Gait asymmetry is often observed in populations with varying degrees of neuromuscular control. While changes in vertical ground reaction force (vGRF) peak magnitude are associated with altered limb loading that can be observed during asymmetric gait, the challenge is identifying techniques with the sensitivity to detect these altered movement patterns. Autoregressive (AR) modeling has successfully delineated between healthy and pathological gait during running; but has been little explored in walking. Thus, AR modeling was implemented to assess differences in vGRF pattern dynamics during symmetric and asymmetric walking. We hypothesized that the AR model coefficients would better detect differences amongst the symmetric and asymmetric walking conditions than the vGRF peak magnitude mean. Seventeen healthy individuals performed a protocol that involved walking on a split-belt instrumented treadmill at different symmetric (0.75m/s, 1.0 m/s, 1.5 m/s) and asymmetric (Side 1: 0.75m/s-Side 2:1.0 m/s; Side 1:1.0m/s-Side 2:1.5 m/s) gait conditions. Vertical ground reaction force peaks extracted during the weight-acceptance and propulsive phase of each step were used to construct a vGRF peak time series. Then, a second order AR model was fit to the vGRF peak waveform data to determine the AR model coefficients. The resulting AR coefficients were plotted on a stationarity triangle and their distance from the triangle centroid was computed. Significant differences in vGRF patterns were detected amongst the symmetric and asymmetric conditions using the AR modeling coefficients (p = 0.01); however, no differences were found when comparing vGRF peak magnitude means. These findings suggest that AR modeling has the sensitivity to identify differences in gait asymmetry that could aid in monitoring rehabilitation progression.
Gait asymmetry is often observed in populations with varying degrees of neuromuscular control. While changes in vertical ground reaction force (vGRF) peak magnitude are associated with altered limb loading that can be observed during asymmetric gait, the challenge is identifying techniques with the sensitivity to detect these altered movement patterns. Autoregressive (AR) modeling has successfully delineated between healthy and pathological gait during running; but has been little explored in walking. Thus, AR modeling was implemented to assess differences in vGRF pattern dynamics during symmetric and asymmetric walking. We hypothesized that the AR model coefficients would better detect differences amongst the symmetric and asymmetric walking conditions than the vGRF peak magnitude mean. Seventeen healthy individuals performed a protocol that involved walking on a split-belt instrumented treadmill at different symmetric (0.75m/s, 1.0 m/s, 1.5 m/s) and asymmetric (Side 1: 0.75m/s-Side 2:1.0 m/s; Side 1:1.0m/s-Side 2:1.5 m/s) gait conditions. Vertical ground reaction force peaks extracted during the weight-acceptance and propulsive phase of each step were used to construct a vGRF peak time series. Then, a second order AR model was fit to the vGRF peak waveform data to determine the AR model coefficients. The resulting AR coefficients were plotted on a stationarity triangle and their distance from the triangle centroid was computed. Significant differences in vGRF patterns were detected amongst the symmetric and asymmetric conditions using the AR modeling coefficients (p = 0.01); however, no differences were found when comparing vGRF peak magnitude means. These findings suggest that AR modeling has the sensitivity to identify differences in gait asymmetry that could aid in monitoring rehabilitation progression.
Gait asymmetry is often observed in individuals as a compensatory response to neuromuscular deficits [1-5]. Yet despite the fact that gait asymmetry is commonly exhibited in populations with varying degrees of neuromuscular control, differences in asymmetric gait patterns exist even amongst these affected groups [1-5]. Thus, it suggests that not all gait asymmetry patterns are the same. However, uncovering these changes in gait patterns requires the identification of appropriate metrics and analysis techniques that possess the sensitivity to delineate between healthy and abnormal movement patterns. The vertical ground reaction force (vGRF) waveform is a valuable source to extract gait metrics from because it functions as a surrogate for one’s center of mass motion with the vGRF peaks reflecting limb loading [4, 6, 7]. Previous studies observed that elevated vGRF peak magnitude and variability, particularly during the weight-acceptance and propulsion phases of gait, are often reported in individuals with detrimental joint loading and gait asymmetry [6-9]. Thus, this suggests that the investigation of vGRF peak gait patterns could yield valuable information about differences in gait dynamics.Autoregressive (AR) modeling is a statistical technique that can evaluate waveform pattern variability to both quantitatively and visually denote changes in pattern dynamics [10, 11]. While a multitude of techniques can quantify gait pattern variability, AR modeling provides a way to graphically cluster times series with similar pattern dynamics together to help aid in the delineation between normal and abnormal movement patterns. The advantage of this approach is that the graphical component of this technique will allow researchers and clinicians to rapidly identify individuals with alternate and/or adverse gait patterns based on where they reside in the triangle. Often approaches compare the means between groups to assess gait differences; however, the problem is that comparing group means requires conducting studies that require large populations each time. Yet with AR modeling, while traditional studies with large populations will initially be needed to determine the regions on the AR triangle associated with specific conditions, once the regions have been defined researchers and clinicians can assess individuals gait based on their location in the triangle without the need to conduct a large scale study. Therefore, the AR modeling approach has the potential to serve as a valuable gait diagnostic tool. Thus, this study employed AR modeling to denote differences in symmetric and asymmetric walking pattern dynamics through both quantitative and graphical means as a first step to evaluate AR modeling’s capability as a gait diagnostic tool.Altered limb loading is often a characteristic of asymmetric gait and can