Post-stroke deficits in mediolateral foot placement accuracy depend on the prescribed walking task

Abstract

People with chronic stroke (PwCS) are susceptible to mediolateral losses of balance while walking, possibly due in part to inaccurate control of mediolateral paretic foot placement. We hypothesized that mediolateral foot placement errors when stepping to stationary or shifting visual targets would be larger for paretic steps than for steps taken by neurologically-intact individuals, hereby referred to as controls. Secondarily, we hypothesized that paretic foot placement errors would be correlated with previously identified deficits in isolated paretic hip abduction accuracy. 34 PwCS and 12 controls walked overground on an instrumented mat used to quantify foot placement location relative to parallel lines separated by various widths (10, 20, 30 cm). With stationary step width targets, foot placement errors were larger for paretic steps than for either non-paretic or control steps, most notably for the narrowest prescribed step width (mean absolute errors of 3.9, 2.3, and 1.9 cm, respectively). However, no differences in foot placement accuracy were observed immediately following visual target shifts, as all groups required multiple steps to achieve the new prescribed step width. Paretic hip abduction accuracy was moderately correlated with mediolateral foot placement accuracy when stepping to stationary targets (r=0.49), but not shifting targets (r=0.16). The present results suggest that a reduced ability to accurately abduct the paretic leg contributes to inaccurate paretic foot placement. However, the need to ensure mediolateral walking balance through mechanically-appropriate foot placement may often override the prescribed goal of stepping to visual targets, a concern of particular importance for narrow steps.

Introduction

People with chronic stroke (PwCS) often exhibit balance deficits, contributing to a high rate of falls during walking (Weerdesteyn et al., 2008). Many falls among PwCS can be attributed to “intrinsic factors” such as self-generated movement errors, rather than external perturbations (e.g. slips, trips, pushes) (Jørgensen et al., 2002). A substantial proportion of these falls occur sideways toward the paretic leg (Hyndman et al., 2002), suggesting that the typical strategy of ensuring mediolateral balance is disrupted.

Mediolateral walking balance among PwCS may be compromised by deficits in a foot placement stabilization strategy. Briefly, neurologically-intact individuals (controls) modulate their swing leg hip abductor activity based on pelvis dynamics, contributing to more lateral foot placement when the pelvis has a larger mediolateral displacement or velocity away from the stance foot (Bruijn and van Dieën, 2018; Rankin et al., 2014; Wang and Srinivasan, 2014). This strategy is often disrupted for steps taken with the paretic leg by PwCS (Dean and Kautz, 2015; Stimpson et al., 2019). While the relationship between pelvis motion and foot placement is likely influenced by both active control and passive body dynamics (Patil et al., 2019), post-stroke deficits in the foot placement stabilization strategy may be partially due to reduced accuracy of actively placing the paretic swing foot in the intended mediolateral location. Consistent with this idea, we previously found that a reduced ability to accurately abduct the paretic hip was correlated with wider steps (Dean et al., 2017), possibly reflecting a goal of avoiding losses of balance; errors in which the foot is placed too far laterally are less likely to cause an immediate loss of balance than errors in which the foot is placed too far medially (Heeren et al., 2013). However, our prior work did not directly assess foot placement accuracy or assess a possible link between foot placement accuracy and hip abduction accuracy.

Foot placement accuracy can be quantified by comparing foot location with visual targets on the walking surface. During overground walking with stationary visual targets prescribing step width, the accuracy and precision of mediolateral foot placement are lower for PwCS than for controls, although no significant differences have been observed between paretic and non-paretic steps (Zissimopoulos et al., 2014). Foot placement variability during treadmill walking with stationary step width targets was also higher for PwCS than controls, and was higher for paretic than non-paretic steps (Reissman and Dhaher, 2015). Foot placement accuracy has been further challenged by shifting the location of visual targets while participants walk on a treadmill, revealing that PwCS miss more targets than controls (van der Veen et al., 2020).

