Verbal Feedback Enhances Motor Learning During Post-Stroke Gait Retraining

Abstract

Background:

Fast treadmill walking combined with functional electrical stimulation to the ankle muscles (FastFES) is a well-studied gait training intervention that improves post-stroke walking function. Although individualized verbal feedback is commonly incorporated during clinical gait training, and a variable practice structure is posited to enhance motor learning, the influence of these 2 factors on motor learning during locomotor training interventions such as FastFES is poorly understood.

Objectives:

To determine if the addition of individualized verbal feedback or variable practice to FastFES training enhances motor learning of targeted gait patterns.

Methods:

Nine individuals with post-stroke hemiparesis completed a crossover study comprising exposure to 3 dose-matched types of gait training with >3 weeks between training sessions: (1) FastFES (FF), comprising five 6-minute bouts of training with intermittent FES, (2) FF with addition of individualized verbal instructions and faded feedback delivered by a physical therapist (FF+PT), (3) FF with variable gait speed and timing of FES (FF+Var). Gait biomechanics data were collected before (Pre), immediately after (Post), and 24-hours following each type of training (Retention). Within-session and retention change scores of 3 targeted gait variables were calculated to assess locomotor learning.

Results:

FF+PT resulted in larger improvements within-session and at retention in trailing limb angle, and a trend for larger improvements in paretic leg pushoff compared to FF; whereas FF+Var failed to show greater learning of biomechanical gait variables compared to FF (control).

Conclusions:

Addition of individualized verbal feedback (FF+PT) to a single session of standardized gait training may enhance both within- and across-session learning of targeted gait variables in people post-stroke, and merits more investigation.

INTRODUCTION

Stroke is the leading cause of long-term adult disability in the United States, with residual gait dysfunction observed in a majority of stroke survivors.^1–4 Gait deficits such as reduced paretic propulsion during terminal stance, foot drop during swing, and inter-limb kinematic or spatiotemporal asymmetry can increase energy expenditure of gait, reduce gait speed and endurance, and increase the risk of falls post-stroke.^4–8 There has been considerable emphasis in rehabilitation research toward development of new treatments and refinement of existing treatments to restore post-stroke gait function.^7–9 However, there is scarcity of research on the biomechanical and motor learning processes underlying gait interventions. Individuals with unilateral stroke are capable of learning or re-learning motor skills^10–14. Previous studies have evaluated motor learning by tracking movement performance within training sessions, and retention of improved performance after a gap of minutes, hours, or days.^10–17 A majority of motor learning studies, however, focus on upper limb activities and laboratory-based, well-controlled training tasks involving repetitive practice of isolated movements, with limited clinical relevance.^{10–14,18,19} There is a research gap pertaining to evaluation of learning in the context of functionally-relevant tasks and clinically-relevant interventions directed at lower limb activities such as walking.

Instructions and verbal feedback regarding ongoing task performance are commonly used by rehabilitation clinicians to enhance learning.^20,21 Feedback provides information based on previous movement attempts intended to decrease movement errors and facilitate achievement of the movement goal in subsequent attempts.^18,20–24 Instructions can include statements imparting knowledge of performance or results pertinent to the task.^{18,20–22,25,26} During stroke gait rehabilitation, physical therapists deliver verbal instructions and feedback with high frequency, with one observational study documenting an instruction or feedback statement delivered approximately every 14 seconds.²¹ During gait training, to enhance patient engagement and learning, the clinician tries to effectively transfer knowledge to the client regarding the specific deficits that are being targeted by an intervention.^18–20,23 A faded or summary feedback schedule can also enhance learning^27–32, as providing feedback too frequently may hinder the person’s ability to process internal information and create dependency on external cues.^33,34 In addition, other factors shown to influence motor performance and motor learning include presence of an observer, type of feedback, attentional focus, and the specific verbiage used.^{18,21–23,35–37} Here, we aim to determine if and to what extent do verbal instructions and faded feedback added to a standardized, well-studied gait training paradigm enhance motor learning.

Neurorehabilitation research evidence supports task-specific training with large doses of repetitive, high-intensity stepping practice to improve mobility and walking speed post-stroke.³⁸ Neuroplasticity studies indicate that repetitive practice of appropriate movement patterns may aid with restitution of normal motor behavior, instead of relying on compensatory movement patterns.^11,19,39 Furthermore, motor learning studies, albeit largely comprising upper limb movements, suggest that incorporation of variable practice can enhance long-term retention when compared with block or constant practice.^40–44 Recent data suggest that incorporating variability in gait training programs may improve gait biomechanics and balance confidence in individuals post-stroke^45–47 and locomotor capacity in individuals with spinal cord injury⁴⁸; however, other studies fail to show an advantage of variable practice in the context of clinical gait training.⁴⁹ For instance, Rhea et. al.⁴⁹ failed to detect changes in movement variability in sagittal plane hip and knee angles immediately following a single 20-minute variable speed gait training session post-stroke. Here, we capitalized on a treadmill- and stimulation-based intervention to evaluate the effects of variable practice structure on motor learning of post-stroke gait biomechanics.