be observed from the vGRF waveform peaks as they represent maximum loading on the limb [1, 3, 4, 7]. Furthermore, changes in vGRF peaks during walking have been associated with collagen degradation and synthesis in the knee; therefore, vGRF peak dynamics could potentially function as an important non-invasive measure of movement performance and joint health [12]. AR modeling is an ideal technique to quantify vGRF peak pattern dynamics because the AR model coefficients reveal the behavior and stationarity of the time series [10, 11]. Statistically, stationarity is used to determine if the mean, variance and autocorrelation structure of the waveform is constant over time; thus, in the context of gait analysis it can help evaluate the behavior of gait dynamics over time [10, 11]. Previous studies have successfully used AR modeling to assess differences in running dynamics in rearfoot runners [13], identify alterations in gait patterns in individuals with neurological deficits [14] and evaluate wheelchair propulsion asymmetry [15]. Morgan (2019) used second order AR modeling to identify differences in running dynamics between controls and post anterior cruciate ligament reconstruction (ACLR) individuals from vGRF peaks [16]. Their study showed that post ACLR individuals exhibited different between-limb running dynamics than controls based on their AR model coefficients which placed the groups in different regions of the stationarity triangle [16]. A substantial finding of this work was that significant differences in gait dynamics were detected using the AR model coefficients, yet no differences in vGRF peak magnitude were found between the two groups [16]. This suggests that AR modeling has the sensitivity to detect changes in pattern dynamics that may be difficult to quantify using traditional discrete metrics.The objective of this study was to investigate how AR modeling could be used to both quantitatively and visually identify differences in gait pattern dynamics during symmetric and asymmetric walking based on the graphical interpretation of the AR model analysis. Individuals walked at symmetric and two different asymmetric walking conditions with between-limb speed differences of 0.25m/s and 0.50 m/s to evaluate differences in gait dynamics from time series constructed from vGRF peaks. We hypothesized that the AR model coefficients will detect differences in gait patterns based on the vGRF peak patterns amongst the symmetric and asymmetric walking conditions better than an analysis of the vGRF peak magnitude means. We compared the AR model coefficients to the vGRF peak magnitude means because the vGRF peak means are often used to assess limb loading during gait and provide a standard metric to help understand, interpret and evaluate differences in gait dynamics. Overall employing AR modeling to differentiate between symmetric and asymmetric gait patterns could lead to this method being used more readily in rehabilitation to track the progression of changing gait dynamics.
Materials and methods
Instrumented gait analysis
Seventeen healthy participants (mean ± standard deviation; age: 20.8 ± 1.1 yrs; height: 1.68 ± 0.11 m; mass: 71.6 ± 10.9 kg; males:females: 10:7) performed a walking protocol. The participants ranged in age from 18 to 30 years old and provided written informed consent as required by the University of Connecticut institutional review board. All participants could not have had any previous knee surgery and were injury free for the past six months. Participants were asked to perform a walking protocol where walking speed and gait asymmetry were varied. First, participants performed a five-minute warm-up where they walked on the instrumented split-belt treadmill at a self-selected speed (Bertec Corporation, Columbus, Ohio). The participants’ self-selected speed was determined by starting the participants at 1.0 m/s and then increasing the speed 0.1 m/s until the participant provided verbal feedback that the speed was their preferred self-selected speed. Once acclimated, participants walked at 0.75 m/s, 1.0 m/s, and 1.5 m/s, respectively for one minute each. These were the symmetric walking conditions because both limbs moved at the same speed. Then the participants walked with a between-limb gait speed difference of 0.25 m/s where the right limb was held at 1.0 m/s and the left limb was set to 0.75 m/s. The participants walked at this speed for one minute and then they once again walked with both limbs at 0.75 m/s for one minute to de-adapt. The asymmetric gait was repeated but this time the right limb was held at 1.0 m/s and the left limb was set to 1.5 m/s. The participants walked at this speed for one minute and then once again de-adapted at 0.75 m/s for one minute. The pattern of asymmetric gait followed by de-adaptation period was repeated on the left limb where it was held constant and the other was set to 0.75 m/s and 1.5 m/s, respectively. The gait speed was increased and decreased between trials at 0.1 m/s and then once at the desired speed individuals walked at least 10 strides before data was collected to reduce the potential of collecting spurious data. While the aforementioned gait protocol described how the asymmetric walking conditions were interspersed with de-adaptation, symmetric walking; we did randomize the order of the asymmetric walking trials. This protocol was adopted from previous studies [5, 17–19].
Vertical ground reaction force peak extraction
Vertical GRF data were analyzed for each minute of the symmetric and asymmetric walking conditions. GRF data was collected from the instrumented treadmill at 1200 Hz and filtered at 35 Hz using a fourth-order Butterworth low-pass filter. Heel strike was denoted when 50N of force was detected from the vertical GRF. The vertical GRF waveform was normalized to body weight and the impact and second peaks were extracted from each step to construct the GRF peak patterns (Figs 1 and 2). The GRF peak extraction, the generation of the AR model and the stationarity analysis were performed using a custom MATLAB algorithm (MATLAB R2017a, The MathWorks, Inc., Natick, Massachusetts, USA).
Fig 1
(a) Normalized vertical ground reaction force (GRF) waveform data collected from one individual while walking at 1.0 m/s during a 15 second time interval. The GRF peaks are denoted by the pink and green markers. (b) Normalized vertical GRF peaks extracted from the vertical GRF waveform. (c) Time series generated from the extracted vertical GRF peaks.
