Mapping Trajectories of Gait Recovery in Clinical Stroke Rehabilitation

Abstract

Background

How gait changes during the early stages of stoke rehabilitation, and which patient characteristics are associated with these changes is still largely unknown.

Objective

he first objective was to describe the changes in gait during stroke rehabilitation. Secondly, we determined how various patient characteristics were associated with the rate of change of gait over time.

Methods

Participants were measured every 3 weeks during stroke rehabilitation. The assessment consisted of an inertial measurement unit (IMU) based 2-minute walk test (2MWT), 3 IMU-based balance tests, and standard clinical tests. In the 2MWT, participants were equipped with 3 IMUs, from which speed, variability, asymmetry, and smoothness were calculated. The changes in gait were examined from admission to discharge at an individual level. The effect of patient characteristics on the rate of change of the gait features over time was assessed with growth models.

Results

A total of 81 Trajectories from 72 participants were analyzed. On an individual basis, speed increased in 32 trajectories. Only a few trajectories exhibited significant changes in variability, asymmetry, and smoothness over the clinical rehabilitation period. The growth models revealed a significant increase in speed and decrease in variability and smoothness. Only the Berg Balance Scale and gait speed at onset were (negatively) associated with the rates of change of speed and smoothness, respectively.

Conclusion

We found a substantial variability in the gait-feature outcomes and their progression in individuals after stroke during clinical rehabilitation. The patient characteristics studied had limited associations with the rate of change of gait features over time.

Keywords

accelerometers stroke recovery stroke rehabilitation gait quality growth curve modeling walking ability

Introduction

Stroke is a leading cause of long-term disability worldwide, often resulting in sensorimotor impairments including reduced motor control, spasticity, muscle weakness, and proprioceptive deficits.¹ These impairments can significantly impact people after stroke, hamper their gait, and subsequently reduce walking performance, mobility, independence, and overall quality of life.^2
-4 Stroke rehabilitation aims to improve functional walking, that is, one’s ability to walk in daily life, by applying the most effective strategy to facilitate recovery and training individuals to cope with their impairments.^5
-11 Currently, in clinical rehabilitation, gait function is predominantly assessed via a measurement of gait speed.¹² Gait speed is a valid, reliable, accessible, and sensitive measure of functional status and overall health in the stroke population.^13
-15 Other gait features, for example, stability, variability, symmetry, are not yet routinely measured during clinical rehabilitation, as this is typically expensive and time-consuming, requires specific expertise, and the applicability of the results is still largely unknown.^16,17 Nevertheless, evidence suggests that these other gait features provide clinically relevant information, as they are associated with the degree of recovery and the fall risk after stroke.^18
-20 Moreover, it is conceivable that these other gait features yield information regarding the potential to improve functional walking.

Gait can be measured with multiple tools, such as motion capture systems, force plates, and inertial measurement units (IMUs).^20
-28 Among these, IMUs might be the most feasible in clinical practice, given their small size, low cost, and ease of operation. Various studies assessed the relationship between gait features and functional walking in people after stroke.^21
-23 Two recent studies indicated that the quality of joint movements, measured with the gait profile score (deviation of kinematic gait data relative to normative data) and the knee joint flexion during pre-swing and swing phase were associated with walking performance.^21,22 Contrary, Kim and Eng²³ found that gait features closely resembling those of healthy individuals do not necessarily coincide with better functional walking. The findings of Kim and Eng (2004) might be explained by the fact that people after stroke often use behavioral compensation strategies to improve their functional walking.^6
-9 Behavioral compensation strategies help individuals to cope with their impairments during walking, which varies between individuals depending on their specific impairments. Consequently, gait features might change in opposing directions for individuals during clinical stroke rehabilitation, both resulting in an improvement of functional walking. To get insight into the changes in gait features over time, and how these changes are related to functional walking, studies with a longitudinal design are required.

Several studies have investigated changes in gait features over time after stroke.^27
-29 For instance, Chow and Stokic,²⁷ assessed the changes in spatio-temporal gait characteristics during the first-year after stroke, including gait speed and symmetry. They found an improvement in the majority of the spatio-temporal gait characteristics in the first 6 months. However, only a few individuals showed significant improvement in these gait characteristics in the following 6 months. Kollen et al²⁹ determined changes in functional walking, measured with the Functional Ambulation Categories (FAC), over time, and assessed the relationship between functional walking and time and time-dependent patient characteristics during the first-year post stroke. They concluded that the more pronounced improvements in functional walking occurred in the earlier stages after stroke. Furthermore, they found that improvements in standing balance were associated with improvements in functional walking. In addition to standing balance, evidence suggests that there are other time-dependent and time-independent characteristics that are associated with or have predictive value for the level of functional walking after stroke.^30
-32 For example, age and functional level at admission seem to be linked to the level of motor function after stroke.³⁰ Accordingly, these characteristics may be associated with the rate of change of gait features as well.

To gain a more comprehensive insight into the dynamics of gait recovery in the early phase after stroke, this study had 2 objectives. The first objective was to describe the alterations in gait, measured with IMUs, during clinical stroke rehabilitation. Secondly, we investigated how various characteristics at baseline affect the rate of change of gait over time, aiming to identify and understand potential facilitators, and barriers of gait recovery in stroke rehabilitation. This research contributes to the refinement of our understanding of gait recovery after stroke and therefore potentially provides directions to further personalize stroke rehabilitation by tailoring interventions, providing better prognosis, and managing patient expectations.

