University of Southern California Dissertations and Theses

The effects of fast walking, biofeedback, and cognitive impairment on post-stroke gait

(USC Thesis Other)

The effects of fast walking, biofeedback, and cognitive impairment on post-stroke gait

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content THE EFFECTS OF FAST WALKING, BIOFEEDBACK, AND COGNITIVE IMPAIRMENT
ON POST-STROKE GAIT
by
Sarah Kettlety
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(BIOKINESIOLOGY)
May 2024

ii
DEDICATION
To my late father, whose love I always carry with me.

iii
ACKNOWLEDGMENTS
I want to extend my gratitude to the many wonderful individuals who made navigating this
Ph.D. far less daunting and much more enjoyable.
I want to thank my advisor, Dr. Kristan Leech. She is a brilliant, dedicated, great mentor,
who is remarkably kind-hearted. Beyond her outstanding research and clinical skills, I want to
emphasize how her empathy and support for my personal well-being made my time in the program
much more manageable. Throughout the years, she has helped me become a better writer,
presenter, and researcher. Without her support and guidance, none of this would have been
possible.
I extend my sincere appreciation to all of my committee members. Dr. James Finley's
thoughtful feedback helped me think more deeply about my work, and his genuine enthusiasm for
science is a constant source of inspiration. Dr. Nicolas Schweighofer's guidance in modeling,
statistics, and data analysis has been invaluable to me. Dr. Carolee Winstein's diverse perspectives
and emphasis on the importance of language have helped me think more deeply about my work.
Dr. Sook-Lei Liew's feedback and analysis suggestions provided a different perspective, and her
dedication to reproducible science serves as an inspiration to me.
Thank you to past members of the Locomotor Control Lab, Dr. Russell Johnson and Dr.
Natalia Sánchez, for both providing feedback on my research and general advice. I would also like
to thank the past and present members of the Gait Rehabilitation and Motor Learning Lab and the
Locomotor Control Lab: Morgan Kelly, Maryana Bonilla Yanez, Christina Holl, Amelia Cain,
Sylwia Lipior, Giuliet Kibler, Aria Haver-Hill, Shreya Jain, Ryan Novotny, Pouria Nozari,
Catherine Yunis, Isaiah Lachica, Tara Cornwell, Aram Kim, Chang Liu, and Natalie McLain
(honorary lab member). You all have been excellent scientists to work with and great friends who

iv
make coming to work much more fun. I have learned so much working with all of you. Thank you
to Morgan, Christina, and Amelia for sharing their clinical wisdom and perspectives.
Thank you to everyone who assisted with participant recruitment: Isabel Munoz Orozco,
Christina Holl, Jason Lewis, and Elizabeth Cortez. Thank you to Maryana Bonilla Yanez, Morgan
Kelly, Leana Mosesian, and Sylwia Lipior for their assistance with data collection and scoring
clinical tests. I would also like to thank my participants, without whom this work would not be
possible.
Thank you to the Division of Biokinesiology and Physical Therapy for funding my Ph.D.
journey and to Dr. Leech for funding many of my semesters as a Research Assistant, which
provided me with the time and resources to think deeply about my research.
I thank my family and friends for their unwavering support and belief in my abilities. A
special thank you to my parents, who always believed in my dreams and provided the support
needed for me to thrive. I also would like to acknowledge my cat, Dakota. Though she doesn’t
know it, her playful nature and unconditional affection have been a source of comfort and greatly
improved my life.
Last but certainly not least, I want to express my gratitude to my husband, Adam. From the
moment we met, you have been my greatest supporter, unwavering in your belief in my
capabilities. Your encouragement has provided me with a solid foundation throughout my journey
in the program. You continuously remind me of life's priorities and bring light into every moment.
You are my ray of sunshine, and I am thankful for your presence in my life.

v
TABLE OF CONTENTS
DEDICATION ................................................................................................................................ ii
ACKNOWLEDGMENTS .............................................................................................................iii
LIST OF TABLES ........................................................................................................................ vii
LIST OF FIGURES .....................................................................................................................viii
ABBREVIATIONS......................................................................................................................xiii
ABSTRACT.................................................................................................................................. xv
CHAPTER 1 Overview................................................................................................................... 1
Specific Aims.............................................................................................................................. 3
CHAPTER 2 Background............................................................................................................... 5
Gait dysfunction post-stroke....................................................................................................... 5
Capturing overall gait biomechanics of individuals post-stroke ................................................ 8
Biofeedback-based gait training – an approach to target biomechanical impairments ............ 11
Post-stroke cognitive impairment ............................................................................................. 13
Cognition and explicit locomotor learning ............................................................................... 15
CHAPTER 3 Speed-dependent biomechanical changes vary across individual gait metrics
post-stroke relative to neurotypical adults.................................................................................... 18
Abstract..................................................................................................................................... 18
Background............................................................................................................................... 19
Methods .................................................................................................................................... 21
Results....................................................................................................................................... 26
Discussion................................................................................................................................. 37
Conclusions............................................................................................................................... 43
CHAPTER 4 Within-session changes in propulsion asymmetry have a limited effect on
overall gait asymmetry in individuals with chronic stroke........................................................... 46
Abstract..................................................................................................................................... 46
Introduction............................................................................................................................... 47
Methods .................................................................................................................................... 48
Results....................................................................................................................................... 56
Discussion................................................................................................................................. 58
Conclusion ................................................................................................................................ 64

vi
CHAPTER 5 Visuospatial skills explain differences in the ability to use propulsion
biofeedback post-stroke ................................................................................................................ 66
Abstract..................................................................................................................................... 66
Introduction............................................................................................................................... 67
Methods .................................................................................................................................... 69
Results....................................................................................................................................... 76
Discussion................................................................................................................................. 80
Conclusion ................................................................................................................................ 86
CHAPTER 6 Discussion............................................................................................................... 88
Findings from our work ............................................................................................................ 88
Impact of dissertation................................................................................................................ 89
Limitations and future work ..................................................................................................... 91
REFERENCES ............................................................................................................................. 97
APPENDIX A Minimal step length asymmetry distribution in our sample ............................... 119
APPENDIX B Test-retest reliability and minimal detectable change of the combined gait
asymmetry metric........................................................................................................................ 120
APPENDIX C Sensitivity of CGAM to variable selection ........................................................ 121

vii
LIST OF TABLES
Table 2.1. Common biomechanical impairments post-stroke………………………………..........6
Table 2.2. Definitions of commonly affected cognitive domains post-stroke…………………….14
Table 3.1. Clinical demographics of participants post-stroke…………………………………….27
Supplemental Table 3.1. Demographics of neurotypical adults…………………………………45
Table 4.1. Participant demographics…………………………………………………………......56
Table 4.2. Average normalized propulsion and change in normalized propulsion across trials......57
Table 5.1. Clinical demographics of participants included in the analyses……………………….77
Table 5.2. Model results………………………………………………………………………….79
Supplemental Table 5.1. Performance best-subsets variable selection results…………………86
Supplemental Table 5.2. Performance variability best-subsets variable selection results……...86
Supplemental Table 5.3. Immediate retention best-subsets variable selection results…………87

viii
LIST OF FIGURES
Figure 3.1. Spatiotemporal asymmetries by group across gait speeds. The figure displays
the step length asymmetry (A), double-limb support time asymmetry (B), and single-limb
support time asymmetry (C) of individual participants (red: post-stroke; black: neurotypical)
across gait speeds. The thinner, lighter lines represent the data for an individual participant
walking at 3 or 4 different gait speeds. The thicker, darker lines represent the group fits from
the robust mixed-effects model….……………………………..……………………...…………28
Figure 3.2. Kinematic gait parameters by group across gait speeds. The thinner, lighter lines
represent the data for an individual participant (red: post-stroke; black: neurotypical) walking
at 3 or 4 different gait speeds. The thicker, darker lines represent the group fits from the robust
mixed-effects model. A) Peak swing knee angle across gait speeds. Positive values indicate
knee flexion, negative values indicate knee extension. B) Trailing limb angle across gait
speeds. C) Hip hiking across gait speeds. Two participants were excluded from this analysis
due to missing iliac crest marker data. D) Circumduction across gait
speeds……………………………………………………………………………………………30
Figure 3.3. K-means clustering results across gait speeds. Individual scores for principal
component 1 vs. principal component 2 are plotted to allow for visualization of two out of
seven dimensions of the data included in the cluster analysis. Due to missing hip hiking data,
three neurotypical participants were excluded from the k-means analysis. Marker colors
represent the true group for the individual (black: neurotypical, red: post-stroke). The marker
shape represents the assigned cluster (diamond: neurotypical gait behavior cluster, circle:
stroke gait behavior cluster). Two participants post-stroke switched clusters at faster gait
speeds. One participant moved from the neurotypical gait behavior cluster to the stroke gait

ix
behavior cluster (cyan marker), and one participant moved the opposite way (royal blue
marker) ………………………………………………………………….………...……….…….32
Figure 3.4. Cluster centroids (A), and variable importance results (B & C). A) Cluster
centroids (means) for each cluster at self-selected and fast speeds. The red data represents
the stroke gait behavior cluster, the black data represents the neurotypical gait behavior
cluster. The circle markers represent the self-selected speed, and the asterisk represents the
fast speed. Data were scaled (mean = 0, SD = 1) before calculating the mean value. B)
Variable importance results at self-selected speeds. C) Variable importance results at fast
speeds. Abbreviations: TLA, trailing limb angle; SLA, step length asymmetry; DSTA,
double-limb support time asymmetry; SLSTA, single-limb support time asymmetry………........34
Figure 3.5. Lower Extremity Fugl-Meyer scores (A) and gait speeds (B) of participants poststroke in each cluster. The white-filled boxplots represent participants post-stroke in the
neurotypical cluster. The red-filled boxplots represent participants post-stroke………………….35
Figure 3.6. Between sum of squares bootstrap analysis………………………………………….37
Figure 4.1. Experimental paradigm. A) Treadmill walking paradigm. Participants walked on
the treadmill at their comfortable speed for baseline and biofeedback trials and rested as long
as needed between trials. B) Schematic anterior-posterior GRF trace and corresponding
components of the visual biofeedback display. The anterior-posterior GRF signals
represented by the dashed lines are not shown in the biofeedback display. The solid purple
section of the curve corresponds to the paretic GRF signal that was displayed in the realtime biofeedback. The blue dot indicates the peak paretic propulsion in value that was
provided as end-point biofeedback. The goal (orange dashed line) was set at +50% of the

x
difference between peak paretic propulsion and peak non-paretic propulsion at baseline. The
bounds of the goal zone were + 5 N (orange rectangle). C) Snap-shot of real-time paretic
propulsion biofeedback display. The purple bar represents the real-time anterior-posterior
GRF signal at a given time. The blue dot represents peak propulsion and appears after every
step to provide end-point feedback to the participant. The orange rectangle represents the
goal zone. Abbreviations: GRF; ground reaction force…………………………………………...51
Figure 4.2. Normalized propulsion across trials and the relationship between change in
normalized propulsion and change in propulsion asymmetry. A) Normalized propulsion
across trials. * denotes a significant increase in normalized propulsion compared to baseline.
B) Change in normalized propulsion from baseline for biofeedback trials. C) Change in
normalized propulsion vs. change in propulsion asymmetry. The black line is the group-level
model fit (fixed effect). The dashed lines represent individual model fits from the robust
mixed-effects model. Data points are data for individual strides, colored by participant.
Abbreviations: BL, baseline; BFB, biofeedback; % bw, percent body weight………..…………..57
Figure 4.3. Relationship between change in propulsion asymmetry magnitude and change
in the combined gait asymmetry metric (CGAM). The black line is the group-level model fit
(fixed effect). The dashed lines represent individual model fits from the robust mixed-effects
model. Data points are data for an individual stride, colored by participant……………………...58
Supplemental Figure 4.1. Propulsion asymmetry for all strides taken during the
biofeedback trials, coded by where the peak propulsion value was in relation to the
propulsion biofeedback goal. The green strides were within the propulsion goal zone, the
blue strides were above the goal, and the brown strides were below the goal……………………..65

xi
Figure 5.1. Experimental paradigm. A) Treadmill walking paradigm. Participants walked on
the treadmill at their comfortable speed for baseline, biofeedback, and retention trials and
rested as long as needed between trials. B) Schematic anterior-posterior GRF trace and
corresponding components of the visual biofeedback display. The anterior-posterior GRF
signals represented by the dashed lines are not shown in the biofeedback display. The solid
purple section of the curve corresponds to the paretic GRF signal that was displayed in the
real-time biofeedback. The blue dot indicates the peak paretic propulsion in value that was
provided as end-point biofeedback. The goal (orange dashed line) was set at +50% of the
difference between peak paretic propulsion and peak non-paretic propulsion at baseline. The
bounds of the goal zone were + 5 N (orange rectangle). C) Snap-shot of real-time paretic
propulsion biofeedback display. The purple bar represents the real-time anterior-posterior
GRF signal at a given time. The blue dot represents peak propulsion and appears after every
step to provide end-point feedback to the participant. The orange rectangle represents the
goal zone. Abbreviations: GRF; ground reaction force…………………………………………...72
Figure 5.2. Normalized propulsion error calculation for a representative participant. A)
Paretic propulsion across strides. The horizontal dashed lines represent the goal propulsion.
B) Propulsion error across strides. Propulsion error was calculated for each stride as the
absolute distance between peak propulsion and the closest goal zone border. Propulsion error
was then averaged over the final 30 baseline strides (between vertical dashed lines). C)
Normalized propulsion error across strides. The propulsion error value for each stride was
divided by the average baseline propulsion error and multiplied by 100 to obtain normalized
propulsion error. A normalized propulsion error of 100% is equivalent to baseline error
(horizontal dashed line). The average of the final 30 biofeedback strides was our

xii
performance outcome measure (between vertical dashed lines). The difference between the
average final thirty biofeedback and the average first immediate retention strides was our
immediate retention outcome measure…………………………………………………………...74
Figure 5.3. Biofeedback performance results. A) Relationship between
visuospatial/constructional skills, LE-FM, and normalized propulsion error averaged over
the final 30 strides of biofeedback training (n = 18). A normalized propulsion value of 100%
is equivalent to baseline propulsion error (horizontal dashed line). B) Added variable plots
for relationship between visuospatial/constructional skills (top) and LE-FM (bottom) and
normalized propulsion error. Abbreviations: RBANS, Repeatable Battery for the Assessment
of Neuropsychological Status; LE-FM, Lower-Extremity Fugl-Meyer…………………………..78
Figure 5.4. Performance variability and immediate retention results. A) Relationship
between attention and coefficient of variation (n = 18). B) Relationship between language
skills and recall error (n = 13). A recall error of 0% represents exact recall (horizontal dashed
line). Negative recall error means that participants were closer to the goal at immediate
retention than at the end of biofeedback training. Positive recall error means that participants
were further from the goal at immediate retention than at the end of biofeedback training.
The linear model fit is plotted in red, with 95% confidence intervals in gray. Abbreviation:
RBANS, Repeatable Battery for the Assessment of Neuropsychological Status…………………79

xiii
ABBREVIATIONS
% bw Percent bodyweight
AFO Ankle-foot orthosis
BFB Biofeedback
BL Baseline
CGAM Combined gait asymmetry metric
DM Delayed memory
DSTA Double-limb support time asymmetry
EF Executive function
F Female
GRF Ground reaction force
IM Immediate memory
L Left
LE-FM Lower-Extremity Fugl-Meyer
M Male
MDC Minimal detectable change
R Right
RBANS Repeatable Battery for the Assessment of Neuropsychological Status
SLA Step length asymmetry
SLSTA Single-limb support time asymmetry
SPC Single-point cane
TLA Trailing limb angle
V/C Visuospatial/constructional

xiv
VIF Variance inflation factor
Δ |PA| Change in propulsion asymmetry magnitude
Δ CGAM Change in the combined gait asymmetry metric

xv
ABSTRACT
Biomechanical gait impairments are common in individuals post-stroke. Both fast walking
(Tyrell et al., 2011) and paretic propulsion visual biofeedback (Genthe et al., 2018) improve select
post-stroke biomechanical impairments. However, it is unclear whether the improvements in these
select gait impairments lead to an improvement in overall gait biomechanics. In this work, we
aimed to understand how different rehabilitation approaches (fast walking and paretic propulsion
biofeedback) impacted overall gait biomechanics. Additionally, cognitive impairment is a common
consequence of stroke (Nys et al., 2007), and previous work has demonstrated that a measure of
global cognition is related to the acquisition and retention of a new, biofeedback-driven gait
pattern. Therefore, we also investigated the role of cognitive domain impairments in explicit
locomotor learning.
In Chapter 3, we evaluated the effect of walking speed on overall gait biomechanics in
people post-stroke compared to neurotypical adults using a k-means clustering analysis. We
hypothesized that there would be two clusters (neurotypical and post-stroke), and that walking
faster would cause the clusters to move further apart. We performed a secondary analysis with data
from 28 individuals post-stroke and 50 neurotypical adults walking at a range of speeds on a
treadmill. We performed k-means clustering at both self-selected and fast speeds using seven gait
metrics (step length asymmetry, double-limb support time asymmetry, single-limb support time
asymmetry, peak swing knee flexion angle, trailing limb angle, hip hiking, and circumduction).
We found two distinct clusters representative of neurotypical gait behavior (comprised of both
neurotypical and post-stroke participants) and stroke gait behavior (comprised of only post-stroke
participants). People post-stroke in the stroke gait behavior cluster walked at similar speeds but
had greater motor impairment compared to the people post-stroke in the neurotypical gait behavior

xvi
cluster. At faster speeds, the distance between clusters did not change, suggesting that fast walking
did not improve nor degrade overall gait biomechanics after stroke. These analyses demonstrate
that while fast walking may reduce the magnitude of some kinematic impairments relative to one’s
habitual walking pattern, the resulting gait kinematics are not more similar to neurotypical adults
walking at like speeds.
In Chapter 4, we investigated how changes in propulsion asymmetry impacted overall gait
asymmetry, measured by the combined gait asymmetry metric (CGAM) in individuals post-stroke.
The CGAM provides a single comprehensive and easily interpretable measure of overall kinematic
and spatiotemporal gait asymmetry (Ramakrishnan et al., 2018). We hypothesized that reducing
propulsion asymmetry would lead to a reduction in CGAM. Twenty-three participants completed
a baseline treadmill walking trial, then twenty minutes of biofeedback treadmill training to increase
paretic propulsion. We calculated the change in propulsion asymmetry magnitude (Δ |PA|) and
change in CGAM (Δ CGAM) from baseline during biofeedback training. We examined the
relationship between propulsion asymmetry magnitude and the corresponding change in CGAM
by fitting a robust linear mixed-effects model with Δ CGAM as the outcome, a fixed effect for Δ
|PA|, and a random intercept and random slope for each participant. We found a positive association
between Δ |PA| and Δ CGAM (intercept β = -1.6, p = 0.49; CGAM β = 3.0, p = 0.002). The average
change in propulsion asymmetry magnitude in our sample was -0.07, suggesting that, on average,
we would expect to see a CGAM reduction of 1.8. A CGAM change of 1.8 is relatively small,
particularly since the average CGAM of our sample during baseline was 41.6 ± 30.1. These results
indicate that while propulsion biofeedback can be used to reduce propulsion asymmetry, it is
unlikely to produce meaningful reductions in overall gait asymmetry. Therefore, propulsion

xvii
asymmetry may not be an ideal target for biofeedback interventions designed to improve overall
gait asymmetry.
In Chapter 5, we aimed to understand which cognitive domains were associated with
locomotor performance during a visual biofeedback task and immediate recall of the newly learned
walking pattern in individuals post-stroke. Because there is little research informing our
understanding of the relationship between specific cognitive domains and biofeedback-driven
changes in gait impairments, we performed an exploratory analysis to understand which cognitive
domains contribute to locomotor performance, variability during performance, and immediate
retention when using visual biofeedback. We considered the following six commonly impacted
domains of cognition: immediate memory, visuospatial/constructional skills, language, attention,
delayed memory, and executive function. Participants post-stroke completed cognitive testing,
which provided scores for different cognitive domains, including executive function, immediate
memory, visuospatial/constructional skills, language, attention, and delayed memory. Next,
participants completed a single session of paretic propulsion biofeedback training, where we
collected treadmill-walking data for twenty minutes with biofeedback and two minutes without
biofeedback. We fit separate regression models to determine if cognitive domain scores, motor
impairment (measured with the lower-extremity Fugl-Meyer), and gait speed could explain
propulsion error during biofeedback use and recall error during immediate retention. We found
that visuospatial/constructional skills and motor impairment best-explained propulsion error
during biofeedback use, and attention best-explained performance variability. Language skills
best-explained recall error during immediate retention. These results demonstrate that information
on specific cognitive domain impairments explains inter-individual variability in locomotor
learning outcomes.

xviii
Together, these studies demonstrate 1) rehabilitation approaches that improve single
biomechanical impairments do not necessarily lead to improvements in overall gait biomechanics
and 2) cognitive impairment, specifically visuospatial/constructional skills, attention, and
language skills, are related to explicit locomotor learning. Further work is needed to understand
how to effectively manipulate overall gait biomechanics, and how to use personal factors, such as
cognitive impairment, to predict individual responsiveness to an intervention.

CHAPTER 1
OVERVIEW
In the United States, stroke is a leading cause of long-term disability (Centers for Disease
Control and Prevention (CDC), 2009), affecting approximately 7.6 million people (Tsao et al.,
2022). Walking activity limitations (i.e., reduced gait speed and distance) and biomechanical gait
impairments (i.e., step length and propulsion asymmetries) are common after a stroke (Olney &
Richards, 1996). Biomechanical impairments are associated with increased metabolic cost (Awad,
Palmer, et al., 2015; Finley & Bastian, 2017; Stoquart et al., 2008) and fall risk (Burpee & Lewek,
2015; Matsuda et al., 2017; Wei et al., 2017). Activity limitations are related to reduced community
ambulation and quality of life (Grau-Pellicer et al., 2019). Individuals post-stroke cite improving
overall gait biomechanics and activity limitations as important rehabilitation goals (Bohannon et
al., 1991).
To reduce activity limitations, one common treatment approach is high-aerobic intensity
gait training, which can be achieved by walking faster on a treadmill (Hornby et al., 2020). While
this approach does not explicitly prioritize biomechanical impairment reduction, walking faster
improves select biomechanical impairments, such as reducing step length asymmetry and
increasing trailing limb angle (Tyrell et al., 2011). However, since biomechanical impairments
rarely occur in isolation, it is difficult to understand the effect of fast walking on overall gait
biomechanics from prior work that reports changes in select impairments that are examined
separately (Tyrell et al., 2011). Investigating the effect of fast walking on a single measure that
captures multiple dimensions of gait will enable a more comprehensive understanding of how fast
walking affects overall gait biomechanics after a stroke.

2
Biofeedback-based gait training is one approach to specifically target post-stroke
biomechanical impairments (Tate & Milner, 2010). Biofeedback leverages an explicit form of
motor learning by providing extrinsic information about a selected gait variable for participants to
use to change an aspect of their gait (Giggins et al., 2013). Participants have successfully used
visual biofeedback of step length and paretic propulsion to improve the targeted gait variable while
the biofeedback is displayed (Padmanabhan et al., 2019; Sánchez & Finley, 2018), and after the
biofeedback is removed (French et al., 2021; Genthe et al., 2018). However, most studies only
measure the change in the selected biofeedback variable (i.e., propulsion or step length) and do
not investigate the biofeedback’s effect on other biomechanical impairments (French et al., 2021;
Genthe et al., 2018; Sánchez & Finley, 2018). Providing biofeedback of a single gait variable may
only alter the targeted variable and not address the remaining impairments or may exacerbate
existing impairments, highlighting the importance of understanding how altering a single
biomechanical impairment influences overall gait biomechanics.
Using visual biofeedback to alter gait impairments is an explicit, cognitively demanding
motor learning task (Leech et al., 2022). Participants are given external information about their
task performance and asked to reduce their movement error (Spencer et al., 2021). This requires
cognitive processing, as participants must understand the information contained in the
biofeedback, its relationship to their movements, and the instructions for the task. Cognitive
impairment is common post-stroke (Nys et al., 2007), with deficits in visuospatial skills, attention,
executive function, language, and memory among the most commonly affected domains (Jokinen
et al., 2015; Pinter et al., 2019; Yang et al., 2020). Previous work found that participants with
higher fluid cognition scores had better acquisition and 24-hour retention of a biofeedback-driven
gait pattern (French et al., 2021). However, fluid cognition is a global measure of cognition

3
reflecting someone’s ability to adapt to new information and solve problems (Heaton et al., 2014);
therefore, it is still unknown which specific cognitive domains are important for acquiring and
retaining an explicit locomotor skill.
This work aims to understand the immediate effects of fast walking and visual biofeedback
on overall gait biomechanics in individuals post-stroke. Additionally, we will investigate which
cognitive domains are associated with performance during a visual biofeedback task and
immediate recall of the newly learned walking pattern. Collectively, these efforts will inform a
more effective and personalized rehabilitation approach for individuals post-stroke.
Specific Aims
Aim 1: Evaluate the effect of gait speed on overall gait biomechanics in people poststroke compared to neurotypical adults. We performed a secondary analysis of data from three
previously published studies (Finley & Bastian, 2017; Fukuchi et al., 2018; Tyrell et al., 2011) of
participants walking at multiple speeds. We evaluated the effect of speed on all kinematic gait
metrics simultaneously using a k-means clustering analysis (overall gait biomechanics measure)
to determine how the difference in overall gait biomechanics between people post-stroke and
neurotypical adults changes with speed. We hypothesized that there would be two clusters
(neurotypical and post-stroke). Additionally, based on studies that demonstrate abnormal muscle
coactivation patterns post-stroke during walking (Clark et al., 2010; Knutsson & Richards, 1979;
Shiavi et al., 1987) and maximal hip extension torque production (Sánchez et al., 2017), we
hypothesized that walking faster would cause the clusters to move further apart.
Aim 2: Understand how changes in propulsion asymmetry affect the combined gait
asymmetry metric in people post-stroke. We used propulsion biofeedback to manipulate
propulsion asymmetry in people post-stroke. We evaluated the change in combined gait asymmetry

4
metric (CGAM; overall gait biomechanics measure) relative to the change in propulsion
asymmetry. Because of propulsion’s relationship with numerous biomechanical impairments
(Dean et al., 2020; Lewek & Sawicki, 2019) and sagittal kinematic gait asymmetry (Padmanabhan
et al., 2020), we hypothesized that changes in propulsion asymmetry would reduce CGAM.
Aim 3: Understand which cognitive domains were associated with locomotor
performance during a visual biofeedback task and immediate recall of the newly learned
walking pattern in individuals post-stroke. Because there is little research informing our
understanding of the relationship between specific cognitive domains and biofeedback-driven
changes in gait impairments, we performed an exploratory analysis using the data collected for
Aim 2 to understand which cognitive domains contribute to locomotor performance and immediate
recall during a visual biofeedback task. We considered the following six domains of cognition,
immediate memory, visuospatial/constructional skills, language, attention, delayed memory, and
executive function.