Fig 2
(a) Normalized vertical ground reaction force (GRF) waveforms for three different walking conditions during a one minute interval. The GRF peaks are denoted by the pink and green markers. (b) Plots of the vertical GRF peaks extracted and transformed into time series waveforms that were fit with second order autoregressive (AR) models. (c) Stationarity triangle with the locations of the second order AR model coefficients for the symmetric and asymmetric walking trials.
(a) Normalized vertical ground reaction force (GRF) waveform data collected from one individual while walking at 1.0 m/s during a 15 second time interval. The GRF peaks are denoted by the pink and green markers. (b) Normalized vertical GRF peaks extracted from the vertical GRF waveform. (c) Time series generated from the extracted vertical GRF peaks.(a) Normalized vertical ground reaction force (GRF) waveforms for three different walking conditions during a one minute interval. The GRF peaks are denoted by the pink and green markers. (b) Plots of the vertical GRF peaks extracted and transformed into time series waveforms that were fit with second order autoregressive (AR) models. (c) Stationarity triangle with the locations of the second order AR model coefficients for the symmetric and asymmetric walking trials.
Autoregressive modeling and analysis
Time series were generated for each participant from the normalized vertical GRF peaks extracted from the normalized vertical GRF waveform. Second order AR, AR(2), models were fit to the vertical GRF peaks time series’ to evaluate the stationarity of the GRF peak pattern. The AR(2) models were fit to the data using an established procedure outlined by Box and Jenkins (1976) [10]. First, the mean was subtracted from the vertical GRF peak time series, which centers the data about zero and ensures that the AR model is fit to the data pattern. Then the autocorrelation (ACF) and partial autocorrelation (PACF) functions were used to determine the order of the AR model that best fit the vertical GRF peaks time series. The ACF and PACF plots indicate how well a time series is correlated with a delayed or lagged version of itself [10, 11]. The ACF and PACF plots supported that the AR(2) model was the best fit given that the ACF exhibited a slow decay while the PACF displayed a drop off after lag order 2. To further validate that an AR(2) was appropriate, the estimated model was plotted against the original time series (Fig 3a). The goodness of the AR(2) model fit was also evaluated by plotting the histogram of the residuals, which result from subtracting the estimated model time series from the original time series. The histogram plots of the residuals were normally distributed about zero confirming that the AR(2) model was the appropriate model for the vertical GRF peak time series’ [10, 11] (Fig 3b). The Anderson-Darling normality test was also conducted and indicated that the residuals were normally distributed. This model selection and validation criteria were applied to each GRF time series. Computing an R-squared value is not appropriate here because the time series model errors are not independent but serially correlated, therefore, we evaluated the residuals [10, 11].
Fig 3
(a) Comparison of original normalized ground reaction force (GRF) time series plotted against a second order autoregressive, AR(2), derived times series fit to the original time series. (b) Histogram of the residuals obtained by subtracting the original and fitted waveforms.
(a) Comparison of original normalized ground reaction force (GRF) time series plotted against a second order autoregressive, AR(2), derived times series fit to the original time series. (b) Histogram of the residuals obtained by subtracting the original and fitted waveforms.An AR model models time series as a function of the previous values of the same time series. Specifically, an AR(2) model indicates that the times series is modeled as a function of the two previous values that proceed the current value (1).Here ∅1 and ∅2 represent the AR model coefficients, y is the current time series value, y and y represent the time series values at the two proceeding time intervals t-1 and t-2, δ is a constant and ε is white noise. For an AR(2) model the stability for the time series can be determined from the model coefficients using the characteristic eq (2) and solving for the roots of the polynomial, m1 and m2
(3).In lieu of having to perform these calculations the stationarity triangle provides a graphical and visual means to evaluate the stationarity of the time series. To assess the time series stationarity, the AR(2) coefficients, AR1 (∅1) and AR2 (∅2), are plotted on the x and y axes respectively, and if the resulting location of the coefficients resides inside of the triangle the time series is stationary (Montgomery et al. 2015; Box and Jenkins 1976). However, if the resulting point lies outside of the triangle the time series is unstable [10, 11]. While all points that lie inside of the triangle represent stable time series’, a model is deemed as less stable as the point moves closer towards the edge of the triangle region. To quantify differences in model stability using the stationarity triangle, we computed the distance of the point from the centroid of the triangle (0, -1/3).
Statistical analysis
A one-way ANOVA and Tukey post-hoc analysis was also performed to compare the mean magnitude of the vertical GRF peaks and distance of the AR model points from the centroid of the triangle for the symmetric and asymmetric gait conditions. All statistical analyses were conducted in SPSS (Version 23, IBM, Amonk, NY, USA.).
Results
A comparison of the vertical GRF peak time series when both limbs were moving at 0.75m/s, 1.0 m/s and 1.5 m/s presented no differences in vertical GRF peak patterns (Fig 4a, Table 1). A comparison of the vertical GRF peak time series for when both limbs were moving at 1.0 m/s and the two asymmetric conditions determined that the individuals exhibited a significantly different vertical GRF peak pattern during the asymmetric walking condition than the symmetric walking condition (both limbs at 1.0 m/s) (p = 0.001) (Table 1). These differences were also observed on the stationarity triangle as the smaller between limb asymmetric walking conditions patterns were clustered towards the center, more stable region of the triangle (Fig 4b). Conversely, the larger between limb asymmetric walking patterns were clustered at the edge of the triangle, closer to the unstable region (Fig 4b). The larger asymmetric walking condition patterns was located a significant distance away from the smaller asymmetric walking condition patterns (p<0.01) (Fig 4b). The symmetric walking condition where both limbs were moving at 1.0 m/s was located in between the two asymmetric walking conditions (Fig 4b).