Methods

Participants

Longitudinal data were collected from individuals after stroke between January 1, 2021, and July 1, 2023, in 5 clinical rehabilitation centers in the Netherlands. The inclusion criteria for the study were as follows: (1) participants aged 18 years or older; (2) capable of understanding and signing the informed consent document; (3) able to perform simple tasks; and (4) first-ever or recurrent stroke. Participants were excluded from the study if only a single assessment was obtained, ensuring that analyses were based on individuals with multiple data points. Participants provided written informed consent before participating. The Medical Ethical Review Committee of Utrecht approved the protocol (20-462/C).

Assessment

Before the assessment, demographic and stroke-specific characteristics were collected, including sex, age, stroke type and side, and pre-morbid walking aid use. The assessment consisted of 3 parts and was administered by an experienced physiotherapist or trained research assistant every 3 weeks. Data were collected from participants as soon as they were able to walk a minimum of 8 meters within 2 minutes without physical assistance, extending up to the point of discharge.²⁴

In the first part, participants walked for 2 minutes at a self-selected speed on a 14-m walking path. Three IMUs consisting of a triaxial accelerometer and gyroscope, were used (created by Aemics b.v. Oldenzaal, The Netherlands). Two IMUs were attached on top of the left and right foot at the height of the cuneiform bones, and 1 IMU was attached to the lower back at the height of L5 using elastic bands. The IMUs measured with a sampling frequency of 104 Hertz. Participants were allowed to walk with a walking aid during the walking test.²⁴ If participants with a walking aid were also able to safely walk without walking aid, the walking test was administered under both conditions.

In the second part, the 3 balance conditions were used to measure postural sway. The balance conditions were arranged based on difficulty and executed in the following order (1) Sitting unsupported on a balance cushion with feet touching the ground and knees in a 90° angle for 60 seconds. (2) Standing with feet in a self-selected position for 60 seconds. (3) Standing with eyes closed and feet in a self-selected position for 30 seconds.³³ The sensor was placed at the back, at the level of T7 during sitting and L5 during standing, to represent the center of mass during the test.

In the third part, several standard clinical tests were administered. These included the Berg Balance Scale (BBS)³⁴ and Barthel Index (BI)³⁵ at admission, and the FAC, both with and without a walking aid.³⁶

Data Processing

The collected IMU-based gait and balance data underwent digital processing on a custom-made online platform, where data were processed and stored. The IMU data were processed by, first, resampling the data to 100 Hz and correcting the gyroscope offset. Next, gait and balance features in various domains were calculated.^24,33

Gait Data

Gait was described using 4 gait features, namely: speed (m/s), asymmetry, smoothness, and variability (s). These 4 gait features were selected because they are reliable and represent different aspects of gait.^24,37 A description of the gait features is given below. The equations for the 4 variables are described in Supplemental Table 1 in the appendix.

Speed is often used as a measure for the functional capacity and in stroke rehabilitation an indication of someone’s ability to walk in daily life, that is, functional walking.¹³ First, a sensor-fusion algorithm was used to transform acceleration from a local to a global reference frame by combining the accelerometer and gyroscope data.³⁸ Second, the linear acceleration in the anterior-posterior direct was integrated once to determine the gait speed. Third, a Zero-Velocity Update was applied to set velocity to zero during stationary phases of walking, thereby reducing estimation errors.³⁹ Fourth, the corrected gait speed was integrated to determine position. This calculation enabled us to measure the total distance covered in the 2-minute assessment, and thus, the gait speed.

Asymmetry is an expression of the asymmetry between the left and right side of an individual’s body during walking. The asymmetry ratio was calculated by dividing the stance-swing phase ratio of the paretic leg by the stance-swing phase ratio of the non-paretic leg. The stance-phase duration was calculated using the acceleration signal of the left and right foot sensor. In normal gait, there is typically a high degree of asymmetry, resulting in an asymmetry ratio close to 1. Evidence suggests that temporal asymmetry is negatively associated with preferred speed and motor recovery of the leg and foot.⁴⁰

Variability was expressed as the variability in stride time. The variability was calculated as the standard deviation of the time per stride. A lower gait variability score is related to a more stereotypical gait. Variability appears to be an indicator for decline in locomotor function.⁴¹

Smoothness characterizes the regularity of the movement patterns during walking. Smoothness was expressed as the index of harmonicity of the pelvis acceleration in the medio-lateral direction during walking. The index of harmonicity indicates the harmonics in the data by comparing the amplitude of the dominant frequency to the sum of the amplitude of the 5 subsequent dominant frequencies. A lower index of harmonicity in the medio-lateral is related to a lower smoothness. Previous work suggests that smoothness is associated with lower limb spasticity and falls.^20,42 The minimal-detectable-change values for these features were determined in a previous study and are as follows: 0.137 m/s for the speed, 1.217 for the asymmetry, 0.144 s for the variability, and 0.160 for the smoothness.²⁴

Postural Sway Data

The postural sway data of the 3 balance conditions was described using 1 feature per measurement, namely: path. The path was computed as the length of the postural sway trajectory determined from the pelvis IMU (m/s²).³³ A higher path score indicates more movement during the balance assessment, which is associated with a worse performance.

Statistical Analysis

Outliers were identified and removed based on z-scores. For each gait feature, z-scores were calculated to standardize the values. Observations exceeding an absolute value of 4.0 were considered outliers and removed from further analysis. The statistical analysis consisted of 2 parts. In the first part, we compared gait features between admission and discharge on an individual level. In the second part, we evaluated the change of gait features relative to time on a group level.

Individual Changes in Gait Features

The changes in gait features were examined from admission to discharge at an individual level across trajectories with and without walking aid. A difference was made between with and without a walking aid, since a walking aid can significantly affect spatio-temporal gait features.⁴³ Individual trajectories were categorized as showing a significant increase, remaining stable, or decreasing over time, by comparing change scores between admission and discharge to the minimal detectable change (MDC).