FastFES is a fairly well-studied post-stroke gait intervention that combines fast treadmill walking with functional electrical stimulation (FES) to the ankle dorsi-flexor and plantar-flexor muscles to target specific gait deficits such as reduced paretic propulsion, trailing limb angle (a measure of overall limb orientation at the stance to swing transition)⁵⁰, and foot drop.^51–53 Twelve weeks of FastFES training induced greater improvements in over ground gait endurance and energy efficiency compared to a control intervention without FES.^52,53 A single-session of FastFES has shown promise for promotion of within-⁵⁴ and across-session gait improvements (case study)¹⁵; however, factors that modulate motor learning during gait treatments such as FastFES remain unknown. Here, to evaluate strategies that can enhance motor learning during FastFES training, we systematically compared motor learning (assessed via within-session changes in gait performance and across-session retention of improved gait biomechanics 24 hours after training) during FastFES (FF), which served as the control intervention, to two other types of training: (1) FastFES with addition of individualized, targeted verbal instructions and faded feedback by an experienced physical therapist (FF+PT), and (2) FastFES modified to incorporate variable practice (FF+Var). We hypothesized that addition of verbal instructions and feedback by a physical therapist or variable practice structure would augment motor learning of gait patterns induced by FastFES in participants post-stroke.

METHODS

A repeated-measures crossover design was used to evaluate changes in gait biomechanics induced by three different types of gait training treatment sessions (FF, FF+PT, FF+Var) in participants with chronic post-stroke hemiparesis. Gait biomechanics data were collected before (Pre), immediately after (Post), and 24 hours after (Retention) each type of gait training session.

Study Participants

Nine individuals with post-stroke hemiparesis (4 males, 5 females, aged 59.0 ± 7.4 years, 35.9 ± 21.1 months post-stroke) completed the study (Table 1). All participants were ≥ 6 months post-stroke, able to walk for 5 minutes at a self-selected speed with or without an assistive device, without an ankle foot orthosis, and able to communicate with the study investigators. Because the current study data were collected before the new NIH clinical trial registration guidelines were revised and finalized (FDAAA, Section 801, 42 CFR Part 11, Final Rule for Clinical Trials Registration), the current study was not pre-registered as a clinical trial. All study procedures were approved by the institutional human subjects review board, and all participants provided written informed consent.

Table 1.

Participant Demographics and Clinical Characteristics.

Subject	Sex	Age (y)	Months Post-Stroke	Side of Hemiparesis	Type of Stroke	Berg Balance Score	Fugl-Meyer Score	OG SS Gait Speed (m/s)	TUG Time (s)	TM SS Speed (m/s)	TM Fast Speed (m/s)
SS06	F	61	14	Right	I	48	23	0.69	13.41	0.55	0.85
SS07	M	69	51	Right	H	50	22	0.43	28.05	0.42	0.57
SS08	M	52	23	Left	I	55	19	1.15	6.71	0.90	1.30
SS09	M	66	43	Right	H	45	20	0.79	13.63	0.60	0.80
SS10	F	61	72	Right	H	43	23	0.63	11.08	0.40	0.56
SS11	F	63	19	Left	I	48	20	0.52	20.49	0.40	0.54
SS12	F	55	27	Left	H	34	18	--	--	0.25	0.30
SS13	M	59	15	Left	I	27	--	0.33	38.70	0.20	0.25
SS15	F	45	59	Left	H	50	--	0.83	13.11	0.25	0.32
Mean		59.0	35.9			44.4	20.7	0.67	18.1	0.44	0.61
SD		7.4	21.1			8.8	2.0	0.26	10.5	0.22	0.33

Abbreviations: y = years. F = female. M = Male. I = Ischemic. H = Hemorrhagic. OG = over ground. SS = self-selected. TUG = Timed Up And Go. TM = Treadmill.

Overview of Study Procedures

Each participant completed a total of 6 sessions in the motion analysis laboratory. Prior to the first session, participants completed a screening session, which included informed consent procedures and a clinical examination. The clinical evaluation included a lower extremity Fugl-Meyer Score, Berg Balance Score, 10-m Walk Test to assess the overground walking speed, and the Timed Up and Go test (Table 1). The study was organized into 3 blocks of two sessions each. Each block consisted of one gait training session (FF, FF+PT, or FF+Var) with biomechanical data measured before (Pre) and immediately after (Post) each type of training session, and one retention test session conducted 24 hours after training (Retention). Pre, Post, and Retention gait tests were collected at the participant’s self-selected (SS) comfortable treadmill gait speed, with the same SS speed being used for gait tests for all subsequent sessions for each participant.

Each of the three types of gait training sessions constituted five 6-minute bouts of treadmill walking at a fast speed with FES. Prior to the first training session, the participant’s SS and fast speeds were determined. The SS speed was evaluated as a comfortable walking speed that the participant could maintain on the treadmill for 30 seconds. The fast speed was determined as the fastest treadmill speed that the participant could safely maintain for 4-minutes without a break.^51–53 Note that the Fast (training) speed was determined on the treadmill with the criterion to maintain a prolonged duration of walking, which could contribute to a mismatch between the treadmill and overground gait speeds (assessed using a short distance walking test), especially for lower-functioning stroke survivors, a methodological feature that affects treadmill-based stroke gait training studies For each of the 3 training sessions, the same Fast speed was used as the training speed initially, which could be modulated during the training session upon the participant’s request. The modulation of training speeds during or across bouts was also conducted in a similar manner across the 3 gait training sessions.

The first training block employed FastFES (FF) training as a control intervention. After at least a 3-week period, participants completed the second training block, which was randomized to either of 2 types of training: FastFES training combined with verbal faded feedback from a physical therapist (FF+PT) or FastFES training modified to include variable practice (FF+Var). Except for the systematic addition of the PT’s verbal feedback or modulation of practice variability, respectively, all other procedures used for training were identical during the 3 types of gait training sessions. At least a 3 week period was also provided between the second and third training blocks. Here, the FastFES training was always conducted first because these data were collected as part of a larger study, and to prevent the PT’s instructions from influencing the participants’ performance during the control (FF) session. Based on our previous unpublished data from a single training session of FF, which showed no retention of gait biomechanics parameters at a 48-hour retention test, in the current study, while a 1-week washout period may have been more than sufficient, we opted to include a longer 3-week washout to minimize carryover effects across the 3 types of training. All participants completed the FF and FF+PT training blocks (n = 9); however, two participants did not complete the FF+Var training block (n = 7).