5
CHAPTER 2
BACKGROUND
In the United States, stroke is a leading cause of long-term disability (Centers for Disease
Control and Prevention (CDC), 2009), affecting approximately 7.6 million people (Tsao et al.,
2022). A stroke occurs either when blood flow to the brain is blocked (ischemic stroke) or a
ruptured artery causes a bleed in the brain (hemorrhagic stroke). Both types of strokes cause
damage and death to neurons in the brain. This brain damage may lead to various outcomes,
including hemiparesis, sensory deficits, speech difficulties, anxiety, depression, and cognitive
decline. Stroke is a growing problem in the United States; an estimated 3.4 million additional
adults will have had a stroke by 2030 (Tsao et al., 2022). The disability from a stroke causes
reduced work productivity, leading to high indirect costs ($19.4 billion in 2017; Tsao et al., 2022).
Stroke survivors also report lower quality of life than neurotypical adults (Jaracz & Kozubski,
2003), which declines over time (Dhamoon et al., 2009, 2010, 2012).
Gait dysfunction post-stroke
A common manifestation of post-stroke disability is gait dysfunction, affecting
approximately 80% of stroke survivors (Wade et al., 1987). Common components of gait
dysfunction include activity limitations and biomechanical impairments; improving both are top
goals for individuals post-stroke (Bohannon et al., 1991). Activity limitations include decreased
gait speed and distance. Measuring and tracking gait speed via the 10-meter walk test and distance
via the 6-minute walk test are feasible and reliable measures (Fulk et al., 2008; Fulk & Echternach,
2008; J. L. Moore et al., 2018). Decreased gait speed and distance are associated with reduced
community ambulation and quality of life (Grau-Pellicer et al., 2019). Improvements in gait speed
are associated with increased independence (Schmid et al., 2007) and ability to return to work

6
(Jarvis et al., 2019). Additionally, stroke survivors with slower walking speeds and shorter walking
distances take fewer steps per day (Miller et al., 2021) and have higher mortality rates (Newman
et al., 2006).
The current gold standard treatment approach for addressing activity limitations in chronic
stroke survivors is moderate- to high-aerobic intensity gait training, specifically to increase gait
speed and distance (Hornby et al., 2020). Moderate-aerobic intensity can be defined as between
40 – 60% of the heart rate reserve, and high-aerobic intensity can be defined as greater than 60%
(Boyne et al., 2023). These heart rate reserve ranges can be achieved through various tactics, such
as increasing walking speed or incline on a treadmill. High-intensity interval training over twelve
weeks improved gait speed and distance individuals post-stroke (Boyne et al., 2023). The
hypothesized mechanism for the meaningful improvements in walking outcomes seen with highintensity training is from improvements in VO2 peak (Pang et al., 2013) and locomotor efficiency
(J. L. Moore et al., 2010). High-aerobic intensity walking also increases brain-derived neurotrophic
factor and corticospinal excitability, which may enhance use-dependent learning post-stroke
(Boyne et al., 2019; Kleim & Jones, 2008; Leech et al., 2018; Nepveu et al., 2017).
Individuals post-stroke also commonly present with biomechanical gait impairments (see
Table 2.1 for common biomechanical impairments; Chen et al., 2005; Olney & Richards, 1996).
Biomechanical impairments are quantified using a marker-based motion capture system and force
plates (S. A. Moore et al., 2022).
Table 2.1. Common biomechanical impairments post-stroke (Chen et al., 2005; Olney & Richards, 1996).
Kinematic Kinetic Spatiotemporal
↑ hip hiking ↓ propulsion ↑ step length asymmetry
↓ trailing limb angle ↑ braking impulse ↑ swing time asymmetry
↓ hip flexion ↑ stance time asymmetry
↓ swing knee flexion ↑ double-support time asymmetry
↑ circumduction ↑ single-limb support time asymmetry
↓ ankle dorsiflexion

7
Relationship between biomechanical impairments and activity limitations
Numerous biomechanical impairments have been associated with activity limitations and
other important outcome measures related to walking activity. For example, step length asymmetry
and propulsion, which are common targets for clinical and research interventions, have been
related to metabolic cost (Awad, Palmer, et al., 2015; Finley & Bastian, 2017; Penke et al., 2019),
gait speed (Balasubramanian et al., 2007; Bowden et al., 2006), gait distance (Awad, BinderMacleod, et al., 2015; Awad et al., 2020; Ryan et al., 2020), balance (Lewek et al., 2014), and
hemiparetic severity (Balasubramanian et al., 2007; Bowden et al., 2006; Roelker et al., 2019).
Falls have been related to single limb support time asymmetry (Wei et al., 2017) and knee flexion
(Burpee & Lewek, 2015; Matsuda et al., 2017). Metabolic cost has been related to stance time
asymmetry (Ryan et al., 2020). Walking speed has been related to hip hiking, circumduction, ankle
dorsiflexion, and knee flexion (Stanhope et al., 2014), and balance has been related to swing time
asymmetry (Lewek et al., 2014). The results of these studies suggest that there may be a link
between outcomes related to walking activity and biomechanical impairments, and that targeting
biomechanical impairments may impact these measures.
However, recent evidence has highlighted that the relationships between biomechanical
impairments and outcomes related to walking activity are inconsistent across studies and
heterogeneous across individuals (McCain et al., 2021; Nguyen et al., 2020; Padmanabhan et al.,
2020; Sánchez & Finley, 2018; Wonsetler & Bowden, 2017). For example, Nguyen et al. (2020)
found that increasing and decreasing step length asymmetry did not alter metabolic cost in
individuals post-stroke. Sanchez and Finley (2018) found that individual characteristics such as
the direction of the step length asymmetry affect the relationship between step length asymmetry
changes and metabolic cost. Additionally, improvements in biomechanical impairments, such as

8
step length asymmetry, are not associated with improved outcomes related to walking activity such
as balance and metabolic cost, after a long-term intervention (Ryan et al., 2020). The inconsistency
in results across the literature suggests that we still do not fully understand the link between
biomechanical impairments and outcomes related to walking activity and how to best leverage this
relationship to make meaningful improvements to an individual’s gait pattern.
Capturing overall gait biomechanics of individuals post-stroke
Asymmetric gait is common post-stroke and is often quantified by calculating asymmetry
in a single biomechanical impairment (i.e., step length asymmetry). However, a single
biomechanical impairment rarely presents in isolation after a stroke; rather, numerous gait
biomechanical impairments are evident. For example, after a stroke, someone may walk with
increased paretic hip hiking, reduced paretic knee flexion, and reduced paretic trailing limb angle.
Describing the symmetry between paretic and non-paretic sides will not give you information
about the other impairments. The individual may be using a compensatory measure (e.g., increased
hip hiking) to achieve the desired rehabilitation goal (e.g., increased toe clearance). In this
scenario, the targeted biomechanical impairment has improved, but the changes in other
biomechanical impairments (i.e., hip hiking) may have an adverse effect on overall gait
biomechanics. If you only measured the targeted biomechanical impairment, you would miss these
compensations. One possible solution is to quantify the changes in multiple biomechanical
impairments that are not directly targeted by an intervention. However, it is challenging to interpret
the changes in multiple biomechanical impairments, particularly if the direction of change is
different across the impairments. For example, if your intervention increased toe clearance
(typically a desired outcome; Matsuda et al., 2017) but also increased hip hiking (typically a less

9
desired outcome; Awad et al., 2017), how do you interpret the effect of your intervention on overall
gait biomechanics?
To simplify the interpretation of the numerous biomechanical impairments that can be
present in post-stroke gait, the third Stroke Recovery and Rehabilitation Roundtable recommends
including multivariate gait measures to provide a more comprehensive and interpretable measure
of walking recovery (Van Criekinge et al., 2023). Researchers have developed numerous ways to
more comprehensively quantify gait biomechanics such as the Gait Deviation Index (Schwartz &
Rozumalski, 2008), Gait Profile Score (Baker et al., 2009), interlimb asymmetry (Padmanabhan
et al., 2020), k-means clustering analysis (Abbasi et al., 2021), and the combined gait asymmetry
metric (Ramakrishnan et al., 2018, 2019). Throughout my dissertation, I have chosen to use two
measures to quantify overall gait biomechanics, a k-means clustering analysis and the combined
gait asymmetry metric (Ramakrishnan et al., 2018, 2019). I have chosen these two options for
numerous reasons. First, they do not require large amounts of normative data (as the Gait Deviation
Index and Gait Profile Score do; Baker et al., 2009; Schwartz & Rozumalski, 2008). Second, they
capture kinematic, kinetic, and spatiotemporal impairments (interlimb asymmetry only captures
sagittal plane kinematic impairment; Padmanabhan et al., 2020). I use a k-means clustering
analysis when I only have data for the paretic limb in individuals post-stroke and also have data
for neurotypical adults. I use the combined gait asymmetry metric when I have data for both paretic
and non-paretic limbs, but do not have data for neurotypical adults. An additional benefit of the
CGAM is that it has excellent test-retest reliability, and a small minimal detectable change
(Appendix B).
The combined gait asymmetry metric (CGAM) is a multidimensional gait asymmetry
metric developed by Ramakrishnan et al. (2018) that is based on a modified Mahalanobis distance.

10
The CGAM allows the inclusion of any biomechanical impairments of interest and provides a
single measure for multidimensional gait asymmetry that is bounded between 0 (no asymmetry)
and 200 (complete asymmetry). A value of zero represents perfect symmetry, and the greater the
magnitude of the CGAM value, the greater asymmetry that is present in the gait pattern. However,
the ability to choose which impairments to include limits comparison between studies because it
is not appropriate to compare CGAM values with different biomechanical impairments included
(Ramakrishnan et al., 2018, 2019). The impact of including different variables in the CGAM
calculation is demonstrated in Appendix C.
When a consistent set of biomechanical impairments is used to calculate it, the CGAM has
a powerful utility to evaluate the biomechanical effects of an intervention. Consider a study
designed to investigate the impact of a powered exosuit that supplements paretic propulsion in
individuals after stroke. Traditionally, this study's primary outcome measures may focus on paretic
propulsion and trailing limb angle since these are outcomes directly related to the goal of the
intervention. However, there is often no attempt to understand the impact of the intervention on
other important aspects of the gait pattern. By adding the CGAM as an outcome measure, you can
include other variables of interest in your population (e.g., step length, knee flexion, hip angles,
etc., for individuals post-stroke) and develop a more comprehensive understanding of the impact
of the intervention on overall gait biomechanics. The CGAM can also help us begin to understand
the relationship between overall gait asymmetry and outcomes related to walking activity after
stroke. Previous work has shown that a measure of sagittal gait asymmetry is related to the
metabolic cost of walking (Padmanabhan et al., 2020), and that CGAM (calculated using step
length, step time, swing time, double limb support, and vertical ground reaction force) is
moderately correlated with gait distance and speed (Ramakrishnan et al., 2019). This suggests that

11
overall gait biomechanics may be associated with outcomes related to walking activity after stroke;
however, more work is needed to understand these associations.
Biofeedback-based gait training – an approach to target biomechanical impairments
Biofeedback-based gait training is one approach used to specifically target post-stroke
biomechanical impairments (Tate & Milner, 2010). Biofeedback leverages an explicit form of
motor learning by providing extrinsic information about a selected gait variable for participants to
make voluntary changes in their gait pattern (Giggins et al., 2013). There are various options for
biofeedback targets including electromyography, spatiotemporal, kinematic, or kinetic variables
(Spencer et al., 2021). Biofeedback can be provided in a variety of modes such as visual, auditory,
or haptic (Spencer et al., 2021), all of which can be used by individuals post-stroke (J. Liu et al.,
2020; Ma et al., 2018). To date, no studies have compared the different modes of biofeedback;
therefore, it is unclear whether one mode may be more advantageous. Regardless of the mode
provided, biofeedback is typically delivered continuously, meaning the participant has constant
knowledge of their performance. However, this may cause participants to become reliant on the
biofeedback (Schmidt et al., 2019; Winstein et al., 1996; Winstein & Schmidt, 1990). There is
evidence that individuals post-stroke can use intermittent biofeedback (one-minute with the
biofeedback on, one-minute with the biofeedback off) to increase paretic propulsion within a single
session (Genthe et al., 2018). Additionally, individuals post-stroke can use biofeedback that fades
out over fifteen sessions, meaning that the biofeedback is available to the participant for a shorter
amount of time each session (Jonsdottir et al., 2007). Yet, more work is required to understand
how to optimally fade out biofeedback over a training program to maximize retention.
Visual gait biofeedback is a well-studied approach to target biomechanical impairments in
individuals post-stroke (Spencer et al., 2021). Visual biofeedback provided on two commonly

12
targeted gait impairments, step length and paretic propulsion, have successfully improved the
targeted gait variable while the biofeedback is displayed (French et al., 2021; Genthe et al., 2018;
Padmanabhan et al., 2019; Sánchez & Finley, 2018), and after the biofeedback is removed (French
et al., 2021; Genthe et al., 2018). However, there is great methodological heterogeneity between
studies, making the optimal approach to improve biomechanical impairments unclear (Spencer et
al., 2021). Additionally, there is considerable inter-individual variability in response to
biofeedback training, possibly due to factors such as cognitive impairment (French et al., 2020).
Results of long-term studies are also inconsistent, with some finding improvements over
conventional therapy (Jonsdottir et al., 2007) and some finding no difference compared to treadmill
training without biofeedback (Drużbicki et al., 2016).
Quantifying the effectiveness of biofeedback-based interventions
There are a few different ways to quantify the effectiveness of a biofeedback paradigm.
The first is to measure performance, or the behavior during acquisition or practice (Kantak &
Winstein, 2012; Schmidt & Bjork, 1992). Measuring performance quantifies whether participants
are using the biofeedback to execute the behavior as intended. It is also important to measure
retention, which is when the behavior can be retrieved and repeated during a recall task without
the biofeedback (Kantak & Winstein, 2012; Schmidt & Bjork, 1992). Measuring retention is
critical when considering motor learning approaches, as it is how we can infer learning has
occurred and the observed change in motor behavior is not simply a change in performance that
relies on the presence of biofeedback (Kantak & Winstein, 2012). If the participant cannot
demonstrate retention without the biofeedback, the participant likely will not be able to alter their
gait pattern to that of the biofeedback outside of the lab setting. Retention tests can be performed
at different time points after the completion of practice. Immediate retention is measured the same

13
day after acquisition trials, typically within a few minutes of completing practice. Delayed
retention is measured days to weeks after the acquisition period. These two types of retention tests
have different consolidation periods, which can impact the interpretation of what was learned
(Kantak & Winstein, 2012). Consolidation is a time-dependent process where the learned behavior
becomes more stable with time (Kantak & Winstein, 2012; Krakauer & Shadmehr, 2006);
therefore, an immediate retention test may not appropriately measure learning.
Post-stroke cognitive impairment
Post-stroke cognitive impairment is commonly defined as any cognitive impairment seen
after a stroke, regardless of the etiology (El Husseini et al., 2023). Some researchers and clinicians
use more stringent definitions that require defining the etiology of the cognitive impairment
(McDonald et al., 2019). For this reason, there is variability in the reported prevalence of poststroke cognitive impairment, ranging from 22 – 80% (El Husseini et al., 2023). Global cognitive
impairment (a measure taking into account task performance over multiple cognitive domains) is
common and persists many years after a stroke (Delavaran et al., 2017; Douiri et al., 2013). Global
cognition also declines faster in individuals post-stroke than neurotypical adults (Levine et al.,
2015). Cognitive impairment (specifically attention and visuospatial ability impairments) is
associated with lower quality of life (Cumming et al., 2014), and quality of life declines more
quickly over time in individuals with post-stroke cognitive impairment (Dhamoon et al., 2010).
Cognition has also been associated with measures such as lower extremity functioning (measured
using the Short Physical Performance Battery) and grip strength post-stroke (Einstad et al., 2021).
While global cognitive impairment is common post-stroke, the impairments can also be
broken into specific cognitive domains (Jokinen et al., 2015). The most frequently affected
cognitive domains post-stroke are memory, visuospatial function, executive function, attention,

14
and language (Jokinen et al., 2015). Definitions for these domains are listed in Table 2.2 (APA
Dictionary of Psychology, n.d.).
Table 2.2. Definitions of commonly affected cognitive domains post-stroke.
Cognitive domain Definition from the APA Dictionary of Psychology
Memory “the ability to retain information or a representation of past experience, based on the mental
processes of learning or encoding, retention across some interval of time, and retrieval or
reactivation of the memory”
Visuospatial function “the ability to perceive the spatial (relational) aspects of a figure or object in two and three
dimensions”
Executive function “higher level cognitive processes of planning, decision making, problem solving, action
sequencing, task assignment and organization, effortful and persistent goal pursuit, inhibition of
competing impulses, flexibility in goal selection, and goal-conflict resolution”
Attention “state in which cognitive resources are focused on certain aspects of the environment rather than
on others and the central nervous system is in a state of readiness to respond to stimuli”
Language Receptive language: “the language perceived and mentally processed by a person, as opposed to
that which he or she originates”
Expressive language: “the language produced by a speaker or writer, as opposed to that received
by a listener or reader”
All definitions were obtained from the APA Dictionary of Psychology (APA Dictionary of Psychology, n.d.)
Characterizing post-stroke cognitive impairment
There is no single standard diagnostic tool used to screen for post-stroke cognitive
impairment (El Husseini et al., 2023); however, various shorter instruments have been used for
screening purposes. Two commonly used instruments are the Montreal Cognitive Assessment
(Nasreddine et al., 2005) and the Mini-Mental State Exam (Folstein et al., 1975). Researchers and
clinicians often use these because they are easy to administer and have a low time burden (both ~
ten minutes). However, due to the brevity of the tests, they lack information about impairments in
specific cognitive domains. Additionally, these tests may also categorize an individual as having
normal cognitive function despite the presence of impairments in specific cognitive domains
(McDowd et al., 2003). The clinical purpose of these screening instruments is to identify
individuals with cognitive impairment and refer them to a neuropsychologist who will administer
a full battery of neuropsychological tests to thoroughly quantify cognitive impairment across
multiple domains of cognition. However, in many research settings, it is often difficult to quantify

15
cognitive impairment this way due to the time and expertise required to administer a full
neuropsychological battery. An alternative screening test is the Repeatable Battery for the
Assessment of Neuropsychological Status (RBANS), which screens for and characterizes
impairments in specific cognitive domains (Randolph, 1998).
The RBANS was developed to screen for and characterize cognitive decline in older adults
(Randolph et al., 1998) and has been found to be an appropriate test to detect domain-specific poststroke cognitive impairment (Green et al., 2013). It takes approximately thirty minutes to
administer and provides five subscale scores for the following cognitive domains: immediate
memory, visuospatial/constructional, language, attention, and delayed memory, as well as a
measure of global cognition (Randolph, 1998). The benefits of the RBANS include 1) the minimal
time required to administer, 2) characterization of specific cognitive domains along with global
cognition, 3) robust normalization procedures across different demographic factors, and 4) the
ability to be repeated to characterize cognitive impairment longitudinally. A drawback is that it
does not measure executive function, which is common post-stroke (Green et al., 2013; Nys et al.,
2007). Additionally, it requires training to administer, score, and interpret and is only available in
English and Spanish. Another potential limitation of many cognitive tests, including the RBANS,
is that in populations with motor dysfunction, the motor demands of the cognitive tests could
artificially lower the scores (Lee et al., 2021).
Cognition and explicit locomotor learning
During an explicit motor learning task, participants are provided information about their
movement performance and asked to use that information to make conscious changes to their
movements using cognitive resources (Kleynen et al., 2014). However, to date, the role of
cognition in motor learning has not been highly emphasized (VanGilder et al., 2020). This is

16
particularly important in the rehabilitation context, as many rehabilitation approaches attempt to
leverage explicit motor learning mechanisms (Leech et al., 2022). Someone with cognitive
impairment may be less able to use the cognitive resources required to acquire and retain the new
movement pattern from training. Supporting this idea, individuals post-stroke who have lower
fluid cognition (someone’s ability to adapt to new information and solve problems; Heaton et al.,
2014) are less able to use visual biofeedback to improve performance and retention of the gait
pattern (French et al., 2021). Additionally, cognition (specifically visuospatial working memory)
explains the retention of an upper extremity task in neurotypical adults (Hooyman et al., 2022;
Lingo VanGilder, Lohse, et al., 2021; VanGilder et al., 2022; Wang et al., 2022) and individuals
post-stroke (Lingo VanGilder, Hooyman, et al., 2021).
Utilizing cognitive status information to personalize rehabilitation approaches
The relationship between cognitive status and motor learning suggests that we may be able
to use information on cognitive impairment to individually tailor gait rehabilitation interventions
to optimize learning. This could be achieved by identifying training parameters, such as the mode,
frequency, or practice schedule, that can be individualized based on specific cognitive
impairments. For example, random practice in neurotypical adults shows poorer performance but
better retention than blocked practice, called the contextual interference effect (Kantak & Winstein,
2012). However, participants post-stroke with visuospatial working memory deficits learned a
hand grip task better with a blocked practice schedule than a random one (Schweighofer et al.,
2011), highlighting the importance of considering cognitive impairment status when designing
motor learning paradigms. Additionally, fading biofeedback results in better behavior retention in
neurotypical adults (Schmidt et al., 2019; Winstein et al., 1996; Winstein & Schmidt, 1990), and
previous work has demonstrated that participants post-stroke can use EMG biofeedback that fades

17
out over fifteen sessions(Jonsdottir et al., 2007). However, no work has systematically investigated
how to fade out real-time biofeedback to maximize explicit locomotor learning in individuals after
stroke, and whether post-stroke cognitive impairment may impact what type of biofeedback
schedule may be most appropriate. More research is necessary to identify how to use information
on cognitive status to best optimize training parameters to maximize learning.

18
CHAPTER 3
SPEED-DEPENDENT BIOMECHANICAL CHANGES VARY ACROSS INDIVIDUAL GAIT
METRICS POST-STROKE RELATIVE TO NEUROTYPICAL ADULTS
This work is adapted from:
Kettlety SA, Finley JM, Reisman DS, Schweighofer N, Leech KA. Speed-dependent
biomechanical changes vary across individual gait metrics post-stroke relative to neurotypical
adults. J NeuroEngineering Rehabil. 2023;20(1):14. doi:10.1186/s12984-023-01139-2
Abstract
Background. Gait training at fast speeds is recommended to reduce walking activity limitations
post-stroke. Fast walking may also reduce gait kinematic impairments post-stroke. However, it is
unknown if differences in gait kinematics between people post-stroke and neurotypical adults
decrease when walking at faster speeds. Objective. To determine the effect of faster walking speeds
on gait kinematics post-stroke relative to neurotypical adults walking at similar speeds. Methods.
We performed a secondary analysis with data from 28 people post-stroke and 50 neurotypical
adults treadmill walking at multiple speeds. We evaluated the effects of speed and group on
individual spatiotemporal and kinematic metrics and performed k-means clustering with all
metrics at self-selected and fast speeds. Results. People post-stroke decreased step length
asymmetry and trailing limb angle impairment, reducing between-group differences at fast speeds.
Speed-dependent changes in peak swing knee flexion, hip hiking, and temporal asymmetries
exaggerated between-group differences. Our clustering analyses revealed two clusters. One
represented neurotypical gait behavior, composed of neurotypical and post-stroke participants. The
other characterized stroke gait behavior – comprised entirely of participants post-stroke with
smaller lower extremity Fugl-Meyer scores than the post-stroke participants in the neurotypical

19
gait behavior cluster. Cluster composition was largely consistent at both speeds. Conclusions. The
biomechanical effect of fast walking post-stroke varied across individual gait metrics. For
participants within the stroke gait behavior cluster, walking faster led to an overall gait pattern
more different than neurotypical adults compared to the self-selected speed. This suggests that to
potentiate the biomechanical benefits of walking at faster speeds and improve the overall gait
pattern post-stroke, gait metrics with smaller speed-dependent changes may need to be specifically
targeted within the context of fast walking.
Background
In the United States, approximately 795,000 people have a stroke each year (Virani et al.,
2020), and gait dysfunction is a common outcome (Olney & Richards, 1996). Two domains of gait
dysfunction are kinematic impairments (e.g., increased circumduction and reduced knee flexion;
Awad, Binder-Macleod, et al., 2015; Chen et al., 2005) and activity limitations (e.g., decreased
gait speed and independence; Hornby et al., 2020). Kinematic and spatiotemporal impairments are
associated with increased metabolic cost (Awad, Palmer, et al., 2015; Finley & Bastian, 2017) and
fall risk (Wei et al., 2017), whereas activity limitations, particularly gait speed, are associated with
reduced community ambulation and quality of life (Grau-Pellicer et al., 2019). Improvements in
both categories are essential goals for people post-stroke and often targets during gait rehabilitation
(Bohannon et al., 1991).
Recently, there has been an increased emphasis on structuring interventions to target
activity limitations in walking post-stroke (Fahey et al., 2020). Current evidence suggests that the
most effective way to do this is with moderate to high aerobic intensity gait training, often achieved
by walking at faster speeds (Hornby et al., 2020). The shift away from interventions that prioritize
reducing kinematic gait impairments post-stroke may be due, in part, to evidence that select

20
kinematic metrics improve while walking faster. People post-stroke reduce step length asymmetry
and increase the magnitudes of paretic swing knee flexion and trailing limb angle with fast walking
(Jonkers et al., 2009; Tyrell et al., 2011). Furthermore, walking faster does not lead to increased
circumduction – a common compensatory movement (Mahtani et al., 2017; Tyrell et al., 2011).
This suggests that improved post-stroke gait kinematics may be a byproduct of walking faster,
even when not specifically targeted.
While studies have demonstrated speed-dependent improvements in select kinematic
metrics in people post-stroke relative to their habitual walking pattern, it is unclear how the
resultant kinematics compare to that of neurotypical adults walking at similar speeds. For example,
we expect both groups to exhibit speed-dependent changes in knee flexion, but the magnitude of
speed-dependent change in each group may differ such that the between-group difference in knee
flexion may be larger at faster speeds. In addition to this, previous work has focused on speeddependent changes in individual gait metrics post-stroke (Beaman et al., 2010; Jonkers et al., 2009;
Mahtani et al., 2017; Tyrell et al., 2011), leaving the impact of fast walking on overall gait behavior
post-stroke relative to neurotypical adults an open question.
This study evaluated the effect of gait speed on select spatiotemporal and kinematic gait
metrics in people post-stroke relative to neurotypical adults walking at similar speeds. To do this,
we performed a secondary analysis of three previously published data sets (Finley & Bastian, 2017;
Fukuchi et al., 2018; Tyrell et al., 2011). Based on studies that demonstrate abnormal muscle coactivation patterns post-stroke during walking (Clark et al., 2010; Knutsson & Richards, 1979;
Shiavi et al., 1987) and maximal hip extension torque production (Sánchez et al., 2017), we
hypothesized that the difference between groups would increase as speed increases for all gait
metrics. We also evaluated all metrics simultaneously in a k-means clustering analysis to determine

21
if the difference in overall gait behavior between people post-stroke and neurotypical adults
changed with speed. We hypothesized that there would be two clusters (one that captured the gait
behavior of neurotypical adults and one that captured post-stroke gait behavior) and that walking
faster would cause the distance between the clusters to increase. This would indicate that the
overall walking patterns of people post-stroke and neurotypical adults became more different at
fast speeds. With this work, we hoped to demonstrate the effect of fast walking on gait kinematics
post-stroke relative to neurotypical adults to further define the advantages and limitations of this
intervention in addressing gait biomechanics post-stroke.
Methods
Participants
We performed a secondary analysis of data sets from three previously published crosssectional studies of people post-stroke and neurotypical adults (Finley & Bastian, 2017; Fukuchi
et al., 2018; Tyrell et al., 2011). Specifically, data from people post-stroke were obtained from
Tyrell et al. (2011) (n = 20), data from people post-stroke (n = 15) and neurotypical adults (n = 15)
from Finley and Bastian (2017), and additional data from neurotypical adults were obtained from
Fukuchi et al. (2018) (n = 42). To evaluate continuous gait patterns, only data from participants
whose slowest walking speed was >0.20 m/s were included. Participants without a standing
calibration file were also excluded from these analyses. All data were de-identified; therefore, this
analysis is not considered human subjects research and did not require review from the University
of Southern California Institutional Review Board.

22
Stroke Data
The post-stroke data set included lower extremity kinematic data of people > six months
post-stroke previously published in Tyrell et al. (2011) (n = 20). Participants walked on a treadmill
at four different speeds in a randomized order: self-selected, fast-as-safely possible, and two
intermediate speeds. The two intermediate speeds were chosen to be as equally distributed as
possible between the self-selected and fast-as-safely possible speeds. Two participants only
completed one intermediate speed; therefore, only three speeds were included in the analyses for
these participants. Marker data were collected using a motion capture system sampling at 100 Hz.
Details about the marker set can be found in the original publication (Tyrell et al., 2011). Data
were collected at each speed for two twenty-second trials, resulting in forty seconds of data.
The post-stroke data set also included data from participants published in Finley and
Bastian (2017) (n = 15). Participants walked at four speeds on a treadmill: self-selected speed,
fastest possible speed they could maintain for five minutes, and 80% and 120% of their selfselected speed. One participant could not complete the trial at 120% of their self-selected speed;
therefore, only three speeds were included in the analysis for this participant. The order of speed
presentation was randomized, and participants walked on the treadmill for five minutes at each
speed. For these analyses, the middle thirty seconds of each trial were analyzed. Marker data were
collected using a motion capture system with infrared-emitting markers sampling at 100 Hz.
Details about the marker set can be found in the original publication (Finley & Bastian, 2017).
Neurotypical Data
Part of the neurotypical data set included age- and speed-matched neurotypical adults to
the people post-stroke in Finley and Bastian (2017) (n = 15). Neurotypical adults in this data set

23
walked on the treadmill at the same four absolute speeds as the people post-stroke. The data
collection procedures were the same as outlined above for the post-stroke group.
The age-matched neurotypical data set also included data from older adults from Fukuchi
et al. (2018) (n = 18). Participants walked on a treadmill at eight speeds ranging from 40 – 145%
(in increments of 15%) of their self-selected speed, in a randomized order. Our statistical analysis
(described below) was not paired and did not require exact speed matching, so for each participant,
we extracted four speeds that reflected a similar speed range to post-stroke speeds in the first stroke
data set (Tyrell et al., 2011). This ensured that all speeds in the stroke data set were represented in
the neurotypical data set (Supplemental Table 3.1). Participants walked for ninety seconds at each
speed, and data were recorded during the final thirty seconds. Marker data were collected using a
motion capture system sampling at 150 Hz. Details about the marker set can be found in the
original publication (Fukuchi et al., 2018).
For the clustering analysis only, we included data from neurotypical young adults (n = 24)
in addition to the post-stroke and neurotypical older adult data to make the analysis more robust.
These data were collected in the same laboratory and manner as the older adults from Fukuchi et
al. (2018). The data from young neurotypical participants were chosen to match the range of selfselected and fastest speeds seen in the first stroke data set (Tyrell et al., 2011).
Data Analyses
We used spatiotemporal and kinematic metrics computed in Visual3D (C-Motion,
Germantown, MD) from Tyrell et al. (2011) in all statistical analyses. The marker data from the
other data sets were processed and analyzed in MATLAB R2020a (MathWorks, Natick, MA).

24
These data were low-pass filtered with a 6 Hz cutoff (Winter, 2009). Foot-strike and toe-off were
defined as the most anterior and posterior positions of the lateral malleoli markers, respectively.
The spatiotemporal and kinematic metrics of interest reported here were selected and
defined to be consistent with those reported in Tyrell et al. (2011) since we could not re-process
those data. Spatiotemporal outcome measures were step length asymmetry, single-limb support
time asymmetry, and double-limb support time asymmetry. Kinematic outcome measures were
peak swing knee flexion angle, trailing limb angle, circumduction, and hip hiking. Markers used
to calculate these metrics were iliac crest, greater trochanter, lateral femoral epicondyle, lateral
malleolus, and fifth metatarsal. All intralimb kinematic outcome measures were calculated on the
paretic (stroke) or the right limb (neurotypical). Mean values across all strides taken within the bin
specified above were used in all statistical analyses.
Statistical Analysis
To evaluate the effect of speed and group on the individual gait metrics, we fit robust linear
mixed-effects models using the rlmer package (Koller, 2016) in R (4.0.2; R Core Team, n.d.). We
used a robust model instead of a traditional mixed-effects model to address normality and
homoscedasticity assumptions violations. Fixed effects for group, speed, and speed by group
interaction were included. Before fitting the models, we removed the mean from the speed values,
which allowed us to interpret the group coefficient at the average speed across the sample. A
random intercept term was included in all models to account for the repeated measures design. A
separate model was fit for each outcome measure. P-values were calculated using the Satterthwaite
approximation (Luke, 2017). Statistical significance was set a priori at 0.05.