Growth Curve Modeling

To identify if patient characteristics influenced the rate of change of gait over time, growth curve models were used. Growth curve models are a statistical technique to analyze how an outcome variable changes over time within individuals. The growth curve models were created using linear mixed models, which incorporate both fixed effects, representing average changes across all individuals, and random effects, allowing individual variations in change patterns. Linear mixed models can handle issues like missing data and time-varying covariates.^44
-46 Initially, we developed a univariate mixed model per gait feature to estimate the average rate of change per unit of time measurement (every 3 weeks). To accommodate the within-participant correlation of repeated measurements, we included a random intercept on the participant level to the model. We evaluated if random slopes improved the model via the likelihood ratio test. Subsequently, we determined which patient characteristics at baseline acted as confounders or effect modifiers in this relationship, thus significantly influencing the rate of change. A variable was identified as a confounder if it altered the time coefficient by more than 10%. An effect modifier was deemed significant if its interaction term in the model was statistically significant.⁴⁶ Finally, we constructed a multivariate mixed model incorporating time, along with all identified confounders and effect modifiers. This model provides insight into the unique contribution of each included variable.

Results

Demographics and Characteristics

Data of 99 people after stroke were collected in this study. Twenty-eight participants were excluded as only 1 gait measurement was collected. Of the remaining 71 participants (Table 1), 9 participants walked with and without walking aid, which were both included as individual trajectories. Three measurements showed z-scores exceeding 4 and were therefore deemed as outliers. In total, 219 walk tests were included in the analysis, of which 110 with walking aid and 109 without walking aid, resulting in an average 6.6 weeks (±3.7, range [3-18]) per trajectory. On average, participants were admitted to rehabilitation at 3.2 weeks after stroke. The speed without a walking aid significantly surpassed the observed speed with a walking aid, measuring 0.85 and 0.50 m/s, respectively.

Table 1.

Demographic and Characteristics.

Participants (N = 71)
		N	Admission	N	Discharge
Demographics
Gender	Male/female	71	38/33
Age (y)	μ (σ)	71	71.2 (13.1)
Pre-morbid walking aid	Yes/no	71	9/62
Stroke type	Ischemic/hemorrhagic/other	71	59/11/1
Stroke side	Left/right/ other	71	30/20/21
Time after stroke (weeks)	μ (σ)	71	3.2 (1.9)
Clinical tests
Berg Balance Scale^a	μ (σ)	68	39.1 (16.2)	64	50.0 (5.8)
Barthel Index^a	μ (σ)	68	13.5 (5.0)	64	18.8 (1.8)
FAC with walking aid^a	μ (σ)	47	2.5 (1.5)	39	4.5 (0.6)
FAC without walking aid^a	μ (σ)	70	2.5 (1.9)	68	3.9 (1.3)
Balance conditions
USIT: Path (m/s²)^a	μ (σ)	71	24.7 (8.3)	68	21.6 (8.2)
EO: Path (m/s²)^a	μ (σ)	71	28.6 (11.9)	68	25.1 (9.4)
EC: Path (m/s²)	μ (σ)	69	22.2 (14.7)	68	21.1 (13.3)
Gait measurements without walking aid
Speed (m/s)	μ (σ)	41	0.85 (0.26)	41	0.92 (0.26)
Asymmetry	μ (σ)	41	1.03 (0.26)	41	1.03 (0.24)
Variability	μ (σ)	41	0.07 (0.10)	41	0.06 (0.05)
Smoothness	μ (σ)	41	0.42 (0.11)	41	0.41 (0.13)
Gait measurements with walking aid
Speed (m/s)^a	μ (σ)	39	0.50 (0.21)	39	0.62 (0.23)
Asymmetry	μ (σ)	39	1.06 (0.32)	39	1.07 (0.37)
Variability	μ (σ)	39	0.13 (0.11)	39	0.09 (0.18)
Smoothness	μ (σ)	39	0.46 (0.12)	39	0.44 (0.16)

Abbreviations: USIT, sitting unsupported on a balance cushion with feet touching the ground and knees in a 90° angle, EO: standing with feet in a self-selected position, EC: standing with closed eyes and feet in a self-selected position.

Note. Not all participants were able to complete the full assessment at every time point, therefore, there is some missing data for the clinical tests and balance measurements.

Variables that were significantly different between admission and discharge were marked with a ^a.

Gait Recovery on an Individual Level From Admission to Discharge

The change scores of the individual recovery trajectories were compared to the MDC to indicate the number of trajectories that increased, decreased, or remained stable over time per gait feature (Figure 1). Out of the 80 analyzed trajectories, 31 displayed a significant change in speed, surpassing the MDC. Thirty-two of these trajectories exhibited an increase, as depicted in Figure 1. No changes over time were found for asymmetry. Regarding variability, 3 trajectories demonstrated an increase, while 5 trajectories showed a decrease larger than the MDC. Finally, in terms of smoothness, 5 trajectories exhibited an increase, whereas 9 trajectories showed a decrease over time.

Figure 1.

Changes in gait features from admission to discharge per individual trajectory with walking aid (bottom panels) and without walking aid (top panels for people after stroke during rehabilitation. The figure offers insight into the individual recovery trajectories that significantly increased, remained stable, and decreased over time. The admission point indicates the initial instance at which participants were able to walk. The trajectories that increased were denoted in green, the stable trajectories in gray, and the trajectories with a decrease in red.

Gait Recovery Over Time

In addition to analyzing the changes in gait from admission to discharge, the gait recovery trajectories can be visualized over time, including all measurements per trajectory (Figure 2). Figure 2 reveals a high individual variability in both the value at admission, and the subsequent changes over the course of rehabilitation. The data from the left panels was used to model the changes over time in the growth curve analysis. A comparison between gait features and clinical tests was made in Supplemental Figure 1.