Gait Analysis

Gait analysis was performed at Pre, Post, and Retention time-points as participants walked on a split-belt treadmill instrumented with 6-degree of freedom force platforms (Bertec Corporation, Columbus, OH, USA). Ground reaction forces (GRF) were recorded independently from each leg. Retroreflective markers were attached to the participants’ pelvis, as well as bilateral thigh, shank, and foot segments.⁵⁰ Marker position data were recorded at 100 Hz using a 7-camera motion capture system (Vicon Motion Systems, Inc. Denver, CO, USA) and were synchronized with GRF data sampled at 1000 Hz. For all gait tests, marker position and GRF data were recorded as the participants walked at their SS speed on the treadmill for 30 seconds without FES. The recording was started after the participant had reached steady state speed. A front handrail and overhead harness without bodyweight support provided safety.

FastFES (FF) Gait Training

The training procedures and methods for FES delivery for FF were consistent with previous FastFES studies^15,51–53. During FF training, electrical stimulation setup, including attachment of surface stimulation electrodes, footswitch setup, and determination of stimulation intensity, was completed at the start of the session to enable delivery of FES to ankle dorsiflexor muscles during paretic swing phase and plantarflexor muscles during paretic terminal double support using procedures described previously.^51–53 Footswitches were used to determine gait events for closed-loop control of FES timing during gait.^15,51–53

The training session consisted five 6-minute bouts of Fast treadmill walking. Participants were informed that the purpose of the dorsiflexor and plantarflexor FES was to assist with “lifting their toes up while the foot is in the air” and to assist with “generating more push-off with their affected leg”, respectively. FES was delivered intermittently, with alternating 1-minute periods with and without FES, totaling to 3-minutes of FES-on time and 3-minutes of FES-off time for each 6-minute bout.⁵⁴ Participants were instructed “to use their own muscle activation to assist the FES, and to practice the gait patterns being trained by FES during intervening periods without FES”. Seated rest breaks were provided between consecutive 6-minute training bouts. Thus, the active therapeutic ingredients of the FF training session were fast treadmill walking that targeted paretic trailing limb angle, FES to ankle plantarflexor muscles that targeted paretic pushoff, and FES to ankle dorsiflexors muscles that targeted paretic ankle angle during swing and initial contact.

FF+PT training:

The FF+PT training was conducted ≥ 3 weeks after the FF retention test or the FF+Var retention test, in a randomized order across participants. The training procedures, session structure, and gait speeds used for FF+PT were identical to those used for FF training. FF+PT training, however, differed in the addition of verbal feedback using a faded-feedback schedule regarding gait performance tailored to each individual participant provided by an experienced PT throughout the FF+PT session. The PT was board certified in neurologic physical therapy with > 10 years clinical experience. The PT was not present during the other 2 training sessions (FF or FF+Var). The PT was informed about the biomechanical impairments targeted by FF training, and instructed to only provide verbal feedback based on her clinical judgement to target those 3 gait parameters, i.e. paretic push-off, trailing limb angle, and foot drop. Before training, the PT had an opportunity to visually observe each participant’s baseline gait. To enhance motor learning while minimizing dependency on external cues,^27–33 the PT feedback was provided in a faded schedule; i.e. throughout the first 6-minute bout, and progressively decreased to 4-minutes, 2-minutes, 1-minute for the next 3 bouts, with no-feedback for the last (5^th) bout.

No specific pre-determined script was used to direct the content of verbal instructions and feedback provided by the PT. The PT was asked to use predominantly verbal cues with minimal or no tactile cues. For some participants, during the verbal feedback, the PT also provided a demonstration or enactment with her own legs or trunk (using actions that matched the words she was using) to enhance the participant’s understanding of the targeted gait patterns. During FF+PT, our goal was to provide feedback that was non-scripted, focused on specific gait deficits, individualized to each participant’s impairments, and determined based on an experienced clinician’s judgement. To enable documentation of the content of the feedback for each FF+PT training session, the dialog between the PT and the participant was recorded using videos and written transcripts. These transcripts were analyzed by a study team member to categorize the content of the feedback/instruction according to gait deficits being targeted (Table 2). Thus, the active therapeutic ingredients during the FF+PT training session (in addition to fast treadmill walking and FES) were individualized verbal instruction and feedback related to the specific gait parameters targeted by FF, which were provided by an experienced neuro-rehabilitation clinician in a faded schedule throughout the training session.

Table 2.

Summary of individualized instructions and feedback provided by physical therapist.