25
To capture the effect of speed on overall gait behavior, we used k-means clustering to
identify subsets of participants with similar overall gait behavior and determined whether walking
faster altered the composition of these subsets. We performed k-means clustering using the kmeans
function in R (4.0.2; R Core Team, n.d.) with data from both groups at the self-selected and fastest
speeds of the participants post-stroke using all seven gait metrics listed above. Because these
metrics have different units, the data were scaled (mean = 0, standard deviation = 1) before
clustering. We determined the number of clusters using the silhouette method. Within sum of
squares was used to assess the between-subjects variability within each cluster and between sum
of squares to determine the distance between clusters. To more robustly understand how the
distance between the clusters changed with gait speed, we performed a bootstrap analysis with
1000 iterations at each gait speed to construct 95% confidence intervals of the between sum of
squares value. To identify the relative importance of each variable in the determination of the
clustering, we used the cluster assignment as the outcome variable in a random forest algorithm
(Cutler & Wiener, 2022). We then extracted the variable importance using the importance function
in the randomForest package (Cutler & Wiener, 2022). For each variable, the prediction error on
the data that was not used to train the model was determined, then the prediction error when the
data for each predictor variable was permuted (shuffled) was determined. The difference between
the two prediction errors represents the mean decrease in accuracy for a specified variable; higher
values represent greater variable importance (Cutler & Wiener, 2022; Han et al., 2016). We also
performed a principal components analysis on the scaled data at both speeds to allow visualization
of the clusters in two dimensions. Finally, we used Mann-Whitney U tests to determine if two key
clinical measures of impairment, Lower-Extremity Fugl-Meyer scores and gait speeds, differed
between the participants post-stroke who were assigned to different clusters.

26
Results
After combining data sets, 28 people post-stroke, 26 neurotypical older adults, and 24
neurotypical younger adults were included in these analyses. Eight participants were excluded due
to their slowest walking speed being <0.20 m/s, and six were excluded due to a missing standing
calibration file. For the robust mixed-effects analysis, we used post-stroke data (n = 28),
neurotypical data from older adults obtained from Fukuchi et al. (2018) (n = 18), and matched
neurotypical controls from Finley and Bastian (2017) (n = 8). For the clustering analysis, we
included additional data from younger neurotypical adults (18 – 39 years; n = 24). Clinical
demographics for participants post-stroke are included in Table 3.1.

27
Table 3.1. Clinical demographics of participants post-stroke.
Data
Set Sex Side of
Paresis
Age
(years)
Orthosis or
Assistive
Device
LowerExtremity
Fugl-Meyer
Speed 1
(m/s)
Speed 2
(m/s)
Speed 3
(m/s)
Speed 4
(m/s)
1 M L 66 AFO 18 0.80 0.90 1.10 1.30
1 M L 50 None 19 0.90 1.00 1.10 1.20
1 M L 74 AFO 17 0.30 0.40 0.50 0.70
1 F L 59 None 31 0.70 0.80 0.90 1.00
1 M R 61 AFO 14 1.00 1.10 1.20 1.30
1 F L 66 None 21 0.40 0.50 0.50 0.60
1 F L 78 None 24 0.70 0.80 0.90 1.00
1 M R 57 None 15 0.60 0.80 0.90 1.10
1 M R 75 None 31 0.70 0.90 1.10 1.30
1 F L 51 AFO 20 0.40 0.50 0.60 0.70
1 M R 47 None 25 0.80 0.90 1.00 1.10
1 M R 60 None 25 0.90 1.00 1.20 1.30
1 M L 52 AFO 20 0.80 1.00 1.20 1.40
1 F R 62 SPC 25 0.50 0.60 0.80 1.00
1 M R 72 None 32 0.80 0.90 1.00 1.20
1 M R 71 AFO, SPC 19 0.70 0.80 0.90 -
1 M R 77 None 22 0.70 0.80 0.90 1.10
1 F L 45 None 23 0.90 1.00 1.10 -
1 M R 73 None 31 0.70 0.80 1.00 1.30
1 M L 78 SPC 23 0.50 0.60 0.70 0.90
2 M L 42 Not reported 22 0.77 0.96 1.15 1.24
2 F R 56 Not reported 33 0.46 0.57 0.68 0.81
2 F R 68 Not reported 23 0.21 0.26 0.31 0.56
2 M R 57 Not reported 17 0.48 0.61 0.72 1.00
2 M L 52 Not reported 32 0.82 1.02 1.22 1.35
2 M L 67 Not reported 27 0.26 0.33 0.40 -
2 M L 54 Not reported 31 0.28 0.35 0.42 0.60
2 M L 60 Not reported 33 0.22 0.28 0.34 0.57
Abbreviations: 1, Tyrell et al., 2011; 2, Finley and Bastian 2017; F, female; M, male; L, left; R, right; AFO, ankle-foot orthosis;
SPC, single-point cane. Ankle-foot orthoses were worn during data collection. Single-point canes were not used during data
collection but were used by participants during daily walking.

28
Spatiotemporal Parameters
As expected (Chen et al., 2005), people post-stroke exhibited greater step length
asymmetry (Figure 3.1A; β = 0.08, p < 0.001) compared to neurotypical adults. With increases in
speed, neurotypical adults decreased step length asymmetry (β = -0.08, p < 0.001). However,
people post-stroke decreased step length asymmetry more than neurotypical adults at faster speeds
(speed by group interaction: β = -0.05, p < 0.001), reducing the difference between the groups at
faster speeds.
We also found that people post-stroke exhibited more double-limb support time asymmetry
relative to neurotypical adults (Figure 3.1B; β = 0.12, p < 0.001). Neurotypical adults decreased
double-limb support time asymmetry at faster speeds (Figure 3.1B; β = -0.06, p < 0.001), while
double-limb support time asymmetry increased in people post-stroke as demonstrated by a
significant speed by group interaction (β = 0.22, p < 0.001). This means the group difference in
double-limb support time asymmetry was exaggerated faster speeds.
Figure 3.1. Spatiotemporal asymmetries by group across gait speeds. The figure displays the step length asymmetry
(A), double-limb support time asymmetry (B), and single-limb support time asymmetry (C) of individual participants
(red: post-stroke; black: neurotypical) across gait speeds. The thinner, lighter lines represent the data for an individual
participant walking at 3 or 4 different gait speeds. The thicker, darker lines represent the group fits from the robust
mixed-effects model.

29
People post-stroke also had greater single-limb support time asymmetry (Figure 3.1C; β =
0.09, p < 0.001) than neurotypical adults. We found a significant speed by group interaction (β =
0.02, p < 0.001), suggesting that the difference in single-limb support time asymmetry between
groups increased at faster speeds. This was driven by a significant decrease in single-limb support
time asymmetry exhibited by neurotypical adults with increased speeds (β = -0.03, p < 0.001).
Kinematic Parameters
We found a significant group effect for peak swing knee flexion angle (Figure 3.2A; p <
0.001), with neurotypical adults having an average of 17° greater peak swing knee flexion angle
than people post-stroke. Peak swing knee flexion angle of neurotypical adults increased
significantly with increases in speed (β = 12.1, p < 0.001). People post-stroke also exhibited speeddependent increases in peak swing knee flexion angle. However, the magnitude of this increase
for a given change in speed was smaller than that observed in neurotypical adults (speed by group
interaction: β = -8.0, p < 0.001), increasing the difference between groups at faster speeds.
People post-stroke exhibited an average of 5° less trailing limb angle compared to
neurotypical adults (Figure 3.2B; p < 0.001). While neurotypical adults increased their trailing
limb angles with faster speeds (β = 9.6, p < 0.001), people post-stroke increased their trailing limb
angle more for a given increase in speed than neurotypical adults (speed by group interaction: β =
2.3, p < 0.001), decreasing the difference in trailing limb angle between the groups at faster speeds.

30
People post-stroke had an average of 6° greater hip hiking (Figure 3.2C; p < 0.001) than
neurotypical adults. This difference was exaggerated at faster speeds, as people post-stroke
exhibited a slight increase in hip hiking relative to neurotypical adults with increased speeds
(speed: β = -0.8, p < 0.001, speed by group interaction: β = 1.6, p < 0.001). Of note, two
neurotypical participants were excluded from the hip hiking analysis due to iliac crest marker
occlusion.
Figure 3.2. Kinematic gait parameters by group across gait speeds. The thinner, lighter lines
represent the data for an individual participant (red: post-stroke; black: neurotypical) walking
at 3 or 4 different gait speeds. The thicker, darker lines represent the group fits from the robust
mixed-effects model. A) Peak swing knee angle across gait speeds. Positive values indicate
knee flexion, negative values indicate knee extension. B) Trailing limb angle across gait
speeds. C) Hip hiking across gait speeds. Two participants were excluded from this analysis
due to missing iliac crest marker data. D) Circumduction across gait speeds.

31
As expected (Chen et al., 2005), we found that people post-stroke had an average of 0.02
meters greater circumduction (Figure 3.2D; p < 0.001) compared to neurotypical adults. There was
a small increase in circumduction in neurotypical adults at faster speeds (β = 0.004, p = 0.03), but
there was no significant speed by group interaction (β = -0.005, p = 0.11).
K-means Clustering Analysis
We used k-means clustering to identify subsets of participants with similar overall gait
behavior and determined whether walking at faster speeds altered the composition of these subsets.
Due to missing hip hiking data, three neurotypical participants were excluded from the k-means
analysis. Two participants were missing hip data at all speeds, and one participant was missing
data at the fastest speed only. We chose two clusters for both the self-selected and fastest speeds
using the silhouette method.
At self-selected speeds, all 47 neurotypical participants and 11/28 people post-stroke were
in the first cluster (Figure 3.3A; diamonds). The remaining 17 people post-stroke were in the
second cluster (circles). The cluster assignments were largely the same at the fastest speeds (Figure
3.3B), except for two participants post-stroke who switched clusters.

32
Figure 3.4A displays the centroids (i.e., scaled mean values) for each gait variable at both speeds
for each cluster. These values indicate that the first cluster captured participants with kinematics
associated with neurotypical gait behavior (e.g., larger peak swing knee flexion, smaller gait
asymmetries, etc.), and the second cluster had kinematics associated with stroke gait behavior (e.g.,
smaller peak swing knee flexion, greater gait asymmetries, etc.). Because of this and the participant
composition of the clusters, we will refer to the first cluster as the “neurotypical gait behavior
cluster” and the second cluster as the “stroke gait behavior cluster.” However, it is important to
note that there are individuals post-stroke in both clusters. The order of variable importance in
determining the cluster assignments changed slightly from self-selected to fast speeds (Figure
3.4B-C; see the ranks of hip hiking and knee flexion). However, single-limb support time
Figure 3.3. K-means clustering results across gait speeds. Individual scores for principal
component 1 vs. principal component 2 are plotted to allow for visualization of two out
of seven dimensions of the data included in the cluster analysis. Due to missing hip
hiking data, three neurotypical participants were excluded from the k-means analysis.
Marker colors represent the true group for the individual (black: neurotypical, red: poststroke). The marker shape represents the assigned cluster (diamond: neurotypical gait
behavior cluster, circle: stroke gait behavior cluster). Two participants post-stroke
switched clusters at faster gait speeds. One participant moved from the neurotypical gait
behavior cluster to the stroke gait behavior cluster (cyan marker), and one participant
moved the opposite way (royal blue marker).

33
asymmetry was the variable most important in determining the cluster assignment regardless of
speed. This is likely because participants in the stroke gait behavior cluster had much higher singlelimb support time asymmetries than participants in the neurotypical gait behavior cluster at both
speeds (Figure 3.4A; right most panel).

34
Figure 3.4. Cluster centroids (A), and variable importance results (B & C). A) Cluster centroids
(means) for each cluster at self-selected and fast speeds The red data represents the stroke gait
behavior cluster, the black data represents the neurotypical gait behavior cluster. The circle
markers represent the self-selected speed, and the asterisk represents the fast speed. Data were
scaled (mean = 0, SD = 1) before calculating the mean value. B) Variable importance results at
self-selected speeds. C) Variable importance results at fast speeds. Abbreviations: TLA, trailing
limb angle; SLA, step length asymmetry; DSTA, double-limb support time asymmetry; SLSTA,
single-limb support time asymmetry.

35
Of note, the participants post-stroke assigned to the neurotypical gait behavior cluster
exhibited less lower extremity motor impairment on the Lower Extremity Fugl-Meyer than the
individuals in the stroke gait behavior cluster at both speeds (Figure 3.5A; self-selected speed: p =
0.004; fast speed: p = 0.008). Yet, there were no differences between the participants' post-stroke
Figure 3.5. Lower Extremity Fugl-Meyer scores (A) and gait speeds (B) of
participants post-stroke in each cluster. The white-filled boxplots represent
participants post-stroke in the neurotypical cluster. The red-filled boxplots
represent participants post-stroke.

36
self-selected or fast speeds in each cluster (Figure 3.5B; self-selected speed: p = 0.59; fast speed:
p = 0.43).
Two participants post-stroke were classified into a different cluster when walking at faster
speeds. One moved from the stroke gait behavior cluster at self-selected speed to the neurotypical
gait behavior cluster at the fast speed (Figures 3.3A and 3.3B; royal blue symbol). The other
switched from the neurotypical gait behavior cluster at self-selected speed to the stroke gait
behavior cluster at the fast speed (Figures 3.3A and 3.3B; cyan symbol).
Finally, we found that the kinematics of the individuals in the stroke gait behavior cluster
were more variable and different than those in the neurotypical gait behavior cluster at the fast
speed relative to self-selected speed. The within sum of squares (capturing within-cluster
variability) of the neurotypical gait behavior cluster decreased at the fast speed (self-selected
speed: 107.3; fast speed: 92.2) while it increased in the stroke gait behavior cluster (self-selected
speed: 156.5; fast speed: 167.7). The two clusters appeared to move further apart at fast speeds,
demonstrated by the increase in between sum of squares (self-selected speed: 187.7; fast speed:
207.6). However, our bootstrap analysis shows that the 95% confidence intervals largely overlap
between speeds (self-selected: [99.8, 257.3]; fast: [102.7, 291.2]; Figure 3.6). It is important to
note that the variability within and between clusters is not well-represented in Figure 3.3 because
the data is plotted in the principal components space, with only two out of seven dimensions
plotted.

37
Discussion
This study aimed to determine the effect of fast walking on gait kinematics in people poststroke relative to neurotypical adults. Consistent with our hypothesis, speed-dependent changes in
peak swing knee flexion, hip hiking, double-limb support time asymmetry, and single-limb support
time asymmetry led to a larger between-group difference at faster speeds. Contrary to our
hypothesis, faster walking did not affect between-group differences in circumduction and reduced
between-group differences in step length asymmetry and trailing limb angle. We found two distinct
clusters when we included all kinematic metrics in clustering analyses during self-selected and fast
walking to understand the effect on overall gait behavior. One cluster represented neurotypical gait
behavior and was composed of all the neurotypical participants (n = 47) and a sub-group of the
participants post-stroke (n = 11/28). The other characterized post-stroke gait behavior and
contained only participants post-stroke (n = 17/28). At fast speeds, cluster assignments largely did
not change. These findings indicate that the biomechanical benefit (i.e., the change of metric in the
Figure 3.6. Between sum of squares bootstrap analysis.

38
same direction as that observed in neurotypical adults) of fast walking post-stroke varies across
individual gait metrics and in a sub-group of participants post-stroke (n = 17/28) these benefits did
not lead to an overall gait pattern more similar to neurotypical adults. This suggests that there is a
need to potentiate speed-dependent biomechanical changes by coupling fast walking with other
interventions if reducing kinematic impairments is a priority for a given patient.
Differences in between-group peak swing knee flexion magnitude grew larger at faster speeds
Our evaluation of individual gait metrics found that the relative speed-dependent changes
in four of the seven aligned with our hypothesis that between-group differences would be larger at
faster speeds. For example, walking at faster speeds increased the absolute magnitude of peak
swing knee flexion in the participants post-stroke. Yet, the magnitude of speed-dependent increase
was much smaller than observed in neurotypical adults. This small speed-dependent change in
swing knee flexion, relative to that observed in neurotypical adults, likely explains the persistence
of compensatory gait deviations, such as hip hiking and circumduction, at faster speeds (Akbas et
al., 2019; Lewek et al., 2012).
A few mechanisms may explain this smaller speed-dependent change in knee flexion in
our participants post-stroke. First, people post-stroke are well-known to exhibit reduced
neuromuscular complexity (i.e., abnormal muscle synergies, merged motor modules) in the lower
extremity during walking (Clark et al., 2010; Knutsson & Richards, 1979; Shiavi et al., 1987) that
may limit their capacity for large changes in knee flexion during swing phase. In addition to this,
studies of the upper extremity post-stroke demonstrate that raising the demands of a movement
task can increase the expression of abnormal synergistic movements (Kline et al., 2007; Sukal et
al., 2007). This suggests that fast walking may increase the expression of lower extremity synergies
that may limit knee flexion, similar to that observed during maximal hip extension on a

39
dynamometer (Sánchez et al., 2017), particularly if participants are attempting to increase their
trailing limb angle to increase propulsion to walk at faster speeds (see trailing limb angle
discussion below; Hsiao, Knarr, et al., 2016). However, a previous study of neuromuscular
complexity in participants post-stroke during walking demonstrated that the organization of motor
modules does not change during fast walking (Routson et al., 2014). This is indirectly supported
by our finding that people post-stroke assigned to the neurotypical gait behavior cluster at selfselected speeds were not assigned to the stroke gait behavior cluster at fast speeds.
Another potential explanation for the smaller speed-dependent changes in peak swing knee
flexion observed in people post-stroke is increased spastic activity of the quadriceps on the paretic
limb (Caty et al., 2008; Lewek et al., 2007; Tok et al., 2012). There is a strong relationship between
the magnitude of peak swing knee flexion and measures of quadriceps muscle hyperactivity in
people post-stroke (Campanini et al., 2013; Lewek et al., 2007). Walking at faster speeds on a
treadmill with larger trailing limb angles will naturally induce a faster stretch of the hip flexors
during late stance than walking slower. In theory, this would lead to velocity-dependent,
involuntary hyperactivation of the quadriceps (Dean et al., 2020) that may contribute to smaller
speed-dependent changes in peak swing knee flexion – particularly if coupled with the activation
of an abduction-dependent reflex that can occur post-stroke (Finley et al., 2008). However, one
previous study demonstrated that only the velocity-independent prolonged stretch reflex response
of the quadriceps (assessed in sitting) – not the velocity-dependent initial stretch reflex magnitude
– is related to swing knee flexion angle during walking post-stroke (Lewek et al., 2007). More
research is needed to better understand the relationship between gait speed and quadriceps
spasticity. Lastly, reduced peak swing knee flexion post-stroke has also been associated with
diminished paretic limb propulsion (Campanini et al., 2013; Dean et al., 2020).

40
Fast walking reduced between-group differences in step length asymmetry and trailing limb angle
In contrast to the speed-dependent exaggeration of between-group differences described
above, speed-dependent reductions in step length asymmetry within this single session led to
smaller between-group differences at faster speeds. This suggests that the biomechanical
byproduct of reduced step length asymmetry during fast walking post-stroke may be sufficient to
reduce the difference relative to neurotypical gait. Yet, people post-stroke have consistently
demonstrated no significant change in step length asymmetry following long-term fast walking
interventions (Ardestani et al., 2020; Mahtani et al., 2017).
People post-stroke also exhibited larger increases in trailing limb angle for a given change
in speed relative to neurotypical adults, decreasing the magnitude of the difference between the
groups at faster speeds. This finding is likely related to the increased propulsion needed to walk
faster. Propulsion during walking can be altered by increasing plantar-flexor moment and trailing
limb angle (Awad et al., 2020). However, people post-stroke tend to increase propulsion during
fast walking through increases in trailing limb angle, without changing plantar-flexor moment
(Hsiao, Knarr, et al., 2016). We posit that people post-stroke were modulating propulsion primarily
through changes in trailing limb angle, while neurotypical participants were using both strategies
to meet the increased propulsion demands of fast walking. However, we did not have kinetic data
to test this theory.
Walking faster did not cause people post-stroke to walk more similarly to neurotypical adults
Given that the relative speed-dependent effects varied across individual gait metrics, we
used k-means clustering to consider all gait metrics simultaneously and gain insight into the
relative effect of speed on overall gait behavior post-stroke. This revealed two clusters at self-

41
selected and fast speeds. One cluster characterized gait behavior typically found in neurotypical
adults, and the other characterized gait behavior typically found in people post-stroke. Our analysis
of variable importance in participants’ cluster assignment found that magnitudes of single-limb
stance time asymmetry, knee flexion, and circumduction were the most influential in determining
the clusters at self-selected speeds (Figure 3.4B). This changed slightly at fast speeds – with single
limb stance time asymmetry, hip hiking, and circumduction being the top three determinants of
cluster classification (Figure 3.4C).
Approximately 40% of the participants post-stroke in our sample (n = 11/28) were
classified as part of the neurotypical gait behavior cluster at both speeds. This indicates that this
subset of post-stroke participants walked with an overall gait pattern more similar to neurotypical
than the other participants post-stroke. The centroids (i.e., scaled mean values) for each gait
variable for each cluster (Figure 3.4A) and the order of variable importance analysis (Figures 3.4B
and 3.4C) suggest that these participants exhibited less single-limb stance time asymmetry, hip
hiking, and circumduction and more knee flexion than the post-stroke participants assigned to the
other cluster. This difference in overall gait behavior was not due to differences in self-selected or
fastest gait speed (Figure 3.5B). However, this subset of participants had significantly less lower
extremity motor impairment (Figure 3.5A) than those classified in the stroke gait behavior cluster.
This highlights the need for future work to 1) determine other clinical characteristics that may
contribute to heterogeneity in overall gait behavior post-stroke and 2) identify subgroups of people
post-stroke that may benefit more from combining fast walking with an approach that strategically
targets gait biomechanics.
Fast walking alone did not lead to biomechanical changes that were large or consistent
enough to cause participants post-stroke to change cluster assignments. Only two participants post-

42
stroke were assigned to the opposite cluster when walking at faster speeds – one to the neurotypical
cluster and one to the post-stroke cluster (royal blue and cyan points, respectively in Figures 3.3A
and 3.3B). Fast walking also did not cause the clusters to move closer together, as we might expect
if overall gait behavior more similar to neurotypical were a byproduct of fast walking. This
suggests that fast walking alone may not adequately address kinematic impairments in people poststroke. These results are inconsistent with upper extremity literature, which found that reaching
for an object at faster speeds improved movement quality (DeJong et al., 2012). However, the
changes in movement quality reported by DeJong et al. (2012) were quantified only by the reach
path, and limb biomechanics were not assessed. Additionally, though DeJong et al. (2012) found
improvements in post-stroke reaching movement quality relative to their baseline performance,
moving faster did not result in individuals post-stroke having reach paths similar to neurotypical
controls, which is consistent with our findings.
The results of this analysis highlight the utility of employing analytical approaches that
examine the effect of an intervention on individual gait metrics and overall gait behavior. In this
study, we found that gait speed had varying effects on individual gait metrics when evaluated
independently. This information is useful to understand the magnitude of speed-dependent changes
within a group and the differences between the groups for a given metric. However, it is difficult
to draw conclusions about the effect of gait speed on the overall gait pattern by looking at these
metrics individually because they do not occur in isolation. The cluster and random forest
algorithm analyses used here, allowed us to 1) account for the speed-dependent changes in all gait
metrics simultaneously, 2) gain insight into the related change in overall gait behavior, and 3)
identify which gait metrics were most influential in the cluster assignment. It is likely for this
reason that the use of outcome measures that provide a more comprehensive assessment of an

43
individual’s overall gait pattern has gained traction in recent years (Fukuchi & Duarte, 2019;
Padmanabhan et al., 2020; Ramakrishnan et al., 2018; Steele & Schwartz, 2022). Given the need
for clinicians to provide effective and efficient care, we anticipate that the focus on assessing and
improving overall gait behavior will continue to increase with the use of comprehensive outcome
measures or analytical approaches like those used here.
Limitations
This study has a few limitations. First, the data were collected at three sites with different
motion capture set-ups and study protocols. Because the participants collected at each site were
not equally balanced between groups (i.e., one site collected data from only stroke participants,
one site collected data from only neurotypical adults, and one site collected data from both), we
could not account for this in our model. Next, we did not have access to ground reaction force or
EMG data. Therefore, speed-dependent differences between groups in kinetic and muscular
activity outcomes are unclear. Finally, we only had data from the paretic limb for 20/28 of the
participants post-stroke. Consequently, we could not study the effect of fast walking on non-paretic
limb kinematics.
Conclusions
This study demonstrated that the biomechanical changes resulting from fast walking poststroke, relative to neurotypical adults, vary across individual gait metrics. We found two distinct
clusters representative of neurotypical gait behavior and stroke gait behavior, which became more
distinct at fast speeds. People post-stroke in the stroke gait behavior cluster walked at similar
speeds but had lower Fugl-Meyer scores compared to those in the neurotypical gait behavior
cluster. These analyses demonstrate that while fast walking may reduce the magnitude of a
kinematic impairment relative to one’s habitual walking pattern, the resulting gait kinematics are

44
not necessarily more similar to neurotypical adults walking at like speeds. This suggests that to
further improve gait kinematics, fast walking may need to be combined with another intervention,
such as verbal cues from a therapist, gait biofeedback, or virtual reality-based exergames, to
strategically target gait metrics with smaller speed-dependent changes. This would allow for a
single intervention to address both activity limitations and kinematic impairments in post-stroke
gait rehabilitation, which could have additive or synergistic effects on the rehabilitation of walking
dysfunction post-stroke.

45
Supplemental Table 3.1. Demographics of neurotypical adults.
Data Set Age
(years) Sex Speed 1
(m/s)
Speed 2
(m/s)
Speed 3
(m/s)
Speed 4
(m/s)
Finley and Bastian 68 M 0.77 0.96 1.15 1.24
Finley and Bastian 63 F 0.46 0.57 0.68 0.81
Finley and Bastian 57 M 0.21 0.26 0.31 0.56
Finley and Bastian 56 F 0.48 0.61 0.72 1.00
Finley and Bastian 43 M 0.82 1.02 1.22 1.35
Finley and Bastian 52 M 0.26 0.33 0.40 -
Finley and Bastian 55 M 0.28 0.35 0.42 0.60
Finley and Bastian 65 M 0.22 0.28 0.34 0.57
Fukuchi et al. 59 M 0.64 0.82 0.99 1.17
Fukuchi et al. 57 F 0.75 0.95 1.16 1.36
Fukuchi et al. 55 M 0.78 0.99 1.2 1.41
Fukuchi et al. 50 M 0.58 0.80 1.02 1.23
Fukuchi et al. 71 M 0.57 0.78 1.00 1.21
Fukuchi et al. 58 F 0.92 1.08 1.24 1.4
Fukuchi et al. 63 F 0.61 0.78 0.95 1.11
Fukuchi et al. 61 F 0.75 0.95 1.15 1.36
Fukuchi et al. 63 M 0.89 1.08 1.27 1.46
Fukuchi et al. 62 M 0.42 0.48 0.73 0.89
Fukuchi et al. 68 M 0.52 0.71 0.90 1.10
Fukuchi et al. 63 M 0.80 1.02 1.24 1.46
Fukuchi et al. 73 M 0.61 0.78 0.95 1.11
Fukuchi et al. 56 F 0.80 1.02 1.24 1.46
Fukuchi et al. 84 M 0.40 0.54 0.69 0.84
Fukuchi et al. 68 F 0.74 0.94 1.14 1.34
Fukuchi et al. 55 F 0.47 0.64 0.82 0.99
Fukuchi et al. 63 F 0.36 0.49 0.63 0.76
Fukuchi et al. 25 M 0.49 - - 1.03
Fukuchi et al. 22 F 0.5 - - 0.69
Fukuchi et al. 33 M 0.39 - - 1.12
Fukuchi et al. 24 M 0.71 - - 0.91
Fukuchi et al. 28 M 0.9 - - 1.1
Fukuchi et al. 25 M 0.9 - - 1.28
Fukuchi et al. 24 F 0.61 - - 1.1
Fukuchi et al. 36 M 0.76 - - 0.97
Fukuchi et al. 25 F 0.82 - - 0.99
Fukuchi et al. 31 F 0.76 - - 1.37
Fukuchi et al. 32 M 0.92 - - 1.31
Fukuchi et al. 24 F 0.58 - - 0.9
Fukuchi et al. 30 M 1.02 - - 1.33
Fukuchi et al. 31 F 0.78 - - 0.62
Fukuchi et al. 23 M 0.85 - - 1.31
Fukuchi et al. 31 M 1.1 - - 1.27
Fukuchi et al. 28 M 0.55 - - 1.16
Fukuchi et al. 28 F 0.5 - - 0.88
Fukuchi et al. 29 M 1.03 - - 1.21
Fukuchi et al. 21 F 0.9 - - 1.09
Fukuchi et al. 22 M 0.99 - - 1.21
Fukuchi et al. 25 F 0.73 - - 1.12
Fukuchi et al. 28 F 0.84 - - 1.06
Fukuchi et al. 37 M 1.02 - - 1.17
Neurotypical adults from Fukuchi et al (2018) were speed-matched to the participants post-stroke in the Tyrell et al (2011) data set.
The shaded region represents the younger, neurotypical adults used only in the k-means clustering analysis.