Figure 2.

Changes in gait features over time per individual trajectory (left panels), on average (middle panels), and corrected for the initial score (right) for people after stroke during rehabilitation. The figure offers insight into both individual and aggregated gait recovery trends during rehabilitation. The reference point of weeks 0 indicates the initial instance at which participants were able to walk. The trajectories with walking aid are denoted in blue, while those without walking aid are in orange. Only a few trajectories extend beyond 12 weeks, therefore the data was visualized up to 5 measurements.

Associations Between Gait Features and Patient Characteristics

The association between patient characteristics (independent variables) and gait features (dependent variables) were examined with growth curve analysis using linear mixed models. In these models, repeated measurements per individual were included. The outcomes of the linear mixed models regarding gait speed are reported in Table 2. The linear mixed models regarding the variability, asymmetry, and smoothness are reported in Supplemental Tables 2, 3, and 4 in the supplementary materials. The FAC at baseline was not included in this analysis due to missing data. Four Measurements were excluded due to an insufficient number of observations at the measured time points, resulting in a total of 215 included measurements in 80 trajectories. First, the relationship between time and gait speed was estimated and resulted in a regression coefficient of 0.041 (SE = 0.005, P < .01) indicating that on average participants increased their gait speed with 0.041 m/s every 3 weeks. Random slopes at the participant level did not improve the model. Next, a bivariate mixed model with time and the other patient characteristics was created. Only gait speed at the first measurement (T₀) was a significant confounder (β = .927, SE = 0.027, P < .01) and an effect modifier (β = −.052, SE = 0.015, P < .01) in the association between gait speed and time. The multivariate mixed model indicated that (1) gait speed increased over time; (2) gait speed at T₀ was related to gait speed at later time points; and (3) a greater gait speed at T₀ was associated with a smaller increase over time. Although gait speed at T0 was the only confounder and effect modifier in the change of gait speed over time, the bivariate mixed-model analysis showed that most patient characteristics were significantly associated with gait speed. For example, each point increase in the BBS score at T0 corresponded, on average, to a 0.011 m/s increase in gait speed. There was no significant association between time and asymmetry, indicating that asymmetry did not change over time. Variability did decrease over time with a regression coefficient of −.01 (SE = 0.004, P = .01). There were no characteristics that affected this association. The gait smoothness also decreased significantly over time (β = −.01, SE = 0.005, P = .04). Only the BBS-score at T₀ influenced the association between gait smoothness and time.

Table 2.

Associations Between Patient Characteristics and Gait Speed.

Gait speed (*100)
		Linear mixed models
		Univariate					Multivariate
		β (SE)	β (SE)	[−CI, CI]	Z	P	β (SE)	[−CI, CI]
Time		—	4.1 (0.5)	[3.1, 5.2]	7.8	<.01	7.3 (1.2)	[5.0, 9.7]
		Bivariate
		Time	Characteristics
		β (SE)	β (SE)	[−CI, CI]	Z	P
Speed (*100)	T₀	3.6 (0.5)	92.7 (2.7)	[87.5, 98.0]	34.8	<.01	99.2 (3.2)	[92.8, 105.6]
Speed * time	T₀	7.3 (1.2)	−5.2 (1.5)	[−8.1, 9.7]	−3.4	<.01	−5.1 (1.5)	[−8.§, −2.1]
Asymmetry	T₀	4.1 (0.5)	−0.9 (11.0)	[−22.5, 20.7]	−0.1	.93
Variability	T₀	4.2 (0.5)	−147.7 (23.8)	[−194.4,−101.0]	−6.2	<.01
Smoothness	T₀	4.1 (0.5)	−105.8 (24.3)	[−153.4,−58.1]	−4.4	<.01
Walking aid	No	4.2 (0.5)	−32.9 (5.0)	[−42.7,−23.1]	−6.6	<.01
Time after stroke (wk)		4.1 (0.5)	−6.7 (1.7)	[−10.1,−3.3]	−3.9	<.01
Age (y)		4.1 (0.5)	−0.9 (0.2)	[−1.3,−0.5]	−4.1	<.01
Sex (Male)		4.1 (0.5)	1.8 (6.2)	[−10.4,14.0]	0.3	.77
Pre-morbid walking aid		4.1 (0.5)	−29.0 (8.8)	[−46.2,−11.7]	−3.3	<.01
Type
Hemorrhagic		4.1 (0.5)	11.8 (8.2)	[−4.5, 28.1]	1.4	.16
Other			19.9 (19.7)	[−18.6, 58.5]	1.0	.31
BBS	T₀	3.8 (0.5)	1.1 (0.2)	[0.8,1.5]	7.1	<.01
BI	T₀	4.1 (0.5)	3.2 (0.5)	[2.2,4.2]	6.4	<.01
USIT	T₀	4.1 (0.5)	−0.2 (0.4)	[−0.9,0.6]	−0.5	.65
EO	T₀	4.1 (0.5)	−0.8 (0.2)	[−1.2, −0.3]	−3.1	<.01
EC	T₀	4.0 (0.5)	−0.4 (0.2)	[−0.8,0.0]	−1.8	.07

Note. Variables marked in bold are confounders or effect modifiers in the rate of change of speed over time.

Discussion

In this study, we described the alterations in IMU-based gait features measured during a 2-minute walk test in clinical stroke rehabilitation. We found a high variance between subjects and within-subjects over time in gait during rehabilitation. On an individual level we found an increase in gait speed in approximately 40% of the participants. Additionally, we assessed the rate of change of speed, asymmetry, variability, and smoothness over time, and evaluated which characteristics affected this rate of change. We found significant changes over time in speed, variability, and smoothness. Of all included patient characteristics, only gait speed and the BBS-score at baseline influenced the rate of change of speed and smoothness, respectively.

