Does Crouch Alter the Effects of Neuromuscular Impairments on Gait? A Simulation Study

Abstract

Cerebral palsy (CP) is a neurologic injury that impacts control of movement. Individuals with CP also often develop secondary impairments like weakness and contracture. Both altered motor control and secondary impairments influence how an individual walks after neurologic injury. However, understanding the complex interactions between and relative effects of these impairments makes analyzing and improving walking capacity in CP challenging. We used a sagittal-plane musculoskeletal model and neuromuscular control framework to simulate crouch and nondisabled gait. We perturbed each simulation by varying the number of synergies controlling each leg (altered control), and imposed weakness and contracture. A Bayesian Additive Regression Trees (BART) model was also used to parse the relative effects of each impairment on the muscle activations required for each gait pattern. By using these simulations to evaluate gait-pattern specific effects of neuromuscular impairments, we identified some advantages of crouch gait. For example, crouch tolerated 13% and 22% more plantarflexor weakness than nondisabled gait without and with altered control, respectively. Furthermore, BART demonstrated that plantarflexor weakness had twice the effect on total muscle activity required during nondisabled gait than crouch gait. However, crouch gait was also disadvantageous in the presence of vasti weakness: crouch gait increased the effects of vasti weakness on gait without and with altered control. These simulations highlight gait-pattern specific effects and interactions between neuromuscular impairments. Utilizing computational techniques to understand these effects can elicit advantages of gait deviations, providing insight into why individuals may select their gait pattern and possible interventions to improve energetics.

1. Introduction

Cerebral palsy (CP) is a motor disorder caused by a brain injury at or near the time of birth (Graham et al., 2016). This primary neurologic injury alters control (i.e., increased co-activation and reduced capacity to selectively activate individual muscles), resulting in less complex control strategies during walking for individuals with CP than nondisabled (ND) peers (Bekius et al., 2020; Steele et al., 2015). Less complex control has been shown to be associated with worse function and treatment outcomes (Cheung et al., 2012; Schwartz et al., 2016). Additionally, individuals with CP often develop secondary, progressive impairments like weakness and contracture (Gage et al., 2009; O’Dwyer et al., 1989). Interactions between neuromuscular impairments impart complex restrictions on gait that are difficult to elicit and understand experimentally (Kuska et al., 2022), inhibiting treatment efficacy.

In silico techniques, like modeling, simulation, and machine learning (ML), whose clinical translation is limited by inherent methodological assumptions, can improve foundational understanding and inform future experiments by offering a means to rapidly evaluate hypotheses (e.g., investigate interactions between neuromuscular impairments and parse their individual effects on gait (Halilaj et al., 2018; Shourijeh et al., 2020)). Prior simulation research primarily focused on identifying gait deviations caused by altered control (Mehrabi et al., 2019) or weakness and contracture (Fox et al., 2018a; Johnson et al., 2022; Ong et al., 2019; van der Krogt et al., 2012). The few studies that investigated the effects of altered control and muscle morphology highlight how the two combinatorially impose greater restrictions on gait than either alone (Falisse et al., 2020; Kuska et al., 2022). Investigating the interactions between neuromuscular impairments and their effects on gait could improve our understanding of the mechanisms inhibiting function in individuals with CP and bolster treatment efficacy.

The purpose of this study was to investigate the interactions between neuromuscular impairments and gait in CP. Specifically, we used modeling, simulation, and ML to examine how altered control, weakness, and contracture interact and impact simulated crouch gait. We then compared our findings to those from simulated ND gait to elicit how crouch gait can alter the effects of neuromuscular impairments, determining whether there are advantages to walking in crouch. We hypothesized that 1) similar to ND gait, altered control would exacerbate the effects of weakness and contracture and increase the muscle activations required to walk in crouch but, 2) when compared to ND gait, crouch gait would reduce the effects of neuromuscular impairments, making crouch potentially advantageous with altered control and muscle morphology.

