistic regression model 2

Group 0–2

Slow

Preferred

Fast

CV of stride length

Chi2; p value

11.18; p < 0.01

6.23; p < 0.05

5.01; n.s.

 

Correct prediction

0.54

0.50

CV of stride time

Chi2; p value

12.56; p < 0.01

5.76; n.s.

4.77; n.s.

 

Correct prediction

0.52

CV of base of support

Chi2; p value

0.15; n.s.

2.55; n.s.

4.44; n.s.

 

Correct prediction

 

Overall correct prediction

Logistic regression model 1 describes the binomial model for the dependent variable “fall category” (fallers vs. non-fallers). Covariates were the coefficients of variation of stride length, stride time and base of support

Logistic regression model 2 describes the multinomial model for the dependent variable “fall frequency” of the preceding 12 months (group 0, no falls; group 1, one fall; group 2, two or more falls)

Table 4

Logistic models for the AUCs of gait variability and the fall frequencies

A

Logistic regression model I

Non-fallers vs. fallers

For CV of stride length

For CV of stride time

For CV of base of support

AUC 0.25–1.75

Chi2; p value

9.64; p < 0.01

11.92; p < 0.01

2.43; n.s.

 

Correct prediction

0.72

0.71

AUC 0.25–0.75

Chi2; p value

28.24; p < 0.001

16.92; p < 0.01

3.56; n.s.

 

Correct prediction

0.94

0.74

AUC 0.75–1.25

Chi2; p value

7.26; p < 0.05

10.92; p < 0.01

4.40; n.s.

 

Correct prediction

0.72

0.73

AUC 1.25–1.75

Chi2; p-value

8.45; p < 0.05

9.21; p < 0.05

2.43; n.s.

 

Correct prediction

0.52

0.61

B

Logistic regression model 2

Group 0–2

For CV of stride length

For CV of stride time

For CV of base of support

AUC 0.25–0.75

Chi2; p-value

20.02; p < 0.001

28.30; p < 0.001

4.87; n.s.

 

Correct prediction

0.63

0.83

AUC 0.75–1.25

Chi2; p-value

8.87; p < 0.05

7.28; p < 0.05

3.51; n.s.

 

Correct prediction

0.67

0.54

AUC 0.75–1.25

Chi2; p value

4.94; n.s.

2.96; n.s.

4.84; n.s.

 

Correct prediction

 

Correct prediction

AUC 0.25–1.75

Chi2; p value

14.71; p < 0.05.

19.34; p < 0.001

3.64; n.s.

 

Correct prediction

0.50

0.72

Logistic regression model 1 describes the binomial model for the dependent variable “fall category” (fallers vs. non-fallers). Covariates were the AUCs (0.25–0.75 of PWS; 0.75–1.25 of PWS; 1.25–1.75 of PWS, 0.25–1.75 of PWS)

Logistic regression model 2 describes the multinomial model for the dependent variable “fall frequency” of the preceding 12 months (group 0, no falls; group 1, one fall; group 2, two or more falls). Covariates were the AUC (0.25–0.75 of PWS; 0.75–1.25 of PWS; 1.25–1.75 of PWS, 0.25–1.75 of PWS)

AUC area under the curve, CV coefficient of variation, PWS preferred walking speed

The logistic regression model also found significant effects for the CV of stride length, the CV of stride time and the AUCs of these parameters in model 2 (patient groups with no-falls, one fall and ≥2 falls). The amount of correct prediction was higher for the AUCs than for the raw values (0.83 vs. 0.54 for CV of stride time) (Tables 3, 4b).

Similar results were found for a subgroup analysis of only SAOA patients. The logistic regression models showed significance for the CV of stride length (p < 0.05), for the CV of stride time (p < 0.01), for the AUC of stride length (during slow walking) (p < 0.01) and the AUC of stride time (for all speed sections (p < 0.01) in both models.

Discussion

Our main findings are:
  1. I.

    Temporal and spatial gait variability in the fore–aft, but not medio-lateral, direction of locomotion depends on walking speed in patients with cerebellar ataxia.

     
  2. II.

    These parameters are associated with the history of falls; thus, they are useful for the identification of patients with high fall risk.

     
  3. III.

    Speed-dependent, integral analysis of gait variability further increases the sensitivity of identification of fallers in CA.

     

Gait variability of the CA cohort

Gait variability is a quantitative measure for walking stability. Impairments of gait variability have been shown to be associated with falling and frailty in neurodegenerative disorders [15, 16, 20, 41]. It is well known that patients with cerebellar ataxia have increased gait variability parameters of spatial and temporal variables [10, 24, 25, 32, 40, 44] and the magnitudes of gait variability measures of the current CA cohort were similar to those reported in previous studies. Previous studies and our data indicate that in CA patients, the CVs of stride length and of stride time are increased during walking at slow and walking at fast speeds [39, 48]. In contrast, gait variability in the medio-lateral direction does not show significant speed-dependent changes in the CA cohort. For healthy subjects and for patients with CA, it has been proposed that medio-lateral adjustments are differentially controlled, in terms of a more actively feedback control, which is based on integrative sensory inputs. Passive dynamics of the walking patterns appear to be stable in the fore–aft direction, but quite unstable in the medio-lateral direction [2, 31].