We expected more individuals to improve during rehabilitation since previous research indicated that both spatio-temporal and asymmetry features significantly improved during rehabilitation.^27,47 Given that the data comprised predominantly short trajectories, with an average of 6.6 weeks per trajectory, the duration from admission to discharge may have been insufficient for many individuals to improve more than the MDC. This hypothesis is further supported the growth curve models, which included all time points in the analysis and indicated a significant change over time for speed, variability, and smoothness. However, the observed changes every 3 weeks were below the MDC. For example, estimated change in gait speed was 0.41 m/s, whereas previous research identified the MDC for gait speed as 0.137 m/s.²⁴ Consequently, the trajectory durations may have been too short to capture improve larger than the MDC.

The individual-level analysis from admission to discharge yielded different results from those observed in the growth curve models. Specifically, only a few individuals exhibited changes in variability and smoothness that exceeded the minimal detectable change (MDC) during rehabilitation. In contrast, the growth curve models demonstrated a significant overall change over time. This discrepancy can be attributed to 2 primary factors. First, the growth curve models provided insights into changes over a fixed time period, whereas the individual-level analysis reflected changes during rehabilitation, which varied in duration among participants. It is likely that individuals with longer rehabilitation periods exhibited greater improvements on average, a trend captured by the growth curve models. Second, the growth curve models utilized data from all available time points and accounted for within-individual variations, thereby increasing statistical power and sensitivity. As a result, the growth curve models were able to detect significant changes that were not evident in the individual-level analysis.

We found large between-subject differences in gait-feature scores at admission, throughout the rehabilitation and discharge. These changes were visualized in Figure 2, which illustrated that some individuals show changes in a positive direction, others in a negative direction, and some do not show changes over time. These differences are unexpected if we assume that (1) all individuals optimize functional walking during rehabilitation and (2) a walking pattern that more closely resembles that of healthy individuals results in improved functional walking. If both assumptions were correct, we would expect most individuals to change in the same direction. However, the data suggests that a linear time coefficient, as estimated in the linear mixed models, might not be applicable to all individual trajectories. This is understandable, considering that people after stroke employ diverse mechanisms to enhance their functional gait. For instance, some individuals primarily improve through neural recovery, while others use compensation strategies to overcome their impairments.^6
-9 These compensation strategies might motivate some individuals to adopt a gait pattern strongly resembling that of healthy individuals, whereas other individuals might achieve better functional walking with more abnormal gait to cope with their impairments. Future studies could take this into account by splitting the study population into subgroups, based on impairments or gait features, before analyzing the data. Secondly, future studies might add impairments as covariates to a multivariate model to identify which impairments affect the gait-features outcomes and their changes over time.

There are several studies that assessed the effect of patient characteristics on functional outcomes in people after stroke.^{29
-33,47
-50} Age appears to have a significant negative relationship with the amount of recovery.^48,49 Our results also indicated that age was negatively associated with gait speed, however, age did not influence the rate of change over time. Kollen et al²⁹ investigated the longitudinal change in walking ability, measured with the FAC. They found that standing balance was highly associated with walking ability. We were unable to replicate these results, as the BI and the IMU-based postural sway measurements did not significantly affect the rate of change. Only the BBS-score at baseline influenced the rate of change of the smoothness over time. However, this might also be due to a measurement error considering the large number of assessed variables. Nevertheless, the BBS, BI, and IMU-based postural sway measurements were significantly associated with the gait-features, indicating that balance is associated with walking performance. This finding agrees with literature that postural control is crucial for functional walking.⁵¹ In the studies by Shin et al,^28,47 the authors reported a significant longitudinal association between gait parameters and functional gait, defined by gait speed, during the early stages of stroke rehabilitation. Similarly, our study observed significant associations between post-stroke gait patterns and walking capacity, as indicated by the p-values in Table 2. Furthermore, Shin et al⁴⁷ found an increase in gait speed of 0.413 m/s over 3 months, which translates to approximately 0.064 m/s every 2 weeks, assuming a linear rate of improvement. In contrast, our study found a smaller increase of 0.041 m/s per 2 weeks. This discrepancy could be attributed to differences in the populations studied, since Shin et al’s⁴⁷ participants had a significantly lower baseline gait speed compared to ours. Since baseline gait speed influences the rate of improvement over time, as found in this study and Shin et al⁴⁷ this may explain the difference in results. Individuals with higher baseline walking speeds tend to have less room for improvement and may not prioritize further increases in speed during rehabilitation, whereas those with lower baseline speeds are more likely to focus on increasing their walking speed as a primary rehabilitation goal.

We used linear mixed models to conduct the growth curve analysis regarding the relationship between the 4 gait features and various patient characteristics in addition to time. Linear mixed models are often used to analyze longitudinal data, since they can account for multiple observations per subject; handle missing measurements and different numbers of measurements per subject; and the coefficients are interpretable in relation to the outcome.^44,45 However, the downside of linear mixed models is that they can only estimate linear relationships between the dependent and independent variables. Considering the course of gait features in individual trajectories as depicted in this study and previous research, and the fact that participants demonstrate the largest recovery early after stroke, it is likely that the gait features do not abide to a linear relationship over a long period. However, we were unable to accurately estimate higher-order polynomials due to limited measurements per individual and a relatively short time frame in which most of the participants were measured.