2. Methods

To evaluate how altered control and neuromuscular impairments interact with crouch gait, we utilized a sagittal-plane musculoskeletal model (Kuska et al., 2022; Mehrabi et al., 2019) built in MapleSim (Maplesoft, Inc) (Figure 1). The model consisted of seven rigid body segments and nine kinematic degrees of freedom actuated by eight Hill-type musculotendinous units per leg: biarticular hamstring (HAM), gluteus maximus (GLU), iliopsoas (IP), rectus femoris (RF), vasti (VAS), gastrocnemius (GAS), soleus (SOL), and tibialis anterior (TA). Activation dynamics were modeled by muscle activation time constants based on Geyer (2010) and Mehrabi (2019). Ground contact was simulated by modeling ten continuous Coulomb friction contact spheres placed in-line and equidistantly along each foot (Brown and McPhee, 2016). Dynamic equations of motion were exported from MapleSim into a direct collocation (DC) optimal control framework in MATLAB (Mathworks, Inc). Within this framework, we used an implicit Euler collocation scheme with a 51-point temporal grid (Ackermann and van den Bogert, 2010) to optimize and simulate a half gait cycle; symmetry was assumed (Ankarali et al., 2015; Mehrabi et al., 2019). ADiGator (Weinstein and Rao, 2017) assisted with automatic differentiation and MATLAB’s interior-point optimizer (IPOPT) (Wächter and Biegler, 2006) solved each optimization. This model and framework were chosen because of their ability to rapidly elicit the effects of altered control (Kuska et al., 2022; Mehrabi et al., 2019).

Figure 1:

A sagittal-plane musculoskeletal model and neuromuscular simulation framework that tracked average nondisabled (ND) kinematics and moderate and severe crouch gait. The model contains nine degrees-of-freedom (pelvic tilt and translation, and right and left hip, knee, and ankle flexion) actuated by eight Hill-type musculotendinous units per leg. The objective function minimized deviations from tracked kinematics and the sum of muscle activations squared (a²). We perturbed each gait simulation with multi-modal neuromuscular impairments—altered control, weakness, and contracture—of varying severities. Altered control was simulated by reducing the number of fixed synergies controlling each leg, and weakness and contracture were simulated by reducing a muscle’s maximum isometric force ($F_o^m$) and tendon slack length ($L_s^t$), respectively. A Bayesian Additive Regression Trees (BART) model then predicted resultant a² from the simulated neuromuscular impairments for crouch and ND gait to evaluate the relative effects of each simulated neuromuscular impairment on the muscle activations required to maintain each gait pattern.

As direct collocation is sensitive to initial guesses, we analyzed variance in simulations from different initializations. We tested initial guess sensitivity with a null guess (i.e., tracked kinematics, muscle lengths estimated from tracked kinematics, and zeroed controls) and a hot start which utilized a baseline simulation (i.e., no neuromuscular impairments) as the initial guess. Analyses revealed that kinematic variance from the different initial guesses were within a single standard deviation of experimentally reported ND sagittal-plane waveforms (Fukuchi et al., 2018). Thus, all reported results were generated by simulations initialized with a hot start to reduce convergence time.

Tracking simulations (minimizing deviations from desired kinematics at all nine kinematic degrees of freedom) were selected for this investigation because of their simplicity, previous success with similar analyses (Kuska et al., 2022; Mehrabi et al., 2019; van der Krogt et al., 2012), and because we required a specific gait pattern to be maintained when perturbed. Tracking simulations without residual forces applied to the pelvis generated gait while minimizing muscle activations—the summation of all muscle activations squared (a²) (Ackermann and van den Bogert, 2010) —and an activation smoothing term:

$$J={\int }_{t=0}^{t=t_f}\left(w_1\sum _{j=1}^9{\left(\text{θ}-{\text{θ}}_{tracked}\right)}^2+w_2\sum _{m=1}^{16}{a}_m^2+w_3\sum _{m=1}^{16}{\stackrel{·}{a}}_m^2\right)dt$$ (1)

Heavy weighting was placed on the tracking term (w₁ = 5000) to force the model to remain in the desired gait pattern. The additional a² (w₂) and derivative (w₃) weightings were set at 35 and 0.05, respectively (Mehrabi et al., 2019). To analyze crouch gait, we tracked average moderate and severe crouch patterns previously identified by k-means clustering from a database of 2000 children with CP (Rozumalski and Schwartz, 2009). Additionally, average ND kinematics were tracked (Liu et al., 2008) for comparison. To minimize the effects of walking speed, both CP and ND gait patterns were fixed to the same gait speed, whereafter, corresponding cadence and step-length were selected based on non-dimensional values in CP and ND age-matched individuals (Brædvik et al., 2020).