While PwCS generally exhibit poorer mediolateral foot placement accuracy than controls, individual differences among PwCS have also been linked to common clinical measures of function. Specifically, worse mediolateral foot placement accuracy is correlated with poorer Berg Balance Scale and Timed Up and Go scores (Ng et al., 2015), clinical tests often used to assess balance performance. Further, worse paretic mediolateral foot placement accuracy is correlated with poorer lower extremity Fugl-Meyer motor scores (Hollands et al., 2016), which in part assess the ability to perform coordinated movements with the paretic leg. These results are consistent with the possibility that deficits in paretic leg coordination can decrease foot placement accuracy, which can in turn contribute to poorer balance.

The primary purposes of this study were to: 1) extend prior investigations of post-stroke mediolateral foot placement accuracy during overground walking to include both stationary and more challenging shifting visual targets; and 2) relate mediolateral paretic foot placement accuracy to paretic hip abduction accuracy. We hypothesized that foot placement accuracy would be reduced for paretic steps in comparison to controls, and that paretic foot placement errors would be correlated with paretic hip abduction errors.

Methods

Participants

34 PwCS and 12 controls with a similar mean age participated in this study, with demographic data provided in Table 1. Inclusion criteria for the PwCS were: a stroke resulting in unilateral paresis ≥6 months before recruitment; ability to walk independently without external bracing or an assistive device; bilateral active frontal plane hip range of motion of at least 20°. Exclusion criteria were: severe visual impairment; inability to follow 3-step commands; depressive symptomology (PHQ-9>5); cardiac instability or significant medical complications limiting functional capabilities. All participants had within the prior week completed the experimental procedures to quantify hip abduction accuracy described previously (Dean et al., 2017). One participant from the prior study did not return. All participants provided informed consent using a form approved by the MUSC Institutional Review Board, and consistent with the Declaration of Helsinki.

Table 1.

Participant demographic characteristics. Age, height, and mass values are presented as means ± standard deviations.

	PwCS	Controls
Gender	14 F / 20 M	4 F / 8 M
Age (yrs)	56 ± 15	56 ± 6
Height (cm)	173 ± 11	175 ± 8
Mass (kg)	81 ± 16	87 ± 16
Paretic side	15 L / 19 R	N/A

Experimental Protocol

For all trials, participants walked over a 7-meter instrumented mat (GAITRite; Franklin, NJ, USA) used to record step location and timing at a sampling rate of 80 Hz. The walking path included 2-meters for unrecorded steps during acceleration and deceleration. Participants were instructed to walk at their comfortable speed, but were free to adjust their speed based on the perceived difficulty of stepping on visual targets (Hollands et al., 2016), unlike during treadmill walking. Participants wore a harness attached to an overhead rail that did not support body weight but would have prevented a fall in case of a loss of balance.

Participants first performed a series of nine Stationary Step Width trials in randomized order. Visual targets were presented in the form of two parallel lines projected along the length of the mat (Fig. 1). Immediately prior to each experimental session, the location of each set of parallel lines with respect to the instrumented mat was confirmed through measurement at each end of the mat. These targets prescribed mediolateral foot placement, while allowing anteroposterior foot placement to vary. Participants were instructed to step such that their midfoot landed on top of the corresponding line, and a small piece of red tape was placed on top of participants’ shoes at the geometric center of the foot to serve as a visual cue. Three trials were performed at each of three prescribed step widths: 10, 20, and 30 cm. These targets were chosen to prescribe step widths both narrower and wider than the average step width previously reported for PwCS (~17.3-18.8 cm; (Chen et al., 2005; Zissimopoulos et al., 2014)).

Figure 1.

Parallel lines projected along the length of the pressure-sensitive mat were separated by 10, 20, or 30 cm. For each step, we quantified step width as the mediolateral displacement between the two foot centers, and foot placement error as the absolute mediolateral displacement between the foot center and the target line. This is illustrated for four sample foot placement locations with the narrowest prescribed step width.

Participants then performed a series of 15 Shifting Step Width trials in randomized order. Participants were informed that one of the target lines may move while they walked and were instructed to continue to match their foot placement to the corresponding line. All trials began with the target lines at a width of 20 cm. Three trials each were performed in which: the left line narrowed to 10 cm; the right line narrowed to 10 cm; the left line widened to 30 cm; the right line widened to 30cm; neither line moved (catch trials). These targets were chosen to match the range tested in the Stationary Step Width trials. The line shifts occurred for a randomly selected step within the center 3-meters of the mat, reducing their predictability. The line shifted near-instantaneously when the contralateral foot hit the ground, as detected using pressure-sensitive switches (Motion Lab Systems; Baton Rouge, LA, USA) on the shoe sole, and recorded using a sync pulse to the instrumented mat. This timing provided participants with a full step to make the targeted adjustment, which is sufficient for clear changes in the mediolateral trajectory of the swing leg among both young and older adults (Young and Hollands, 2012).