Fig 4
Stationarity triangles where the AR1, ∅1, and AR2, ∅2, serve as the x and y axes, respectively.
(a) Zoomed in plot of the autoregressive (AR) model coefficients on the stationarity triangle for the three symmetric conditions when both limbs are moving at the same speed—0.75 m/s (pink x), 1.0 m/s (blue circle) and 1.5 m/s (green triangle). (b) Zoomed in plot of the AR model coefficients on the stationarity triangle for symmetric and asymmetric walking conditions. The blue oval encircles the region for the AR coefficients that represent the individuals walking with both limbs (symmetric) at 1.0 m/s. The + represents the centroid of the triangle from which the AR distances were computed.
Table 1
Comparison of normalized vertical GRF (vGRF) peaks mean and distance at increasing walking speed and symmetric and asymmetric gait for each limb mean (standard deviation).
Comparison of Symmetric Gait
Right Limb
Both 0.75 m/s
Both 1.0 m/s
Both 1.5 m/s
P-Value
Mean Peak vGRF
1.0 (0.0)A
1.1 (0.1)A
1.2 (0.1)B
<0.01
AR Distance
0.6 (0.2)
0.7 (0.2)
0.7 (0.2)
0.23
Left Limb
Both 0.75 m/s
Both 1.0 m/s
Both 1.5 m/s
P-Value
Mean Peak vGRF
1.0 (0.0)A
1.0 (0.1)A
1.2 (0.1)B
<0.01
AR Distance
0.6 (0.3)
0.6 (0.2)
0.7 (0.2)
0.26
Comparison of Symmetric and Asymmetric Gait
Right Limb
Both 1.0 m/s
L 0.75 m/s—R 1.0 m/s
L 1.5 m/s—R 1.0 m/s
P-Value
Mean Peak vGRF
1.1 (0.1)
1.0 (0.0)
1.1 (0.1)
0.32
AR Distance
0.7 (0.2)A
0.6 (0.3)A
0.9 (0.2)B
<0.01
Left Limb
Both 1.0 m/s
L 1.0 m/s—R 0.75 m/s
L 1.0 m/s—R 1.5 m/s
P-Value
Mean Peak vGRF
1.0 (0.1)
1.0 (0.0)
1.0 (0.0)
0.95
AR Distance
0.6 (0.2)A,B
0.5 (0.2)A
0.8 (0.1)B
0.01
A,B,C represent means that do not share the same letter are significantly different (α = 0.05) across each row.
The p-values represent significant differences found when comparing values in the same row.
Stationarity triangles where the AR1, ∅1, and AR2, ∅2, serve as the x and y axes, respectively.
(a) Zoomed in plot of the autoregressive (AR) model coefficients on the stationarity triangle for the three symmetric conditions when both limbs are moving at the same speed—0.75 m/s (pink x), 1.0 m/s (blue circle) and 1.5 m/s (green triangle). (b) Zoomed in plot of the AR model coefficients on the stationarity triangle for symmetric and asymmetric walking conditions. The blue oval encircles the region for the AR coefficients that represent the individuals walking with both limbs (symmetric) at 1.0 m/s. The + represents the centroid of the triangle from which the AR distances were computed.A,B,C represent means that do not share the same letter are significantly different (α = 0.05) across each row.The p-values represent significant differences found when comparing values in the same row.Differences in mean vertical GRF peak magnitude were also found when comparing the three symmetric walking speeds—0.75 m/s, 1.0 m/s, 1.5 m/s—under equivalent limbs speed conditions, but no differences were found when comparing the mean vertical GRF peak magnitudes across the asymmetric walking conditions. Thus, differences in mean vertical GRF peak magnitude is more variable during symmetric walking than asymmetric walking.
Discussion
The objective of this study was to investigate how AR modeling both quantitatively and visually identified differences in gait pattern stability during symmetric and asymmetric walking. The results supported the hypothesis as the AR2 model coefficients were able to differentiate amongst the symmetric and two asymmetric walking conditions quantitatively and visually. The same level of differentiation was not achieved using the mean of the vGRF peaks. Moreover, the locations at which the different symmetric and asymmetric walking groups clustered provided additional insight about the stability of the gait patterns and individuals dynamic response to the asymmetric walking perturbations. The fact that the individuals resided in three different locations on the stationarity triangle suggests that these individuals adopted three different gait patterns in response to the symmetric and the two asymmetric walking conditions. While it may have been anticipated that the individuals would adopt three different vGRF peak patterns in response to the three different walking conditions, it is significant to note that the analysis of the vGRF peak magnitude means did not reveal any differences across the three walking conditions. This suggests that for this data, the AR modeling analysis displayed a greater sensitivity in delineating amongst the symmetric and asymmetric walking conditions than more traditional comparisons of mean peak vGRF data.The advantage of the AR modeling approach is that it provided both quantitative and visual measures to identify differences in gait patterns during the symmetric and asymmetric walking conditions. Visually, these metrics were based on their location in the stationarity triangle. The findings from this study indicated that the participants adopted different gait patterns during the asymmetric walking than the symmetric walking. While often the asymmetric condition placed the individuals at the edge of the triangle near the unstable region, for the 0.25 m/s between-limb difference some individuals adopted gait patterns that were more stable that the symmetric condition. This is possibly attributed to the fact in response to this 0.25 m/s between-limb perturbation, the individuals restrict and constrain their motion resulting in a more stable gait pattern denoted by their placement near the center of the triangle. Given that individuals gait is inherently variable, a restricted, constrained gait pattern would reside in the more stable region on the AR triangle than normal symmetric walking. However, in the case of the larger asymmetric condition where the between-limb difference was 0.50 m/s, this perturbation appeared to be more destabilizing thus placing individuals at the edge of the triangle near the unstable region, which corresponded with larger distance metrics due to the fact that the individuals resided further away from the center of the triangle. Thus, the combination of the AR coefficients and stationarity triangle provided a visual way to both group and differentiate individuals by their gait patterns and we were able to quantify these differences based on their distance from the center of the triangle.AR modeling has been successfully implemented to assess movement dynamics with those studies often using a first order AR model [13, 20]. In those studies, first order AR models were appropriate to capture differences in movement dynamics; however, a second order AR model was the optimal fit for this data. Differences in model selection could be attributed to