A modular neuromuscular controller simulated altered control. To alter control, we varied the number of muscle synergies—grouped patterns of muscle weightings theorized to reflect modular control—controlling each leg (Mehrabi et al., 2019). During walking, fewer synergies are required to explain the muscle excitations of individuals with CP than their ND peers (Bekius et al., 2020; Steele et al., 2015). We used sets of eight, five, and three synergies to simulate each gait cycle. Eight synergies, or individual muscle control (IMC), is commonly used in simulation studies (Falisse et al., 2020; Rosenberg and Steele, 2017). Five and three synergies were selected to represent control of ND children and adults and children with CP, respectively (Bekius et al., 2020; Rozumalski et al., 2017; Shuman et al., 2019; Tang et al., 2015). Muscle excitations from each gait patterns’ IMC simulation were decomposed using non-negative matrix factorization (NNMF) (Lee and Seung, 1999) to determine the synergy weights—fixed ratios of muscle activations—for five and three synergies. Synergy weights were fixed, thereby constraining the controller to modulate only synergy activations. Synergy weights can vary when task demand is altered (McGowan et al., 2010); however, we chose to fix synergies, assuming gait-specific task demand did not change. Imposing neuromuscular impairments may change task demand, but our gait-specific sets of synergies represented the groups of muscles that would explain the greatest variance in simulated muscle activity and were the most robust to a broad range of neuromuscular impairments (Kuska et al., 2022).

We simulated two secondary, progressive impairments common in individuals with CP: weakness and contracture (Graham et al., 2016). Muscle weakness was simulated by reducing maximum isometric force ($F_o^m$) (Fox et al., 2018b; Ong et al., 2019; van der Krogt et al., 2012). Contracture was simulated by reducing tendon slack length ($L_s^t$) (Barrett and Lichtwark, 2010; Fox et al., 2018b; Steele and Lee, 2014) of the muscles commonly impacted in CP: HAM, GAS, SOL, and PFlex (GAS + SOL) (Barber et al., 2011; Handsfield et al., 2016).

Weakness and contracture were incrementally increased in severity from the original value, by 1% for weakness and 0.1% for contracture, until the simulation failed to generate each gait pattern. Weakness and contracture failure thresholds were defined when RMSE between the simulated and any tracked lower-extremity kinematic exceeded 2.5°, or the simulation did not converge after 2500 iterations; 2.5° reflects intra-gait cycle variance for children with CP (Tabard-Fougére et al., 2022). Weakness and contracture thresholds demonstrate how robust a gait pattern is to weakness and contracture and, by varying the number of synergies, we elicit how control complexity alters those thresholds (Kuska et al., 2022).

To evaluate the relative effect of each neuromuscular impairment on a², we developed a Bayesian Additive Regression Trees (BART) model (Chipman et al., 2012). BART is a sum-of-trees machine learning algorithm that uses Bayesian probability to prevent overfitting. We chose BART because of its predictive capabilities and ability to parse complex, nonlinear relationships like those between altered control, weakness, contracture, and gait (Dorie et al., 2019; Hill, 2011; Kapelner and Bleich, 2015). We built a BART model for each gait pattern where simulated neuromuscular impairment severities were input and BART predicted a². Hyperparameters for each BART model were tuned using 10-fold cross-validation (Steele and Schwartz, 2022) and we used pseudo-R² as measure of model quality. Accumulated local effect (ALE) plots were used to determine the relative effects of each simulated neuromuscular impairment on a² while accounting for all other variables in the model. BART model development and ALE plot generation were performed in R (R Core Team, 2022) using the ‘bartMachineCV’ (Kapelner and Bleich, 2016) and ‘ALEPlot’ packages (Apley and Zhu, 2020).

3. Results

Simulations closely tracked ND and crouch kinematics but tracking errors and muscle activations squared (a²) increased with crouch severity and fewer synergies (Figure 2). Average lower-extremity RMSE for ND, and moderate, and severe crouch gait with IMC were 0.15°, 0.28°, and 0.57°, respectively. With three synergy-control, average lower-extremity RMSE values increased to 0.32° for ND gait, 0.51° for moderate crouch, and 1.17° for severe crouch. With IMC, a² was 20% and 190% greater than ND gait, for moderate and severe crouch gait, respectively. Decreasing control complexity from IMC to three synergies, increased a² by 38% for ND gait, 32% for moderate crouch, and 19% for severe crouch.