Data Analysis

Foot placement location was calculated as the geometric center of the mat’s activated pressure sensors during stance. As depicted in Figure 1, these locations were used to calculate absolute mediolateral foot placement error and step width for each step. For the Stationary Step Width trials, we calculated the average value of these metrics across all steps. For the Shifting Step Width trials, we focused on foot placement errors for the step immediately following the shift in the visual target, to provide insight into the ability to adjust foot placement within a single step. The average walking speed for a trial was calculated based on the anteroposterior displacement and time difference from the first to the last recorded footfall.

As one of our goals was to investigate whether paretic mediolateral foot placement accuracy was related to paretic hip abduction accuracy, we compared the present results to results from an isolated hip oscillation task performed within the prior week. This task has been described in detail previously (Dean et al., 2017). Briefly, participants lay reclined on a mat table with their leg supported by a sling hanging from a low-friction track, reducing the task’s strength demands. In a series of nine randomized-order, 30-second trials with each leg, participants rhythmically moved their leg through a hip abduction-adduction range of motion in time to a metronome prescribing a movement period of 3 seconds. Participants’ vision of their legs was blocked, and they were instructed to use real-time visual feedback of their hip angle displayed on a monitor to match the movement end points to visual targets. These targets were always centered around 5° of abduction, and prescribed one of three movement amplitudes (10°, 15°, and 20° peak-to-peak) that varied across trials. With these movement parameters, following the metronome in a sinusoidal pattern with the largest prescribed amplitude would produce a peak angular velocity of ~20°/second, similar to the normative peak swing phase hip abduction velocity (Vaughan, et al., 1999). Additionally, walking with step widths of approximately 10 and 30 cm (our narrow and wide prescribed step targets) produce peak swing phase abduction angles of about 8° and 14° respectively (Kikel et al., 2020), similar to the range of peak abduction angles in our prescribed isolated task. We previously found that the accuracy of matching the prescribed abduction target with the paretic leg (absolute error in degrees, averaged across the three movement amplitudes) was positively correlated with step width (Dean et al., 2017).

Statistics

To assess gait characteristics during the Stationary Step Width and Shifting Step Width trials, we performed a series of repeated measures two-way ANOVA with interactions, as summarized in Table 2. We also performed Pearson correlations to compare individual participants’ paretic hip abduction accuracy (absolute error from a visual target during the reclined hip oscillation task described above) with their paretic mediolateral foot placement accuracy during Stationary Step Width trials (averaged across prescribed step widths), and immediately following the target shift during the Shifting Step Width trials (averaged across narrowing and widening trials). These correlations included only the subset of PwCS who were able to achieve the full prescribed range of motion and match the target movement period (within 0.5 s) during the hip oscillation task (n=24). We used an alpha value of 0.05 for all statistical comparisons. In the case of significant ANOVA effects, post-hoc Tukey tests were performed as appropriate.

Table 2.

We performed a series of repeated measures ANOVA to investigate the effects of matching mediolateral foot placement to stationary visual targets (top two rows), to investigate the effects of matching foot placement to targets that shifted at the start of a step (third row), and to determine if gait speed changed when participants anticipated a shift in target that did not actually occur (last row).

Included Trials	Independent Variable 1	Independent Variable 2	Dependent Variable(s)
Stationary Step Width	Stepping leg (control vs. non-paretic vs. paretic)	Prescribed step width (10 vs. 20 vs. 30 cm)	Absolute mediolateral foot placement error
Stationary Step Width	Group (control vs. PwCS)	Prescribed step width (10 vs. 20 vs. 30 cm)	Step width Walking speed
Shifting Step Width	Stepping leg (control vs. non-paretic vs. paretic)	Visual target shift (narrowing vs. widening)	Absolute mediolateral foot placement error
Combined	Group (control vs. PwCS)	Anticipation of shift (20 cm Stationary trials vs. Shifting “catch” trials)	Walking speed