Participant	Affected Side	Feedback Focus	Phrases Used By PT during Verbal Feedback
SS06	Right	POI	“Lift hip forward and up” “Push back further with right” “Focus on pushing back and lifting” “Lift right pelvis during swing” Tactile cues were given to increase hip flexion phase
SS07	Right	POI	“Push through toes” “Hold knee soft all the way through push” “Nice step length, push through” “Resist left knee into hand, resist and shift forward, keep right knee in tiny bent position with stepping” Tactile cues given to prevent knee hyperextension
SS08	Left	AAIC	“Ease up and reach with heel” “Try not to bring your hip out (circumduction)” “Rotate left pelvis forward when swinging with left foot” “Exaggerate arm swing, step length looks good”
SS09	Right	AAIC	“Reach forward with right foot and land heel first” “Increase heel strike on right leg” “Try to keep right knee bent slightly” “Don’t land flat-footed. Toes Up. Heel Down.”
SS10	Right	POI TLA AAIC	“Think power thoughts. Think of something you want to get to and push through.” “Doing well with lifting your toes” “Longer step with left leg” “Reach with heel”
SS11	Left	POI TLA	“Increase movement at left hip and foot” “When you take steps with the right leg, look at the space between the left and right foot. Try to get the left leg in front of the right leg when you step with the left leg” “Hit ground with heel first” “Focus to step further forward and push-off with left foot”
SS12	Left	TLA	“Land with heel and then toe on right” “Lift right heel” “Try to get right foot to land softly” “Reach with heel and long steps”
SS13	Left	TLA	“Exaggerate heel down first with right leg” “Reach for the ground out front” “Heel out first on both sides” “Longer steps on the left”
SS15	Left	AAIC	“Feel your knee – keep it soft as you land” “Land with your heel first”

FF+Var Training

The FF+Var training was conducted ≥3 weeks after the FF retention test or the FF+PT retention test, in a randomized order across participants. The training procedures used for FF+Var were similar to those used for FF training, with respect to duration of training bouts (five 6-minute bouts) and methods for delivery of FES. The FF+Var session differed, however, because both the training speed and timing of FES delivery were varied throughout. In contrast to FF, which included a consistent (fast) training speed throughout each 6-minute training bout, during FF+Var the participant was informed that the treadmill speed would be varied without warning during the bout, but the speeds would be within the range of SS and Fast speeds selected by the participant at the start of the first session. During the bout, the experimenter manually modified the treadmill speed using a computer-driven speed-control software at random intervals. Because the SS and Fast treadmill speeds varied across participants, the magnitude of speed changes as well as the increment of each stepwise increase or decrease in speed differed for each participant. Additionally, the participant was informed that FES would be switched on and off at variable intervals during the bout, and a second experimenter manually switched the FES on and off using the FES control software at random intervals during the bout. Both speed and FES timing variability were randomized manually by 2 different experimenters. To match training parameters to the FF (control) training, the total duration of FES on-time was maintained at 3 minutes for each bout. Thus, the active therapeutic ingredients of the FF+Var training session (in addition to fast treadmill walking and FES) were variable training speeds and variable or unpredictable timing of FES.

Data analysis

Labeled marker trajectories and GRFs were low-pass filtered at 6 and 30 Hz, respectively (Visual 3D, C-Motion, Rockville, MD, USA). Gait events were determined using a 20-Newton vertical GRF force threshold.⁵⁵ Data from the Pre, Post, and Retention gait trials were used for analysis. Consecutive gait cycles with no marker gaps and no errors in ground reaction forces were analyzed, with the same number of strides for each individual being utilized across the 3 training types. Dependent variables are listed below.

1. Push-off integral (POI): Integral of the ground reaction forces from the onset of terminal double support to toe-off of the paretic limb.

2. Trailing limb angle (TLA): Sagittal plane angle between the lab’s vertical (Z) axis and a vector joining markers located on the lateral malleolus and the greater trochanter of the paretic lower extremity. Trailing limb angle at paretic toe-off was used for analysis

3. Ankle angle at initial contact (AAIC): Ankle dorsi- or plantar-flexion angle at paretic leg’s initial contact.

For the paretic leg post-stroke, a larger POI, larger TLA, and a more dorsiflexed ankle angle would be considered an improvement. Two change scores were computed to evaluate motor learning for each dependent variable: “within-session” = (Post - Pre), and “retention” = (Retention - Pre). A positive within-session change score indicates an improvement in the gait parameter immediately after versus before the training session, indicating online acquisition of improved gait performance. A positive retention change score indicates an improvement in the gait parameter at the 24-hour retention test versus before training, indicating motor learning of an improved gait pattern.

Statistical analysis

Two separate 2-way repeated-measures ANOVAs were used to examine the effects of training type (FF+PT and FF+Var each compared to FF) and time (Pre, Post, Retention) on each of the dependent variables (POI, TLA, AAIC). Planned comparisons were determined a priori to evaluate differences between Pre and Post as well as Pre and Retention for each training type, in the case of significant main effects. Because comparisons between FF+PT and FF+Var were not of interest, and 2 subjects’ data were missing for FF+Var, a 3-way ANOVA was not performed. If Mauchly’s test of sphericity was significant, a Huynh-Feldt adjustment for violation of sphericity was performed.

Paired samples t-tests were conducted to compare the effects of training type (FF versus FF+PT (N=9); FF versus FF+Var (N=7)) on both change scores (within-session and retention). Effect sizes were calculated using paired Hedges’ g.⁵⁶ Statistical significance was set a priori at p < .05 and all statistical analyses were performed using IBM SPSS Statistics Version 24.0 software (International Business Machines Corp., Armonk, NY, USA).

The written transcripts from the FF+PT training session were analyzed by a study team member to categorize and organize the content of the PT instructions and feedback according to the 3 targeted gait variables (Table 2). For each gait variable (POI, TLA, AAIC), we categorized study participants as to whether or not they received feedback targeting that parameter (Figure 4). Secondarily, individual-subject, descriptive analyses were conducted on the motor learning change score data to determine whether the specific gait impairment targeted by the PT instructions and feedback showed a greater magnitude of within-session and retention change scores, with FF change scores used as a comparison (Figure 4).

Figure 4.