46
CHAPTER 4
WITHIN-SESSION CHANGES IN PROPULSION ASYMMETRY HAVE A LIMITED
EFFECT ON OVERALL GAIT ASYMMETRY IN INDIVIDUALS WITH CHRONIC
STROKE
Abstract
Background. Biomechanical gait impairments, such as reduced paretic propulsion, are common
post-stroke. Biofeedback is commonly used to increase paretic propulsion, but it is unclear if
changes in propulsion impact overall gait asymmetry. There is an implicit assumption that reducing
propulsion asymmetry will improve symmetry in the entire gait pattern, as propulsion has been
related to numerous biomechanical impairments. However, no work has investigated the impact
of reducing propulsion asymmetry on overall gait asymmetry. Objective. To understand how
within-session changes in propulsion asymmetry related to changes in overall gait asymmetry,
operationalized as the combined gait asymmetry metric (CGAM). We hypothesized that decreasing
propulsion asymmetry would reduce CGAM. Methods. Participants completed twenty minutes of
biofeedback training designed to increase paretic propulsion. We calculated the change in
propulsion asymmetry magnitude (Δ |PA|) and the change in CGAM (Δ CGAM) during
biofeedback relative to baseline. Then, we fit a robust linear mixed-effects model with Δ CGAM
as the outcome and a fixed effect for Δ |PA|. Results. We found a positive association between Δ
|PA| and Δ CGAM (intercept β=0.47, p=0.78; CGAM β=2.6, p=0.009). The average Δ |PA| in our
sample was -0.09. This suggests that, on average, we would expect to see a CGAM change of 0.2.
Conclusion. Reducing propulsive asymmetry using biofeedback is unlikely to produce meaningful
reductions in overall gait asymmetry. Therefore, propulsion asymmetry may not be an ideal target
for biofeedback interventions designed to improve overall gait asymmetry. Future work should

47
investigate different approaches to improve overall gait asymmetry. Clinical Trial Registration.
NCT04411303.
Introduction
In the United States, stroke is a leading cause of long-term disability, affecting an estimated
seven million people (Virani et al., 2020). Individuals post-stroke commonly present with
biomechanical gait impairments such as increased spatiotemporal asymmetries, reduced paretic
propulsion, and reduced swing knee flexion (Chen et al., 2005; Olney & Richards, 1996). Paretic
propulsion has become a popular target for clinical interventions and research studies because it is
associated with walking speed in individuals post-stroke (Awad et al., 2020; Hsiao, Awad, et al.,
2016; Hsiao, Zabielski, et al., 2016; Roelker et al., 2019) and can be successfully increased using
a variety of interventions (gait biofeedback, functional electrical stimulation robotics, etc.; Genthe
et al., 2018; Awad et al., 2014; Reisman et al., 2013; Alingh et al., 2020). Research studies
investigating these interventions mainly focus on quantifying the changes in paretic propulsion or
propulsion asymmetry and do not consider the intervention’s effect on symmetry in other gait
impairments.
There is an implicit assumption that reducing propulsion asymmetry will result in improved
symmetry in the entire gait pattern, as propulsion has been related to numerous biomechanical
impairments such as knee flexion (Campanini et al., 2013), trailing limb angle (Lewek & Sawicki,
2019), and step length (Roelker et al., 2019). Although there is evidence that propulsion is also
related to sagittal kinematic asymmetry (Padmanabhan et al., 2020), no work has directly
investigated the impact of reducing propulsion asymmetry on overall gait asymmetry. With
numerous degrees of freedom in the lower limb, it is possible that reducing propulsion asymmetry
may not reduce other biomechanical impairments and, therefore, may not improve overall gait

48
asymmetry. It is also possible that in an attempt to reduce propulsion asymmetry, individuals may
exaggerate existing impairments. The relationship between propulsion asymmetry and overall gait
asymmetry in adults with stroke is unclear.
The primary aim of this study was to understand how within-session changes in propulsion
asymmetry related to changes in overall gait asymmetry, operationalized by the combined gait
asymmetry metric (CGAM; Ramakrishnan et al., 2018, 2019). The CGAM provides a single
comprehensive and easily interpretable measure of overall kinematic and spatiotemporal gait
asymmetry (bounded between 0 and 200), allowing for the inclusion of any biomechanical
impairment (Ramakrishnan et al., 2018, 2019). With the CGAM, we can assess the impact of
changing propulsion asymmetry on multiple dimensions of the gait pattern, not just on a single
biomechanical impairment. To manipulate propulsion asymmetry, we used a visual biofeedback
paradigm targeting paretic propulsion. Because of propulsion’s relationship with numerous
biomechanical impairments (Campanini et al., 2013; Lewek & Sawicki, 2019; Roelker et al., 2019)
and sagittal plane kinematic gait asymmetry (Padmanabhan et al., 2020), we hypothesized that a
decrease in propulsion asymmetry would reduce CGAM.
Methods
Participants
Twenty-nine participants at least six months post-stroke participated in this study.
Participants were recruited from the community and through the Registry for Aging and
Rehabilitation Evaluation database at the University of Southern California. Participants were
included if they could walk independently for five minutes on a treadmill, had paresis confined to
one side, were aged 18 – 80, and had no contraindications to exercise. Participants were excluded
if they had damage to the pons, basal ganglia, or cerebellum on an MRI, signs of cerebellar

49
involvement or extrapyramidal symptoms, uncontrolled hypertension, orthopedic or pain
conditions, or Montreal Cognitive Assessment five-minute protocol score less than nineteen. If a
participant regularly wore an ankle-foot orthosis for community ambulation, they wore it during
the study. The University of Southern California Institutional Review Board approved the
experimental procedures, and all participants provided informed consent before beginning the
experiment. This study is registered on clinicaltrials.gov (NCT04411303).
Clinical assessments
We assessed motor impairment using the lower-extremity Fugl-Meyer scale (Fugl-Meyer
et al., 1975) and balance using the Berg Balance Test (Berg et al., 1992). We measured
cardiovascular endurance using the six-minute walk test (Flansbjer et al., 2005) and determined
overground self-selected gait speed using the ten-meter walk test (Sullivan et al., 2007).
Experimental protocol
We collected kinematic and kinetic data while participants walked on an instrumented dualbelt treadmill. Before beginning data collection, we used a custom staircase algorithm to determine
the participant’s comfortable treadmill walking speed (C. Liu et al., 2022). The participants walked
at their comfortable speed during all trials. First, participants walked for two minutes without
biofeedback (Figure 4.1A). Participants then completed four five-minute biofeedback trials.
During the biofeedback trials, participants walked with paretic propulsion biofeedback. Our
custom paretic propulsion biofeedback code provided the paretic limb's real-time anterior ground
reaction force during stance (Figure 4.1B, purple bar & Figure 4.1C, purple curve). It also provided
peak propulsion end-point feedback, allowing participants to visualize their previous peak
propulsion value (Figure 4.1B, Figure 4.1C, blue dot). The goal was set at +50% of the difference

50
between paretic and non-paretic limbs at baseline (Genthe et al., 2018), with a + 5N tolerance goal
zone (Figure 4.1B & Figure 4.1C, orange rectangle). Participants were given the following
instructions before the first biofeedback trial with a static visual of the biofeedback display (Figure
4.1C): “You’ll see on the screen how much you push off the treadmill with your weaker leg while
you are walking. The purple bar represents how hard you are pushing off the treadmill in real-time.
The blue dot shows the hardest you pushed off on the step before. The goal is to make the blue dot
land in the orange line with each step.” Some participants used the handrails during the experiment
for balance aid (n = 11). If participants used the handrails, they were instructed to lightly touch the
handrail. Participants rested as long as needed between trials.

51
Kinematic data were acquired using a ten-camera motion capture system (Qualisys AB,
Goteborg, Sweden; 100 Hz), and kinetic data using a split-belt instrumented treadmill (Bertec
Corporation, Columbus, OH, USA; 1000 Hz). Participants wore a harness for safety; however, the
harness did not provide body weight support. Markers were placed bilaterally on the iliac crest,
greater trochanter, lateral femoral epicondyle, lateral malleolus, and fifth metatarsal head.
Figure 4.1. Experimental paradigm. A) Treadmill walking paradigm. Participants walked on the treadmill at their
comfortable speed for baseline and biofeedback trials and rested as long as needed between trials. B) Schematic
anterior-posterior GRF trace and corresponding components of the visual biofeedback display. The anterior-posterior
GRF signals represented by the dashed lines are not shown in the biofeedback display. The solid purple section of
the curve corresponds to the paretic GRF signal that was displayed in the real-time biofeedback. The blue dot
indicates the peak paretic propulsion in value that was provided as end-point biofeedback. The goal (orange dashed
line) was set at +50% of the difference between peak paretic propulsion and peak non-paretic propulsion at baseline.
The bounds of the goal zone were + 5 N (orange rectangle). C) Snap-shot of real-time paretic propulsion biofeedback
display. The purple bar represents the real-time anterior-posterior GRF signal at a given time. The blue dot represents
peak propulsion and appears after every step to provide end-point feedback to the participant. The orange rectangle
represents the goal zone. Abbreviations: GRF; ground reaction force.

52
Data processing
All kinematic and kinetic data were processed and analyzed in MATLAB R2020a
(MathWorks, Natick, MA). Kinematic data were low pass filtered with a 6 Hz cutoff, and kinetic
data were low pass filtered with a 20 Hz cutoff (Winter, 2009). Foot-strike and toe-off were defined
as the most anterior and posterior positions of the lateral malleoli markers, respectively.
We first calculated peak propulsion (P), the highest value from the anterior/posterior
ground reaction force during the stance phase, for both the right and left limbs. Then, we calculated
propulsion asymmetry magnitude (|PA|; Equation 1; Padmanabhan et al., 2020).
|PA| = ( − )
−
(1)
We then calculated the change in propulsion asymmetry magnitude (Δ |PA|) from baseline for each
stride taken during the biofeedback trials. To do this, we averaged propulsion asymmetry
magnitude across the final thirty strides of the baseline trial. Then, we subtracted this value from
propulsion asymmetry magnitude for each stride taken during the biofeedback trials, reflecting
how much participants changed propulsion asymmetry compared to baseline. We did not account
for the direction of the propulsion asymmetry to allow us to calculate a change score for each
stride.
Combined Gait Asymmetry Metric (CGAM)
The CGAM is an overall gait asymmetry measure developed by Ramakrishnan et al. (2018)
that is based on a modified Mahalanobis distance. This metric allows the inclusion of any
lateralized biomechanical measures of interest and provides a single measure of overall gait
asymmetry between 0 (no asymmetry) and 200 (completely asymmetric). First, a symmetry index

53
is calculated for each biomechanical impairment of interest using Equation 2. Then, the symmetry
indices (si) are combined into a single matrix for each participant (S), with m columns (number of
metrics) and n rows (number of strides). Finally, the symmetry matrix and the covariance of the
symmetry matrix (KS) are used to calculate CGAM for each stride (Equation 3). A value of zero
represents perfect symmetry, and the greater the CGAM magnitude, the greater the asymmetry in
the gait pattern.
= 100 ∗ (ℎ − )
0.5 (ℎ + ) (2)
= � ∗ () ∗ ′
Σ () (3)
We had eight candidate variables to include in the CGAM calculation that encompassed
commonly studied biomechanical gait impairments post-stroke: single-limb support time, doublelimb support time, stance time, step length, peak swing knee flexion, peak hip flexion, trailing limb
angle, and circumduction. Before calculating CGAM using these variables, we checked for
collinearity using a variance inflation factor (VIF) cutoff of ten (James et al., 2013). We found that
stance time and single-limb support time had VIF values greater than ten. After removing stance
time, the remaining variables had a VIF of less than ten. Therefore, we included the following
seven variables in the CGAM calculation: double-limb support time, single-limb support time,
step length, peak swing knee flexion, peak hip flexion, trailing limb angle, and circumduction.
Definitions for the variables included in the CGAM calculation are described in the following
section. We chose not to include propulsion in the CGAM calculation to understand the effect of

54
manipulating propulsion asymmetry on overall gait asymmetry outside of what the biofeedback
was designed to change.
Finally, we calculated the change in CGAM (Δ CGAM) from baseline for each stride. To
do this, we averaged the CGAM values from the final thirty strides of the baseline trial and
subtracted this value from the CGAM of each stride taken during the biofeedback trials.
Definitions of variables included in CGAM calculation
Step length was defined as the anterior-posterior distance between the lateral malleoli
markers at heel strike. Knee flexion was defined as the angle between the thigh segment (between
lateral tibial epicondyle to greater trochanter markers) and shank segment (between lateral
malleolus and lateral tibial epicondyle). Hip flexion was defined as the angle between the thigh
segment and pelvis segment (between greater trochanter and iliac crest markers). Circumduction
was the maximal lateral difference between the ankle marker during swing and the same ankle
marker during stance. Trailing limb angle was defined as the angle between the vertical lab axis
and the vector created by the greater trochanter and lateral malleoli markers. Double-limb support
time was the time from contralateral heel strike to ipsilateral toe off. Single-limb support time was
the time from contralateral toe off to contralateral heel strike.
Statistical analyses
All statistical analyses were performed in R (4.2.2; R Core Team, n.d.) First, we confirmed
that all participants had propulsion asymmetry at baseline. If participants had a baseline propulsion
asymmetry < 0.11 (the minimal detectable change for propulsion asymmetry in our data), they
were removed from all analyses. Then, we wanted to establish that participants used biofeedback
to increase paretic propulsion as intended. To do this, we first normalized paretic propulsion values

55
to body weight and averaged normalized paretic propulsion over the final thirty strides of each
trial. Then, we fit a linear mixed effects model with average normalized paretic propulsion as the
outcome, a fixed effect for trial, and a random intercept using the lme4 package (Bates et al., 2015).
This allowed us to verify that, as a group, participants used the biofeedback to increase paretic
propulsion in one or more feedback trials compared to baseline.
Next, because we only provided biofeedback on the paretic limb, we determined how
increasing paretic propulsion influenced propulsion asymmetry. To do this, we fit a linear mixedeffects model with Δ |PA| as the outcome, a fixed effect for change in normalized paretic
propulsion, and a random intercept for each participant. We also included a random slope because
it improved the model fit (determined using the Bayesian information criterion score). We fit the
model to the final 113 strides of each trial for each participant, which matched the minimum
number of strides per trial taken in the cohort, ensuring that each participant had data from the
same number of strides included in the analyses. After fitting the model, the residuals did not meet
regression assumptions; therefore, we ran robust mixed-effects models to account for those
violations (Koller, 2016) and reported those results.
For our primary analysis, we examined the relationship between Δ |PA| and the
corresponding Δ CGAM. To do this, we fit a linear mixed-effects model with Δ CGAM as the
outcome, a fixed effect for Δ |PA|, and a random intercept for each participant. We also included a
random slope because it improved the model fit (determined using the Bayesian information
criterion score). We fit the model to the last 113 strides of each trial for each participant, which
matched the minimum number of strides per trial taken in the cohort. After fitting the model, the
residuals did not meet regression assumptions; therefore, we ran robust mixed-effects models to
account for those violations (Koller, 2016) and reported those results.

56
Results
Twenty-nine participants post-stroke participated in the study. Two participants were
excluded due to marker occlusion, and four were excluded because they did not have a propulsion
asymmetry > 0.11 at baseline. We excluded an additional two participants because they were not
able to consistently generate paretic propulsion throughout the trials (i.e., paretic propulsion = 0
on many strides). Therefore, we included 21 participants in the analyses. Clinical demographics
for participants included in the analysis are provided in Table 4.1.
Table 4.1. Participant demographics.
Race/Ethnicity Sex Age Affected
side AFO LEFM Berg
6-minute
walk test
(m)
10-meter
walk speed
(m/s)
Treadmill
speed
(m/s)
Asian M 63 R No 28 51 378 1.19 0.61
Asian F 53 R No 29 49 322 1.07 0.55
White M 59 L No 29 45 221 0.70 0.37
White/Hispanic F 35 R No 24 45 424 1.39 0.78
White/Hispanic M 61 L Yes 22 42 223 0.73 0.39
Asian M 59 L No 24 50 370 1.19 0.74
White M 74 L No 26 44 104 0.37 0.25
White F 61 R No 24 52 162 0.51 0.50
Asian M 53 R No 25 52 431 1.17 0.60
Black M 66 R No 27 49 266 1.54 0.84
Asian M 78 L No 24 48 236 0.86 0.49
White/Hispanic F 46 L Yes 24 52 249 0.95 0.54
White/Hispanic F 55 L No 28 55 311 1.04 0.54
White/Hispanic M 63 L Yes 18 53 247 0.92 0.56
Black M 56 L Yes 11 32 243 0.81 0.52
White F 64 R Yes 23 46 315 0.92 0.48
White/Hispanic M 69 L Yes 22 39 98 0.23 0.30
Black M 79 L No 28 50 431 1.38 0.72
White/Hispanic M 44 R Yes 16 54 428 1.07 0.58
White/Hispanic F 31 L No 25 55 280 0.86 0.46
Asian M 37 R Yes 22 49 109 0.26 0.20
LEFM, Lower Extremity Fugl-Meyer; M, male; F, female; R, right; L, left.
Participants used biofeedback to increase paretic propulsion
During the biofeedback trials, peak propulsion was within the provided goal zone for 36%
of strides, below the goal zone for 51% of strides, and above the goal zone for 13% of strides
across all participants (Supplemental Figure 4.1). Despite over half of the strides being below the
propulsion goal, participants still increased normalized paretic propulsion compared to baseline
during biofeedback trials 2 - 4 (Figure 4.2A and Figure 4.2B). This confirms that participants were

57
able to use the biofeedback to increase paretic propulsion as intended. Average normalized
propulsion and change in normalized propulsion for each trial are presented in Table 4.2. For all
remaining analyses, we only included data from trials 2 – 4, where participants used the
biofeedback to increase paretic propulsion.
Table 4.2. Average normalized propulsion and change in normalized propulsion across trials.
Normalized propulsion
(% bw)
Δ normalized propulsion
(% bw)
Baseline 5.6 (2.4)
Trial 1 6.2 (2.4) 0.5 (1.8)
Trial 2 6.6 (2.5)* 1.0 (1.6)
Trial 3 6.7 (2.4)* 1.1 (1.9)
Trial 4 6.7 (2.4)* 1.1 (2.0)
Increasing paretic propulsion reduced propulsion asymmetry magnitude
Next, we wanted to understand whether increasing paretic propulsion led to a decrease in
propulsion asymmetry. We found that increased normalized paretic propulsion was related to
reduced propulsion asymmetry magnitude (β = -0.05, p = 0.0001; Figure 4.2C).
Figure 4.2. Paretic propulsion across trials and the relationship between change in paretic propulsion and change in propulsion
asymmetry. A) Paretic propulsion across trials. * denotes a significant increase in paretic propulsion compared to baseline. B)
Change in paretic propulsion from baseline for biofeedback trials. C) Change in paretic propulsion vs. change in propulsion
asymmetry. The black line is the group-level model fit (fixed effect). The dashed lines represent individual model fits from the
robust mixed-effects model. Data points are data for individual strides, colored by participant. Abbreviations: BL, baseline; BFB,
biofeedback; % bw, percent body weight.

58
Changes in propulsion asymmetry magnitude led to small changes in overall gait asymmetry
We found a
significant association
between Δ |PA| and ΔCGAM
(intercept β = 0.47, p = 0.78;
Δ |PA| β = 2.6, p = 0.009;
Figure 4.3), suggesting that a
reduction in propulsion
asymmetry is associated
with a reduction in CGAM.
However, the average
change in propulsion
asymmetry magnitude in our
sample was -0.09,
suggesting that on average,
we would expect to see a CGAM change of 0.2. This indicates that reducing propulsion asymmetry
with biofeedback is unlikely to generate meaningful changes in overall gait asymmetry.
Discussion
We aimed to understand how changes in propulsion asymmetry related to changes in overall
gait asymmetry, measured by CGAM. We found a significant association between Δ |PA| and Δ
CGAM; however, the average expected CGAM change (0.2) suggests that reducing propulsion
asymmetry likely will not result in meaningful changes to overall gait asymmetry. This suggests
Figure 4.3. Relationship between change in propulsion asymmetry magnitude and change
in the combined gait asymmetry metric (CGAM). The black line is the group-level model
fit (fixed effect). The dashed lines represent individual model fits from the robust mixedeffects model. Data points are data for an individual stride, colored by participant.

59
that propulsion asymmetry may not be an ideal target for biofeedback-based interventions designed
to improve overall gait asymmetry.
Participants used biofeedback to increase paretic propulsion and reduce propulsion asymmetry
Consistent with previous work (Genthe et al., 2018), we found that participants were able
to use paretic propulsion biofeedback to increase paretic propulsion and reduce propulsion
asymmetry. However, our results indicate that hitting the biofeedback goal zone was very
challenging for some participants. Specifically, in only 13% of strides taken across all participants,
the peak propulsion produced was sufficient to reach the propulsion goal zone (i.e., hit the target);
whereas, for 51% of strides, the participants’ peak propulsion was below the propulsion goal zone
(i.e., undershot the target). Previous work has set the difficulty of the propulsion biofeedback goal
in a more personalized manner to attempt to provide a challenging yet attainable goal (Genthe et
al., 2018). This may have led to a higher success rate (hitting the target more often); however,
some participants with more cognitive and motor impairments generally have difficulty using
biofeedback (French et al., 2021). This suggests that changing the goal would not necessarily lead
to an improvement in performance with biofeedback. Future work is needed to determine the most
effective way to set a biofeedback goal to target biomechanics and promote learning.
Though participants walked with peak propulsion values that were below the propulsion
goal for 51% of strides, there was still a group-level increase in paretic propulsion in trials 2-4.
This propulsion increase led to a group-level decrease in propulsion asymmetry. However, the
strength and direction of the relationship between paretic propulsion and propulsion asymmetry
varied across participants. Notably, three participants increased propulsion asymmetry when they
increased paretic propulsion (Figure 4.2C). This may be due, in part, to the biofeedback providing
information only on the paretic limb, leaving the non-paretic limb unconstrained. Therefore,

60
participants’ success using the biofeedback did not depend on their non-paretic propulsion.
Participants could have increased paretic propulsion while also either increasing or decreasing
non-paretic propulsion, resulting in increased propulsion asymmetry. Future work should
investigate the impact of providing bilateral propulsion biofeedback, which may better reduce
propulsion asymmetry.
Changing propulsion asymmetry had minimal effect on overall gait asymmetry
We found that changing propulsion asymmetry had a minimal effect on overall gait asymmetry.
This was somewhat surprising due to the evidence of an association between paretic propulsion
and other metrics of lower limb asymmetry (Padmanabhan et al., 2020) as well as many of the
variables included in our CGAM calculation, such as trailing limb angle (Lewek & Sawicki, 2019),
swing knee flexion (Campanini et al., 2013), and step length asymmetry (Roelker et al., 2019).
The small relationship between Δ |PA| and ΔCGAM may be explained by a few factors. First, there
are multiple degrees of freedom in the lower limb, resulting in a variety of limb configurations that
can reduce propulsion asymmetry without reducing other dimensions of gait asymmetry. For
example, there is evidence that increasing paretic propulsion with an exosuit reduces hip hiking
and circumduction without changing peak swing knee flexion, stance time, or swing time (Awad
et al., 2017). The number of possible limb configurations may have been larger in our study due
to providing paretic propulsion biofeedback, leaving the nonparetic limb unconstrained.
Propulsion biofeedback provided on both limbs may have a larger impact on overall gait
asymmetry, as it would constrain nonparetic propulsion.
Secondly, the specific muscles recruited to increase paretic propulsion (and reduce propulsion
asymmetry) may have an undesirable effect on overall gait asymmetry. Propulsion asymmetry is
inversely associated with paretic plantar flexor activity (Brough et al., 2022). However, individuals

61
post-stroke can still generate relatively high levels of propulsion without plantar flexor activity by
recruiting the hamstrings to compensate (Brough et al., 2022). Recruiting the hamstrings instead
of the plantar flexors to reduce propulsion asymmetry may lead to undesirable changes in other
biomechanical variables. Musculoskeletal simulation work by Sauder et al. (2019) found that
optimizing functional electrical stimulation to the calf while walking reduced propulsive
asymmetry. Yet, the resulting reduction in propulsion asymmetry did not restore symmetry in
kinematic measures (e.g., knee flexion, hip flexion; Sauder et al., 2019). Sauder et al. (2019)
suggest that stimulating additional muscles may improve joint symmetry. However, no simulated
or experimental studies (Awad et al., 2014; Reisman et al., 2013; Sauder et al., 2019) have
investigated the effect of doing so. Additionally, inappropriate recruitment of musculature during
walking can also increase braking forces (Turns et al., 2007). Previous work has shown that greater
braking forces (generated by the vastus medialis; Brough et al., 2022) are associated with reduced
swing knee flexion (Brough et al., 2022; Dean et al., 2020; Ohta et al., 2023), reduced trailing limb
angle (Ohta et al., 2023), and increased step length asymmetry (Sánchez et al., 2021), suggesting
that braking forces may also impact overall gait asymmetry. Future work should incorporate
electromyography measures to help understand muscular contributions to changes in overall gait
asymmetry.
The timing of paretic propulsion may also be an important factor that contributes to the
relationship between propulsion asymmetry and overall gait asymmetry. Propulsion occurs earlier
in the stance phase in individuals post-stroke compared to neurotypical adults, and it occurs earlier
in the paretic limb than the non-paretic limb (Alam et al., 2022). The timing of propulsion
assistance has been related to metabolic cost in individuals post-stroke (Penke et al., 2019),
suggesting that propulsion timing plays an important role in post-stroke gait. It is currently unclear

62
how propulsion timing relates to other biomechanical impairments in individuals post-stroke. We
did not control propulsion timing with our biofeedback scheme; therefore, future work should
investigate the effect of both propulsion magnitude and timing on overall gait asymmetry.
Finally, participants only completed a single session of paretic propulsion biofeedback
training. It is possible that with practice, individuals may learn a different gait strategy to maximize
propulsion that may lead to a more symmetric gait pattern or may increase their capacity to increase
propulsion with longer-term training. There is evidence that multiple training sessions of robotic
gait training (Alingh et al., 2021) and fast functional electrical stimulation training (Awad et al.,
2014) that targets paretic propulsion can increase paretic propulsion, reduce propulsion
asymmetry, and increase paretic trailing limb angle. This suggests that individuals post-stroke have
the capacity to increase paretic propulsion with repeated training. However, it is currently unclear
how repeated sessions of paretic propulsion biofeedback training would impact overall gait
asymmetry. Future work should incorporate overall gait asymmetry measures across multiple days
of paretic propulsion biofeedback training to understand how repeated training impacts overall gait
asymmetry.
The importance of understanding the effect of an intervention on overall gait asymmetry
To date, very few studies in adults with stroke have included comprehensive measurements of
gait asymmetry, even though individuals post-stroke often cite improving gait appearance as an
important rehabilitation goal (Bohannon et al., 1991). The importance of assessing overall gait
asymmetry is reflected in the most recent recommendations from the third Stroke Recovery and
Rehabilitation Roundtable, which recommend including multivariate kinematic measures in
research studies to provide a more comprehensive and interpretable measure of gait asymmetry
(Van Criekinge et al., 2023). Most interventional studies aim to manipulate and measure changes

63
in a single biomechanical impairment or functional outcomes and do not consider the impact of
the intervention on overall gait asymmetry. Recent work has demonstrated the value of
incorporating overall gait asymmetry measures (such as CGAM). For example, Shin et al. (2020)
found that individuals post-stroke increased their gait speed over the first three months of standard
inpatient physical therapy without restoring kinematic symmetry. This suggests that changes in
activity (such as gait speed) may be independent of changes in overall gait biomechanics that we
commonly assume to occur alongside these activity-level changes. The inclusion of overall gait
asymmetry measures, in addition to measures of walking function, in Shin et al. (2020) provides a
more nuanced understanding of the multifaceted nature of walking recovery that could not
appreciated otherwise.
Potential avenues to reduce overall gait asymmetry
It is currently unclear how best to reduce overall gait asymmetry after stroke. As mentioned
above, biofeedback that contains bilateral information on propulsion magnitude and timing may
have a different effect on overall gait asymmetry than the unilateral biofeedback provided in this
study. Another potential avenue is to provide biofeedback that targets multiple dimensions of gait
simultaneously. In Day et al. (2019), participants post-stroke were provided biofeedback on hip
and knee flexion angles simultaneously and successfully improved hip and knee flexion angles
within a single session. This suggests that individuals post-stroke have some ability to use
biofeedback provided on multiple dimensions of gait. However, the biofeedback used by Day et
al. (2019) only had information on two biomechanical impairments. Including more biomechanical
impairments in the biofeedback may make the biofeedback difficult to use. This is particularly
important to consider because cognitive impairment is common post-stroke (Nys et al., 2007), and
cognitive status has been related to the ability to use biofeedback to improve performance on a

64
locomotor task (French et al., 2021). However, more work is needed to understand the efficacy of
overall biofeedback and test other rehabilitation approaches that may be appropriate to improve
overall gait asymmetry post-stroke.
Considering other aspects of walking impairment and function
Improvement in gait quality (i.e., appearance) is a well-known goal of rehabilitation for
people who have had a stroke (Bohannon et al., 1991). This is the foundational motivation to
develop interventions that reduce overall gait asymmetry after stroke. However, it is still unclear
how overall gait asymmetry relates to other important measures of walking function and
impairment (e.g., gait speed or dynamic balance) and whether reducing overall gait asymmetry is
necessary to achieve or maximize progress toward different rehabilitation goals (e.g., faster
walking speeds). Some evidence indicatesthat asymmetric gait may be necessary to optimize some
other aspect of gait, such as stability or metabolic cost of walking. For example, reducing step
length asymmetry impairs dynamic balance in individuals post-stroke (Park et al., 2021), and
predictive simulation work suggests that spatiotemporal asymmetry may be metabolically optimal
for individuals post-stroke (Johnson et al., 2022). Previous work has also demonstrated that
individuals post-stroke increase gait speed without improving kinematic symmetry (Shin et al.,
2020), suggesting that reducing overall gait asymmetry may not be necessary to improve gait
speed. However, no studies have directly investigated the impact of reducing overall gait
asymmetry on metabolic cost, stability, or gait speed; therefore, more work is necessary to define
the relationship between these domains of walking impairment and overall gait asymmetry.
Conclusion
We found that individuals post-stroke were able to use visual biofeedback to increase
paretic propulsion and reduce propulsion asymmetry, which were associated with limited

65
improvements in overall gait asymmetry. This suggests propulsion asymmetry may not be an ideal
target for biofeedback interventions designed to improve overall gait asymmetry. Future work
should investigate different approaches, such as multidimensional biofeedback, to improve overall
gait asymmetry.
Supplemental Figure 4.1. Propulsion asymmetry for all strides taken during the biofeedback trials, coded by where the peak
propulsion value was in relation to the propulsion biofeedback goal. The green strides were within the propulsion goal zone,
the blue strides were above the goal, and the brown strides were below the goal.