Figure 2:

(Top) Simulated kinematics (right leg hip, knee, and ankle) for nondisabled gait and moderate and severe crouch with individual muscle control (IMC), five-synergy control, and three-synergy control. (Bottom) Root-mean squared errors (RMSE) between tracked kinematics and simulated gait at each lower-extremity joint and muscle activations squared (a²) required to generate each gait pattern.

With IMC, ND and crouch gait tolerated nearly 100% reductions (i.e., the muscle could be removed) in GLU, HAM, and GAS strength. ND gait tolerated the greatest reductions in VAS and HFL strength, but the least in PFlex strength. Moderate and severe crouch tolerated the most weakness in PFlex, and the least in the HFL and TA. As fewer synergies were used by the controller, muscles weakness thresholds decreased (Figure 3 & 4). Average weakness threshold decreased from 92% (IMC) to 71% for ND gait when constrained to three-synergy control. Similarly, average weakness threshold for moderate and severe crouch decreased from 96% and 90% with IMC, to 68% and 49% with three-synergy control, respectively.

Figure 3:

Weakness thresholds for nondisabled, moderate crouch, and severe crouch with individual muscle control (IMC), and five- and three-synergy control. A greater threshold indicates that the simulations were less sensitive, i.e., more robust, to weakness of that muscle. (Top) Hip and knee weakness thresholds. (Bottom) Ankle muscle weakness thresholds.

Figure 4:

Change in weakness threshold when changing from IMC to five- and three-synergy control for nondisabled and moderate and severe crouch gait. Negative values indicate decreases in weakness threshold when decreasing control complexity (# of synergies) from IMC to five- or three-synergy control. Larger values indicate greater effects of altered control. (Top) Change in hip and knee muscle weakness thresholds. (Bottom) Change in ankle muscle weakness thresholds.

Crouch gait impacted the effect of altered control on weakness thresholds (Figure 4). Changing from IMC to five-synergy control only affected VAS weakness thresholds in moderate and severe crouch. Decreasing control complexity from IMC to five-synergy control did not affect HAM weakness thresholds for moderate or severe crouch, but did for ND gait. Weakness thresholds (except GAS) decreased with altered control, but the magnitude was gait-pattern specific. For example, when changing from IMC to three-synergy control, VAS and HAM weakness thresholds decreased more for crouch than ND gait, but SOL weakness thresholds decreased more for ND than crouch gait.

With IMC, HAM contracture threshold was greater for ND gait than crouch, but GAS contracture thresholds were greater for crouch than ND gait (Figure 5). Contracture thresholds decreased for all gait patterns when decreasing control complexity to five- and three-synergy control, but the magnitude was gait-pattern specific. For example, decreasing control complexity from IMC to three-synergy control had a larger effect on GAS contracture threshold in severe crouch. Contrastingly, the effect of altered control on HAM contracture threshold was smaller with more severe crouch, e.g., when changing from IMC to three-synergy control, HAM contracture threshold decreased more for ND than crouch gait.

Figure 5:

(Top) Contracture thresholds for nondisabled, and moderate and severe crouch with individual muscle control (IMC), and five- and three-synergy control. A greater contracture threshold indicates that the simulations were less sensitive, i.e. more robust, to contracture of that muscle. (Bottom) Change in contracture threshold when changing from IMC to five- and three-synergy control for nondisabled and moderate and severe crouch gait. Negative values indicate decreases in contracture threshold when decreasing control complexity (# of synergies) from IMC to five- or three-synergy control. Larger values indicate greater effects of altered control.

Muscle activations required to generate gait (a²) were accurately predicted by BART from the simulated neuromuscular impairments for all gait patterns (pseudo-R² > 0.97). Hamstring contracture had the largest effect on a² during ND gait and vasti weakness had the largest effect on a² during crouch (Figure 6). Plantarflexor contracture (except GAS) had a larger effect on a² during crouch when compared to ND gait and plantarflexor weakness had a larger effect on a² during ND gait when compared to crouch gait. The effect of control complexity (i.e., number of synergies) on a² was larger with less severe crouch.