Gait variability and fall events of the CA cohort

One intriguing aspect of gait variability is its relationship to fall risk. Many recent studies implied that this relationship occurs in elderly walkers [18, 19, 21, 23] and in patients with Parkinson’s disease [15, 22], Huntington’s disease [17, 20], or cognitive impairments [7, 27, 43]. These studies mainly investigated gait variability during self-selected walking modes or dual-task paradigms. Fonteyn and coworkers utilized data of patients with cerebellar ataxia from the EuroSCA cohort and identified non-ataxic features, such as pyramidal signs, as predictors for a higher incidence of falls [11, 12]. However, they did not provide information about gait features. Our results indicate for the first time that increased levels of gait variability in the fore–aft direction are associated with a higher incidence of falls in patients with CA. The present study used a retrospective fall assessment, which is in general found to under-report the amount of falls [29]. However, overall incidence of falls in our cohort was comparable to that of previous epidemiological studies [11, 12, 46], with more than half of our cohort reporting falls in the preceding 6 months. We did not find significant correlations between the clinical scores (SARA, FGA, BBS) and the fall history of the patients.

Both group comparison and regression models showed that the amount of fore–aft gait variability is higher in patients with falls than in non-fallers and that these differences can be used to discriminate the fall groups. Gait data during walking at slow speed is most useful, since slow walking data showed the highest differences among the different fall groups and the highest rates of correct prediction in the regression models. Although tested in a heterogeneous group of patients with CA, the subgroup analysis of SAOA patients was in line with the basic findings of the results for the overall patient groups. This might indicate that in patients with cerebellar gait disorders, irrespective of the underlying pathology, there is a relationship between impaired gait variability and fall events. Including slow speed walking in the clinical gait assessment is recommended in order to identify CA patients with a higher fall risk. We suggest that gait measurements in the speed section of 0.25–0.5 of PWS should be performed in order to assess gait variability in a slow walking mode condition.

Although gait variability was also increased during fast walking, we did not find a strong relationship between the gait parameters and the fall status for that speed condition. Previous case series indicated the relationship between the CV of stride time during fast walking and subjective fall-related parameters, e.g., the Activity-specific balance Confidence Scale and the Falls Efficacy Scale-International [37, 38]. Further prospective studies should include the cross-validation of such scores with the risk of falling.

The discriminatory power and the quality of correct prediction were better for CV of stride time than for CV of stride length. This difference might originate from the utilized sensor mat, since the temporal sampling rate is higher than the spatial resolution of the carpet.

Integrals of gait variability and fall events

Preferred walking modes exhibit energy-efficient properties in terms of a constraint step-frequency to step-length ratio that is optimized towards the individual biomechanical and sensorimotor properties [26, 42]. Gait at non-preferred walking speeds does not reveal such constraint characteristics, so measurements at these speed sections are more susceptible to procedural variations. Variations of gait speed result in changes of speed-dependent gait variables, which might influence raw-value gait data. In order to cope with this problem, we analyzed speed-dependent integrals of gait variability, which are preprocessed and normalized to preferred walking speed. Mathematically, we used an area under the curve (AUC) calculation, which is established for the estimation of the bioavailability of drugs or the quality of receiver operating characteristics [6, 14, 34, 35]. The AUC results revealed a similar relationship between fall events and gait variability like raw gait variability procedures; however, AUC results showed a better correct-prediction rate and higher F values in the regression models. Due to their procedural effort, AUC approaches probably remain procedures restricted to specialized balance centers. For the setup of randomized clinical trials, these procedures might serve as important outcome measures in order to minimize confounding procedural variations.

Limitations of the study

One limiting factor of the present study is the utilized fall assessment, which evaluated the fall history by a retrospective self-report of falls within 6 months. This kind of assessment is prone to underestimate fall events [29] and might therefore have an impact on our results. However, whereas the reliability of fall-related injuries and fall circumstances in retrospective questioning is poor compared to prospective designs, the documentation of the incidence of falls shows a moderate to sufficient reliability [8, 9]. Another limitation of this study is that we were restricted to a short-distance gait measurement. Therefore, gait data analysis was based on 15–30 step events, which might especially influence the gait variability outcomes. Patients with ataxia commonly exhibit high variations within their walking mode, which remain undetected during short-distance walk recordings. However, the present results show a high consistency with previous studies using the same short-distance gait assessment [39] as well as long-distance data collection during treadmill walking [48].

Conclusions and future perspectives

The study generated the first evidence that gait variability measures can be used for systematic fall risk studies in patients with CA. Future studies should focus on the prospective recording of fall events (and near-fall events), fall-related morbidity and mobility markers. The study helps in identifying the relevant gait parameters and gait conditions that should be examined for fall risk estimation.

The findings suggest potential fall protection regimens for the management of patients with cerebellar ataxia. Physiotherapeutic interventions should include exercises at different gait speeds and transitions between different gait speed sections.

Acknowledgments

The authors thank Judy Benson for copy-editing the article. The work was supported by the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG JA 1087/1-1), the German Hertie Foundation and the Federal Ministry for Education and Science (BMBF, Nr. 80121000-49) of Germany.

Conflicts of interest

On behalf of all authors, the corresponding author states that there is no conflict of interest.

Supplementary material

415_2013_7189_MOESM1_ESM.docx (16 kb)
Supplementary material 1 (DOCX 15 kb)
415_2013_7189_MOESM2_ESM.docx (15 kb)
Supplementary material 2 (DOCX 14 kb)

Copyright information

© Springer-Verlag Berlin Heidelberg 2013