Results

Stationary Step Width trials

During Stationary Step Width trials, absolute mediolateral foot placement error was influenced by the stepping leg and prescribed step width (Fig. 2A). Stepping leg had a significant main effect on foot placement error (p<0.0001), with paretic steps exhibiting larger errors than either non-paretic or control steps. Prescribed step width also had a significant main effect on foot placement error (p<0.0001), with larger errors for both the 10 cm and 30 cm visual targets than for the 20 cm target. These effects on foot placement error were qualified by a significant interaction between prescribed step width and stepping leg (p=0.022). Among individual post-hoc comparisons, paretic steps exhibited significantly larger errors than non-paretic and control steps only for the narrowest visual target.

Figure 2.

During Stationary Step Width trials, the prescribed step width, stepping leg, and participant group affected various aspects of participants’ walking characteristics. Here, we illustrate the effects on absolute mediolateral foot placement error (A), step width (B), and walking speed (C). Large dots indicate the group mean, rectangular boxes indicate the range of values within one standard deviation of the mean, and small dots indicate individual values. In panel B, the dashed horizontal lines indicate the prescribed step widths. For visual clarity, significant post-hoc comparisons (*) are illustrated only between stepping legs or participant groups for a given prescribed step width. Other significant post-hoc comparisons of interest are stated in the Results text.

Absolute step width varied approximately as prescribed across trials, albeit with a tendency for steps to be too wide with the narrowest target, and too narrow with the widest target (Fig. 2B). While absolute step width did not vary significantly between control and PwCS groups (p=0.14), prescribed step width did have a significant main effect on absolute step width (p<0.0001), with significant post-hoc differences between each prescribed step width. The interaction between group and prescribed step width also had a significant effect on absolute step width (p=0.015), as PwCS walked with significantly wider steps than controls only for the narrowest visual target.

Average walking speed remained fairly constant across prescribed step widths (Fig. 2C). A significant main effect of group on walking speed (p<0.0001) reflected a faster gait speed among controls than among PwCS, while prescribed step width did not have a significant main effect on walking speed (p=0.77). No significant interaction between the effects of prescribed step width and group on walking speed was detected (p=0.90).

Shifting Step Width trials

Anticipation of a visual target shift did not affect walking speed (Fig. 3). While control participants walked significantly faster than PwCS (p<0.0001), there was no difference in speed between walking with stationary medium step width targets and the catch trials in which participants anticipated a visual target shift (p=0.97). We also observed no significant interaction between these variables on walking speed (p=0.86).

Figure 3.

Walking speed differed between control participants and PwCS but did not change when participants expected a shift in the visual target location. The figure follows the same format as described for Figure 2.

Absolute mediolateral foot placement errors for the step immediately following the visual target shift were quite large for both narrowing and widening target shifts (Fig. 4A). We detected no significant main effect of the stepping leg (p=0.62) or target shift direction (p=0.29) on foot placement errors. A possible interaction effect between stepping leg and target shift direction on foot placement errors did not reach our defined level of significance (p=0.052).

Figure 4.

During Shifting Step Width trials, foot placement was generally inaccurate for the first step after the shift in visual target. For this first step, no differences in absolute mediolateral foot placement accuracy were observed between stepping legs or between the narrowing and widening conditions (A). Panel A follows the same format as used for Figures 2 and 3. Panels B and C illustrate the mediolateral foot placement location of the two steps before the visual shift (S₋₂ and S₋₁), the step that immediately followed the visual shift (S₀), and the subsequent two steps (S₊₁ and S₊₂). For visual simplicity, all steps are plotted as if the visual shift occurred immediately preceding a step with the left leg, while participants walked to the right of the figure. The dashed black lines indicate the location of the visual target corresponding to that step, illustrating the shift between steps S₋₁ and S₀. The marker colors indicate the stepping leg (control, non-paretic, or paretic), following the same color scheme as in previous figures. The large marker shape indicates which leg was used to step immediately following the visual target shift (circle = control leg; upward pointing triangle = non-paretic leg; downward pointing triangle = paretic leg). As with previous figures, large markers indicate the group mean, rectangular boxes indicate the range of values within one standard deviation of the mean, and small data points indicate individual values.