Figure 6:

Accumulated local effect (ALE) plots (left) generated from Bayesian Additive Regression Trees (BART) models that predicted muscle activations squared required to generate each gait pattern (a₂) from simulated neuromuscular impairments. Relative, net effects of each neuromuscular impairment parsed by BART and estimated via ALE plots (right). Larger net effects indicate neuromuscular impairments that had a larger effect on muscle activations during simulated nondisabled and crouch gait.

4. Discussion

Our gait simulations indicated that altered control made it more difficult to walk by 1) increasing the muscle activations required to generate gait and 2) decreasing the amount of weakness and contracture that could be tolerated. This supports our first hypothesis that altered control would make it more difficult to remain in crouch with and without secondary neuromuscular impairments. However, our second hypothesis that simulated crouch gait, when compared to simulated ND gait, would lessen the effect of neuromuscular impairments was only partially supported. We found that the effects of, and interactions between, neuromuscular impairments were gait-pattern specific, i.e., the effects of neuromuscular impairments could be lessened and exacerbated by crouch gait, highlighting both advantages and disadvantages of walking in crouch. Thus, our results highlight 1) complex interactions between neuromuscular impairments and gait, emphasizing the importance of including altered control in future analyses of populations with neuromuscular impairments, and 2) the restricting effects of altered control on gait make it a promising, but not often addressed, target for interventions.

Simulated severe crouch necessitated greater a² (Figure 1): a common proxy for energetic cost (Ackermann and van den Bogert, 2010; Miller et al., 2012). Thus, our findings align with the greater amounts of energy children with CP consume during walking when compared to ND peers (Bell and Davies, 2010; Campbell and Ball, 1978). The cause of increased energy consumption in CP still remains unclear (Ries and Schwartz, 2018), but our results suggest that crouch severity would correlate with elevated energetics. This contrasts with reported weak correlations between crouch severity and elevated energy consumption (Steele et al., 2017) but aligns with more recent studies that found that altered gait kinematics, specifically landing in crouch, were associated with elevated walking in CP (Gill et al., 2022; Schwartz et al., 2022).

Fewer synergies (i.e., less complex control) made walking more difficult, creating larger deviations from tracked kinematics and increasing a² (Figure 1). Our prior studies found similar trends in simulated ND gait (Kuska et al., 2022; Mehrabi et al., 2019), but we found a decrease in a² for simulated severe crouch when changing from five- to three-synergy control (Figure 1). This was the result of our multi-term objective function: our three-synergy severe crouch simulation incurred a small decrease in a² for a large increase in ankle tracking error. However, the overall objective function value increased with decreasing control complexity, highlighting that less complex control increased the cost of walking (Falisse et al., 2020). Our findings, along with experimental improvements in walking energy with interventions that improved motor control (Conner et al., 2021, 2020), support that altered control contributes to elevated energy in CP (Gill et al., 2022).

Modeling and simulation enabled us to conduct these point-of-failure analyses, allowing us to highlight interactions between neuromuscular impairments and elicit advantages of different gait patterns. For example, without altered control, simulated crouch gait was less robust to vasti weakness than simulated ND gait (Figure 3). Thus, walking in crouch with vasti weakness would be disadvantageous, i.e., “like driving with your parking brake on” (Hicks et al., 2008; Steele et al., 2013). Then when reducing control complexity, we found that crouch gait increased the impact of altered control on vasti weakness, exacerbating the disadvantages of walking in crouch with knee extensor weakness (Figure 4). Conversely, crouch gait was more robust to plantarflexor weakness— the most prevalent secondary neuromuscular impairment in CP often targeted by interventions (Handsfield et al., 2016)—and decreased the effect of altered control on plantarflexor weakness. These results highlight advantages of walking in crouch in the presence of neuromuscular impairments (Figure 3–5) and why individuals may select to walk in crouch. Similar methods extended to patient-specific analyses could elicit potential advantages and causes of gait deviations for a given individual.