66
CHAPTER 5
VISUOSPATIAL SKILLS EXPLAIN DIFFERENCES IN THE ABILITY TO USE
PROPULSION BIOFEEDBACK POST-STROKE
This work is currently in-revision in the Journal of Neurologic Physical Therapy.
Kettlety SA, Finley JM, Leech KA. Visuospatial skills explain differences in the ability to use
propulsion biofeedback post-stroke.
Abstract
Background and Purpose. Visual biofeedback can be used to help people post-stroke reduce
biomechanical gait impairments. Using visual biofeedback engages an explicit, cognitively
demanding motor learning process. Participants with better overall cognitive function are better
able to use visual biofeedback to promote locomotor learning; however, which specific cognitive
domains are responsible for this effect are unknown. We aimed to understand which cognitive
domains were associated with performance during acquisition and immediate retention when using
visual biofeedback to increase paretic propulsion in individuals post-stroke. Methods. Participants
post-stroke completed cognitive testing, which provided scores for different cognitive domains,
including executive function, immediate memory, visuospatial/constructional skills, language,
attention, and delayed memory. Next, participants completed a single session of paretic propulsion
biofeedback training, where we collected treadmill-walking data for twenty minutes with
biofeedback and two minutes without biofeedback. We fit separate regression models to determine
if cognitive domain scores, motor impairment (measured with the lower-extremity Fugl-Meyer),
and gait speed could explain propulsion error and variability during biofeedback use and recall
error during immediate retention. Results. Visuospatial/constructional skills and motor
impairment best-explained propulsion error during biofeedback use (adjusted R2 = 0.56, p =

67
0.0008), and attention best-explained performance variability (adjusted R2 = 0.26, p = 0.02).
Language skills best-explained recall error during immediate retention (adjusted R2 = 0.37, p =
0.02). Discussion and Conclusions. These results demonstrate that specific cognitive domain
impairments explain variability in locomotor learning outcomes in individuals post-stroke. This
suggests that with further investigation, specific cognitive impairment information may be useful
to predict responsiveness to interventions and personalize training parameters to facilitate
locomotor learning.
Introduction
Reducing biomechanical gait impairments is an important goal for individuals post-stroke
and a common component of gait rehabilitation (Bohannon et al., 1991). This is, in part, because
biomechanical gait impairments, such as step length asymmetry and reduced paretic swing knee
flexion, are associated with increased metabolic cost (Awad, Palmer, et al., 2015; Finley & Bastian,
2017) and fall risk (Burpee & Lewek, 2015; Matsuda et al., 2017; Wei et al., 2017). People poststroke can alter biomechanical gait impairments (including increasing paretic propulsion and
reducing step length asymmetry; Genthe et al., 2018; Padmanabhan et al., 2020; Sánchez & Finley,
2018) using visual biofeedback (Tate & Milner, 2010), but the ability to do so varies from person
to person. Recent evidence suggests that this inter-individual variability may be due to post-stroke
cognitive impairment (French et al., 2021).
Using visual biofeedback to alter gait impairments is an explicit, cognitively demanding
motor learning task (Leech et al., 2022). Participants are given external information about their
task performance and asked to reduce their movement error (Spencer et al., 2021). This requires
cognitive processing, as participants must understand the information contained in the
biofeedback, its relationship to their movements, and the instructions for the task. Cognitive

68
impairment is common post-stroke (Nys et al., 2007), with deficits in visuospatial skills, attention,
executive function, language, and memory among the most commonly affected domains (Jokinen
et al., 2015; Pinter et al., 2019; Yang et al., 2020). Previous work found that participants with
higher fluid cognition scores had better performance when gait biofeedback was present and better
24-hour retention of a biofeedback-driven gait pattern (French et al., 2021). However, fluid
cognition is a global measure of cognition reflecting someone’s ability to adapt to new information
and solve problems (Heaton et al., 2014); therefore, it is still unknown which specific cognitive
domains are important for performance during practice and retention of an explicitly learned
locomotor skill.
There is no evidence linking specific cognitive domains to performance during and
immediate retention of a gait biofeedback task, but there are logical roles for commonly affected
domains. For example, visuospatial skills may play a role in understanding the visual biofeedback
display and how movements map to the display. Attention may be important to direct cognitive
resources to focus on the information contained in the biofeedback display, as opposed to other
aspects of the environment (e.g., the experimenter, feet on the treadmill, etc.). Executive function,
which is the ability to identify goals and adapt plans to achieve those goals (APA Dictionary of
Psychology, n.d.; Bonelli & Cummings, 2007; Leh et al., 2010), could play a role in integrating
the biofeedback information and using it to accomplish the biofeedback goal. Language skills may
be required to understand verbalized instructions. Finally, during a retention test, recalling the
biofeedback-driven gait pattern likely requires intact memory. Currently, we do not understand if
one or a combination of these domains is associated with biofeedback-driven performance and
immediate retention of the biofeedback-driven gait pattern. Identifying specific cognitive domains
involved in motor learning tasks may facilitate the development of more personalized motor

69
rehabilitation approaches by individualizing training parameters based on specific cognitive
impairments (Lingo VanGilder, Hooyman, et al., 2021).
Here, we aimed to understand which cognitive domains were associated with performance,
performance variability, and immediate retention during a paretic propulsion biofeedback task in
individuals post-stroke. Because there is little research informing our understanding of the
relationship between specific cognitive domains and biofeedback-driven changes in gait
impairments, we performed an exploratory analysis to understand which cognitive domains
contribute to locomotor performance, variability during performance, and immediate retention
when using visual biofeedback. We considered the following six commonly impacted domains of
cognition: immediate memory, visuospatial/constructional skills, language, attention, delayed
memory, and executive function.
Methods
Participants
Twenty-nine participants post-stroke completed a single session of paretic propulsion
biofeedback training. Participants were recruited from the community and through the Registry
for Aging and Rehabilitation Evaluation database at the University of Southern California.
Participants were included if they could walk independently for five minutes on a treadmill, had
paresis confined to one side, were aged 18 – 80, and had no contraindications to exercise.
Participants were excluded if they had damage to the pons, basal ganglia, or cerebellum on an
MRI, signs of cerebellar involvement or extrapyramidal symptoms, uncontrolled hypertension,
orthopedic or pain conditions, or a Montreal Cognitive Assessment five-minute protocol score of
less than nineteen (Wong et al., 2018). We chose to exclude participants who had Montreal
Cognitive Assessment five-minute protocol score of less than nineteen to screen for moderate-to-

70
severe cognitive impairment and ensure participants were able to provide informed consent. If a
participant regularly wore an ankle-foot orthosis for community ambulation, they wore it during
the study (n = 9). The University of Southern California Institutional Review Board approved the
experimental procedures, and all participants provided informed consent before beginning the
experiment. This study is registered on clinicaltrials.gov (NCT04411303).
Cognitive testing
Participants completed the Repeatable Battery for the Assessment of Neuropsychological
Status (RBANS; Randolph 1988) and the Trail Making B test (Bowie & Harvey, 2006). The
RBANS is a comprehensive cognitive testing battery that provides normalized scores for
immediate memory, visuospatial/constructional skills, language, attention, and delayed memory.
The RBANS also provides a total score, which accounts for all tested cognitive domains. Data for
all participants were age-normalized using the RBANS scoring manual. We did not perform racebased normalization on these data to align with the recommendation from the American Academy
of Clinical Neuropsychology (The American Academy of Clinical Neuropsychology, 2021). For
all domains, higher scores indicate better cognitive function. All RBANS were independently
double-scored to identify and resolve any discrepancies in scoring. When discrepancies were
found, the two scorers discussed the results and resolved the discrepancies. The Trail Making B
test provided a score for executive function and was scored on time to correct completion in
seconds (Sánchez-Cubillo et al., 2009). We included the Trail Making B Test because the RBANS
does not have a dedicated executive function score, and impairment in executive function is
common post-stroke (Nys et al., 2007). Higher scores indicate worse executive function. The test
was terminated after 300 seconds, which is the highest possible score.

71
Gait analysis procedures
We collected kinematic and kinetic data while participants walked on an instrumented dualbelt treadmill. Before beginning data collection, we used a custom staircase algorithm to determine
the participant’s comfortable treadmill walking speed (C. Liu et al., 2022). The participants walked
at their comfortable speed during all trials. First, participants walked for two minutes without
biofeedback (Figure 5.1A). Participants then completed four five-minute biofeedback trials.
During the biofeedback trials, participants walked with paretic propulsion biofeedback. Our
custom paretic propulsion biofeedback code provided the paretic limb's real-time anterior ground
reaction force during stance (Figure 5.1B, purple bar & Figure 5.1C, purple curve). It also provided
peak propulsion end-point feedback, allowing participants to visualize their previous peak
propulsion value (Figure 5.1B, Figure 5.1C, blue dot). The goal was set at +50% of the difference
between paretic and non-paretic limbs at baseline (Genthe et al., 2018), with a + 5N tolerance goal
zone (Figure 5.1B & Figure 5.1C, orange rectangle). Participants were given the following
instructions before the first biofeedback trial with a static visual of the biofeedback display (Figure
5.1C): “You’ll see on the screen how much you push off the treadmill with your weaker leg while
you are walking. The purple bar represents how hard you are pushing off the treadmill in real time.
The blue dot shows the hardest you pushed off on the step before. The goal is to make the blue dot
land in the orange line with each step.” The participants were not provided with a practice session
using the biofeedback before completing the first biofeedback trial. Finally, participants walked
for two minutes without biofeedback to assess immediate retention. Participants were given the
following instructions prior to starting the immediate retention trial: “You will not be provided any
feedback during this trial. You should try to walk in the same way you did in the trials when you

72
used the feedback. So, whatever you were doing to meet the feedback goal, I want you to recreate
that (French et al., 2020).” Participants rested as long as needed between trials.
Data acquisition
Kinematic data were acquired using a ten-camera motion capture system (Qualisys AB,
Goteborg, Sweden; 100 Hz), and kinetic data using a split-belt instrumented treadmill (Bertec
Corporation, Columbus, OH, USA; 1000 Hz). Participants wore a harness for fall prevention, but
the harness did not provide any body weight support. Markers were placed bilaterally on the iliac
crest, greater trochanter, lateral femoral epicondyle, lateral malleolus, and fifth metatarsal head.
Figure 5.1. Experimental paradigm. A) Treadmill walking paradigm. Participants walked on the treadmill at their comfortable
speed for baseline, biofeedback, and retention trials and rested as long as needed between trials. B) Schematic anterior-posterior
GRF trace and corresponding components of the visual biofeedback display. The anterior-posterior GRF signals represented by the
dashed lines are not shown in the biofeedback display. The solid purple section of the curve corresponds to the paretic GRF signal
that was displayed in the real-time biofeedback. The blue dot indicates the peak paretic propulsion in value that was provided as
end-point biofeedback. The goal (orange dashed line) was set at +50% of the difference between peak paretic propulsion and peak
non-paretic propulsion at baseline. The bounds of the goal zone were + 5 N (orange rectangle). C) Snap-shot of real-time paretic
propulsion biofeedback display. The purple bar represents the real-time anterior-posterior GRF signal at a given time. The blue dot
represents peak propulsion and appears after every step to provide end-point feedback to the participant. The orange rectangle
represents the goal zone. Abbreviations: GRF; ground reaction force.

73
The visual biofeedback code was written in MATLAB (MathWorks, Natick, MA, USA) and used
real-time kinematic and kinetic data streamed from Qualisys Track Manager at 100 Hz.
Data analyses
Gaps of less than thirty frames in the marker trajectory data were filled using Qualisys
Track Manager. Kinematic and kinetic data were processed and analyzed in MATLAB R2020a
(MathWorks, Natick, MA). We used a fourth-order low-pass Butterworth filter on kinematic (6 Hz
cutoff) and kinetic data (20 Hz cutoff). Foot-strike and toe-off were defined as the most anterior
and posterior positions of the lateral malleoli markers, respectively.
We calculated three dependent variables: biofeedback performance, performance
variability, and immediate retention. Biofeedback performance – the ability of the participants to
use the biofeedback to increase paretic propulsion – was represented by normalized propulsion
error averaged over the final thirty strides of biofeedback training (Long et al., 2015). We
completed four steps to calculate normalized propulsion error: 1) Calculate propulsion error for
each stride (Figure 5.2B). This was defined as the absolute distance of peak propulsion to the
closest border of the goal zone (Figure 5.2A). Of note, this method was used to calculate baseline
propulsion error, even though the goal was not visible to participants at this point in the experiment.
We chose to calculate the absolute error (as opposed to a directional error) because we were
interested in the participants' ability to manipulate propulsion to achieve the provided goal.
Therefore, either over- or undershooting the target was considered an error. 2) Average the
propulsion error over the final thirty strides of the baseline trial. 3) Divide the propulsion error
value for each stride by the average baseline propulsion error and multiply by 100 to obtain a
normalized propulsion error (Figure 5.2C). A normalized propulsion error of 100% represents
average baseline error. Any value under 100% reflects reduced normalized propulsion error

74
relative to baseline performance, and any value over 100% reflects increased normalized
propulsion error relative to baseline performance. We chose to normalize each participant’s
propulsion error to their baseline error to allow for comparison between individuals, as the goal
was personalized for each participant. 4) Average normalized propulsion error over the final thirty
strides of biofeedback training. We chose this been to capture the plateau in performance after
ample practice with the biofeedback. This value was used for the performance analysis. Figure 5.2
displays the transformation between peak propulsion, propulsion error, and normalized propulsion
error of a representative participant's data.
We quantified the participants variability during performance by calculating the coefficient
of variation of paretic propulsion across the four biofeedback trials. The coefficient of variation
was calculated by dividing the standard deviation of paretic propulsion by the mean paretic
propulsion across all biofeedback trials and multiplying by 100%, providing a measure of
propulsion variability across biofeedback training.
Figure 5.2. Normalized propulsion error calculation for a representative participant. A) Paretic propulsion across strides. The
horizontal dashed lines represent the goal propulsion. B) Propulsion error across strides. Propulsion error was calculated for each
stride as the absolute distance between peak propulsion and the closest goal zone border. Propulsion error was then averaged over
the final 30 baseline strides (between vertical dashed lines). C) Normalized propulsion error across strides. The propulsion error
value for each stride was divided by the average baseline propulsion error and multiplied by 100 to obtain normalized propulsion
error. A normalized propulsion error of 100% is equivalent to baseline error (horizontal dashed line). The average of the final 30
biofeedback strides was our performance outcome measure (between vertical dashed lines). The difference between the average
final thirty biofeedback and the average first immediate retention strides was our immediate retention outcome measure.

75
To capture immediate retention – the ability of participants to recreate the intended walking
pattern without relying on the biofeedback – we calculated recall error, which we defined as the
difference between the average normalized propulsion error over the final thirty biofeedback
strides and the average of the first thirty strides of the immediate retention trial (similar to previous
work; Kim et al., 2019; VanGilder et al., 2022). This allows us to infer retention of the newly
learned walking pattern based on how much of the gait pattern they can recall without the
biofeedback. A recall error of 0% indicates that participants normalized propulsion error was the
same during immediate retention as their performance at the end of biofeedback training. Positive
recall error values mean they were further away from the goal compared to the end of biofeedback
training, and negative values mean participants were closer to the goal at immediate retention
compared to the end of biofeedback training.
Statistical analyses
All analyses were conducted in R (4.2.2;R Core Team, n.d.). We performed linear
regression with best-subsets selection to determine the cognitive domains associated with
performance, performance variability, and immediate retention. Best-subsets selection evaluates
each possible combination of predictors and identifies the combination of predictors that best
explains the outcome variable, resulting in a single model (James et al., 2013). We included terms
for Lower-Extremity Fugl-Meyer (LE-FM), overground gait speed, immediate memory,
visuospatial/constructional skills, language, attention, delayed memory, and executive function.
We included a LE-FM term as a potential predictor to account for the possibility that motor
impairment plays a role in the participants ability to increase paretic propulsion. We included
overground gait speed as a measure of walking recovery. We performed this process three times,
once with each of our dependent variables (normalized propulsion error, coefficient of variation,

76
and recall error). We chose the models with the lowest Bayesian information criterion score. We
verified that the final models meet linear regression assumptions using the performance package
(Lüdecke et al., 2021).
For all analyses, we excluded participants whose average baseline propulsion magnitude ±
one standard deviation captured the propulsion goal. We did this to ensure that participants had to
increase their propulsion over their baseline propulsion magnitude to meet the goal. For the
immediate retention analysis, we excluded additional participants who did not improve
performance during the biofeedback task (normalized propulsion error > 95% at the end of
biofeedback training).
Results
We included eighteen participants in the performance and performance variability analyses.
Eleven participants were excluded because their baseline propulsion magnitude ± one standard
deviation captured the propulsion goal. Five participants did not improve performance during the
biofeedback task and were excluded from the immediate retention analysis. Clinical demographics
for participants included in these analyses are included in Table 5.1. Our sample had RBANS total
scores ranging from 54 – 106, suggesting we captured a large range of cognitive impairment.
Participants had an average normalized propulsion error of 59.9 (44.0), an average coefficient of
variation of 29.7 (16.2), and an average recall error of 7.6 (25.6).

77
Table 5.1. Clinical demographics of participants included in the analyses.
The maximum Lower-Extremity Fugl-Meyer (LE-FM) score is 34. Scores for visuospatial/constructional (V/C), language,
attention, immediate memory (IM), and delayed memory (DM) are from the Repeatable Battery for the Assessment of
Neuropsychological Status. Executive function (EF) scores are from the Trail Making B Test and are scored in seconds to
completion. Participants excluded from the immediate retention analysis are denoted with a *.
Visuospatial/constructional and LE-FM scores best-explained normalized propulsion error
during biofeedback training (adjusted R2 = 0.59, p = 0.0005; Figure 5.3A), suggesting that
participants with better visuospatial/constructional skills and less motor impairment were better
able to use the information contained in the biofeedback display to improve performance. To
understand the relative contribution of each predictor to normalized propulsion error, we visualized
the relationship between our outcome and one predictor while controlling for the other predictor
using added variable plots (Gallup, 2019). When controlling for the influence of LE-FM, we see a
negative relationship between visuospatial/constructional scores and normalized propulsion error
(Figure 5.3B, top). When controlling for visuospatial/constructional scores, we similarly see a
negative relationship between LE-FM and normalized propulsion error, albeit weaker (Figure
Age Race/
Ethnicity
Affected
side
10m
walk
speed
(m/s)
Treadmill
speed
(m/s)
LEFM IM V/C Language Attention DM EF
63 Asian Right 1.19 0.61 28 94 92 92 82 102 136
*53 Asian Right 1.07 0.55 29 57 69 75 97 68 182
35 White/Hispanic Right 1.39 0.78 24 81 69 112 85 60 122
31 White/Hispanic Left 1.38 0.85 29 65 87 64 75 60 73
61 White/Hispanic Left 0.73 0.39 22 73 75 90 82 92 93
59 Asian Left 1.19 0.74 24 90 89 87 103 108 109
61 White Right 0.51 0.5 24 85 112 92 97 84 66
53 Asian Right 1.17 0.6 25 57 87 79 82 91 189
78 Asian Left 0.86 0.49 24 94 105 96 88 113 95
46 White/Hispanic Left 0.95 0.54 24 94 60 118 56 98 68
55 White/Hispanic Left 1.04 0.54 28 69 378 64 75 75 74
63 White/Hispanic Left 0.92 0.56 18 100 109 98 109 110 71
*56 Black Left 0.81 0.52 11 94 60 87 68 99 300
*64 White Right 0.92 0.48 23 94 87 90 75 112 140
*60 Black Right 0.42 0.23 12 53 58 54 43 64 300
69 White/Hispanic Left 0.23 0.3 22 81 64 85 53 86 198
*44 White/Hispanic Right 1.07 0.58 16 53 66 85 53 60 267
45 Black Left 0.59 0.37 17 103 69 83 85 98 300

78
5.3B, bottom). This suggests that the visuospatial/constructional score has a greater contribution
to normalized propulsion error than the LE-FM score. Individual coefficient estimates for all
models are provided in Table 5.2. It is important to note that the best-subsets selection process
employed here selects the combination of predictors that best-explains the outcome variable,
resulting in a non-significant term being included in our final model (LE-FM).
Figure 5.3. Biofeedback performance results. A) Relationship between visuospatial/constructional skills, LE-FM, and normalized
propulsion error averaged over the final 30 strides of biofeedback training (n = 18). A normalized propulsion value of 100% is
equivalent to baseline propulsion error (horizontal dashed line). B) Added variable plots for relationship between
visuospatial/constructional skills (top) and LE-FM (bottom) and normalized propulsion error. Abbreviations: RBANS, Repeatable
Battery for the Assessment of Neuropsychological Status; LE-FM, Lower-Extremity Fugl-Meyer.

79
Table 5.2. Model results.
β SE 95% CI p-value
Performance model
Intercept 243.5 37.0 [164.7, 322.4] < 0.0001
LE-FM -2.4 1.4 [-5.3, 0.6] 0.1
Visuospatial/Constructional -1.6 0.4 [-2.6, -0.7] 0.002
Performance variability model
Intercept 67.4 14.9 [35.7, 99.1] 0.0004
Attention -0.5 0.2 [-0.9, -0.09] 0.02
Immediate retention model
Intercept 91.3 36.7 [10.5, 172.1] 0.03
Language -0.9 0.4 [-1.8, -0.04] 0.04
Attention scores best explained the coefficient of variation across biofeedback trials
(adjusted R2 = 0.25, p = 0.02; Figure 5.4A). Participants with higher attention scores had a more
Figure 5.4. Performance variability and immediate retention results. A) Relationship between attention and coefficient of variation
(n = 18). B) Relationship between language skills and recall error (n = 13). A recall error of 0% represents exact recall (horizontal
dashed line). Negative recall error means that participants were closer to the goal at immediate retention than at the end of
biofeedback training. Positive recall error means that participants were further from the goal at immediate retention than at the end
of biofeedback training. The linear model fit is plotted in red, with 95% confidence intervals in gray. Abbreviation: RBANS,
Repeatable Battery for the Assessment of Neuropsychological Status.

80
consistent propulsion output throughout performance, suggesting that attention plays a role in the
consistency of the biofeedback-driven walking pattern execution.
Language scores best-explained recall error (adjusted R2 = 0.27, p = 0.04; Figure 5.4B).
Participants with higher language scores were able to maintain or move closer to the propulsion
goal immediately after the biofeedback was removed.
For all analyses, we have included supplemental tables to show which variables were
selected at each stage of the best-subsets selection process (Supplemental Tables 5.1-5.3).
Discussion
We aimed to understand which cognitive domains were associated with performance,
performance variability, and immediate retention of a paretic propulsion biofeedback task in
individuals post-stroke. We found that visuospatial/constructional skills and motor impairment
were associated with performance, attention was associated with performance variability, and
language skills were associated with immediate retention. These results are the first to identify an
association between specific cognitive domains and the biofeedback-driven acquisition,
variability, and immediate retention of a newly learned locomotor behavior.
Our data indicated that participants with better visuospatial/constructional skills exhibited
less error while performing a paretic propulsion biofeedback task compared to those with lower
scores. The RBANS visuospatial/constructional subscale consists of line orientation and figure
copy tasks, and lower scores indicate impairment in processing and using visual information
(Kimbell, 2013). To use the information contained in the biofeedback display, participants must
understand how their propulsion magnitude maps onto the visual display. They must also
internalize the visual error on the screen to adjust their propulsion output. Based on their poorer

81
performance with the biofeedback task, this was more challenging for participants with lower
visuospatial/constructional skills. This suggests that commonly used rehabilitation approaches that
require processing visual information – such as taping step length targets on the floor or practicing
specific movements in front of a mirror – may be less effective for participants with
visuospatial/constructional impairments. These individuals may instead benefit from an alternative
biofeedback method, such as auditory biofeedback, which can also be used to increase paretic
propulsion in individuals post-stroke (J. Liu et al., 2020). However, future research is needed to
understand the cognitive domains that influence motor learning for other modes of biofeedback.
Previous work by French et al. (2021) found that fluid cognition, not visuospatial working
memory, was related to the acquisition and 24-hour retention of a biofeedback-driven walking
pattern in individuals post-stroke. Importantly, the visuospatial measures used in the
aforementioned study measured a different aspect of visuospatial cognition than our study.
Visuospatial working memory reflects someone’s ability to store and manipulate visual
information (McAfoose & Baune, 2009), and visuospatial/constructional skills reflect someone’s
ability to process and use visual information (Kimbell, 2013). Our work expands upon the findings
of French et al. (2021) by identifying the specific cognitive domains associated with both the
acquisition and immediate retention of a biofeedback-driven walking pattern.
In addition to the visuospatial/constructional cognitive domain, our selected performance
model also included the LE-FM score, accounting for motor impairment. This is unsurprising, as
motor impairment likely contributes to an individual’s ability to increase paretic propulsion. It is
possible that the participant is unable to physically increase propulsion and, therefore, cannot reach
the biofeedback goal regardless of cognitive impairment. Including the LE-FM score in the model
partially accounts for this possibility. However, it is likely that it does not entirely capture motor

82
impairment, particularly since the LE-FM is not measured during gait. Adding other gait-specific
measures of motor impairment may better address this concern in future studies.
We found that participants with better attention scores had lower performance variability
during biofeedback training, indicating that attention plays a role in the execution consistency of
the biofeedback-driven walking pattern. This is consistent with work relating attention and gait
variability in individuals with Parkinsons Disease (Morris et al., 2019). Additionally, functional
connectivity between the dorsal attentional and default networks, which reflect the ability to
allocate and sustain attention during a task, are related to gait variability in older adults and
individuals with Parkinsons Disease (Lo et al., 2021). To use the propulsion biofeedback,
participants had to direct cognitive resources to sustain attention on the display. This may explain
why participants with poorer attentional abilities exhibited greater variability during the
acquisition of the biofeedback-driven walking pattern.
Our work is the first to find that language skills play a role in the immediate retention of a
biofeedback-driven walking pattern. The language subscale on the RBANS consists of picture
naming and semantic fluency tasks, and lower scores indicate expressive (producing language;
APA Dictionary of Psychology, n.d.) or receptive (perceiving and understanding language; APA
Dictionary of Psychology, n.d.) impairments (Kimbell, 2013). Based on these results, we posit that
the participants in our study with lower language scores had more difficulty comprehending the
verbal instructions for the immediate retention test, which previous work has shown to impact
retention of a biofeedback-driven walking behavior (French et al., 2018). It is also important to
consider that we performed an immediate retention test (approximately two minutes after
completion of biofeedback training), which did not provide a consolidation period for long-term
learning (Kantak & Winstein, 2012). Therefore, while we found an association between language

83
skills and immediate retention, future work is needed to understand if language skills are also
associated with the long-term retention of a biofeedback-driven walking pattern.
Although we observed an association between language skills and immediate retention of
our biofeedback-driven walking pattern, retention of upper extremity tasks has most often been
related to visuospatial function (Hooyman et al., 2022; Lingo VanGilder et al., 2018; Lingo
VanGilder, Hooyman, et al., 2021; Lingo VanGilder, Lohse, et al., 2021; VanGilder et al., 2022;
Wang et al., 2022). For example, the RBANS visuospatial/constructional score has been associated
with one-week retention of an upper extremity task in neurotypical older adults (Lingo VanGilder
et al., 2018). In addition, previous studies have found that visuospatial working memory (not
captured in our visuospatial/constructional score) is associated with one-month retention of an
upper extremity task in neurotypical adults (Hooyman et al., 2022; Lingo VanGilder, Lohse, et al.,
2021; VanGilder et al., 2022; Wang et al., 2022) and individuals post-stroke (Lingo VanGilder,
Hooyman, et al., 2021). Interestingly, French et al. (2021) found that visuospatial working memory
did not explain any more variance than the fluid cognition score in 24-hour retention of a
biofeedback-driven locomotor behavior. This suggests that visuospatial working memory may not
play the same role in the retention of a visually guided walking pattern as it does in a reaching
task. This difference could be explained by the distinct types of motor skills required for each task,
as recent work demonstrated that participants with better visuospatial working memory have better
retention of fine, but not gross, upper extremity motor skills (Hooyman et al., 2022). Altering gait
impairments requires improvement and retention of gross motor skills.
Interestingly, two participants performed considerably better on the task without
biofeedback than with biofeedback (reflected as a negative recall error in Figure 5.4B). These
participants had high language and low visuospatial scores (see rows three and ten in Table 5.1).