The large foot placement errors (~6 cm for a 10 cm visual shift) and lack of difference between stepping legs motivated post-hoc visual inspection of foot placement location for the steps spanning the visual shift. Immediately following the narrowing target shift (Fig. 4B), participants undershot the prescribed shift irrespective of stepping leg, placing the foot more laterally than was prescribed. For the subsequent two steps, control and non-paretic steps generally matched the target, while paretic steps appeared to be placed more laterally. Immediately following the widening target shift (Fig. 4C), participants again consistently undershot for all stepping legs, with the foot placed more medially than prescribed. However, subsequent steps generally matched the targets, with no clear differences between stepping legs. Statistical analyses were not performed for these unplanned comparisons.

Relationship between hip abduction accuracy and paretic foot placement accuracy

Across individual PwCS, isolated hip abduction accuracy was associated with foot placement accuracy during Stationary Step Width trials. Specifically, hip abduction errors exhibited a moderate positive association (r=0.49; p=0.015) with the average absolute mediolateral foot placement error for trials without a visual target shift (Fig. 5A). In contrast, no significant association (r=0.16; p=0.45) was observed between abduction errors and mediolateral foot placement errors immediately following a visual target shift (Fig. 5B).

Figure 5.

Isolated hip abduction error was moderately correlated with average absolute foot placement error for steps during Stationary Step Width trials (A), but not for steps immediately following the visual target shift during Shifting Step Width trials (B). Each dot represents an individual PwCS, and the solid lines indicate the linear best fit.

Discussion

Mediolateral foot placement was less accurate for paretic steps, albeit only for a subset of our experimental conditions. Our hypotheses were partially supported, as paretic foot placement errors relative to stationary targets were larger than control values and were correlated with isolated paretic hip abduction accuracy. However, all stepping legs were equally inaccurate when the visual target either narrowed or widened at the start of a step.

Across prescribed step widths, foot placement errors were larger for paretic steps than either non-paretic or control steps, although this individual post-hoc comparison only reached significance for the narrowest prescribed step width. This paretic step inaccuracy is consistent with prior observations of reduced foot placement accuracy and precision during overground walking (Zissimopoulos et al., 2014), and increased foot placement variability during treadmill walking with visual targets (Reissman and Dhaher, 2015). While our finding of poorer accuracy for paretic than non-paretic steps was not previously reported during overground walking, this may be attributable to our larger sample size of PwCS (n=34) than used previously (n=15) (Zissimopoulos et al., 2014). Combined across stepping legs, foot placement errors were larger for the narrow and wide targets than for the medium targets due to undershooting (Zissimopoulos et al., 2014). Such behavior may reflect a trade-off between a participant’s willingness to follow the prescribed behavior and their inherent step width preference; overly narrow steps are likely perceived as riskier due to a possible lateral loss of balance (Nonnekes et al., 2010), whereas overly wide steps increase the mechanical (Donelan et al., 2001), muscular (Kubinski et al., 2015), and metabolic (Donelan et al., 2001) demands of the task. Despite these potential challenges of walking with the narrowest and widest prescribed steps, average self-selected walking speed did not vary significantly across prescribed step widths, suggesting that the primary effects of the visual targets were in the frontal plane.

Step accuracy was notably worse for steps that immediately followed a visual target shift than for the stationary step width trials. With 10-cm target shifts, average foot placement was adjusted by only about half this amount, with no clear differences between control, non-paretic, and paretic steps. This result does not support our hypothesis, and appears inconsistent with recent findings of less accurate within-step foot placement adjustments in populations believed to have poorer motor control; specifically, while even young adults only partially adjust foot placement to mediolateral visual target shifts during a step (Zhang et al., 2020), foot placement errors during this task are notably larger for older adults (Zhang et al., 2021). However, the visual target shifts in these prior studies (Zhang et al., 2021, 2020) were intentionally small (2.5 cm) to reduce the challenge to walking balance. Perhaps the larger target shifts applied in the present study were simply too challenging to be achieved in a single step, irrespective of participant group and stepping leg. Although speculative, our exploratory analyses depicted in Figure 4B-C are consistent with this idea. By the second step after the visual shift, the foot placement location essentially matched the target location. Consistent with our observations for stationary visual targets, paretic foot placement errors for the second and third steps after the visual shift appeared to increase only for the condition that involved a decrease in prescribed step width.