the differences in the variables used to construct the time series data. Winter and Challis (2017) constructed their time series from vGRF peaks during running whereas this study used both the vGRF peaks during the weight-acceptance and propulsive phases of walking [13]. However, Kucynski (1999) used a second order AR model to assess postural stability from center of pressure displacement [21]. Here the second AR peak played a significant role in differentiating between the three walking conditions as there was vertical separation amongst the three groups with the asymmetric walking conditions flanking the symmetric condition. Significant differences between the asymmetric conditions were observed in the AR2 coefficient as smaller AR2 coefficients were measured for the asymmetric walking condition with the smaller between-limb speed difference. Interestingly, an increase in the magnitude of the AR2 coefficients corresponded with a shift towards the unstable region. The damped or diminished presence of the AR2 coefficient during the smaller between-limb speed difference asymmetric walking trials compared to the larger asymmetric between-limb speed difference could suggest that the emergence of strong AR2 coefficient contribution could represent a shift toward more unstable gait dynamics. While additional work is needed to substantiate this observation, the results of this study highlight the importance of model selection and support the use of second order AR model to delineate amongst symmetric and asymmetric gait patterns.A limitation of this work was that gait was analyzed during one-minute intervals. Traditional temporal analyses tend to require large time series of five minutes or more to evaluate temporal patterns [20, 22]. These alternate techniques are appropriate in their use of extended time series; however, AR modeling does not require such extensive time series for its analysis [11]. While tracking gait changes over larger time intervals could further delineate differences in gait patterns, we were able to successfully identify three different motor control strategies from the one-minute intervals. The ability to extract such information in a short time period suggests that this analysis technique could be used as a diagnostic tool to identify altered gait strategies and neuromuscular function.
Conclusion
Autoregressive modeling can be a valuable tool to aid in the differentiation of gait pattern dynamics and stability via both quantitative and visual means. The location of the AR model coefficients on the stationarity triangle provided insight about changes in gait pattern dynamics based on the vertical GRF peak patterns individuals adopted in response to the gait disturbances. The results indicated that individuals adopted different gait patterns during the symmetric and asymmetric walking conditions. Given this techniques ability to differentiate between altered gait patterns, it suggests that AR modeling could potentially serve as a valuable clinical tool to help monitor injury and rehabilitation progression.(XLSX)Click here for additional data file.10 Sep 2020PONE-D-20-21892Characterizing Gait Pattern Dynamics during Symmetric and Asymmetric Walking using Time Series AnalysisPLOS ONEDear Dr. Morgan,Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.Overall the reviewers expressed some enthusiasm for this manuscript, but highlighted several areas that will benefit from additional clarity. Please carefully address each comment in your revision.Please submit your revised manuscript by Oct 25 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocolsWe look forward to receiving your revised manuscript.Kind regards,Eric R. AnsonAcademic EditorPLOS ONEJournal Requirements:When submitting your revision, we need you to address these additional requirements.1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found athttps://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf andhttps://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf[Note: HTML markup is below. Please do not edit.]Reviewers' comments:Reviewer's Responses to QuestionsComments to the Author1. Is the manuscript technically sound, and do the data support the conclusions?The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.Reviewer #1: YesReviewer #2: Yes**********2. Has the statistical analysis been performed appropriately and rigorously?Reviewer #1: NoReviewer #2: Yes**********3. Have the authors made all data underlying the findings in their manuscript fully available?The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.Reviewer #1: YesReviewer #2: Yes**********4. Is the manuscript presented in an intelligible fashion and written in standard English?PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.Reviewer #1: YesReviewer #2: Yes**********5. Review Comments to the AuthorPlease use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)Reviewer #1: In this manuscript, the authors use vertical ground reaction forces to differentiate symmetric and asymmetric walking patterns. They induce asymmetric walking using a split-belt treadmill. Their autoregressive (AR) modeling revealed differences between the symmetric and asymmetric walking conditions.TITLEI suggest that the title could be made clearer with revision, as “time series analysis” is quite vague (i.e., many different variables could be subject to time series analysis when investigating gait dynamics).ABSTRACTLines 33 and 34: the authors suggest that new methods are needed to detect to detect changes in vertical GRFs during asymmetric gait, but also state that these changes are “commonly observed”. These statements seem to contradict one another.Line 42: I suggest using a colon rather than a dash to indicate the split-belt treadmill speeds.Line 44-45: this sentence requires revision for grammar and punctuation.INTRODUCTIONThe justification for this study is somewhat unclear to me. While I certainly agree with the authors that it is important to be able to quantify the asymmetry of gait in clinical populations. there are already many well-established ways to do this. It would be helpful if the authors could provide some perspective about why this particular approach may be better than existing approaches, or at least how this approach may provide unique information relative to others.The authors provide a broad introduction with regard to gait asymmetry, but it seems as though the primary interest is in leg loading and likely patients with knee injuries (knee injury is listed as specific exclusion criteria). If this is true, I suggest that they could provide a more focused introduction about how this technique could specifically be useful for