This study has several strengths and limitations. The first strength was that many different types of data were collected in this study, including patient characteristics, standard clinical tests, and IMU-based balance and gait measurements. This resulted in a large variation of information, which could provide new insight into the factors that are associated with the rate of change of gait features over time in people after stroke. Secondly, participants were measured on a regular basis during rehabilitation in the early stages after stroke. This allowed us to accurately map changes in gait in the first few weeks after stroke. A limitation of this study is that not all participants were able to complete the whole assessment at every measurement, which resulted in some missing data. Linear mixed models can deal with missing data using maximum likelihood estimations if the data were missing at random.⁴⁵ However, it is difficult to determine if the data was really missing at random, or a factor caused the data to be missing. The amount of missing data in this study was limited, except for the FAC, which was excluded from further analysis. Therefore, we expect the effect of the missing values to be limited. A second limitation is that it is unclear if the 4 selected gait features fully capture the gait pattern. As indicated by previous research, there are countless gait features that can be calculated from an IMU-based gait measurement.²⁴ However, to obtain interpretable results, we selected 4 gait features with a good reliability.³⁷ As an alternative to a theoretical approach of extracting gait features from IMU-data, deep learning algorithms, such as the variational autoencoder, might be capable of obtaining relevant information from raw sensor data instead of using the conventional methods.⁵² The third limitation is that many participants were measured over a relatively short period (on average 6.6 weeks) which might be insufficient to demonstrate changes on a group level. Furthermore, because of the short trajectories, we could not fit higher-order time coefficients to estimate non-linear changes over time. The fourth limitation is that we included both trajectories from 9 individuals with and without walking aid. It is possible that there is an association between the values and their development over time within subjects both with and without walking aid, which might enhance model outcomes.

We did not measure or quantify the impairments underlying the walking disabilities. Nevertheless, these specific impairments might affect the rate of recovery since some impairments might be more persistent and difficult to recover from or compensate for than others. Future studies might take this into account by also assessing specific impairments and evaluating their association with the rate of recovery over time.

Conclusion

In this study, we described the temporal changes of gait features in clinical stroke rehabilitation, focusing on speed, asymmetry, variability, and smoothness. Our analysis revealed considerable variations in IMU-based gait features at admission, during rehabilitation, and at discharge. Growth curve analysis demonstrated that speed, variability, and smoothness exhibited significant changes over time. The effect of patient characteristics, such as age and balance, on the rate of change of gait features was minimal.

Supplemental Material

sj-docx-1-nnr-10.1177_15459683241304350 – Supplemental material for Mapping Trajectories of Gait Recovery in Clinical Stroke Rehabilitation

Supplemental material, sj-docx-1-nnr-10.1177_15459683241304350 for Mapping Trajectories of Gait Recovery in Clinical Stroke Rehabilitation by Richard A. W. Felius, Michiel Punt, Natasja C. Wouda, Marieke Geerars, Sjoerd M. Bruijn, Harriet Wittink and Jaap H. van Dieën in Neurorehabilitation and Neural Repair

Supplemental Material

sj-docx-2-nnr-10.1177_15459683241304350 – Supplemental material for Mapping Trajectories of Gait Recovery in Clinical Stroke Rehabilitation

Supplemental material, sj-docx-2-nnr-10.1177_15459683241304350 for Mapping Trajectories of Gait Recovery in Clinical Stroke Rehabilitation by Richard A. W. Felius, Michiel Punt, Natasja C. Wouda, Marieke Geerars, Sjoerd M. Bruijn, Harriet Wittink and Jaap H. van Dieën in Neurorehabilitation and Neural Repair

Footnotes

Author Contributions

Richard Felius: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Software; Validation; Visualization; Writing—original draft. Michiel Punt: Conceptualization; Data curation; Formal analysis; Funding acquisition; Methodology; Project administration; Resources; Supervision; Writing—review & editing. Natasja Wouda: Data curation; Investigation; Methodology; Writing—review & editing. Marieke Geerars: Data curation; Investigation; Methodology; Writing—review & editing. Sjoerd Bruijn: Conceptualization; Formal analysis; Investigation; Methodology; Supervision; Writing—review & editing. Harriet Wittink: Funding acquisition; Project administration; Writing—review & editing. Jaap van Dieën: Conceptualization; Investigation; Methodology; Supervision; Writing—review & editing.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study is independent research and was funded by: SIA-RAAK (RAAK.PRO.03.006). SMB was funded by a VIDI grant (016.Vidi.178.014) from the Dutch Organization for Scientific Research (NWO).

Ethics Approval and Consent to Participate

All methods were performed in accordance with the declaration of Helsinki. The medical ethical review committee of Utrecht (METC number: 20-462/C) approved this study. Informed consent was obtained from all participants involved in the study.

Consent for Publication

Not Applicable.

Availability of Data and Materials

Data will be made available on request in 2024, after the research project ‘making sense of sensor data for personalized healthcare’ is finished.

ORCID iDs

Richard Anton Willem Felius

Natasja C. Wouda

Supplementary material for this article is available on the Neurorehabilitation & Neural Repair website along with the online version of this article.

References

Francisco

Zhou

Post-stroke hemiplegic gait: new perspective and insights. Front Physiol. 2018;9:1021. doi:10.3389/fphys.2018.01021

Lord

McPherson

McNaughton

Rochester

Weatherall

Community ambulation after stroke: how important and obtainable is it and what measures appear predictive?

Arch Phys Med Rehabil. 2004;85(2):234-239. doi:10.1016/j.apmr.2003.05.002

Park

Kim

TH.

The effects of balance and gait function on quality of life of stroke patients. NeuroRehabilitation. 2019;44(1):37-41. doi:10.3233/NRE-182467

Grau-Pellicer

Chamarro-Lusar

Medina-Casanovas

Serdà Ferrer

BC.