Using BART with synthetic data from simulation let us compare the relative effects of neuromuscular impairments on gait. BART results emphasized how the effects of neuromuscular impairments on gait are gait-pattern specific. Additionally, BART highlighted how vasti weakness in crouch, with or without altered control, is a primary driver of muscle activations (Figure 6). Targeting vasti weakness may be an effective intervention to reduce muscle demand, and perhaps energetics, for individuals with CP who walk in crouch. Conversely, GAS contracture had a relatively small effect on muscle activations during crouch gait, highlighting why corrective surgeries, like tendon lengthenings that impact the gastrocnemius, may not improve energetics (Dahlbäck and Norlin, 1985). Surprisingly, altered control had a small relative effect on a² compared to weakness and contracture. This contrasts the minimal changes in energetics in CP post-strength training and corrective surgeries (Dahlbäck and Norlin, 1985; Damiano and Abel, 1998; Scholtes et al., 2006), as well as retrospective causal analyses that identified selective and dynamic motor control as clinical measures associated with energetics in CP (Gill et al., 2022). The small net effect of control relative to weakness and contracture, and this results deviation from prior literature (Gill et al., 2022), may be due to the range of values evaluated in this study. Weakness and contracture were simulated to failure, while only three levels of altered control were evaluated. Additionally, our model and its control are relatively simple: we do not model feedback control (e.g., proprioception), out-of-sagittal-plane motion, nor balance, and do not incorporate things like uncertainty or noise. All of which are present during motion and many of which are altered during disabled gait (Bell et al., 2002; Hsue et al., 2009; Tracy et al., 2019; Wingert et al., 2009). Thus, increased model and simulation complexity would likely increase control complexity and its effect size on energetics.

Interestingly, our simulations of ND and crouch gait did not reflect experimental observations of changes in control common in CP. Children with CP who walk in crouch typically demonstrate fewer synergies and a larger total variance accounted for by one synergy (tVAF₁) than ND peers (Bekius et al., 2020; Kim et al., 2018; Shuman et al., 2019; Steele et al., 2015). In contrast, from our simulations, the tVAF₁ for ND gait and moderate and severe crouch were 0.63, 0.58, and 0.51 respectively. Thus, our simulations of crouch required more complex activation strategies when compared to ND gait. These findings could stem from several causes: 1) our model is tuned to generate ND gait (Kuska et al., 2022; Mehrabi et al., 2019), 2) disabled individuals use different control strategies than their ND peers (Song and Geyer, 2018), 3) the altered demand of crouch dynamics are not enough to simplify control (Spomer et al., 2022), and 4) constraints imposed by weakness and contracture force individuals with CP to operate in different, lower-dimensional control spaces. These all warrant further investigation but emphasize the importance of incorporating control complexity into simulations. Using muscle synergies to constrain simulations provides a method for 1) improving lower-dimensional control representation (Berniker et al., 2009; Michaud et al., 2020) typical in individuals with CP (Bekius et al., 2020), 2) personalizing control based on an individual’s EMG patterns (Ferrante et al., 2016), and 3) understanding the implications of altered control (Falisse et al., 2020; Kuska et al., 2022; Mehrabi et al., 2019).

Our simulations were simplified representations of ND and crouch gait—e.g., only sagittal plane motion was simulated lacking out-of-sagittal-plane motion common in CP (Bell et al., 2002)—and used simplified muscle paths which likely influenced results (Kuska et al., 2022). Additionally, we assumed symmetry because the added complexity of asymmetric gait was not necessary to investigate how crouch impacted the effect of altered control. We also only simulated average gait patterns meaning our results may not extend to patient-specific analyses. However, our results were gait-pattern specific and aligned with literature (e.g., landing in crouch increases cost of transport (Schwartz et al., 2022)), highlighting our methods potential applicability to patient-specific analyses. Additionally, our objective function minimized tracking errors and an energetic approximation—a₂ defined as the summation of all muscle activations squared (Ackermann and van den Bogert, 2010; Miller et al., n.d.). However, what individuals with CP (and ND individuals) optimize during gait is still unknow and our results highlight advantages and possible rationale for why individuals with CP choose to walk in crouch.

By utilizing modeling and simulation with machine learning this study investigated interactions between altered control, weakness, and contracture and their complex effects on gait. We found that the interactions between, and the effects of, neuromuscular impairments were gait-pattern specific. These results emphasize potential advantages of walking in gait patterns like crouch, but are dependent on an individual and their unique neuromuscular impairments. Additionally, the inclusion of altered control enabled us to discover possibly new and even greater advantages of crouch gait (e.g., simulated crouch was more robust than simulated ND gait to hamstring contracture only with altered control). Thus, the inclusion of altered control can create quantitative and even qualitative differences in results. Future studies investigating populations with neurologic injuries should include altered control and consider utilizing in silico techniques to elicit advantages and possible rationale for gait deviations and to identify primary mechanisms affecting gait.