certain populations.They authors make a couple of references to the ability of AR modeling to “visually” detect differences in gait parameters. Further clarification would be helpful, as it is not clear to me what this means or why it is useful to detect a result “visually” when the same result can be expressed quantitatively.METHODSLine 115 – how was self-selected speed determined?Lines 159-161 – was any quantitative analysis performed to determine that AR(2) was the best fit model or was this done by visual inspection? A quantitative analysis would provide much stronger support for this model selection.RESULTSWere the left and right legs subjected to separate analyses? This is appears to be the case in Table 1. If the authors wish to analyze the legs separately, I suggest a two-way (leg x condition) ANOVA.DISCUSSIONNo comments.Reviewer #2: 1. Abstracta. Primary analysis: autoregressive modeling (previously used for running but not walking).b. Hypothesize that autoregressive model coefficients better detect gait asymmetries than peak vGRF magnitude.c. Used second order AR modeld. The asymmetric condition notation is confusing and could be presented more clearly. Perhaps (Side 1: XXX & Side 2: XXX).2. Introductiona. Gave good justification for the importance of quantifying gait asymmetry.b. Gave compelling reasons for using AR and its success in previous studies.c. Objective: “investigate how AR modeling could be used to both quantitatively and visually identify differences in gait pattern dynamics during symmetric and asymmetric walking.”i. Can you please clarify if by “visually”, you are referring to graphical interpretations or clinical assessments?ii. Induced asymmetric walking with instrumented split belt treadmill.d. Hypothesis: ARM model coefficients will detect differences in giat patterns based on vGRF peak patterns than peak mean vGRF magnitudes.i. Did you consider comparing RMS differences of interlimb vGRF?ii. Are there other measures that may isolate differences in separate phases of the gait cycle? Although mean peak vGRF is often used, it’s easy to see that it may not be the most sensitive measure. It might be good to address why you chose to only consider the mean magnitude.3. Methodsa. Gait Analysisi. Were the trial orders randomized? If not, how do you mitigate potential learning effects?ii. Line 118: “walked at an asymmetric gait” is awkward.iii. From my understanding, individuals only have 10 strides in an asymmetric condition before data is collected. Is the goal to collect data in a transient or steady state condition? If steady-state is the goal, I would be hesitant to say this is steady-state.b. Peak vGRF Extractionc. Autoregressive Modeling & Analysisi. Modeling was described very clearly!d. Statistical Analysis4. Resultsa. Figure 4b: interesting that some people have “more stable” gait patterns in the 0.75 m/s Left condition than they are in the symmetric condition. Do you have an explanation for that?b. Lines 233 -235: are these interlimb differences, differences across conditions, or differences between your calculation methods. Would be helpful to explain.c. Table 1: Can you provide more explanation to make the information in this table more clear?i. Are these Mean and distance parameters form the AR models? Or are they differences in mead vGRF magnitude?ii. Which comparisons are the p values for? Is this symmetric vs both asymmetric?iii. I see the note about ABC for grouping variables that are not significantly different from one another. It might be more useful to just see the asymmetric conditions compared back to the symmetric.5. Discussiona. Lines 244 – 249 & figure 4b: I see pretty clear clusters for the left 1.5 and symmetric conditions, but the left 0.75 seems to be all over the place. How do you explain that?b. Discussion section would benefit from a more thorough comparison of the different symmetry conditions and model parameters. For example, discussing why some models were more or less stable. Did the AR modeling differentiate all asymmetric conditions from the symmetric? Was this method able to differentiate differences between the asymmetric conditions? Are there any clear metrics or measures that a clinical could take away from this when treating patients (i.e. if they are asymmetric, but “more stable” than most symmetric adults, how does that affect their treatments)?6. Conclusionsa. I agree with the overall conclusion and think this work supports it. However, it would be nice to have more clarification/interpretation of the results between numbers on a graph and the big picture (AR modeling is good) at the end.7. Figuresa. The figures appear blurry in the pdf but clear in the download. Be sure to confirm hat the clear ones are used in the final publication.b. Table 1: It’s best practice to keep significant figure or decimal places consistent in tables.**********6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.If you choose “no”, your identity will remain anonymous but your review may still be made public.Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.Reviewer #1: NoReviewer #2: No[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.30 Oct 2020Review: PONE-S-20-27379Title: "Characterizing Gait Pattern Dynamics during Symmetric and Asymmetric Walking using Time Series Analysis"We thank the Editor and Reviewers for the opportunity to revise the manuscript. The revised manuscript provides stronger justification of the metrics used for the analysis as well as greater clarity regarding the interpretation of the results. We believe these changes make for a clearer and more concise manuscript and we thank the Reviewers for taking the time to review the manuscript.Reviewer #1: In this manuscript, the authors use vertical ground reaction forces to differentiate symmetric and asymmetric walking patterns. They induce asymmetric walking using a split-belt treadmill. Their autoregressive (AR) modeling revealed differences between the symmetric and asymmetric walking conditions.TITLEI suggest that the title could be made clearer with revision, as “time series analysis” is quite vague (i.e., many different variables could be subject to time series analysis when investigating gait dynamics).Thank you for this suggestion. We have revised the title and substituted autoregressive modeling for time series analysis.ABSTRACTLines 33 and 34: the authors suggest that new methods are needed to detect to detect changes in vertical GRFs during asymmetric gait, but also state that these changes are “commonly observed”. These statements seem to contradict one another.Yes, thank you for identifying this. We have revised the sentence to address the contradictory