Individual participant data for within-session change scores (top row) and across-session retention change scores (bottom row) for the 3 dependent variables in our study (POI, TLA, AAIC shown in columns from left to right); where a positive change score indicates an improvement in the gait variable. For each participant, FF change scores (as a control or reference comparison) are indicated with the white bar. If the individual participant received specific instruction targeting the dependent variable, the FF+PT change score is indicated with a black bar. If the individual participant did not receive specific instruction regarding the specific dependent variable, the FF+PT change score is indicated with a striped bar. Pie charts (top right) specific to each graph summarize the proportion of participants who showed gait improvements when they did receive specific feedback targeted to that gait variable (black) versus did not receive variable-specific feedback (striped). Note that for all 3 gait variables, participants were more likely to show improvements in a gait variable when they did (black pie chart) versus did not (striped pie chart) receive verbal feedback focused on that same variable. The symbol * against a participant indicates an improvement in the selected variable (i.e. larger change score during FF+PT than FF) with feedback and instruction (FF+PT) for that particular participant.

RESULTS

Push-Off Integral

The ANOVA comparing FF and FF+PT detected a significant main effect of training type on POI (F_(1,8) = 6.93, p = .030, hp2 = .464), where FF+PT (pooled across time points) showed significantly greater POI than the FF training type (M_diff = .0024, SE = .0009, p = .030, g = 0.27, Figure 1A). No main effect of time (F_(2,16) = 1.43, p = .269, hp2 = .152) and no training type x time interaction (F_(2,16) = 2.02, p = .165, hp2 = .202) were identified between FF and FF+PT. The 2-way repeated measures ANOVA evaluating the effect of FF or FF+Var across time (Pre, Post, Retention) revealed no main effects of training type (F_(1,6) = 1.74, p = .236, hp2 = .224), time (F_(2,12) = .33, p = .728, hp2 = .051), or interaction (F_{(2, 6.86)} = 1.02, p = .360, hp2 = .145, Figure 1B). Planned post-hoc paired t-tests showed a statistical trend for greater POI for FF+PT at Post versus Pre (M_diff = .0029, SE = .0014, p = 0.072, g = 0.28) and Retention versus Pre (M_diff = .0022, SE = .0012, p = 0.110, g = 0.24). No other paired comparisons were significant.

Figure 1.

Group means and standard deviations for the 3 dependent variables evaluated in our study (POI, TLA, and AAIC) at 3 time points (Pre, Post, and Retention). The top row of graphs show comparisons between FF and FF+PT, the bottom row shows comparisons between FF and FF+Var. * indicates a significant main effect of training type pooled across time points. A significant training by time interaction was observed for TLA. ^ and + indicate significant difference between time-points. ✝indicates a statistical trend for difference compared to Pre.

Trailing Limb Angle

A significant training type x time interaction was identified between FF and FF+PT on TLA (F_(2,16) = 4.66, p = .025, η_p² = .368, Figure 1C), as well as a significant main effect for training type (F_(1,8) = 7.90, p = .023, η_p² = .497) and time (F_(2,16) = 5.18, p = .018, η_p² = .393). For FF+PT, TLA was significantly greater at Post (M_diff = 2.69, SE = 0.741, p = .007, g = .30) and Retention (M_diff = 2.39, SE = 0.90, p = .029, g = .28) compared to Pre; whereas no differences between time-points were identified for FF training (p > .05). A significant main effect of training type was observed between FF and FF+Var (F_(1,6) = 10.13, p = .019, η_p² = .628), where FF+Var pooled across time demonstrated significantly larger TLA than FF training (M_diff = 1.81, SE = 0.57, p = .019, g = 0.19, Figure 1D). No main effect of time (F_(2,12) = 1.37, p = .292, η_p² = .185) and no interaction (F_(2,12) = .58, p = .574, η_p² = .088) were found between FF and FF+Var.

Ankle Angle at Initial Contact

A statistical trend for main effect of time on AAIC was identified for the comparison between FF and FF+PT (F_(2,16) = 3.534, p = .054, η_p² = .306, Figure 1E). A significant main effect of time was also identified in the comparison of FF and FF+Var (F_(2,12) = 15.19, p = .001, η_p² = .717, Figure 1F), where AAIC pooled across both training types at Post was more plantarflexed than at Pre (M_diff = −2.85, SE = .54, p = .002, g = .44), and at Retention (M_diff = −2.74, SE = .58, p = .003, g = 0.43). Planned comparisons revealed AAIC was more plantarflexed at Post versus Pre for FF (M_diff = 2.89, SE = .92, p = .02, g = 0.42) and FF+Var (M_diff = 2.81, SE = 1.05, p = .037. g = 0.40). No main effect for training type or interaction was identified for either FF and FF+PT or FF and FF+Var repeated measures ANOVAs (p > .05)

Effect of Training Type on Motor Learning Change Scores

The paired samples t-tests for POI change scores revealed a statistical trend for greater within-session change score (t(8) = 1.96, p = 0.086, g = 0.63) and greater retention change score (t(8) = 1.85, p = 0.102, g = 0.52) for FF+PT compared to FF (Figure 2A, Figure 2B). No significant differences between FF and FF+Var (Figure 3A, Figure 3B) were observed for either within-session or retention change scores (p> .05).

Figure 2.