Walking speed as a predictor of community mobility and quality of life after stroke. Top Stroke Rehabil. 2019;26(5):349-358. doi:10.1080/10749357.2019.1605751

Langhorne

Coupar

Pollock

Motor recovery after stroke: a systematic review. Lancet Neurol. 2009;8(8):741-754. doi:10.1016/S1474-4422(09)70150-4

Kwakkel

Kollen

Lindeman

Understanding the pattern of functional recovery after stroke: facts and theories. Restor Neurol Neurosci. 2004;22(3-5):281-299.

Stanhope

Knarr

Reisman

Higginson

JS.

Frontal plane compensatory strategies associated with self-selected walking speed in individuals post-stroke. Clin Biomech. 2014;29(5):518-522. doi:10.1016/j.clinbiomech.2014.03.013

Cirstea

Levin

MF.

Compensatory strategies for reaching in stroke. Brain 2000;123(5):940-953. doi:10.1093/brain/123.5.940

Buurke

Nene

Kwakkel

Erren-Wolters

Ijzerman

Hermens

HJ.

Recovery of gait after stroke: what changes?

Neurorehabil Neural Repair. 2008;22(6):676-683. doi:10.1177/1545968308317972

10.

Langhorne

Bernhardt

Kwakkel

Stroke rehabilitation. Lancet. 2011;377:1693-1702. doi:10.1016/S0140-6736(11)60325-5

11.

Eng

Tang

PF.

Gait training strategies to optimize walking ability in people with stroke: a synthesis of the evidence. Expert Rev Neurother. 2007;7(10):1417-1436. doi:10.1586/14737175.7.10.1417

12.

Winstein

Stein

Arena

, et al. American Heart Association Stroke Council, Council on Cardiovascular and Stroke Nursing, Council on Clinical Cardiology, and Council on Quality of Care and Outcomes Research. Guidelines for adult stroke rehabilitation and recovery: a guideline for healthcare professionals from the American Heart Association/American Stroke Association. Stroke. 2016;47(6):e98-e169. doi:10.1161/STR.0000000000000098

13.

Tasseel-Ponche

Delafontaine

Godefroy

, et al. Walking speed at the acute and subacute stroke stage: a descriptive meta-analysis. Front Neurol. 2022;13:989622. doi:10.3389/fneur.2022.989622

14.

Middleton

Fritz

Lusardi

Walking speed: the functional vital sign. J Aging Phys Act. 2015;23(2):314-322. doi:10.1123/japa.2013-0236

15.

Harvey

RL.

Predictors of functional outcome following stroke. Phys Med Rehabil Clin N Am. 2015;26(4):583-598.

16.

Hobani

Alhakami

Uddin

Ahmad

Alsobayel

Perceived application and barriers for gait assessment in physical therapy practice in Saudi Arabia. Life (Basel, Switzerland). 2022;13(1):50. doi:10.3390/life13010050

17.

Routhier

Duclos

Lacroix

, et al. Clinicians’ perspectives on inertial measurement units in clinical practice. PLoS One. 2020;15(11):e0241922. doi:10.1371/journal.pone.0241922

18.

Balaban

Tok

Gait disturbances in patients with stroke. PM R. 2014;6(7):635-642. doi:10.1016/j.pmrj.2013.12.017

19.

Wonsetler

Bowden

MG.

A systematic review of mechanisms of gait speed change post-stroke. Part 1: spatiotemporal parameters and asymmetry ratios. Top Stroke Rehabil. 2017;24(6):435-446.

20.

Punt

Bruijn

van Schooten

, et al. Characteristics of daily life gait in fall and non fall-prone stroke survivors and controls. J Neuroeng Rehabil. 2016;13(1):67. doi:10.1186/s12984-016-0176-z

21.

Ohta

Tanabe

Katsuhira

Tamari

Kinetic and kinematic parameters associated with late braking force and effects on gait performance of stroke patients. Sci Rep. 2023;13(1):7729. doi:10.1038/s41598-023-34904-3

22.

Jarvis

Brown

Butterworth

, et al. The gait profile score characterises walking performance impairments in young stroke survivors. Gait Posture. 2022;91:229-234. doi:10.1016/j.gaitpost.2021.10.037

23.

Kim

Eng

JJ.

Magnitude and pattern of 3D kinematic and kinetic gait profiles in persons with stroke: relationship to walking speed. Gait Posture. 2004;20(2):140-146. doi:10.1016/j.gaitpost.2003.07.002

24.

Felius

RAW

Geerars

Bruijn

van Dieën

Wouda

Punt

. Reliability of IMU-based gait assessment in clinical stroke rehabilitation. Sensors. 2022;22(3):908. doi:10.3390/s22030908

25.

Gujarathi

Bhole

. Gait analysis using IMU sensor. Paper presented at: 2019 10th International Conference on Computing, Communication and Networking Technologies (ICCCNT); December 30, 2019; Kanpur, India:1-5. doi:10.1109/ICCCNT45670.2019.8944545

26.

Renggli

Graf

Tachatos

, et al. Wearable inertial measurement units for assessing gait in real-world environments. Front Physiol. 2020;11:90. doi:10.3389/fphys.2020.00090

27.

Chow

Stokic

DS.

Longitudinal changes in temporospatial gait characteristics during the first year post-stroke. Brain Sci. 2021;11(12):1648. doi:10.3390/brainsci11121648. Erratum in: Brain Sci. 2022;12(6).

28.

Shin

Lee

Spicer

Sulzer

Does kinematic gait quality improve with functional gait recovery? A longitudinal pilot study on early post-stroke individuals. J Biomech. 2020;105:109761. doi:10.1016/j.jbiomech.2020.109761

29.