nature of the sentence (Line 35).Line 42: I suggest using a colon rather than a dash to indicate the split-belt treadmill speeds.Thank you. The dashed lines have been removed (Line 42).Line 44-45: this sentence requires revision for grammar and punctuation.Agreed, the sentence was not clear, and we have modified it to provide greater clarity (Lines 44-45).INTRODUCTIONThe justification for this study is somewhat unclear to me. While I certainly agree with the authors that it is important to be able to quantify the asymmetry of gait in clinical populations. there are already many well-established ways to do this. It would be helpful if the authors could provide some perspective about why this particular approach may be better than existing approaches, or at least how this approach may provide unique information relative to others.The authors provide a broad introduction with regard to gait asymmetry, but it seems as though the primary interest is in leg loading and likely patients with knee injuries (knee injury is listed as specific exclusion criteria). If this is true, I suggest that they could provide a more focused introduction about how this technique could specifically be useful for certain populations.They authors make a couple of references to the ability of AR modeling to “visually” detect differences in gait parameters. Further clarification would be helpful, as it is not clear to me what this means or why it is useful to detect a result “visually” when the same result can be expressed quantitatively.Yes, the previous version of the manuscript did not provide a clear justification for the purpose and advantage of AR modeling. The introduction has been revised to address this oversight (Lines 75 – 86 and Line 108).METHODSLine 115 – how was self-selected speed determined?We have revised the manuscript to include greater detail about how we determined the participants self-selected speed (Lines 129-131).Lines 159-161 – was any quantitative analysis performed to determine that AR(2) was the best fit model or was this done by visual inspection? A quantitative analysis would provide much stronger support for this model selection.Yes, a quantitative analysis was performed to evaluate if the AR(2) model was the best model to fit to the data. The appropriateness of the model fit was discussed in lines 180-185 and Figure 3b. However, we also updated the manuscript to also include why computing the R-Squared is not appropriate for this analysis; however, the results of the Anderson-Darling normality test yielded a p-value, a quantitative metric, that allowed us to confirm that the residuals were normally distributed (Lines 185-187).RESULTSWere the left and right legs subjected to separate analyses? This is appears to be the case in Table 1. If the authors wish to analyze the legs separately, I suggest a two-way (leg x condition)ANOVA.We ran a two-way ANOVA that compared the leg x condition and it showed that the results were not altered based on the limb. It was consistent the AR distance metric was different between the symmetric and asymmetric condition for both the right and left limb, meaning the results were the same for both limbs.DISCUSSIONNo comments.Reviewer #2:1. Abstracta. Primary analysis: autoregressive modeling (previously used for running but not walking).b. Hypothesize that autoregressive model coefficients better detect gait asymmetries than peak vGRF magnitude.c. Used second order AR modeld. The asymmetric condition notation is confusing and could be presented more clearly. Perhaps (Side 1: XXX & Side 2: XXX).Yes, the asymmetric walking notation was confusing and has been corrected in the manuscript (Lines 42-43).2. Introductiona. Gave good justification for the importance of quantifying gait asymmetry.b. Gave compelling reasons for using AR and its success in previous studies.c. Objective: “investigate how AR modeling could be used to both quantitatively and visually identify differences in gait pattern dynamics during symmetric and asymmetric walking.”i. Can you please clarify if by “visually”, you are referring to graphical interpretations or clinical assessments?By visual, we were referring to the graphical interpretation of the results and revised the manuscript to clarify that (Line 108).ii. Induced asymmetric walking with instrumented split belt treadmill.d. Hypothesis: ARM model coefficients will detect differences in giat patterns based on vGRF peak patterns than peak mean vGRF magnitudes.i. Did you consider comparing RMS differences of interlimb vGRF?We did not consider comparing RMS differences of interlimb vGRF. We decided on comparing the AR modeling to the mean of the vGRF peak magnitude because that is readily used metric for analyzing gait. However, we acknowledge that RMS could have also been used and will consider using RMS for future studies.ii. Are there other measures that may isolate differences in separate phases of the gait cycle? Although mean peak vGRF is often used, it’s easy to see that it may not be the most sensitive measure. It might be good to address why you chose to only consider the mean magnitude.Yes, that is true. We should have discussed why we decided to use the mean peak vGRF and have included an explanation in the manuscript (Lines 113-115).3. Methodsa. Gait Analysisi. Were the trial orders randomized? If not, how do you mitigate potential learning effects?Yes, the trial orders were randomized. We did not mention that in the previous version of the manuscript, but the revised version of the manuscript does acknowledge that the trials were randomized (Lines 142-144).ii. Line 118: “walked at an asymmetric gait” is awkward.This sentence has been corrected to provide greater clarity (Line 133).iii. From my understanding, individuals only have 10 strides in an asymmetric condition before data is collected. Is the goal to collect data in a transient or steady state condition? If steady-state is the goal, I would be hesitant to say this is steady-state.Yes, we agree the individuals would not reach a steady-state condition after 10 strides and we did not want them to reach a steady-state. Our goal was to capture the dynamics of individuals during the various asymmetric and symmetric walking conditions. Capturing the data after 10 strides is a standard operating procedure in our laboratory, and in the case of this study eliminated any major stumbles that can occur due to changing the speed of the treadmill. But we agree that the participants would not have reached a steady-state after 10 strides and no it was not the goal that they reached steady state for this analysis.b. Peak vGRF Extractionc. Autoregressive Modeling & Analysisi. Modeling