84
It is possible that the participants’ impaired visuospatial skills led to worse performance with the
biofeedback, and when the biofeedback was removed, they were able to use the verbal instructions
to better produce the intended gait pattern. These data highlight the potential importance of
considering the distribution of cognitive impairment to individually tailor rehabilitation
interventions.
Collectively, these results suggest that neuropsychological testing may be valuable for
interpreting motor learning and rehabilitation outcomes (French et al., 2021; VanGilder et al.,
2020). Collecting detailed information about the distribution of cognitive impairment across
domains of cognition – beyond that provided by standard screening tools – may help to better
explain individual variation in response to experimental paradigms. Standard screening tools such
as the Montreal Cognitive Assessment and Mini-Mental State Exam may categorize a participant
as having normal cognitive function despite impairments in specific cognitive domains (McDowd
et al., 2003). While full neuropsychological testing may not be feasible in many rehabilitation
research settings, the RBANS can be administered in thirty minutes or less – providing sub-scale
scores for five different cognitive domains with minimal time burden. Collecting detailed data on
specific cognitive domain impairments will allow for a deeper understanding of how these
impairments affect post-stroke rehabilitation outcomes. With further investigation, we may be able
to identify training parameters, such as the mode, frequency, or practice schedule, that can be
individualized based on specific cognitive impairments. Further defining the relationship between
cognitive impairments and rehabilitation responsiveness would also inform the design of future
studies that explore using cognitive training to enhance motor rehabilitation. Recent evidence
suggests that visuospatial skills can be trained in young neurotypical adults (Schaefer et al., 2022),

85
but more research is needed to understand if the same is observed in people post-stroke or if
visuospatial training influences motor skill retention.
Limitations
This study has a few limitations. First, we included a relatively small number of
participants in our exploratory analyses. Due to the smaller number of participants, there is a
possibility that the associations found in this work may not generalize to a larger sample. However,
this work provides the foundation to inform future, hypothesis-driven research with larger sample
sizes. Second, we did not include all factors that may contribute to individual variability in
locomotor learning outcomes in our model. Future work should investigate potential physical,
psychological, or social factors that may contribute to locomotor learning. Finally, we considered
both over- and undershooting the propulsion goal to be a movement error. It is possible that
overshooting the goal may have resulted in improved gait biomechanics (i.e., more symmetric
propulsion between limbs). This may have caused participants to overshoot the goal despite the
instructions of the task, as they may have been implicitly biased towards walking with more
symmetric propulsion.

86
Conclusion
We found that visuospatial/constructional skills and motor impairment were associated
with acquiring a biofeedback-driven locomotor behavior, attention was associated with
performance variability, and language skills were associated with the immediate retention of that
behavior. These results demonstrate that information on specific cognitive domain impairments
explains variability in locomotor learning outcomes. This suggests that with further investigation,
we may be able to use information on specific cognitive impairments to predict responsiveness to
interventions and personalize training parameters to facilitate locomotor learning after a stroke.
Supplemental Table 5.1. Performance best-subsets variable selection results.
IM V/C Language Attention DM EF Speed LE-FM BIC
* -9.1
* * -9.5
* * * -8.2
* * * * -7.6
* * * * * -8.1
* * * * * * -5.4
* * * * * * * -2.6
* * * * * * * * 0.3
Supplemental Table 5.2. Performance variability best-subsets variable selection results.
IM V/C Language Attention DM EF Speed LE-FM BIC
* -0.5
* * 0.01
* * * 2.0
* * * * 4.3
* * * * * 6.9
* * * * * * 9.7
* * * * * * * 12.5
* * * * * * * * 15.4

87
Supplemental Table 5.3. Immediate retention best-subsets variable selection results.
IM V/C Language Attention DM EF Speed LE-FM BIC
* -0.02
* * 1.2
* * * 2.9
* * * * 4.9
* * * * * 5.8
* * * * * * 7.2
* * * * * * * 9.4
* * * * * * * * 11.7

88
CHAPTER 6
DISCUSSION
Findings from our work
In this work, we aimed to understand how different rehabilitation approaches impacted
overall gait biomechanics and investigated the role of cognitive domain impairments in an explicit
locomotor learning task. In Aim 1, we evaluated the effect of gait speed on overall gait
biomechanics in people post-stroke compared to neurotypical adults using a k-means clustering
analysis. We hypothesized that there would be two clusters (neurotypical and post-stroke), and that
walking faster would cause the clusters to move further apart. We found two distinct clusters
representative of neurotypical gait behavior and stroke gait behavior. People post-stroke in the
stroke gait behavior cluster walked at similar speeds but had greater motor impairment compared
to those in the neurotypical gait behavior cluster. At faster speeds, the distance between clusters
did not change, suggesting that fast walking did not improve nor degrade overall gait
biomechanics. These analyses demonstrate that while fast walking may reduce the magnitude of
some kinematic impairments relative to one’s habitual walking pattern, the resulting gait
kinematics are not more similar to neurotypical adults walking at like speeds. This suggests that
to improve overall gait biomechanics, fast walking may need to be combined with another
intervention that targets overall gait biomechanics.
In Aim 2, we investigated how changes in propulsion asymmetry impacted overall gait
asymmetry (measured by CGAM) in individuals post-stroke. We hypothesized that reducing
propulsion asymmetry would lead to a reduction in CGAM. We found a positive association
between Δ |PA| and Δ CGAM (intercept β = 0.47, p = 0.78; Δ |PA| β = 2.6, p = 0.009). The average
change in propulsion asymmetry magnitude in our sample was -0.09, suggesting that on average,

89
we would expect to see a CGAM change of 0.2. This suggests that propulsion asymmetry may not
be an ideal target for biofeedback-based interventions designed to improve overall gait asymmetry.
In Aim 3, we aimed to understand which cognitive domains predicted locomotor
performance during a visual biofeedback task and immediate recall of the newly learned walking
pattern in individuals post-stroke. We found that visuospatial/constructional skills and motor
impairment were associated with acquiring a biofeedback-driven locomotor behavior and language
skills were associated with the immediate retention of that behavior. These results demonstrate that
information on specific cognitive domain impairments explains variability in locomotor learning
outcomes. With further investigation, we may be able to predict responsiveness to interventions
and personalize training parameters based on an individual’s cognitive status to facilitate
locomotor learning after a stroke.
Impact of dissertation
Improving overall gait biomechanics likely requires a targeted rehabilitation approach
Together, the results from Aim 1 and Aim 2 demonstrate that rehabilitation approaches that
target single gait aspects (i.e., fast walking and paretic propulsion biofeedback) likely will not
result in substantial improvements in overall gait biomechanics. Our work builds upon the work
of Padmanabhan et al. (2020) who found that reducing step length asymmetry did improve sagittal
plane gait asymmetry. Additionally, Shin et al. (2020) found that increased gait speed after 12
weeks of physical therapy was not accompanied by improvements in many semi-comprehensive
gait biomechanical measures. Combined with our work, this indicates that improving overall gait
biomechanics likely requires targeted interventions, rather than assuming improvements in other
aspects of gait (e.g., gait speed, step length asymmetry, propulsion asymmetry) will be
accompanied by an improvement in overall gait biomechanics. In line with this idea, the third

90
Stroke Recovery and Rehabilitation Roundtable recommends including multivariate kinematic
measures in research studies to provide a more comprehensive understanding of the impact of an
intervention on overall gait biomechanics (Van Criekinge et al., 2023).
The influence of cognitive impairment on post-stroke motor learning
The Aim 3 results highlight that considering individual characteristics (such as cognitive
impairment) may aid in personalizing rehabilitation interventions. Outcomes from post-stroke
rehabilitation studies are well-known to be variable between individuals (Bowden et al., 2013;
Dobkin et al., 2014). However, clinical practice guidelines are based on group effects from
rehabilitation research studies and do not account for individual variability in the response to an
intervention. For example, the LEAPS trial (Duncan et al., 2011), found that body weightsupported treadmill training and home-based exercises had similar effects on walking outcomes,
suggesting that body weight-supported treadmill training does not provide any additional benefit
over traditional physical therapy. However, in a follow-up analysis the authors found personal
factors such as baseline walking speed, age, and balance influenced individual response to body
weight-supported treadmill training. Our work further emphasizes that cognition plays a role in
explicit locomotor learning and that there is value in considering individual factors when designing
motor learning-based interventions for individuals post-stroke to optimize learning. For example,
a visual-based explicit learning intervention may be disadvantageous to individuals with
visuospatial impairments. Therefore, it may be more appropriate to provide an auditory-based
intervention or leverage another learning mechanism, such as reinforcement or implicit motor
learning. By further understanding the factors that contribute to individual responses to a given
intervention, an individually tailored rehabilitation intervention can be designed for the patient to
maximize their recovery potential.

91
Limitations and future work
Single sessions of fast walking and paretic propulsion biofeedback training
We only completed a single session of fast walking (Aim 1) and paretic propulsion
biofeedback training (Aim 2 and Aim 3). As a result, we do not have the ability to evaluate the
potential effects of longer-term fast walking or paretic propulsion biofeedback training on overall
gait biomechanics using our data. It is possible that with repeated practice of either intervention,
individuals may learn a different movement strategy that leads to improved overall gait
biomechanics. However, previous work has shown that after two months of high-intensity gait
training (achieved by training at faster speeds on a treadmill), participants only improved singlelimb support time asymmetry, step length asymmetry, and intralimb hip-knee consistency, without
improvements in other spatiotemporal or kinematic measures (Ardestani et al., 2020). This
suggests that repeated sessions of fast walking are unlikely to lead to improvements in overall gait
biomechanics. There is evidence that multiple training sessions targeting paretic propulsion can
increase paretic propulsion, reduce propulsion asymmetry, and increase paretic trailing limb angle
(Alingh et al., 2021; Awad et al., 2014); however, there is little work examining other
biomechanical impairments after repeated propulsion training. It is possible that longer-term
training would increase an individual’s capacity to increase propulsion which may lead to larger
changes in overall gait biomechanics. Future work should investigate the effect of repeated fast
walking and paretic propulsion biofeedback training on overall gait biomechanics. Additionally,
because we only completed a single session of paretic propulsion biofeedback training, it is still
unclear what cognitive domains are associated with long-term retention or transfer of a
biofeedback-driven gait behavior. Understanding how cognitive status relates to long-term

92
learning will help tailor rehabilitation interventions to maximize learning and transfer outside of
the laboratory setting.
Considering other clinical factors that may predict motor learning
In Aim 3, we only considered cognitive impairment, motor impairment, and walking
function as potential predictors of acquisition and immediate retention of a biofeedback-driven
gait pattern. It would also be fruitful to investigate other individual characteristics that may predict
the acquisition and retention of a biofeedback task. Our work demonstrated that cognitive domain
impairment (i.e., visuospatial/constructional skills, attention, and language skills) explained a
proportion of the variance in these outcomes; however, there is still much of the variance
unaccounted for by our model. This suggests that there may be other clinical characteristics that
could be used to further tailor rehabilitation interventions to maximize learning in individuals poststroke. Other potential predictors to investigate may include motivation and self-efficacy
(Gangwani et al., 2022). While no work has investigated the effect of self-efficacy or motivation
on the performance or retention of a gait biofeedback task, there is evidence that self-efficacy
predicts performance on a goal-directed reaching task in individuals post-stroke (Stewart et al.,
2019) and that motivational factors, specifically intrinsic motivation, may contribute to motor
learning (Bacelar et al., 2022). Enhancing learners' autonomy (e.g., allowing them to select their
feedback schedule) was previously thought to increase motivation and enhance motor learning
(Wulf & Lewthwaite, 2016); however, recent evidence suggests enhancing autonomy likely does
not influence motivation or motor learning, as the results of previous work in this area are
underpowered (McKay et al., 2022) and not replicable (Bacelar et al., 2022). Additionally, it is
likely that the Lower-Extremity Fugl-Meyer score is not completely capturing motor impairment,
particularly since the Fugl-Meyer is not measured during walking. Capturing motor impairment

93
during walking may better account for the variance in our motor learning outcomes; however, there
are currently no motor impairment measures that are taken during walking.
The impact of cognition on acquisition and retention of other modes of biofeedback.
We only investigated the impact of cognition on acquisition and retention of a visually
guided gait pattern and found that individuals with visuospatial/constructional impairments were
less able to use the visual biofeedback to improve performance. It is possible that these individuals
may be better able to use other modes of biofeedback, such as auditory or haptic, that are less
reliant on visuospatial skills. Future work should also investigate the cognitive domains involved
in using other modes of biofeedback.
Targeting overall gait biomechanics
Our findings suggest that single sessions of fast walking and paretic propulsion biofeedback
did not improve overall gait biomechanics, making it unclear what intervention is most appropriate
to improve overall gait biomechanics with longer-term training paradigms. There has been
increased emphasis on developing rehabilitation methods that also focus on gait biomechanics (S.
A. Moore et al., 2022). Therefore, future work should investigate what interventions are
appropriate to improve post-stroke overall gait biomechanics. Potential lines of investigation
include biofeedback that targets multiple dimensions of gait (similar to Day et al., 2019), virtual
reality-based games, exosuits, or a combination of multiple interventions. For example, if an
approach to specifically target overall gait biomechanics could be combined with fast walking, this
would allow clinicians to simultaneously target two domains of gait dysfunction simultaneously –
activity limitations and biomechanical impairments.

94
The impact of changing overall gait biomechanics on other relevant gait measures
We investigated how changing propulsion asymmetry impacted overall gait biomechanics,
but the relationship between overall gait biomechanics and other relevant gait measures remains
an open question. Future work should also investigate whether improvements in overall gait
biomechanics are related to improvements in outcomes related to walking activity, such as gait
speed, gait distance, and metabolic cost. Recent evidence suggests that changes in gait speed may
be independent of changes in overall gait biomechanics that we commonly assume to occur
alongside these activity-level changes (Shin et al., 2020). However, it is still unclear what the
impact of changing overall gait biomechanics is on outcomes related to walking activity, such as
gait speed. Additionally, predictive simulation work suggests that spatiotemporal asymmetry may
be metabolically optimal for individuals post-stroke (Johnson et al., 2022), however, it is unclear
how metabolic cost relates to overall gait biomechanics.
Selecting biofeedback parameters
We found in our work that reaching the biofeedback goal was difficult for many participants
(51% of strides were below the paretic propulsion goal; Aim 2). This is likely due, in part, to the
lack of consensus on how to develop a challenging and effective biofeedback paradigm for
individuals post-stroke (Spencer et al., 2021). There are various methods that previous studies have
used to set the biofeedback goal. In the post-stroke literature, biofeedback goals are often selected
in reference to the participants’ non-paretic limb. Calculating a goal in reference to someone’s
nonparetic limb can provide a personalized, attainable goal for participants to strive for during
training. However, how the non-paretic limb is used to set the goal is inconsistent across studies.
For example, with step length biofeedback, previous work has let the participant self-select
symmetric step lengths (Padmanabhan et al., 2020), lengthened the shorter step to the length of the

95
longer step (Park et al., 2021), or set the goal as the average of the two step lengths (Nguyen et al.,
2020). Each of these options would result in a different step length goal and may have different
impacts on gait biomechanics or clinical measures (i.e., metabolic cost, gait speed, etc.). An
alternative way to set the goal is to calculate a step length goal that is matched to speed- and
demographic-matched neurotypical adults (Bonilla Yanez et al., 2023), though no work has tested
the feasibility and the biomechanical outcomes of this biofeedback goal setting method . Future
work should investigate the impact of different goals to inform more thoughtful goal setting for
biofeedback paradigms.
It is important to consider whether to provide information from one or both limbs. Typically,
step length biofeedback is provided on both limbs (Nguyen et al., 2020; Padmanabhan et al., 2020;
Park et al., 2021), and propulsion biofeedback is provided on a single limb (Genthe et al., 2018).
This likely occurs because step length biofeedback is used to reduce step length asymmetry, which
is an inter-limb measure, whereas, propulsion biofeedback is used to increase propulsion of the
paretic limb, not necessarily propulsion asymmetry. Providing feedback only on the paretic limb
leaves the nonparetic limb unconstrained. This could lead to undesirable compensations on the
nonparetic limb to meet the goal, suggesting it may be beneficial to target propulsion on both
limbs. Additionally, anecdotally, participants who participated in our study reported that they
would prefer to receive propulsion biofeedback on both limbs. Future work should consider
implementing bilateral propulsion biofeedback.
Moving forward: targeted and personalized rehabilitation
The post-stroke rehabilitation field is moving towards personalized precision
rehabilitation, which aims to provide rehabilitation to the right person at the right time (French et
al., 2022; S. A. Moore et al., 2022). To begin personalizing post-stroke rehabilitation methods,

96
future research studies need to intentionally measure an individual’s physical, cognitive, and
psychosocial abilities (French et al., 2022). In the past, cognition was typically only measured to
use as an exclusion criterion (VanGilder et al., 2020). Now it is recommended that all post-stroke
clinical studies include measurements of cognition to begin to better understand post-stroke
cognitive impairment (Rost et al., 2022) and how it impacts rehabilitation (VanGilder et al., 2020).
This, in turn, will allow investigation into how to potentiate rehabilitation methods based on
individual characteristics. This is a critical step in moving the field forward and assisting in
maximizing patients’ recovery in an effective and efficient manner.
Individuals post-stroke commonly report that improving their gait appearance is a desired
rehabilitation goal (Bohannon et al., 1991), yet, this goal is less emphasized in rehabilitation
research. This could be due in part to an implicit assumption that improving other aspects of gait
(e.g., gait speed, step length asymmetry, propulsion asymmetry) will be accompanied by an
improvement in overall gait biomechanics. However, our work indicates that improving overall
gait biomechanics likely requires targeted interventions rather than assuming improvements in
other aspects of gait (e.g., gait speed, step length asymmetry, propulsion asymmetry) will be
accompanied by an improvement in overall gait biomechanics. Developing interventions that
simultaneously target walking function and overall gait biomechanics will allow clinicians to better
support patients in the achievement of multiple walking-based rehabilitation goals.

97
REFERENCES
Abbasi, L., Rojhani-Shirazi, Z., Razeghi, M., & Raeisi-Shahraki, H. (2021). Kinematic cluster
analysis of the crouch gait pattern in children with spastic diplegic cerebral palsy using
sparse K-means method. Clinical Biomechanics, 81.
https://doi.org/10.1016/j.clinbiomech.2020.105248
Akbas, T., Prajapati, S., Ziemnicki, D., Tamma, P., Gross, S., & Sulzer, J. (2019). Hip
circumduction is not a compensation for reduced knee flexion angle during gait. Journal
of Biomechanics, 87, 150–156. https://doi.org/10.1016/j.jbiomech.2019.02.026
Alam, Z., Rendos, N. K., Vargas, A. M., Makanjuola, J., & Kesar, T. M. (2022). Timing of
propulsion-related biomechanical variables is impaired in individuals with post-stroke
hemiparesis. Gait & Posture, 96, 275–278. https://doi.org/10.1016/j.gaitpost.2022.05.022
Alingh, J. F., Groen, B. E., Kamphuis, J. F., Geurts, A. C. H., & Weerdesteyn, V. (2021). Taskspecific training for improving propulsion symmetry and gait speed in people in the
chronic phase after stroke: A proof-of-concept study. Journal of NeuroEngineering and
Rehabilitation, 18, 69. https://doi.org/10.1186/s12984-021-00858-8
Alingh, J. F., Groen, B. E., Van Asseldonk, E. H. F., Geurts, A. C. H., & Weerdesteyn, V.
(2020). Effectiveness of rehabilitation interventions to improve paretic propulsion in
individuals with stroke – A systematic review. Clinical Biomechanics, 71, 176–188.
https://doi.org/10.1016/j.clinbiomech.2019.10.021
APA Dictionary of Psychology. (n.d.). Retrieved July 12, 2023, from https://dictionary.apa.org/
Ardestani, M. M., Henderson, C. E., Mahtani, G., Connolly, M., & Hornby, T. G. (2020).
Locomotor Kinematics and Kinetics Following High-Intensity Stepping Training in
Variable Contexts Poststroke. Neurorehabilitation and Neural Repair, 34(7), 652–660.
https://doi.org/10.1177/1545968320929675
Awad, L. N., Binder-Macleod, S. A., Pohlig, R. T., & Reisman, D. S. (2015). Paretic Propulsion
and Trailing Limb Angle Are Key Determinants of Long-Distance Walking Function
After Stroke. Neurorehabilitation and Neural Repair, 29(6), 499–508.
https://doi.org/10.1177/1545968314554625
Awad, L. N., Jaehyun Bae, Kudzia, P., Long, A., Hendron, K., Holt, K. G., OʼDonnell, K., Ellis,
T. D., & Walsh, C. J. (2017). Reducing Circumduction and Hip Hiking During
Hemiparetic Walking Through Targeted Assistance of the Paretic Limb Using a Soft
Robotic Exosuit: American Journal of Physical Medicine & Rehabilitation, 96, S157–
S164. https://doi.org/10.1097/PHM.0000000000000800

98
Awad, L. N., Lewek, M. D., Kesar, T. M., Franz, J. R., & Bowden, M. G. (2020). These legs
were made for propulsion: Advancing the diagnosis and treatment of post-stroke
propulsion deficits. Journal of NeuroEngineering and Rehabilitation, 17(1), 139.
https://doi.org/10.1186/s12984-020-00747-6
Awad, L. N., Palmer, J. A., Pohlig, R. T., Binder-Macleod, S. A., & Reisman, D. S. (2015).
Walking Speed and Step Length Asymmetry Modify the Energy Cost of Walking After
Stroke. Neurorehabilitation and Neural Repair, 29(5), 416–423.
https://doi.org/10.1177/1545968314552528
Awad, L. N., Reisman, D. S., Kesar, T. M., & Binder-Macleod, S. A. (2014). Targeting Paretic
Propulsion to Improve Poststroke Walking Function: A Preliminary Study. Archives of
Physical Medicine and Rehabilitation, 95(5), 840–848.
https://doi.org/10.1016/j.apmr.2013.12.012
Bacelar, M. F. B., Parma, J. O., Cabral, D., Daou, M., Lohse, K. R., & Miller, M. W. (2022).
Dissociating the contributions of motivational and information processing factors to the
self-controlled feedback learning benefit. Psychology of Sport and Exercise, 59, 102119.
https://doi.org/10.1016/j.psychsport.2021.102119
Baker, R., McGinley, J. L., Schwartz, M. H., Beynon, S., Rozumalski, A., Graham, H. K., &
Tirosh, O. (2009). The Gait Profile Score and Movement Analysis Profile. Gait &
Posture, 30(3), 265–269. https://doi.org/10.1016/j.gaitpost.2009.05.020
Balasubramanian, C. K., Bowden, M. G., Neptune, R. R., & Kautz, S. A. (2007). Relationship
Between Step Length Asymmetry and Walking Performance in Subjects With Chronic
Hemiparesis. Archives of Physical Medicine and Rehabilitation, 88(1), 43–49.
https://doi.org/10.1016/j.apmr.2006.10.004
Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting Linear Mixed-Effects Models
Using lme4. Journal of Statistical Software, 67(1). https://doi.org/10.18637/jss.v067.i01
Beaman, C. B., Peterson, C. L., Neptune, R. R., & Kautz, S. A. (2010). Differences in selfselected and fastest-comfortable walking in post-stroke hemiparetic persons. Gait &
Posture, 31(3), 311–316. https://doi.org/10.1016/j.gaitpost.2009.11.011
Berg, K. O., Maki, B. E., Williams, J. I., Holliday, P. J., & Wood-Dauphinee, S. L. (1992).
Clinical and laboratory measures of postural balance in an elderly population. Archives of
Physical Medicine and Rehabilitation, 73(11), 1073–1080.

99
Bohannon, R. W., Morton, M. G., & Wikholm, J. B. (1991). Importance of four variables of
walking to patients with stroke. Journal of Rehabilitation Research, 14(3), 246–250.
https://doi.org/10.1097/00004356-199109000-00010
Bonelli, R. M., & Cummings, J. L. (2007). Frontal-subcortical circuitry and behavior. Dialogues
in Clinical Neuroscience, 9(2), 141–151.
Bonilla Yanez, M., Kettlety, S. A., Finley, J. M., Schweighofer, N., & Leech, K. A. (2023). Gait
speed and individual characteristics are related to specific gait metrics in neurotypical
adults. Scientific Reports, 13(1), Article 1. https://doi.org/10.1038/s41598-023-35317-y
Bowden, M. G., Balasubramanian, C. K., Neptune, R. R., & Kautz, S. A. (2006). AnteriorPosterior Ground Reaction Forces as a Measure of Paretic Leg Contribution in
Hemiparetic Walking. Stroke, 37(3), 872–876.
https://doi.org/10.1161/01.STR.0000204063.75779.8d
Bowden, M. G., Behrman, A. L., Neptune, R. R., Gregory, C. M., & Kautz, S. A. (2013).
Locomotor Rehabilitation of Individuals With Chronic Stroke: Difference Between
Responders and Nonresponders. Archives of Physical Medicine and Rehabilitation, 94(5),
856–862. https://doi.org/10.1016/j.apmr.2012.11.032
Bowie, C. R., & Harvey, P. D. (2006). Administration and interpretation of the Trail Making
Test. Nature Protocols, 1(5), 2277–2281. https://doi.org/10.1038/nprot.2006.390
Boyne, P., Billinger, S. A., Reisman, D. S., Awosika, O. O., Buckley, S., Burson, J., Carl, D.,
DeLange, M., Doren, S., Earnest, M., Gerson, M., Henry, M., Horning, A., Khoury, J. C.,
Kissela, B. M., Laughlin, A., McCartney, K., McQuaid, T., Miller, A., … Dunning, K.
(2023). Optimal Intensity and Duration of Walking Rehabilitation in Patients With
Chronic Stroke: A Randomized Clinical Trial. JAMA Neurology.
https://doi.org/10.1001/jamaneurol.2023.0033
Boyne, P., Meyrose, C., Westover, J., Whitesel, D., Hatter, K., Reisman, D. S., Cunningham, D.,
Carl, D., Jansen, C., Khoury, J. C., Gerson, M., Kissela, B., & Dunning, K. (2019).
Exercise intensity affects acute neurotrophic and neurophysiological responses
poststroke. Journal of Applied Physiology (Bethesda, Md.: 1985), 126(2), 431–443.
https://doi.org/10.1152/japplphysiol.00594.2018
Brough, L. G., Kautz, S. A., & Neptune, R. R. (2022). Muscle contributions to pre-swing
biomechanical tasks influence swing leg mechanics in individuals post-stroke during
walking. Journal of NeuroEngineering and Rehabilitation, 19(1), 55.
https://doi.org/10.1186/s12984-022-01029-z

100
Burpee, J. L., & Lewek, M. D. (2015). Biomechanical gait characteristics of naturally occurring
unsuccessful foot clearance during swing in individuals with chronic stroke. Clinical
Biomechanics, 30(10), 1102–1107. https://doi.org/10.1016/j.clinbiomech.2015.08.018
Campanini, I., Merlo, A., & Damiano, B. (2013). A method to differentiate the causes of stiffknee gait in stroke patients. Gait & Posture, 38(2), 165–169.
https://doi.org/10.1016/j.gaitpost.2013.05.003
Caty, G. D., Detrembleur, C., Bleyenheuft, C., Deltombe, T., & Lejeune, T. M. (2008). Effect of
simultaneous botulinum toxin injections into several muscles on impairment, activity,
participation, and quality of life among stroke patients presenting with a stiff knee gait.
Stroke, 39(10), 2803–2808. Scopus. https://doi.org/10.1161/STROKEAHA.108.516153
Centers for Disease Control and Prevention (CDC). (2009). Prevalence and most common causes
of disability among adults—United States, 2005. MMWR. Morbidity and Mortality
Weekly Report, 58(16), 421–426.
Chen, G., Patten, C., Kothari, D. H., & Zajac, F. E. (2005). Gait differences between individuals
with post-stroke hemiparesis and non-disabled controls at matched speeds. Gait &
Posture, 22(1), 51–56. https://doi.org/10.1016/j.gaitpost.2004.06.009
Clark, D. J., Ting, L. H., Zajac, F. E., Neptune, R. R., & Kautz, S. A. (2010). Merging of Healthy
Motor Modules Predicts Reduced Locomotor Performance and Muscle Coordination
Complexity Post-Stroke. Journal of Neurophysiology, 103(2), 844–857.
https://doi.org/10.1152/jn.00825.2009
Cumming, T. B., Brodtmann, A., Darby, D., & Bernhardt, J. (2014). The importance of cognition
to quality of life after stroke. Journal of Psychosomatic Research, 77(5), 374–379.
https://doi.org/10.1016/j.jpsychores.2014.08.009
Cutler, F. original by L. B. and A., & Wiener, R. port by A. L. and M. (2022). randomForest:
Breiman and Cutler’s Random Forests for Classification and Regression (4.7-1)
[Computer software]. https://CRAN.R-project.org/package=randomForest
Day, K. A., Cherry-Allen, K. M., & Bastian, A. J. (2019). Individualized feedback to change
multiple gait deficits in chronic stroke. Journal of NeuroEngineering and Rehabilitation,
16(1), 158. https://doi.org/10.1186/s12984-019-0635-4
Dean, J. C., Bowden, M. G., Kelly, A. L., & Kautz, S. A. (2020). Altered post-stroke propulsion
is related to paretic swing phase kinematics. Clinical Biomechanics, 72, 24–30.
https://doi.org/10.1016/j.clinbiomech.2019.11.024