Kollen

van de Port

Lindeman

Twisk

Kwakkel

Predicting improvement in gait after stroke: a longitudinal prospective study. Stroke. 2005;36(12):2676-2680. doi:10.1161/01.STR.0000190839.29234.50

30.

Meyer

Pereira

McClure

, et al. A systematic review of studies reporting multivariable models to predict functional outcomes after post-stroke inpatient rehabilitation. Disabil Rehabil. 2015;37(15):1316-1323. doi:10.3109/09638288.2014.963706

31.

Selves

Stoquart

Lejeune

Gait rehabilitation after stroke: review of the evidence of predictors, clinical outcomes and timing for interventions. Acta Neurol Belgica. 2020;120(4):783-790. doi:10.1007/s13760-020-01320-7

32.

Taylor-Piliae

Latt

Hepworth

Coull

BM.

Predictors of gait velocity among community-dwelling stroke survivors. Gait Posture. 2012;35(3):395-399. doi:10.1016/j.gaitpost.2011.10.358

33.

Felius

RAW

Geerars

Bruijn

Wouda

Van Dieën

Punt

. Reliability of IMU-based balance assessment in clinical stroke rehabilitation. Gait Posture. 2022;98:62-68. doi:10.1016/j.gaitpost.2022.08.005

34.

Berg

Measuring balance in the elderly: preliminary development of an instrument. Physiother Can. 1989;41:304-311.

35.

Collin

Wade

Davies

Horne

VT.

The Barthel ADL index: a reliability study. Int Disabil Stud. 1988;10:61-63.

36.

Holden

Gill

Magliozzi

Nathan

Piehl-Baker

Clinical gait assessment in the neurologically impaired reliability and meaningfulness. Phys Ther. 1984;64:35-40.

37.

Felius

RAW

Wouda

Geerars

Bruijn

van Dieën

Punt

. Beyond gait speed: the clinical relevance of measuring the way people walk with inertial measurement units in clinical stroke rehabilitation. BMC Neurol. 2024;24(1):129.

38.

Madgwick

SOH

Harrison

AJL

Sharkey

Vaidyanathan

Harwin

. Measuring motion with kinematically redundant accelerometer arrays: theory, simulation and implementation. Mechatronics. 2013;23(5):518-529. doi:10.1016/j.mechatronics.2013.04.003. https://centaur.reading.ac.uk/31540/. Accessed June 6, 2024.

39.

Suresh

Sridhar

Pramod

Talasila

Zero velocity potential update (ZUPT) as a correction technique. Paper presented at: 2018 3rd International Conference on Internet of Things: Smart Innovation and Usages (IoT-SIU); February 23-24, 2018; Bhimtal, India, pp.1-8. doi:10.1109/IoT-SIU.2018.8519902

40.

Patterson

Parafianowicz

Danells

, et al. Gait asymmetry in community-ambulating stroke survivors. Arch Phys Med Rehabil. 2008;89(2):304-310. doi:10.1016/j.apmr.2007.08.142

41.

Patel

Enzastiga

Casamento-Moran

Christou

Lodha

. Increased temporal stride variability contributes to impaired gait coordination after stroke. Sci Rep.2022;12(1):12679.

42.

Garcia

FDV

da Cunha

Schuch

Schifino

Balbinot

Pagnussat

. Movement smoothness in chronic post-stroke individuals walking in an outdoor environment: a cross-sectional study using IMU sensors. PLoS One. 2021;16(4):e0250100. doi:10.1371/journal.pone.0250100

43.

Härdi

Bridenbaugh

Gschwind

Kressig

RW.

The effect of three different types of walking aids on spatio-temporal gait parameters in community-dwelling older adults. Aging Clin Exp Res. 2014;26(2):221-228. doi:10.1007/s40520-014-0204-4

44.

Hair

Anderson

Babin

Multivariate Data Analysis. 7 edn. Prentice Hall; 2009.

45.

Twisk

Applied Multilevel Analysis: A Practical Guide for Medical Researchers (Practical Guides to Biostatistics and Epidemiology). Cambridge University Press; 2006. doi:10.1017/CBO9780511610806

46.

Curran

Obeidat

Losardo

Twelve frequently asked questions about growth curve modeling. J Cogn Dev. 2010;11(2):121-136. doi:10.1080/15248371003699969

47.

Shin

Lee

Spicer

Sulzer

Quantifying dosage of physical therapy using lower body kinematics: a longitudinal pilot study on early post-stroke individuals. J NeuroEngineering Rehabil. 2020;17(1):15. doi:10.1186/s12984-020-0655-0

48.

Bagg

Pombo

Hopman

Effect of age on functional outcomes after stroke rehabilitation. Stroke. 2002;33(1):179-185. doi:10.1161/hs0102.101224

49.

Kugler

Altenhöner

Lochner

Ferbert

; Hessian Stroke Data Bank Study Group ASH. Does age influence early recovery from ischemic stroke? A study from the Hessian Stroke Data Bank. J Neurol. 2003;250(6):676-681. doi:10.1007/s00415-003-1054-8

50.

Gialanella

Santoro

Ferlucci

Predicting outcome after stroke: the role of basic activities of daily living predicting outcome after stroke. Eur J Phys Rehabil Med. 2013;49(5):629-637.

51.

Winter

DA.

Human balance and posture control during standing and walking. Gait Posture. 1995;3:193-214. doi:10.1016/0966-6362(96)82849-9

52.

Felius

RAW

Punt

Wouda

, et al. Exploring unsupervised feature extration of IMU-based gait data in stroke rehabilitation using a variational autoencoder. PLoS One. 2024;19(10):e0304558.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.21 MB

0.26 MB