was described very clearly!d. Statistical Analysis4. Resultsa. Figure 4b: interesting that some people have “more stable” gait patterns in the 0.75 m/s Left condition than they are in the symmetric condition. Do you have an explanation for that?Yes, we believe that the Left: 0.75 m/s – Right 1.0 m/s asymmetric walking condition was more stable than the symmetric walking condition because during the smaller 0.25 m/s asymmetry the participants adopted a more conservative, restrictive gait pattern. They likely adopt this conservative gait pattern because they recognize that they are experiencing an unstable perturbation they try to minimize the instability by restricting their motion. However, during normal walking individuals there is an inherent level of gait variability because individuals gait needs to be “pliable” and adapt to various, unexpected perturbations. Thus, in order to be able to adapt to different gait patterns they cannot walk in a restrictive motion which likely contributes to greater variability in their pattern than the 0.25 m/s between-limb gait speed difference condition.b. Lines 233 -235: are these interlimb differences, differences across conditions, or differences between your calculation methods. Would be helpful to explain.Yes, that was not clear and we revised that section to better explain what the differences were comparing (Lines 255-259).c. Table 1: Can you provide more explanation to make the information in this table more clear?i. Are these Mean and distance parameters form the AR models? Or are they differences in mead vGRF magnitude?The mean was for the vertical GRF peak magnitudes and the distance was for the AR model. This was not clear and Table 1 has been updated to address this and the other concerns raised.ii. Which comparisons are the p values for? Is this symmetric vs both asymmetric?The p-values are for the data in the specific row so for the first set, we are comparing the means for the vertical GRF peak magnitude across the three symmetric conditions – 0.75 m/s, 1.0 m/s, and 1.5 m/s for the right leg. Thus, the p-value at the end of the row represents the comparison of the data in that row.iii. I see the note about ABC for grouping variables that are not significantly different from one another. It might be more useful to just see the asymmetric conditions compared back to the symmetric.I apologize if we are not understanding the question. Our intent was to show differences in the mean vertical GRF peak magnitudes across the different gait conditions as well as to show the differences in AR distance metric across the different gait conditions.5. Discussiona. Lines 244 – 249 & figure 4b: I see pretty clear clusters for the left 1.5 and symmetric conditions, but the left 0.75 seems to be all over the place. How do you explain that?This is a great question and we revised the manuscript and added a paragraph to the discussion to explain this (Lines 276-292).b. Discussion section would benefit from a more thorough comparison of the different symmetry conditions and model parameters. For example, discussing why some models were more or less stable. Did the AR modeling differentiate all asymmetric conditions from the symmetric? Was this method able to differentiate differences between the asymmetric conditions? Are there any clear metrics or measures that a clinical could take away from this when treating patients (i.e. if they are asymmetric, but “more stable” than most symmetric adults, how does that affect their treatments)?These are all great questions and we did our best to address them in the additional paragraphs added to the discussion.6. Conclusionsa. I agree with the overall conclusion and think this work supports it. However, it would be nice to have more clarification/interpretation of the results between numbers on a graph and the big picture (AR modeling is good) at the end.We worked to provide greater clarity between the quantitative and graphical measures in Lines 276-292 as well as Lines 325-326.7. Figuresa. The figures appear blurry in the pdf but clear in the download. Be sure to confirm hat the clear ones are used in the final publication.Thank you, we will work with the journal to make sure the best quality figures are used.b. Table 1: It’s best practice to keep significant figure or decimal places consistent in tables.Yes, you are correct, and Table 1 has been adjusted to correct this error.Submitted filename: Response to Reviewers _vFinal.docxClick here for additional data file.18 Nov 2020Characterizing Gait Pattern Dynamics during Symmetric and Asymmetric Walking using Autoregressive ModelingPONE-D-20-21892R1Dear Dr. Morgan,We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.Kind regards,Eric R. AnsonAcademic EditorPLOS ONEAdditional Editor Comments (optional):Reviewers' comments:Reviewer's Responses to QuestionsComments to the Author1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.Reviewer #1: All comments have been addressed**********2. Is the manuscript technically sound, and do the data support the conclusions?The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.Reviewer #1: Yes**********3. Has the statistical analysis been performed appropriately and rigorously?Reviewer #1: Yes**********4. Have the authors made all data underlying the findings in their manuscript fully available?The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.Reviewer #1: Yes**********5. Is the manuscript presented in an intelligible fashion and written in standard English?PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.Reviewer #1: Yes**********6. Review Comments to the AuthorPlease use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)Reviewer #1: The authors have addressed my prior comments. I have no further suggestions and thank the authors for sharing their nice work.**********7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.If you choose “no”, your identity will remain anonymous but your review may still be made public.Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.Reviewer #1: No23 Nov 2020PONE-D-20-21892R1Characterizing Gait Pattern Dynamics during Symmetric and Asymmetric Walking using Autoregressive ModelingDear Dr. Morgan:I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.If we can help with anything else, please email us at plosone@plos.org.Thank you for submitting your work to PLOS ONE and supporting open access.Kind regards,PLOS ONE Editorial Office Staffon behalf ofDr. Eric R. AnsonAcademic EditorPLOS ONE
Authors: Brian Pietrosimone; Matthew K Seeley; Christopher Johnston; Steven J Pfeiffer; Jeffery T Spang; J Troy Blackburn Journal: Med Sci Sports Exerc Date: 2019-02
Fig 1
Fig 2
Fig 3
Fig 4