Within-session change scores (top row) and across-session retention change scores (bottom row) for the FF and FF+PT (n = 9) training types for the 3 gait variables (POI, TLA, and AAIC shown in 3 columns). Motor learning change scores during FF and FF+PT training are plotted on the left axes as a Tufte slopegraph, and each paired set of observations is connected by a line, where a positive change score indicates a gait improvement. Note that a majority of participants showed a larger change score for FF+PT compared to FF (indicated by a positive slope for the lines in the slopegraph). Additionally, for each graph, Gardner-Altman estimation plots⁶⁶ show the paired hedge’s g effect size for differences between FF+PT versus FF training on a floating axis on the right, with a zero between-training difference shown as reference. The paired mean difference is plotted as a bootstrap sampling distribution. The mean difference is depicted as a dot and the 95% confidence interval is indicated by the ends of the vertical bar. The horizontal line that crosses zero (for the right axis) intersects with the average of the control (FF) change score. The horizontal line that aligns with the center of the distribution aligns with the average of the FF+PT change score. Thus, the estimation plot shows individual subject data, group averages, the effect size, and the 95% confidence interval for the between-group effect. Note that the between-training differences showed medium to large effect sizes in favor of FF+PT. *indicates significantly larger change score for FF+PT versus FF. ✝indicates a statistical trend for significant difference between FF+PT versus FF.

Figure 3.

Within-session change scores (top row) and across-session retention change scores (bottom row) for the FF and FF+Var (n = 7) training types for the 3 gait variables (POI, TLA, and AAIC shown in 3 columns). Motor learning change scores during FF and FF+Var training are plotted on the left axes as a Tufte slopegraph, and each paired set of observations is connected by a line, where a positive change score indicates gait improvement. For each graph, Gardner-Altman estimation plots⁶⁶ show the paired hedge’s g effect size for differences between FF+Var versus FF change scores on a floating axis on the right, with zero between-training difference shown as reference. The paired mean difference is plotted as a bootstrap sampling distribution with the mean difference depicted as a dot with the 95% confidence interval indicated by the ends of the vertical bar. The horizontal line that crosses zero (for the right axis) intersects with the average of the control (FF) change score. The horizontal line that aligns with the center of the distribution aligns with the average of the FF+Var change score. Note that no significant differences in change scores were observed between FF and FF+Var, and the effect sizes were smaller. ✝indicates a statistical trend for significant difference between FF+Var versus FF.

For TLA, Retention change scores were significantly greater for the FF+PT compared to FF training type (t(8) = 2.41, p = .043, g = 1.19, Figure 2D), and within-session change scores showed a statistical trend for greater values for FF+PT versus FF (t(8) = 1.84, p = 0.103, g = 0.92, Figure 2C). Additionally, a statistical trend was identified for a difference between FF versus FF+Var for within-session change (t(6) = 2.00, p = .092, g = −0.29 Figure 3C). There was no difference in retention change scores between FF and FF+Var (p > .05, Figure 3D).

For AAIC, no significant differences between training types were observed for either the within-session or retention change scores (p > .05, Figure 2E, Figure 2F, Figure 3E, Figure 3F).

Individual Subject Analysis Evaluating PT Instructions and Feedback

Using transcripts of verbal instructions and feedback from each FF+PT training session, we identified recurring themes and quotes pertinent to the targeted gait parameters for each individual participant, and summarized whether the PT’s verbal feedback focused on each of the 3 dependent gait variables (Table 2). Below, we summarize individual subject results regarding whether or not the gait variable that was the focus of verbal feedback indeed demonstrated improvements with the FF+PT training type, indicated by positive within-session and retention change scores for that parameter. For these analyses, we used the control training condition (FF) as reference.

Push-Off Integral:

Four of 9 participants received PT instruction and feedback targeting POI (Table 2). All 4 participants (100%) who received instructions and feedback focused on POI showed greater within-session improvement and greater retention change scores during FF+PT compared to FF (Figure 4A, Figure 4B). Of those 5 individuals who did not receive verbal instructions and feedback focused on POI, 3 of 5 (60%) showed greater within-session change scores and greater retention change scores during FF+PT feedback versus FF (Figure 4A, Figure 4B).

Peak Trailing Limb Angle:

Four of 9 participants received instructions and feedback targeting TLA, 3 (75%) of whom displayed greater within-session learning change scores during FF+PT versus FF (Figure 4C). All 4 participants (100%) with targeted instruction of TLA showed greater Retention change scores during FF+PT versus FF (Figure 4D). Of the remaining 5 participants who did not receive instructions and feedback targeting TLA, a majority showed greater within-session change (3 of 5 or 60% participants, Figure 4C) and greater retention scores (80% participants, Figure 4D) for TLA following FF+PT versus FF training.

Ankle Angle at Initial Contact:

Four participants were given instructions and feedback focused on improving AAIC. Three of these participants (75%) demonstrated greater within-session and 3 participants demonstrated greater retention change scores for FF+PT versus FF (Figure 4E, Figure 4F). Of the 5 participants who did not get instructions related to AAIC, only 2 (40%) presented with greater within-session (Figure 4E) and retention (Figure 4F) change scores with FF+PT versus FF.

DISCUSSION

In the context of a clinically-relevant and evidence-supported stroke gait training intervention (FastFES)^51–53, we evaluated within-session performance changes and across-session locomotor learning. For FF as well as other common gait training treatments, the time course and magnitude of motor learning of gait quality, i.e. biomechanical gait deficits, are poorly understood. Here, for the first time, we demonstrated that the addition of individual-specific verbal instructions and feedback by an experienced physical therapist (FF+PT) to a single FastFES training session (FF) resulted in larger paretic leg TLA within-session and retention change scores. Also, unlike the other 2 training types, FF+PT training demonstrated greater paretic leg TLA at Post and Retention compared to Pre. FF+PT also showed trends for greater learning of paretic POI compared to FF. For the gait variables studied, modification of the FF training session to incorporate variable practice (FF+Var) failed to show significant differences in within-session or retention change scores compared to FF. Additionally, AAIC displayed an effect of time, where, regardless of intervention, a decrement of ankle angle (greater plantarflexion or worsening of footdrop) was observed at Post compared to Pre, which was restored at the Retention test.

Our comparison of FF+PT with a rigorously controlled and dose-matched training session (FF) suggests that incorporation of individual-specific verbal instructional feedback by an experienced rehabilitation clinician may enhance motor learning of targeted gait patterns (POI and TLA) post-stroke. Notably, we showed an augmentation of learning of post-stroke gait patterns in context of only one training session, with PT feedback implemented in a way that mimics clinical settings. A significant interaction between training type and time was observed for TLA; and FF+PT was the only type of training to demonstrate significantly larger paretic leg TLA at both Post and Retention tests compared to Pre-training. We also showed greater retention change scores for TLA and POI (statistical trend) in response to FF+PT versus FF. For POI, the group data demonstrated a marked increase (statistical trend) from Pre to Post, as well as Pre to Retention for FF+PT, but not for FF. For both TLA and POI, the effect sizes for the difference in within-session and retention change scores between FF and FF+PT are high (Hedges’ effect sizes g >0.52). Because this was a preliminary, small-sample study, our results serve to show promise and the need for similar investigations. Our study findings provide preliminary support for the motor learning advantages of customized verbal instruction as an adjunct to post-stroke gait training.

Interestingly, while FF+PT showed enhancement of motor learning for paretic leg POI and TLA, we did not show improvements in AAIC in response to FF+PT. AAIC was selected as a dependent variable to determine whether dorsiflexor FES, an active ingredient of FF, induces short-term learning of improved ankle dorsiflexion at initial contact, a measure of foot drop in the paretic leg. Instead of within-session performance improvements in AAIC, irrespective of training type, we observed a decrement in AAIC (greater ankle plantarflexion angles suggesting worsening of foot drop) immediately after (Post) compared to before (Pre) training. The within-session decrement in AAIC could reflect training-induced fatigue for the dorsiflexion task. While other gait variables (POI and TLA) did not show within-session decrement, AAIC may be more susceptible to fatigue because the ankle dorsiflexors are a more fatigable muscle group compared to plantarflexors.^58,59 Furthermore, training-induced dorsiflexor muscle fatigue may be compounded by the fact that due to synchronous and random recruitment order of motor units, FES-induced contractions induce more rapid fatigue compared to physiological or volitional contractions.^60–62 These within-session decrements observed at the Post time-point ‘recovered’ by the 24-hour Retention time point, but likely interfered with motor learning, contributing to lack of improvements in AAIC at Retention compared to Pre. Potentially, a different training structure with greater rest between bouts, shorter bout-duration, or more dosage may be needed to induce within-session improvements in AAIC. Also, given that we only evaluated learning in the context of a single session, cumulative learning over multiple sessions may eventually improve AAIC. During gait training, fatigue may limit the participant’s ability to generate the muscle force required to produce appropriate dorsiflexion kinematics, which may adversely impact motor learning. However, if and to what extend muscle fatigue interacts with or influences motor learning, as well as whether there are differential effects of fatigue across multiple stroke gait deficits, are unanswered questions that merit more in-depth investigation.

Here, during the FF+Var training, while maintaining a matched duration of stepping practice, we made two modifications to the structure of the FF training session to incorporate variable practice: randomly varying the training speeds, and randomly varying the time periods with FES on versus off during each 6-minute bout. Contrary to our hypothesis, our results failed to show a significant change in motor learning of selected gait variables during FF+Var compared to FF. A major factor contributing to the lack of enhancement of motor learning by FF+Var may be that although FF+Var benefitted from the challenge imposed by variable and unpredictable gait speeds, training intensity was compromised due to the average training speed being slower (variable speeds between SS and Fast for FF+Var in contrast to the Fast speed for FF). Given that training intensity (largely determined by gait speed) is considered an important parameter influencing treatment gains in response to locomotor training³⁸, future studies should compare effects of variable practice using locomotor training paradigms that are both dose- and intensity-matched. During FF+Var, we also varied the timing of FES to add variability and unpredictability to FES-induced perturbations on gait. Based on the current results, we cannot conclude that our strategy of varying the timing of FES was helpful. Notably, because our FF+Var condition comprised simultaneous modulation of both speed and FES timing, the current study design cannot make a conclusive inference about individual contributions of these variables. However, we posit that varying the timing of the FES to make it more unpredictable may be a valuable strategy to promote motor learning, especially as participants transition from an early to a more advanced learner level. Furthermore, we suggest that in future studies, only one variable be modified at a time to tease out individual effects of speed variability and FES timing delivery. It remains to be determined whether for a study constituting a more dose- and intensity-matched comparison of FF+Var versus a control training session, and for other gait variables such as temporo-spatial parameters, stride-to-stride variability, and the functional ability to walk at a range of speeds, FF+Var will be found to have an advantage in future investigations.

A unique methodological strength of the FF+PT training was that we incorporated verbal feedback in a systematic pattern (standardized periods of feedback, faded feedback schedule, feedback focused on the specific gait deficits that FF targets), while also providing the PT flexibility to provide non-scripted, individual-specific feedback, in an attempt to imitate clinical rehabilitation settings. To further unpack individual-specific effects of feedback during FF+PT, our analyses of verbal feedback transcripts and motor learning change scores demonstrated that in individuals for whom PT verbal feedback focused on a specific gait variable (e.g. POI), positive within-session and retention change scores were also more likely to be observed in that same gait variable. These individual-specific analyses comparing FF+PT and FF data complement our statistical comparisons between FF and FF+PT, and further strengthen our conclusions regarding the motor learning benefits of verbal feedback.

Our study has several strengths as well as limitations. First, we studied the effects of variable practice and verbal feedback not as stand-alone variables, but superimposed on the FF training paradigm, which served as our control. FF is a well-studied, fairly well-controlled but complex training intervention, and it remains to be determined if similar results would be obtained for other commonly used or less standardized gait treatments. Second, in an attempt to maintain a training protocol that could be matched to the FF (control) training session, our study varied both the timing of FES and the training speed during FF+Var, preventing inferences about individual contributions of each of these two parameters. Also, several other strategies could be incorporated to implement variable stepping practice during gait training such as multidirectional stepping, variable incline, and obstacle-avoidance. Future studies should consider incorporation of greater variation of stepping practice and their effects on functional or real-world walking tasks. Third, during intermittent walking practice with FES, a feature of each of the 3 training types, we instructed the participants to utilize the periods without FES to practice the movement patterns being entrained by the FES. Whether the participants actually tried to mimic the muscle activation patterns being trained by the FES is challenging to evaluate, and could influence inter-individual variability in motor learning. Fourth, because the current study involved a single-training session, our dependent variables did not include clinical tests such as overground gait speed or TUG. Future research directions related to motor learning during gait training would benefit from collection of both functional and biomechanical outcome measures in parallel.

During the FF+PT session, we provided the PT flexibility to use her clinical judgment in determining the appropriate verbal feedback, mimicking conventional clinical training sessions, but this brings limitations related to lack of standardization or scripted verbal cues. As illustrated in Table 2, the PT used variable verbal commands, which likely added variability to the effects of FF+PT. However, this variability in and limited standardization of verbal feedback content should have likely reduced and not enhanced our chances of finding significant effects. Additionally, simply the presence of a physical therapist and research team may affect gait parameters in participants with stroke (Hawthorne effect).³⁷ Yet, though a physical therapist was present and interacting with the participant during the FF+PT training condition, investigators and observers were present during all three training conditions, minimizing the potential for the Hawthorne effect.³⁷ Other interesting comparisons of future interest may involve generic verbal motivation, feedback provided by a team member who is not a trained rehabilitation clinician, and verbal feedback by experienced versus novice rehabilitation clinicians.

As stated previously, we selected FastFES (FF) as the control intervention because a series of previous studies from our lab and others that have rigorously studied the short-term and long-term effects of FastFES training. Our control treatment (FF) itself is a fairly complex training paradigm, which includes fast walking and intermittent FES to 2 groups of muscles. Learning associated with FastFES itself is of interest, but has been studied, albeit in single-session, non-controlled paradigms. Previously, a single session of FF demonstrated improved POI within-session (at Post versus Pre)⁵⁴, and a case study on FF showed across-session retention of POI¹⁵. In contrast, a single session of FF failed to influence online changes or offline retention of biomechanics in the current study. Further investigation into the modulation of motor learning underlying single and cumulative sessions of FF is warranted. Here, our objective was to evaluate the effects of manipulating verbal feedback and practice structure on locomotor learning during FF, which is an important, clinically-impactful, and under-investigated research question. To accomplish our objective, our cross-over design capitalized on the statistical strengths of a repeated-measures design, and minimized the confounding effects of between-subject variability in baseline impairments. The FF+PT condition was designed to evaluate whether there were additive effects of targeted verbal feedback provided by a PT, and the protocol for feedback was selected based on the best current motor learning evidence, which includes a faded feedback schedule. For FF+Var, incorporating only variable speeds in one session and variable FES delivery in a separate session would have entailed too many additional crossover sessions, beyond the scope of our current study. While our study’s unique design was able to conduct a preliminary evaluation of the effects of systematically manipulating variables that affect post-stroke locomotor learning, future larger-scale studies are needed to modify one independent variable at a time to parse out individual effects of training parameters.

Our study results have potential implications for clinical neurorehabilitation focused on restoring normal gait function post-stroke. We demonstrated that verbal feedback commands, purported to be used as frequently as every 14 seconds in the clinic²¹, may indeed enhance learning of more normal gait patterns, even in the context of a single training session. Animal studies have shown that short-term neuroplasticity can be induced by a single session of practice.^63–65 Our results provide some preliminary support for similar single-session fast motor learning processes in a human clinical model using behavioral gait measures. Further, 24-hours following the gait training session, retention of gait biomechanics improvements was observed, suggesting there is potential for cumulative learning across multiple sessions, which may be ‘building blocks’ for long-term therapeutic effects. With healthcare costs and insurance imposing constraints the dosage and frequency of gait treatment sessions, developing clinically-applicable strategies to enhance or maximize motor learning within each treatment session will have significant rehabilitation impact.

In summary, we demonstrated that the addition of individualized verbal feedback and instructions (FF+PT) to a single session of a well-controlled gait training session (FF) induced both online performance improvements and motor learning (retention) of improved gait biomechanics in a small sample of individuals with post-stroke hemiparesis. Therefore, our results suggest that healthcare practitioners can confidently continue to incorporate targeted individual-specific feedback while treating in clinical rehabilitation settings. Though the incorporation of variable practice has shown promise in promoting motor learning^40–49, a single session comprising variable practice gait training in our study failed to demonstrate improvements in chronic, post-stroke gait parameters, specifically for POI, TLA, or AAIC. Additional research is warranted to parse out the effects of different types of feedback and variable practice paradigms on gait retraining, as well as to extend our present findings on a larger sample and over multiple training sessions.