We have proposed that post-stroke deficits in walking balance may be partially due to an inability to accurately modulate mediolateral paretic foot placement based on fluctuating pelvis dynamics. The absence of increased foot placement errors when stepping to challenging shifting targets with the paretic leg may seem to contradict this idea. However, the large errors observed across participant groups and stepping legs may reflect a consistent unwillingness to step to the shifted target, rather than an inability to do so. Participants may have been unwilling to make large foot placement adjustments that could cause a mismatch with body dynamics (Barton et al., 2019) (e.g. placing the foot too far medially, causing the body’s center of mass to move lateral to the base of support during the subsequent stance phase). Instead, participants may choose to “sacrifice task performance for stability” (Heeren et al., 2013; Nonnekes et al., 2010), resulting in larger errors but less of a risk of an immediate loss of balance. Future investigations of stepping accuracy should consider whether performance of a prescribed task may be worsened by participants’ internal goal of controlling the relationship between body dynamics and foot placement, which requires coordinated control of the swing and stance legs (Barton et al., 2019; Reimann et al., 2018; Zhang et al., 2020).

Among PwCS, the accuracy of isolated hip abduction was moderately correlated with the mediolateral accuracy of stepping to stationary visual targets. The strength of this correlation (r=0.49) is similar to the previously reported correlation between poorer hip abduction accuracy and wider preferred step widths (r=0.58) (Dean et al., 2017), and is consistent with the proposal that a reduced ability to accurately position the paretic limb contributes to deficits in foot placement accuracy that can influence walking balance. However, only a small proportion of the inter-participant variability in foot placement accuracy can be attributed to poor abduction accuracy. In part, this may be because hip abduction of the swing leg is only one contributor to mediolateral foot placement, which can also be influenced by transverse plane rotations of the swing leg hip and stance leg knee (Reimann et al., 2018). In addition, many PwCS exhibit abnormal co-activation of the hip abductors and knee extensors during walking (Sulzer et al., 2010), which may limit the accuracy of mediolateral foot placement but not influence the performance of an isolated hip abduction task.

A limitation of this study is the unclear functional relevance of stepping accurately on visual targets. During unperturbed walking, control participants tend not to look at their feet, and thus are unlikely to depend on visual feedback of the walking surface to modulate their foot placement during a step (Matthis et al., 2015). However, the importance of such visual feedback increases when walking near obstacles (Matthis and Fajen, 2014). Additionally, PwCS appear to have an increased reliance on visual feedback to control their balance, at least during standing posture (Marigold and Eng, 2006; Slaboda et al., 2009). While many PwCS indeed seem to look down at their feet while walking (Lamontagne and Fung, 2009), little work has systematically investigated this observation. A related study limitation is that we did not quantify participants’ visual acuity, which could have affected their ability to see the targets. To address this concern, we excluded any participants with severe visual impairment, and ensured that all participants were able to accurately match their feet to the presented targets while in a standing posture as part of the data collection orientation.

While the present study used visual targets as an assessment tool, a growing body of work applies similar methods as part of an intervention to improve walking balance in PwCS. Most notably, the C-Mill instrumented treadmill has been used to deliver an intervention termed “adaptability training” (Heeren et al., 2013) in which visual stepping targets and obstacles are presented to users, requiring the adjustment of both mediolateral and anteroposterior foot placement on a step-by-step basis. Initial results have been promising, as PwCS exhibit improved obstacle avoidance, improved scores on various clinical assessments of function, and increased physical activity following multiple training sessions (Heeren et al., 2013; van Ooijen et al., 2015). It is presently unclear whether observed improvements in foot placement accuracy can be attributed to improved accuracy of hip abduction, as investigated in the present study. Alternatively, the addition of hip abduction accuracy training to the present intervention paradigm could potentially further increase the observed benefits.

In conclusion, PwCS exhibit poorer mediolateral foot placement accuracy when stepping with their paretic leg, particularly for stationary narrow step width targets. While the assessment of foot placement accuracy can be complicated by the competing demand to ensure walking balance, such accuracy appears to be correlated with the ability to accurately abduct the paretic hip. Interventions able to overcome this inaccurate control to allow smaller foot placement errors may have benefits for improving post-stroke walking balance.