101
DeJong, S. L., Schaefer, S. Y., & Lang, C. E. (2012). Need for Speed: Better Movement Quality
During Faster Task Performance After Stroke. Neurorehabilitation and Neural Repair,
26(4), 362–373. https://doi.org/10.1177/1545968311425926
Delavaran, H., Jönsson, A.-C., Lövkvist, H., Iwarsson, S., Elmståhl, S., Norrving, B., &
Lindgren, A. (2017). Cognitive function in stroke survivors: A 10-year follow-up study.
Acta Neurologica Scandinavica, 136(3), 187–194. https://doi.org/10.1111/ane.12709
Dhamoon, M. S., Moon, Y. P., Paik, M. C., Boden-Albala, B., Rundek, T., Sacco, R. L., &
Elkind, M. S. V. (2009). Long-Term Functional Recovery After First Ischemic Stroke.
Stroke, 40(8), 2805–2811. https://doi.org/10.1161/STROKEAHA.109.549576
Dhamoon, M. S., Moon, Y. P., Paik, M. C., Boden-Albala, B., Rundek, T., Sacco, R. L., &
Elkind, M. S. V. (2010). Quality of life declines after first ischemic stroke: The Northern
Manhattan Study. Neurology, 75(4), 328–334.
https://doi.org/10.1212/WNL.0b013e3181ea9f03
Dhamoon, M. S., Moon, Y. P., Paik, M. C., Sacco, R. L., & Elkind, M. S. V. (2012). Trajectory
of Functional Decline Before and After Ischemic Stroke. Stroke, 43(8), 2180–2184.
https://doi.org/10.1161/STROKEAHA.112.658922
Dobkin, B. H., Nadeau, S. E., Behrman, A. L., Wu, S. S., Rose, D., PhD, Bowden, M.,
Studenski, S., Lu, X., & Duncan, P. W. (2014). Prediction of responders for outcome
measures of Locomotor Experience Applied Post Stroke trial. Journal of Rehabilitation
Research and Development, 51(1), 39–50. https://doi.org/10.1682/JRRD.2013.04.0080
Douiri, A., Rudd, A. G., & Wolfe, C. D. A. (2013). Prevalence of Poststroke Cognitive
Impairment. Stroke, 44(1), 138–145. https://doi.org/10.1161/STROKEAHA.112.670844
Drużbicki, M., Guzik, A., Przysada, G., Kwolek, A., Brzozowska-Magoń, A., & Sobolewski, M.
(2016). Changes in Gait Symmetry After Training on a Treadmill with Biofeedback in
Chronic Stroke Patients: A 6-Month Follow-Up From a Randomized Controlled Trial.
Medical Science Monitor, 22, 4859–4868. https://doi.org/10.12659/MSM.898420
Duncan, P. W., Sullivan, K. J., Behrman, A. L., Azen, S. P., Wu, S. S., Nadeau, S. E., Dobkin, B.
H., Rose, D. K., Tilson, J. K., Cen, S., & Hayden, S. K. (2011). Body-Weight–Supported
Treadmill Rehabilitation after Stroke. New England Journal of Medicine, 364(21), 2026–
2036. https://doi.org/10.1056/NEJMoa1010790

102
Einstad, M. S., Saltvedt, I., Lydersen, S., Ursin, M. H., Munthe-Kaas, R., Ihle-Hansen, H.,
Knapskog, A.-B., Askim, T., Beyer, M. K., Næss, H., Seljeseth, Y. M., Ellekjær, H., &
Thingstad, P. (2021). Associations between post-stroke motor and cognitive function: A
cross-sectional study. BMC Geriatrics, 21(1), 103. https://doi.org/10.1186/s12877-021-
02055-7
El Husseini, N., Katzan, I. L., Rost, N. S., Blake, M. L., Byun, E., Pendlebury, S. T., Aparicio,
H. J., Marquine, M. J., Gottesman, R. F., Smith, E. E., & null, null. (2023). Cognitive
Impairment After Ischemic and Hemorrhagic Stroke: A Scientific Statement From the
American Heart Association/American Stroke Association. Stroke, 54(6), e272–e291.
https://doi.org/10.1161/STR.0000000000000430
Fahey, M., Brazg, G., Henderson, C. E., Plawecki, A., Lucas, E., Reisman, D. S., Schmit, B. D.,
& Hornby, T. G. (2020). The Value of High Intensity Locomotor Training Applied to
Patients With Acute-Onset Neurologic Injury. Archives of Physical Medicine and
Rehabilitation. https://doi.org/10.1016/j.apmr.2020.09.399
Finley, J. M., & Bastian, A. J. (2017). Associations Between Foot Placement Asymmetries and
Metabolic Cost of Transport in Hemiparetic Gait. Neurorehabilitation and Neural
Repair, 31(2), 168–177. https://doi.org/10.1177/1545968316675428
Finley, J. M., Perreault, E. J., & Dhaher, Y. Y. (2008). Stretch reflex coupling between the hip
and knee: Implications for impaired gait following stroke. Experimental Brain Research.
Experimentelle Hirnforschung. Experimentation Cerebrale, 188(4), 529–540.
https://doi.org/10.1007/s00221-008-1383-z
Flansbjer, U.-B., Holmbäck, A. M., Downham, D., Patten, C., & Lexell, J. (2005). Reliability of
gait performance tests in men and women with hemiparesis after stroke. Journal of
Rehabilitation Medicine, 37(2), 75–82. https://doi.org/10.1080/16501970410017215
Folstein, M. F., Folstein, S. E., & McHugh, P. R. (1975). “Mini-mental state”. A practical
method for grading the cognitive state of patients for the clinician. Journal of Psychiatric
Research, 12(3), 189–198. https://doi.org/10.1016/0022-3956(75)90026-6
French, M. A., Cohen, M. L., Pohlig, R. T., & Reisman, D. S. (2021). Fluid Cognitive Abilities
Are Important for Learning and Retention of a New, Explicitly Learned Walking Pattern
in Individuals After Stroke. Neurorehabilitation and Neural Repair, 35(5), 419–430.
https://doi.org/10.1177/15459683211001025
French, M. A., Morton, S. M., Charalambous, C. C., & Reisman, D. S. (2018). A locomotor
learning paradigm using distorted visual feedback elicits strategic learning. Journal of
Neurophysiology, 120(4), 1923–1931. https://doi.org/10.1152/jn.00252.2018

103
French, M. A., Morton, S. M., & Reisman, D. S. (2020). Use of explicit processes during a
visually guided locomotor learning task predicts 24-h retention after stroke. Journal of
Neurophysiology, 125(1), 211–222. https://doi.org/10.1152/jn.00340.2020
French, M. A., Roemmich, R. T., Daley, K., Beier, M., Penttinen, S., Raghavan, P., Searson, P.,
Wegener, S., & Celnik, P. (2022). Precision Rehabilitation: Optimizing Function, Adding
Value to Health Care. Archives of Physical Medicine and Rehabilitation, 103(6), 1233–
1239. https://doi.org/10.1016/j.apmr.2022.01.154
Fugl-Meyer, A. R., Jääskö, L., Leyman, I., Olsson, S., & Steglind, S. (1975). The post-stroke
hemiplegic patient. 1. A method for evaluation of physical performance. Scandinavian
Journal of Rehabilitation Medicine, 7(1), 13–31.
Fukuchi, C. A., & Duarte, M. (2019). Gait Profile Score in able-bodied and post-stroke
individuals adjusted for the effect of gait speed. Gait & Posture, 69, 40–45.
https://doi.org/10.1016/j.gaitpost.2019.01.018
Fukuchi, C. A., Fukuchi, R. K., & Duarte, M. (2018). A public dataset of overground and
treadmill walking kinematics and kinetics in healthy individuals. PeerJ, 6, e4640.
https://doi.org/10.7717/peerj.4640
Fulk, G. D., & Echternach, J. L. (2008). Test-retest reliability and minimal detectable change of
gait speed in individuals undergoing rehabilitation after stroke. Journal of Neurologic
Physical Therapy: JNPT, 32(1), 8–13. https://doi.org/10.1097/NPT0b013e31816593c0
Fulk, G. D., Echternach, J. L., Nof, L., & O’Sullivan, S. (2008). Clinometric properties of the
six-minute walk test in individuals undergoing rehabilitation poststroke. Physiotherapy
Theory and Practice, 24(3), 195–204. https://doi.org/10.1080/09593980701588284
Gallup, J. L. (2019). Added-variable plots with confidence intervals. The Stata Journal, 19(3),
598–614. https://doi.org/10.1177/1536867X19874236
Gamer, M., Lemon, J., Fellows, I., & Singh, P. (2019). irr: Various Coefficients of Interrater
Reliability and Agreement (0.84.1) [R]. https://cran.r-project.org/web/packages/irr/irr.pdf
Gangwani, R., Cain, A., Collins, A., & Cassidy, J. M. (2022). Leveraging Factors of SelfEfficacy and Motivation to Optimize Stroke Recovery. Frontiers in Neurology, 13,
823202. https://doi.org/10.3389/fneur.2022.823202

104
Genthe, K., Schenck, C., Eicholtz, S., Zajac-Cox, L., Wolf, S., & Kesar, T. M. (2018). Effects of
real-time gait biofeedback on paretic propulsion and gait biomechanics in individuals
post-stroke. Topics in Stroke Rehabilitation, 25(3), 186–193.
https://doi.org/10.1080/10749357.2018.1436384
Giggins, O. M., Persson, U. M., & Caulfield, B. (2013). Biofeedback in rehabilitation. Journal of
NeuroEngineering and Rehabilitation, 10, 60. https://doi.org/10.1186/1743-0003-10-60
Grau-Pellicer, M., Chamarro-Lusar, A., Medina-Casanovas, J., & Serdà Ferrer, B.-C. (2019).
Walking speed as a predictor of community mobility and quality of life after stroke.
Topics in Stroke Rehabilitation, 26(5), 349–358.
https://doi.org/10.1080/10749357.2019.1605751
Green, S., Sinclair, E., Rodgers, E., Birks, E., Lincoln, N., Hofgren, C., & Ihle-Hansen, H.
(2013). The Repeatable Battery for the Assessment of Neuropsychological Status
(RBANS) for post-stroke cognitive impairment screening...including commentary by
Hofgren C and Ihle-Hansen H. International Journal of Therapy & Rehabilitation,
20(11), 536–542. https://doi.org/10.12968/ijtr.2013.20.11.536
Han, H., Guo, X., & Yu, H. (2016). Variable selection using Mean Decrease Accuracy and Mean
Decrease Gini based on Random Forest. 2016 7th IEEE International Conference on
Software Engineering and Service Science (ICSESS), 219–224.
https://doi.org/10.1109/ICSESS.2016.7883053
Heaton, R. K., Akshoomoff, N., Tulsky, D., Mungas, D., Weintraub, S., Dikmen, S., Beaumont,
J., Casaletto, K. B., Conway, K., Slotkin, J., & Gershon, R. (2014). Reliability and
Validity of Composite Scores from the NIH Toolbox Cognition Battery in Adults.
Journal of the International Neuropsychological Society : JINS, 20(6), 588–598.
https://doi.org/10.1017/S1355617714000241
Hooyman, A., Lingo VanGilder, J., & Schaefer, S. Y. (2022). Mediation Analysis of the Effect
of Visuospatial Memory on Motor Skill Learning in Older Adults. Journal of Motor
Behavior, 1–10. https://doi.org/10.1080/00222895.2022.2105793
Hornby, T. G., Reisman, D. S., Ward, I. G., Scheets, P. L., Miller, A., Haddad, D., Fox, E. J.,
Fritz, N. E., Hawkins, K., Henderson, C. E., Hendron, K. L., Holleran, C. L., Lynskey, J.
E., & Walter, A. (2020). Clinical Practice Guideline to Improve Locomotor Function
Following Chronic Stroke, Incomplete Spinal Cord Injury, and Brain Injury: Journal of
Neurologic Physical Therapy, 44(1), 49–100.
https://doi.org/10.1097/NPT.0000000000000303

105
Hsiao, H., Awad, L. N., Palmer, J. A., Higginson, J. S., & Binder-Macleod, S. A. (2016).
Contribution of Paretic and Nonparetic Limb Peak Propulsive Forces to Changes in
Walking Speed in Individuals Poststroke. Neurorehabilitation and Neural Repair, 30(8),
743–752. https://doi.org/10.1177/1545968315624780
Hsiao, H., Knarr, B. A., Pohlig, R. T., Higginson, J. S., & Binder-Macleod, S. A. (2016).
Mechanisms used to increase peak propulsive force following 12-weeks of gait training
in individuals poststroke. Journal of Biomechanics, 49(3), 388–395.
https://doi.org/10.1016/j.jbiomech.2015.12.040
Hsiao, H., Zabielski, T. M., Palmer, J. A., Higginson, J. S., & Binder-Macleod, S. A. (2016).
Evaluation of measurements of propulsion used to reflect changes in walking speed in
individuals poststroke. Journal of Biomechanics, 49(16), 4107–4112.
https://doi.org/10.1016/j.jbiomech.2016.10.003
James, G., Witten, D., Hastie, T., & Tibshirani, R. (Eds.). (2013). An introduction to statistical
learning: With applications in R. Springer.
Jaracz, K., & Kozubski, W. (2003). Quality of life in stroke patients. Acta Neurologica
Scandinavica, 107(5), 324–329. https://doi.org/10.1034/j.1600-0404.2003.02078.x
Jarvis, H. L., Brown, S. J., Price, M., Butterworth, C., Groenevelt, R., Jackson, K., Walker, L.,
Rees, N., Clayton, A., & Reeves, N. D. (2019). Return to Employment After Stroke in
Young Adults: How Important Is the Speed and Energy Cost of Walking? Stroke, 50(11),
3198–3204. https://doi.org/10.1161/STROKEAHA.119.025614
Johnson, R. T., Bianco, N. A., & Finley, J. M. (2022). Patterns of asymmetry and energy cost
generated from predictive simulations of hemiparetic gait. PLOS Computational Biology,
18(9), e1010466. https://doi.org/10.1371/journal.pcbi.1010466
Jokinen, H., Melkas, S., Ylikoski, R., Pohjasvaara, T., Kaste, M., Erkinjuntti, T., & Hietanen, M.
(2015). Post-stroke cognitive impairment is common even after successful clinical
recovery. European Journal of Neurology, 22(9), 1288–1294.
https://doi.org/10.1111/ene.12743
Jonkers, I., Delp, S., & Patten, C. (2009). Capacity to increase walking speed is limited by
impaired hip and ankle power generation in lower functioning persons post-stroke. Gait
& Posture, 29(1), 129–137. https://doi.org/10.1016/j.gaitpost.2008.07.010

106
Jonsdottir, J., Cattaneo, D., Regola, A., Crippa, A., Recalcati, M., Rabuffetti, M., Ferrarin, M., &
Casiraghi, A. (2007). Concepts of Motor Learning Applied to a Rehabilitation Protocol
Using Biofeedback to Improve Gait in a Chronic Stroke Patient: An A-B System Study
With Multiple Gait Analyses. Neurorehabilitation and Neural Repair, 21(2), 190–194.
https://doi.org/10.1177/1545968306290823
Kantak, S. S., & Winstein, C. J. (2012). Learning–performance distinction and memory
processes for motor skills: A focused review and perspective. Behavioural Brain
Research, 228(1), 219–231. https://doi.org/10.1016/j.bbr.2011.11.028
Kesar, T. M., Binder-Macleod, S. A., Hicks, G. E., & Reisman, D. S. (2011). Minimal detectable
change for gait variables collected during treadmill walking in individuals post-stroke.
Gait & Posture, 33(2), 314–317. https://doi.org/10.1016/j.gaitpost.2010.11.024
Kim, A., Schweighofer, N., & Finley, J. M. (2019). Locomotor skill acquisition in virtual reality
shows sustained transfer to the real world. Journal of NeuroEngineering and
Rehabilitation, 16(1), 113. https://doi.org/10.1186/s12984-019-0584-y
Kimbell, A.-M. (2013, July 24). Cognitive Testing Using the RBANS Update.
http://images.pearsonclinical.com/images/pdf/webinar/rbansjuly2013webinarhandout.pdf
Kleim, J. A., & Jones, T. A. (2008). Principles of Experience-Dependent Neural Plasticity:
Implications for Rehabilitation After Brain Damage. Journal of Speech, Language, and
Hearing Research, 51(1), S225–S239. https://doi.org/10.1044/1092-4388(2008/018)
Kleynen, M., Braun, S. M., Bleijlevens, M. H., Lexis, M. A., Rasquin, S. M., Halfens, J., Wilson,
M. R., Beurskens, A. J., & Masters, R. S. W. (2014). Using a Delphi technique to seek
consensus regarding definitions, descriptions and classification of terms related to
implicit and explicit forms of motor learning. PloS One, 9(6), e100227.
https://doi.org/10.1371/journal.pone.0100227
Kline, T. L., Schmit, B. D., & Kamper, D. G. (2007). Exaggerated interlimb neural coupling
following stroke. Brain, 130(1), 159–169. https://doi.org/10.1093/brain/awl278
Knutsson, E., & Richards, C. (1979). Different types of disturbed motor control in gait of
hemiparetic patients. Brain, 102(2), 405–430. https://doi.org/10.1093/brain/102.2.405
Koller, M. (2016). robustlmm: An R Package for Robust Estimation of Linear Mixed-Effects
Models. Journal of Statistical Software, 75(6). https://doi.org/10.18637/jss.v075.i06

107
Krakauer, J. W., & Shadmehr, R. (2006). Consolidation of motor memory. Trends in
Neurosciences, 29(1), 58–64. https://doi.org/10.1016/j.tins.2005.10.003
Lee, J., Dudley-Javoroski, S., & Shields, R. K. (2021). Motor demands of cognitive testing may
artificially reduce executive function scores in individuals with spinal cord injury. The
Journal of Spinal Cord Medicine, 44(2), 253–261.
https://doi.org/10.1080/10790268.2019.1597482
Leech, K. A., Kim, H. E., & Hornby, T. G. (2018). Strategies to augment volitional and reflex
function may improve locomotor capacity following incomplete spinal cord injury.
Journal of Neurophysiology, 119(3), 894–903. https://doi.org/10.1152/jn.00051.2017
Leech, K. A., Roemmich, R. T., Gordon, J., Reisman, D. S., & Cherry-Allen, K. M. (2022).
Updates in Motor Learning: Implications for Physical Therapist Practice and Education.
Physical Therapy, 102(1), pzab250. https://doi.org/10.1093/ptj/pzab250
Leh, S. E., Petrides, M., & Strafella, A. P. (2010). The Neural Circuitry of Executive Functions
in Healthy Subjects and Parkinson’s Disease. Neuropsychopharmacology, 35(1), Article
1. https://doi.org/10.1038/npp.2009.88
Levine, D. A., Galecki, A. T., Langa, K. M., Unverzagt, F. W., Kabeto, M. U., Giordani, B., &
Wadley, V. G. (2015). Trajectory of Cognitive Decline After Incident Stroke. JAMA,
314(1), 41–51. https://doi.org/10.1001/jama.2015.6968
Lewek, M. D., Bradley, C. E., Wutzke, C. J., & Zinder, S. M. (2014). The Relationship Between
Spatiotemporal Gait Asymmetry and Balance in Individuals With Chronic Stroke.
Journal of Applied Biomechanics, 30(1), 31–36.
Lewek, M. D., Hornby, T. G., Dhaher, Y. Y., & Schmit, B. D. (2007). Prolonged Quadriceps
Activity Following Imposed Hip Extension: A Neurophysiological Mechanism for StiffKnee Gait? Journal of Neurophysiology, 98(6), 3153–3162.
https://doi.org/10.1152/jn.00726.2007
Lewek, M. D., Osborn, A. J., & Wutzke, C. J. (2012). The Influence of Mechanically and
Physiologically Imposed Stiff-Knee Gait Patterns on the Energy Cost of Walking.
Archives of Physical Medicine and Rehabilitation, 93(1), 123–128.
https://doi.org/10.1016/j.apmr.2011.08.019
Lewek, M. D., & Sawicki, G. S. (2019). Trailing limb angle is a surrogate for propulsive limb
forces during walking post-stroke. Clinical Biomechanics, 67, 115–118.
https://doi.org/10.1016/j.clinbiomech.2019.05.011

108
Lingo VanGilder, J., Hengge, C. R., Duff, K., & Schaefer, S. Y. (2018). Visuospatial function
predicts one-week motor skill retention in cognitively intact older adults. Neuroscience
Letters, 664, 139–143. https://doi.org/10.1016/j.neulet.2017.11.032
Lingo VanGilder, J., Hooyman, A., Bosch, P. R., & Schaefer, S. Y. (2021). Generalizing the
predictive relationship between 1-month motor skill retention and Rey–Osterrieth
Delayed Recall scores from nondemented older adults to individuals with chronic stroke:
A short report. Journal of NeuroEngineering and Rehabilitation, 18(1), 94.
https://doi.org/10.1186/s12984-021-00886-4
Lingo VanGilder, J., Lohse, K. R., Duff, K., Wang, P., & Schaefer, S. Y. (2021). Evidence for
associations between Rey-Osterrieth Complex Figure test and motor skill learning in
older adults. Acta Psychologica, 214, 103261.
https://doi.org/10.1016/j.actpsy.2021.103261
Liu, C., McNitt-Gray, J. L., & Finley, J. M. (2022). Impairments in the mechanical effectiveness
of reactive balance control strategies during walking in people post-stroke. Frontiers in
Neurology, 13. https://www.frontiersin.org/articles/10.3389/fneur.2022.1032417
Liu, J., Kim, H. B., Wolf, S. L., & Kesar, T. M. (2020). Comparison of the Immediate Effects of
Audio, Visual, or Audiovisual Gait Biofeedback on Propulsive Force Generation in AbleBodied and Post-stroke Individuals. Applied Psychophysiology and Biofeedback.
https://doi.org/10.1007/s10484-020-09464-1
Lo, O.-Y., Halko, M. A., Devaney, K. J., Wayne, P. M., Lipsitz, L. A., & Manor, B. (2021). Gait
Variability Is Associated With the Strength of Functional Connectivity Between the
Default and Dorsal Attention Brain Networks: Evidence From Multiple Cohorts. The
Journals of Gerontology: Series A, 76(10), e328–e334.
https://doi.org/10.1093/gerona/glab200
Long, A. W., Finley, J. M., & Bastian, A. J. (2015). A marching-walking hybrid induces step
length adaptation and transfers to natural walking. Journal of Neurophysiology, 113(10),
3905–3914. https://doi.org/10.1152/jn.00779.2014
Lüdecke, D., Ben-Shachar, M. S., Patil, I., Waggoner, P., & Makowski, D. (2021). performance:
An R Package for Assessment, Comparison and Testing of Statistical Models. Journal of
Open Source Software, 6(60), 3139. https://doi.org/10.21105/joss.03139
Luke, S. G. (2017). Evaluating significance in linear mixed-effects models in R. Behavior
Research Methods, 49(4), 1494–1502. https://doi.org/10.3758/s13428-016-0809-y

109
Ma, C. Z.-H., Zheng, Y.-P., & Lee, W. C.-C. (2018). Changes in gait and plantar foot loading
upon using vibrotactile wearable biofeedback system in patients with stroke. Topics in
Stroke Rehabilitation, 25(1), 20–27. https://doi.org/10.1080/10749357.2017.1380339
Mahtani, G. B., Kinnaird, C. R., Connolly, M., Holleran, C. L., Hennessy, P. W., Woodward, J.,
Brazg, G., Roth, E. J., & Hornby, T. G. (2017). Altered Sagittal- and Frontal-Plane
Kinematics Following High-Intensity Stepping Training Versus Conventional
Interventions in Subacute Stroke. Physical Therapy, 97(3), 320–329.
https://doi.org/10.2522/ptj.20160281
Matsuda, F., Mukaino, M., Ohtsuka, K., Tanikawa, H., Tsuchiyama, K., Teranishi, T., Kanada,
Y., Kagaya, H., & Saitoh, E. (2017). Biomechanical factors behind toe clearance during
the swing phase in hemiparetic patients. Topics in Stroke Rehabilitation, 24(3), 177–182.
https://doi.org/10.1080/10749357.2016.1234192
McAfoose, J., & Baune, B. T. (2009). Exploring Visual–Spatial Working Memory: A Critical
Review of Concepts and Models. Neuropsychology Review, 19(1), 130–142.
https://doi.org/10.1007/s11065-008-9063-0
McCain, E. M., Berno, M. E., Libera, T. L., Lewek, M. D., Sawicki, G. S., & Saul, K. R. (2021).
Reduced joint motion supersedes asymmetry in explaining increased metabolic demand
during walking with mechanical restriction. Journal of Biomechanics, 126, 110621.
https://doi.org/10.1016/j.jbiomech.2021.110621
McDonald, M. W., Black, S. E., Copland, D. A., Corbett, D., Dijkhuizen, R. M., Farr, T. D.,
Jeffers, M. S., Kalaria, R. N., Karayanidis, F., Leff, A. P., Nithianantharajah, J.,
Pendlebury, S., Quinn, T. J., Clarkson, A. N., & O’Sullivan, M. J. (2019). Cognition in
stroke rehabilitation and recovery research: Consensus-based core recommendations from
the second Stroke Recovery and Rehabilitation Roundtable. International Journal of
Stroke, 14(8), 774–782. https://doi.org/10.1177/1747493019873600
McDowd, J. M., Filion, D. L., Pohl, P. S., Richards, L. G., & Stiers, W. (2003). Attentional
Abilities and Functional Outcomes Following Stroke. The Journals of Gerontology:
Series B, 58(1), P45–P53. https://doi.org/10.1093/geronb/58.1.P45
McKay, B., Yantha, Z., Hussien, J., Carter, M., & Ste-Marie, D. (2022). Meta-Analytic Findings
of the Self-Controlled Motor Learning Literature: Underpowered, Biased, and Lacking
Evidential Value. Meta-Psychology, 6. https://doi.org/10.15626/MP.2021.2803

110
Miller, A., Pohlig, R. T., Wright, T., Kim, H. E., & Reisman, D. S. (2021). Beyond Physical
Capacity: Factors Associated With Real-world Walking Activity After Stroke. Archives
of Physical Medicine and Rehabilitation, 102(10), 1880-1887.e1.
https://doi.org/10.1016/j.apmr.2021.03.023
Moore, J. L., Potter, K., Blankshain, K., Kaplan, S. L., O’Dwyer, L. C., & Sullivan, J. E. (2018).
A Core Set of Outcome Measures for Adults With Neurologic Conditions Undergoing
Rehabilitation: A CLINICAL PRACTICE GUIDELINE. Journal of Neurologic Physical
Therapy, 42(3), 174. https://doi.org/10.1097/NPT.0000000000000229
Moore, J. L., Roth, E. J., Killian, C., & Hornby, T. G. (2010). Locomotor Training Improves
Daily Stepping Activity and Gait Efficiency in Individuals Poststroke Who Have
Reached a “Plateau” in Recovery. Stroke, 41(1), 129–135.
https://doi.org/10.1161/STROKEAHA.109.563247
Moore, S. A., Boyne, P., Fulk, G., Verheyden, G., & Fini, N. A. (2022). Walk the Talk: Current
Evidence for Walking Recovery After Stroke, Future Pathways and a Mission for
Research and Clinical Practice. Stroke, 53(11), 3494–3505.
https://doi.org/10.1161/STROKEAHA.122.038956
Morris, R., Martini, D. N., Smulders, K., Kelly, V. E., Zabetian, C. P., Poston, K., Hiller, A.,
Chung, K. A., Yang, L., Hu, S.-C., Edwards, K. L., Cholerton, B., Grabowski, T. J.,
Montine, T. J., Quinn, J. F., & Horak, F. (2019). Cognitive associations with
comprehensive gait and static balance measures in Parkinson’s disease. Parkinsonism &
Related Disorders, 69, 104–110. https://doi.org/10.1016/j.parkreldis.2019.06.014
Nasreddine, Z. S., Phillips, N. A., Bédirian, V., Charbonneau, S., Whitehead, V., Collin, I.,
Cummings, J. L., & Chertkow, H. (2005). The Montreal Cognitive Assessment, MoCA:
A brief screening tool for mild cognitive impairment. Journal of the American Geriatrics
Society, 53(4), 695–699. https://doi.org/10.1111/j.1532-5415.2005.53221.x
Nepveu, J.-F., Thiel, A., Tang, A., Fung, J., Lundbye-Jensen, J., Boyd, L. A., & Roig, M. (2017).
A Single Bout of High-Intensity Interval Training Improves Motor Skill Retention in
Individuals With Stroke. Neurorehabilitation and Neural Repair, 31(8), 726–735.
https://doi.org/10.1177/1545968317718269
Newman, A. B., Simonsick, E. M., Naydeck, B. L., Boudreau, R. M., Kritchevsky, S. B., Nevitt,
M. C., Pahor, M., Satterfield, S., Brach, J. S., Studenski, S. A., & Harris, T. B. (2006).
Association of Long-Distance Corridor Walk Performance With Mortality,
Cardiovascular Disease, Mobility Limitation, and Disability. JAMA, 295(17), 2018.
https://doi.org/10.1001/jama.295.17.2018

111
Nguyen, T. M., Jackson, R. W., Aucie, Y., de Kam, D., Collins, S. H., & Torres-Oviedo, G.
(2020). Self-selected step length asymmetry is not explained by energy cost minimization
in individuals with chronic stroke. Journal of NeuroEngineering and Rehabilitation,
17(1), 119. https://doi.org/10.1186/s12984-020-00733-y
Nys, G. M. S., van Zandvoort, M. J. E., de Kort, P. L. M., Jansen, B. P. W., de Haan, E. H. F., &
Kappelle, L. J. (2007). Cognitive disorders in acute stroke: Prevalence and clinical
determinants. Cerebrovascular Diseases (Basel, Switzerland), 23(5–6), 408–416.
https://doi.org/10.1159/000101464
Ohta, M., Tanabe, S., Katsuhira, J., & Tamari, M. (2023). Kinetic and kinematic parameters
associated with late braking force and effects on gait performance of stroke patients.
Scientific Reports, 13(1), Article 1. https://doi.org/10.1038/s41598-023-34904-3
Olney, S. J., & Richards, C. (1996). Hemiparetic gait following stroke. Part I: Characteristics.
Gait & Posture, 4(2), 136–148. https://doi.org/10.1016/0966-6362(96)01063-6
Padmanabhan, P., Rao, K. S., Gulhar, S., Cherry-Allen, K. M., Leech, K. A., & Roemmich, R. T.
(2020). Persons post-stroke improve step length symmetry by walking asymmetrically.
Journal of NeuroEngineering and Rehabilitation, 17(1), 105.
https://doi.org/10.1186/s12984-020-00732-z
Padmanabhan, P., Sreekanth Rao, K., Gulhar, S., Cherry-Allen, K. M., Leech, K. A., &
Roemmich, R. T. (2019). Persons post-stroke restore step length symmetry by walking
asymmetrically [Preprint]. Neuroscience. https://doi.org/10.1101/799775
Pang, M. Y. C., Charlesworth, S. A., Lau, R. W. K., & Chung, R. C. K. (2013). Using Aerobic
Exercise to Improve Health Outcomes and Quality of Life in Stroke: Evidence-Based
Exercise Prescription Recommendations. Cerebrovascular Diseases, 35(1), 7–22.
https://doi.org/10.1159/000346075
Park, S., Liu, C., Sánchez, N., Tilson, J. K., Mulroy, S. J., & Finley, J. M. (2021). Using
Biofeedback to Reduce Step Length Asymmetry Impairs Dynamic Balance in People
Poststroke. Neurorehabilitation and Neural Repair, 35(8), 738–749.
https://doi.org/10.1177/15459683211019346
Penke, K., Scott, K., Sinskey, Y., & Lewek, M. D. (2019). Propulsive Forces Applied to the
Body’s Center of Mass Affect Metabolic Energetics Poststroke. Archives of Physical
Medicine and Rehabilitation, 100(6), 1068–1075.
https://doi.org/10.1016/j.apmr.2018.10.010

112
Pinter, D., Enzinger, C., Gattringer, T., Eppinger, S., Niederkorn, K., Horner, S., Fandler, S.,
Kneihsl, M., Krenn, K., Bachmaier, G., & Fazekas, F. (2019). Prevalence and short-term
changes of cognitive dysfunction in young ischaemic stroke patients. European Journal
of Neurology, 26(5), 727–732. https://doi.org/10.1111/ene.13879
R Core Team. (n.d.). R: A language and environment for statistical computing [Computer
software]. https://www.R-project.org/.
Ramakrishnan, T., Kim, S. H., & Reed, K. B. (2019). Human Gait Analysis Metric for Gait
Retraining. Applied Bionics and Biomechanics, 2019, 1–8.
https://doi.org/10.1155/2019/1286864
Ramakrishnan, T., Lahiff, C.-A., & Reed, K. B. (2018). Comparing Gait with Multiple Physical
Asymmetries Using Consolidated Metrics. Frontiers in Neurorobotics, 12, 2.
https://doi.org/10.3389/fnbot.2018.00002
Randolph, C. (1998). Repeatable Battery for the Assessment of Neuropsychological Status
(RBANS) Manual. The Psychological Corporation.
Randolph, C., Tierney, M. C., Mohr, E., & Chase, T. N. (1998). The Repeatable Battery for the
Assessment of Neuropsychological Status (RBANS): Preliminary clinical validity.
Journal of Clinical and Experimental Neuropsychology, 20(3), 310–319.
https://doi.org/10.1076/jcen.20.3.310.823
Reisman, D. S., Kesar, T. M., Perumal, R., Roos, M. A., Rudolph, K. S., Higginson, J., Helm, E.,
& Binder-Macleod, S. (2013). Time course of functional and biomechanical
improvements during a gait training intervention in persons with chronic stroke. Journal
of Neurologic Physical Therapy : JNPT, 37(4), 159–165.
https://doi.org/10.1097/NPT.0000000000000020
Roelker, S. A., Bowden, M. G., Kautz, S. A., & Neptune, R. R. (2019). Paretic propulsion as a
measure of walking performance and functional motor recovery post-stroke: A review.
Gait & Posture, 68, 6–14. https://doi.org/10.1016/j.gaitpost.2018.10.027
Rost, N. S., Brodtmann, A., Pase, M. P., Van Veluw, S. J., Biffi, A., Duering, M., Hinman, J. D.,
& Dichgans, M. (2022). Post-Stroke Cognitive Impairment and Dementia. Circulation
Research, 130(8), 1252–1271. https://doi.org/10.1161/CIRCRESAHA.122.319951
Routson, R. L., Kautz, S. A., & Neptune, R. R. (2014). Modular organization across changing
task demands in healthy and poststroke gait. Physiological Reports, 2(6), e12055.
https://doi.org/10.14814/phy2.12055

113
Ryan, H. P., Husted, C., & Lewek, M. D. (2020). Improving Spatiotemporal Gait Asymmetry
Has Limited Functional Benefit for Individuals Poststroke. Journal of Neurologic
Physical Therapy, 44(3), 197–204. https://doi.org/10.1097/NPT.0000000000000321
Sánchez, N., Acosta, A. M., Lopez-Rosado, R., Stienen, A. H. A., & Dewald, J. P. A. (2017).
Lower Extremity Motor Impairments in Ambulatory Chronic Hemiparetic Stroke:
Evidence for Lower Extremity Weakness and Abnormal Muscle and Joint Torque
Coupling Patterns. Neurorehabilitation and Neural Repair, 31(9), 814–826.
https://doi.org/10.1177/1545968317721974
Sánchez, N., & Finley, J. M. (2018). Individual Differences in Locomotor Function Predict the
Capacity to Reduce Asymmetry and Modify the Energetic Cost of Walking Poststroke.
Neurorehabilitation and Neural Repair, 32(8), 701–713.
https://doi.org/10.1177/1545968318787913
Sánchez, N., Schweighofer, N., & Finley, J. M. (2021). Different Biomechanical Variables
Explain Within-Subjects Versus Between-Subjects Variance in Step Length Asymmetry
Post-Stroke. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 29,
1188–1198. https://doi.org/10.1109/TNSRE.2021.3090324
Sánchez-Cubillo, I., Periáñez, J. A., Adrover-Roig, D., Rodríguez-Sánchez, J. M., Ríos-Lago,
M., Tirapu, J., & Barceló, F. (2009). Construct validity of the Trail Making Test: Role of
task-switching, working memory, inhibition/interference control, and visuomotor
abilities. Journal of the International Neuropsychological Society, 15(3), 438–450.
https://doi.org/10.1017/S1355617709090626
Sauder, N. R., Meyer, A. J., Allen, J. L., Ting, L. H., Kesar, T. M., & Fregly, B. J. (2019).
Computational Design of FastFES Treatment to Improve Propulsive Force Symmetry
During Post-stroke Gait: A Feasibility Study. Frontiers in Neurorobotics, 13.
https://www.frontiersin.org/articles/10.3389/fnbot.2019.00080
Schaefer, S. Y., Hooyman, A., Haikalis, N. K., Essikpe, R., Lohse, K. R., Duff, K., & Wang, P.
(2022). Efficacy of Corsi Block Tapping Task training for improving visuospatial skills:
A non-randomized two-group study. Experimental Brain Research, 240(11), 3023–3032.
https://doi.org/10.1007/s00221-022-06478-5
Schmid, A., Duncan, P. W., Studenski, S., Lai, S. M., Richards, L., Perera, S., & Wu, S. S.
(2007). Improvements in Speed-Based Gait Classifications Are Meaningful. Stroke,
38(7), 2096–2100. https://doi.org/10.1161/STROKEAHA.106.475921

114
Schmidt, R. A., & Bjork, R. A. (1992). New Conceptualizations of Practice: Common Principles
in Three Paradigms Suggest New Concepts for Training. Psychological Science, 3(4),
207–218. https://doi.org/10.1111/j.1467-9280.1992.tb00029.x
Schmidt, R. A., Lee, T. D., Winstein, C. J., Wulf, G., & Zelaznik, H. N. (2019). Motor control
and learning: A behavioral emphasis (Sixth edition). Human Kinetics.
Schwartz, M. H., & Rozumalski, A. (2008). The gait deviation index: A new comprehensive
index of gait pathology. Gait & Posture, 28(3), 351–357.
https://doi.org/10.1016/j.gaitpost.2008.05.001
Schweighofer, N., Lee, J.-Y., Goh, H.-T., Choi, Y., Kim, S. S., Stewart, J. C., Lewthwaite, R., &
Winstein, C. J. (2011). Mechanisms of the contextual interference effect in individuals
poststroke. Journal of Neurophysiology, 106(5), 2632–2641.
https://doi.org/10.1152/jn.00399.2011
Shiavi, R., Bugle, H. J., & Limbird, T. (1987). Electromyographic gait assessment, Part 2:
Preliminary assessment of hemiparetic synergy patterns. Journal of Rehabilitation
Research and Development, 24(2), 24–30.
Shin, S. Y., Lee, R. K., Spicer, P., & Sulzer, J. (2020). Does kinematic gait quality improve with
functional gait recovery? A longitudinal pilot study on early post-stroke individuals.
Journal of Biomechanics, 105, 109761. https://doi.org/10.1016/j.jbiomech.2020.109761
Spencer, J., Wolf, S. L., & Kesar, T. M. (2021). Biofeedback for Post-stroke Gait Retraining: A
Review of Current Evidence and Future Research Directions in the Context of Emerging
Technologies. Frontiers in Neurology, 12, 637199.
https://doi.org/10.3389/fneur.2021.637199
Stanhope, V. A., Knarr, B. A., Reisman, D. S., & Higginson, J. S. (2014). Frontal plane
compensatory strategies associated with self-selected walking speed in individuals poststroke. Clinical Biomechanics, 29(5), 518–522.
https://doi.org/10.1016/j.clinbiomech.2014.03.013
Steele, K. M., & Schwartz, M. H. (2022). Causal Effects of Motor Control on Gait Kinematics
After Orthopedic Surgery in Cerebral Palsy: A Machine-Learning Approach. Frontiers in
Human Neuroscience, 16.
https://www.frontiersin.org/articles/10.3389/fnhum.2022.846205

115
Stewart, J. C., Lewthwaite, R., Rocktashel, J., & Winstein, C. J. (2019). Self-efficacy and Reach
Performance in Individuals With Mild Motor Impairment Due to Stroke.
Neurorehabilitation and Neural Repair, 33(4), 319–328.
https://doi.org/10.1177/1545968319836231
Stoquart, G. G., Detrembleur, C., Palumbo, S., Deltombe, T., & Lejeune, T. M. (2008). Effect of
Botulinum Toxin Injection in the Rectus Femoris on Stiff-Knee Gait in People With
Stroke: A Prospective Observational Study. Archives of Physical Medicine and
Rehabilitation, 89(1), 56–61. https://doi.org/10.1016/j.apmr.2007.08.131
Sukal, T. M., Ellis, M. D., & Dewald, J. P. A. (2007). Shoulder abduction-induced reductions in
reaching work area following hemiparetic stroke: Neuroscientific implications.
Experimental Brain Research, 183(2), 215–223. https://doi.org/10.1007/s00221-007-
1029-6
Sullivan, K. J., Brown, D. A., Klassen, T., Mulroy, S., Ge, T., Azen, S. P., Winstein, C. J., & for
the Physical Therapy Clinical Research Network (PTClinResNet). (2007). Effects of
Task-Specific Locomotor and Strength Training in Adults Who Were Ambulatory After
Stroke: Results of the STEPS Randomized Clinical Trial. Physical Therapy, 87(12),
1580–1602. https://doi.org/10.2522/ptj.20060310
Tate, J. J., & Milner, C. E. (2010). Real-Time Kinematic, Temporospatial, and Kinetic
Biofeedback During Gait Retraining in Patients: A Systematic Review. Physical Therapy,
90(8), 1123–1134. https://doi.org/10.2522/ptj.20080281
The American Academy of Clinical Neuropsychology. (2021). Position Statement on Use of
Race as a Factor in Neuropsychological Test Norming and Performance Prediction.
https://theaacn.org/official-position-papers-and-statements/
Tok, F., Balaban, B., Yaşar, E., Alaca, R., & Tan, A. K. (2012). The Effects of Onabotulinum
Toxin A Injection into Rectus Femoris Muscle in Hemiplegic Stroke Patients with StiffKnee Gait: A Placebo-Controlled, Nonrandomized Trial. American Journal of Physical
Medicine & Rehabilitation, 91(4), 321–326.
https://doi.org/10.1097/PHM.0b013e3182465feb
Tsao, C. W., Aday, A. W., Almarzooq, Z. I., Alonso, A., Beaton, A. Z., Bittencourt, M. S.,
Boehme, A. K., Buxton, A. E., Carson, A. P., Commodore-Mensah, Y., Elkind, M. S. V.,
Evenson, K. R., Eze-Nliam, C., Ferguson, J. F., Generoso, G., Ho, J. E., Kalani, R.,
Khan, S. S., Kissela, B. M., … on behalf of the American Heart Association Council on
Epidemiology and Prevention Statistics Committee and Stroke Statistics Subcommittee.
(2022). Heart Disease and Stroke Statistics—2022 Update: A Report From the American
Heart Association. Circulation, 145(8). https://doi.org/10.1161/CIR.0000000000001052

116
Turns, L. J., Neptune, R. R., & Kautz, S. A. (2007). Relationships Between Muscle Activity and
Anteroposterior Ground Reaction Forces in Hemiparetic Walking. Archives of Physical
Medicine and Rehabilitation, 88(9), 1127–1135.
https://doi.org/10.1016/j.apmr.2007.05.027
Tyrell, C. M., Roos, M. A., Rudolph, K. S., & Reisman, D. S. (2011). Influence of Systematic
Increases in Treadmill Walking Speed on Gait Kinematics After Stroke. Physical
Therapy, 91(3), 392–403. https://doi.org/10.2522/ptj.20090425
Van Criekinge, T., Heremans, C., Burridge, J., Deutsch, J. E., Hammerbeck, U., Hollands, K.,
Karthikbabu, S., Mehrholz, J., Moore, J. L., Salbach, N. M., Schröder, J., Veerbeek, J.
M., Weerdesteyn, V., Borschmann, K., Churilov, L., Verheyden, G., & Kwakkel, G.
(2023). Standardized measurement of balance and mobility post-stroke: Consensus-based
core recommendations from the third Stroke Recovery and Rehabilitation Roundtable.
Neurorehabilitation and Neural Repair, 15459683231209154.
https://doi.org/10.1177/15459683231209154
VanGilder, J. L., Bergamino, M., Hooyman, A., Fitzhugh, M. C., Rogalsky, C., Stewart, J. C.,
Beeman, S. C., & Schaefer, S. Y. (2022). Using whole-brain diffusion tensor analysis to
evaluate white matter structural correlates of delayed visuospatial memory and one-week
motor skill retention in nondemented older adults: A preliminary study. PLOS ONE,
17(9), e0274955. https://doi.org/10.1371/journal.pone.0274955
VanGilder, J. L., Hooyman, A., Peterson, D. S., & Schaefer, S. Y. (2020). Post-stroke cognitive
impairments and responsiveness to motor rehabilitation: A review. Current Physical
Medicine and Rehabilitation Reports, 8(4), 461–468. https://doi.org/10.1007/s40141-020-
00283-3
Virani, S. S., Alonso, A., Benjamin, E. J., Bittencourt, M. S., Callaway, C. W., Carson, A. P.,
Chamberlain, A. M., Chang, A. R., Cheng, S., Delling, F. N., Djousse, L., Elkind, M. S.
V., Ferguson, J. F., Fornage, M., Khan, S. S., Kissela, B. M., Knutson, K. L., Kwan, T.
W., Lackland, D. T., … On behalf of the American Heart Association Council on
Epidemiology and Prevention Statistics Committee and Stroke Statistics Subcommittee.
(2020). Heart Disease and Stroke Statistics—2020 Update: A Report From the American
Heart Association. Circulation, 141(9). https://doi.org/10.1161/CIR.0000000000000757
Wade, D. T., Wood, V. A., Heller, A., Maggs, J., & Hewer, R. L. (1987). Walking After Stroke.
Measurement and Recovery Over the First 3 Months. Journal of Rehabilitation Medicine,
19(1), Article 1. https://doi.org/10.2340/1650197787192530

117
Wang, P., Lingo VanGilder, J., Schweighofer, N., & Schaefer, S. Y. (2022). Rey-Osterrieth
complex figure recall scores and motor skill learning in older adults: A non-linear mixed
effect model-based analysis. Human Movement Science, 86, 103004.
https://doi.org/10.1016/j.humov.2022.103004
Wei, T.-S., Liu, P.-T., Chang, L.-W., & Liu, S.-Y. (2017). Gait asymmetry, ankle spasticity, and
depression as independent predictors of falls in ambulatory stroke patients. PloS One,
12(5), e0177136. https://doi.org/10.1371/journal.pone.0177136
Winstein, C. J., Pohl, P. S., Cardinale, C., Green, A., Scholtz, L., & Waters, C. S. (1996).
Learning a Partial-Weight-Bearing Skill: Effectiveness of Two Forms of Feedback.
Physical Therapy, 76(9), 985–993. https://doi.org/10.1093/ptj/76.9.985
Winstein, C. J., & Schmidt, R. A. (1990). Reduced frequency of knowledge of results enhances
motor skill learning. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 16(4), 677–691. http://dx.doi.org.libproxy2.usc.edu/10.1037/0278-
7393.16.4.677
Winter, D. (2009). Biomechanics and Motor Control of Human Movement (4th ed.). John Wiley
& Sons, Inc.
Wong, A., Black, S. E., Yiu, S. Y. P., Au, L. W. C., Lau, A. Y. L., Soo, Y. O. Y., Chan, A. Y.
Y., Leung, T. W. H., Wong, L. K. S., Kwok, T. C. Y., Cheung, T. C. K., Leung, K.-T.,
Lam, B. Y. K., Kwan, J. S. K., & Mok, V. C. T. (2018). Converting MMSE to MoCA
and MoCA 5-minute protocol in an educationally heterogeneous sample with stroke or
transient ischemic attack. International Journal of Geriatric Psychiatry, 33(5), 729–734.
https://doi.org/10.1002/gps.4846
Wonsetler, E. C., & Bowden, M. G. (2017). A systematic review of mechanisms of gait speed
change post-stroke. Part 2: Exercise capacity, muscle activation, kinetics, and kinematics.
Topics in Stroke Rehabilitation, 24(5), 394–403.
https://doi.org/10.1080/10749357.2017.1282413
Wulf, G., & Lewthwaite, R. (2016). Optimizing performance through intrinsic motivation and
attention for learning: The OPTIMAL theory of motor learning. Psychonomic Bulletin &
Review, 23(5), 1382–1414. https://doi.org/10.3758/s13423-015-0999-9

118
Yang, Y.-M., Zhao, Z.-M., Wang, W., Dong, F.-M., Wang, P.-P., Jia, Y.-J., Han, N., Jia, Y.-L.,
& Wang, J.-H. (2020). Trends in cognitive function assessed by a battery of
neuropsychological tests after mild acute ischemic stroke. Journal of Stroke and
Cerebrovascular Diseases, 29(7), 104887.
https://doi.org/10.1016/j.jstrokecerebrovasdis.2020.104887

119
APPENDIX A
MINIMAL STEP LENGTH ASYMMETRY DISTRIBUTION IN OUR SAMPLE
For the Aim 2 data collection, we also collected data on a separate day where participants
used step length biofeedback designed to reduce step length asymmetry (Figure A.1A). We wanted
to understand how altering step length asymmetry impacted overall gait asymmetry, as step length
asymmetry is a popular intervention target. We ultimately chose not to include the step length
asymmetry data in the final analysis due to the minimal distribution of step length asymmetry of
participants in our sample. Only 14 out of 24 participants who completed the step length
biofeedback had step lengths greater than 0.06 (Figure A.1B), which is what previous work has
used as a threshold for inclusion in analyses investigating the effect of step length changes (Park
et al., 2021). There was a group reduction in step length asymmetry across trials (-0.03 reduction
from baseline for each trial; Figure A.1C); however, the reduction did not exceed the minimal
detectable change calculated for this dataset (0.08). Therefore, because of the minimal distribution
of the data and the small changes in step length asymmetry, we did not move forward with the
main analysis.
Figure A.1. Step length biofeedback schematic and step length data distribution. A) Our custom step length biofeedback
code provided real-time left and right step lengths (purple bars) by tracking the lateral malleoli markers. It also provided
step length end-point feedback at right and left foot-strike to allow participants to visualize their previous step length
(blue dots). The goal was set at the average between the right and left step lengths at baseline, with a + 5% tolerance goal
zone (orange rectangles). B) Step length asymmetry magnitude at baseline for each participant. C) Step length asymmetry
magnitude for each participant across trials.

120
APPENDIX B
TEST-RETEST RELIABILITY AND MINIMAL DETECTABLE CHANGE OF THE
COMBINED GAIT ASYMMETRY METRIC
To assess the test-retest reliability of CGAM, we used baseline data collected for Aim 2
(Chapter 3) for participants who completed two baseline sessions (n = 23). The second testing
session was collected to examine the impact of changing step length on overall gait asymmetry;
however, due to the lack of step length asymmetry distribution in our sample (Appendix A), we
did not include the step length asymmetry data in the main analysis. Sessions were completed a
minimum of one week apart. We calculated the intraclass correlation coefficient (ICC) of average
baseline CGAM across sessions using the irr package in R (Gamer et al., 2019). To calculate the
within-session minimal detectable change (MDC), we used the final thirty strides of the baseline
propulsion trial for each participant (n = 29). To calculate the between-session MDC, we used
baseline data for participants who completed two sessions (n = 23). We used the following formula
to calculate MDC both within-session and between-sessions (Kesar et al., 2011):
= � ∗ √1 − � ∗ 1.962 ∗ √2
The CGAM showed excellent levels of agreement across days (ICC [95% confidence
interval]: 0.93 [0.84, 0.97]), suggesting that CGAM is a reliable measure across days in individuals
post-stroke. The within-session MDC for CGAM was 0.66, and the between-session MDC was
1.7. Together, these results suggest that CGAM is a consistent measure across days and may be an
adequate measure to track changes in overall gait asymmetry post-stroke over the course of an
intervention.

121
APPENDIX C
SENSITIVITY OF CGAM TO VARIABLE SELECTION
The CGAM allows the inclusion of any biomechanical variable of interest (Ramakrishnan
et al., 2018, 2019), which allows flexibility and the ability to analyze specific variables of interest.
However, this flexibility also limits the ability to compare between two CGAM values that were
calculated based on different variables (Ramakrishnan et al., 2018, 2019). Therefore, we
performed a sensitivity analysis to examine how including different variable combinations in the
calculation impacted the value and interpretation of the CGAM.
The CGAM calculated for the Chapter 4 analyses included seven biomechanical variables:
peak swing knee flexion, peak hip flexion, trailing limb angle, circumduction, single-limb support
time, double-limb support time, and step length. Here, we calculated CGAM values for each
possible combination of variables (n = 120) using baseline data from Aim 2.
We found that across possible variable combinations, the average CGAM value across
participants was heterogenous, ranging from 13.1 to 66.7 (Figure C.1A). When the CGAM values
are averaged across the number of variables included in the calculation, the average CGAM values
are similar (ranging from 37.4 to 38.8; Figure C.1B). However, as the number of variables included
in the CGAM increased, CGAM variability decreased (ranging from 15.2 to 5.5; Figure C.1B).

122
The biomechanical characteristics of the participants included will also impact how
variable selection will influence the CGAM interpretation. To illustrate this, we have plotted two
representative participants’ average CGAM values across variable combinations in Figure C.2.
One participant had low asymmetry in all possible gait metrics (Table C.1; representative
participant 1) and the other had generally higher asymmetry indices that varied widely between
gait metrics (Table C.1; representative
participant 2). For representative
participant 1 (Figure C.2; gray), the
CGAM value is relatively consistent
regardless of the variables included in
the calculation. This is unsurprising, as
the participant has minimal asymmetry
Table C.1. Symmetry index values for two representative participants
Representative
participant 1
Representative
participant 2
Circumduction 42.0 96.0
Peak hip flexion 11.0 127.1
Peak swing knee flexion 9.9 119.0
Trailing limb angle 4.7 105.4
Double-limb support time 18.6 112.2
Single-limb support time 6.2 11.6
Step length 14.6 72.5
Figure C.1. Average CGAM across variable combinations. A) CGAM values were averaged across participants for each possible
variable combination. The dashed lines represent . B) Average CGAM across the number of variables included in the CGAM
calculation. The error bars represent the standard deviation.

123
across metrics. However, the
variables included in the
calculation for representative
participant two (Figure C.2;
green) made a large impact on
the CGAM value. To illustrate
this, two points are highlighted
in yellow and red on Figure C.2.
The yellow point is CGAM
calculated using circumduction,
double-limb support time, peak
swing knee flexion, trailing
limb angle, peak hip flexion,
and step length. As you can see,
the resulting CGAM value is quite high (125.7). However, the red point is calculated using
circumduction, double-limb support time, single-limb support time, trailing limb angle, peak hip
flexion, and step length. This is different from the previous iteration by including single-limb
support time instead of peak swing knee flexion angle. This participant has much lower singlelimb support asymmetry than knee flexion asymmetry (Table C.1), which reduces the CGAM
value dramatically (55.4). If you compared this same participant across studies that used different
variables, you would draw different conclusions about their multidimensional gait asymmetry
from each.
Figure C.2. CGAM across variable combinations for two representative participants.
Representative participant 1 (gray) had low asymmetry values in all possible metrics,
and representative participant 2 (green) had higher and more variable asymmetry
across possible metrics. The red and yellow points highlight two different CGAM
values that are vastly different (125.7 vs. 55.4) because of the variables included in
the calculation.

124
This analysis highlights the importance of 1) understanding the differences in the variables
included in the CGAM calculation across studies and 2) choosing which variables to include in
your own CGAM calculation. While there is no definitive “truth” to which variables should be
included, the population and importance of each variable to that population should be considered.

Abstract (if available)

Abstract Fast walking and propulsion visual biofeedback improve select post-stroke biomechanical impairments; however, their effect on overall gait biomechanics is unclear. We examined how fast walking and propulsion biofeedback impacted overall gait biomechanics. Additionally, given that cognitive impairment is a common post-stroke, we investigated the role of cognitive impairment in explicit locomotor learning. We evaluated the effect of speed on overall gait biomechanics in people post-stroke compared to neurotypical adults using k-means clustering analysis. We found two clusters representing neurotypical gait behavior (comprised of neurotypical and post-stroke participants) and stroke gait behavior (comprised of post-stroke participants). The distance between clusters did not change at faster speeds, suggesting that fast walking did not improve overall gait biomechanics. Next, we investigated how biofeedback-driven changes in propulsion asymmetry impacted overall gait asymmetry, measured by the combined gait asymmetry metric (CGAM). We found a positive association between propulsion asymmetry magnitude change and CGAM change. However, we would only expect an average CGAM reduction of 1.8, suggesting that reducing propulsion asymmetry is unlikely to meaningfully reduce overall gait asymmetry. Finally, we aimed to understand which cognitive domains were associated with visual biofeedback performance and immediate recall. We found that visuospatial/constructional skills and motor impairment best-explained performance, and language skills best-explained immediate recall, suggesting specific cognitive domain impairments explain variability in locomotor learning. These studies demonstrate that approaches that improve select biomechanical impairments do not necessarily lead to improvements in overall gait biomechanics and that visuospatial/constructional and language skills are related to explicit locomotor learning post-stroke.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Development and implementation of a modular muscle-computer interface for personalized motor rehabilitation after stroke

PDF

Understanding reactive balance control strategies in non-disabled and post-stroke gait

PDF

Locomotor skill learning in virtual reality in healthy adults and people with Parkinson disease

PDF

Relationship between brain structure and motor behavior in chronic stroke survivors

PDF

Arm choice post-stroke

PDF

Hemisphere-specific deficits in the control of bimanual movements after stroke

PDF

Learning reaching skills in non-disabled and post-stroke individuals

PDF

Feasibility, acceptability, and implementation context of a complex telerehabilitation intervention for post-stroke upper extremity recovery

PDF

Deficits and rehabilitation of upper-extremity multi-joint movements in individuals with chronic stroke

PDF

Effects of a 12-week golf training program on gait and cognition: the Golf Intervention for Veterans Exercise (GIVE) study

PDF

Model-based approaches to objective inference during steady-state and adaptive locomotor control

PDF

Using ecological momentary assessment to study the impact of social-cognitive factors on paretic hand use after stroke

PDF

Computational model of stroke therapy and long term recovery

PDF

Gait flexibility following anterior cruciate ligament reconstruction: implications for early gait retraining

PDF

Manipulating cognitive intensity during aerobic exercise: clinical proof of concept

PDF

The influence of risk on decision-making during walking

PDF

Development of postural strategies for the control of forward momentum

PDF

Trunk control during dynamic balance: effects of cognitive dual-task interference and a history of recurrent low back pain

PDF

The effects of prolonged fasting/ fasting mimicking diet (FMD) on CNS protection, regeneration, and treatment

PDF

The footprint of pain: investigating persistence of altered trunk control in recurrent low back pain

Asset Metadata

Creator Kettlety, Sarah A. (author)

Core Title The effects of fast walking, biofeedback, and cognitive impairment on post-stroke gait

School School of Dentistry

Degree Doctor of Philosophy

Degree Program Biokinesiology

Degree Conferral Date 2024-05

Publication Date 03/28/2024

Defense Date 02/21/2024

Publisher Los Angeles, California (original), University of Southern California (original), University of Southern California. Libraries (digital)

Tag biofeedback,cognition,fast walking,gait,OAI-PMH Harvest,overall gait biomechanics,stroke

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Leech, Kristan A. (committee chair), Finley, James M. (committee member), Liew, Sook-Lei (committee member), Schweighofer, Nicolas (committee member), Winstein, Carolee J. (committee member)

Creator Email kettlety@usc.edu,sarahkettlety@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113862150

Unique identifier UC113862150

Identifier etd-KettletySa-12732.pdf (filename)

Legacy Identifier etd-KettletySa-12732

Document Type Dissertation

Format theses (aat)

Rights Kettlety, Sarah A.

Internet Media Type application/pdf

Type texts

Source 20240401-usctheses-batch-1133 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu