Persons post-stroke can walk with more symmetric step lengths, but this does not change the cost of transport
All participants walked with asymmetric step lengths during preferred walking (Figure 1B, left) and successfully adjusted their step lengths during the symmetric stepping condition (Figure 1B, right) to reduce step length asymmetry (t(8)=3.99, p<0.01; Figure 1C). When comparing step lengths across leg and conditions, we expectedly observed a significant main effect of leg (F(1,8)=18.264, p=0.003) and a significant interaction (F(1,8)=33.72, p<0.001). Post-hoc analyses revealed that there was a significant increase in step length in the shorter limb (p=0.02) and a significant decrease in step length in the longer limb (p=0.04) from the preferred to symmetric stepping conditions. We replicated prior findings16,21 showing that improving step length symmetry with visual feedback had no significant effect on cost of transport (t(8)=0.92, p=0.38; Figure 1D). Furthermore, when comparing the preferred walking trials with feedback off versus feedback on, the presence of visual feedback did not affect step length asymmetry (t(8)=0.07, p=0.95) or metabolic cost (t(8)=0.49, p=0.64). Importantly, we also did not observe any between-participant correlation between step length symmetry and increased cost of transport in either condition (preferred walking, r = -0.22, n=9, p = 0.53; symmetric walking, r = -0.32, n=9, p=0.36).
Persons post-stroke exhibit marked IA even when walking with symmetric step lengths
A conceptual illustration of how we expected IA may differ between healthy symmetric walking and symmetric stepping after stroke is shown in Figure 2A. We hypothesized that healthy walking consists of symmetric step lengths and similar contributions of each segment to step lengths bilaterally, resulting in small IA (Figure 2A, left). On the contrary, we expected that symmetric stepping after stroke consists of symmetric step lengths but asymmetric segment contributions, resulting in high IA (Figure 2A, right).
We show the limb segment orientations during representative steps of symmetric stepping for each participant (Figure 2B). We did not observe a significant reduction in IA during symmetric stepping as compared to preferred walking (t(8)=2.12, p=0.066; Figure 2C). While it is possible that there is a trend toward a mild decrease in IA with symmetric step lengths, IA during symmetric stepping remained markedly increased when compared to healthy symmetric gait (for reference, data from eight healthy adults (age: 26±5 years) walking at 1.25 m/s are shown in Figure 2C). IA during preferred walking correlated strongly with IA during symmetric stepping (r=0.98, n=9, p<0.01; Figure 2D, left) and, qualitatively, the data fell near the unity line (Figure 2D, left), suggesting that persons post-stroke showed similar IA during preferred walking and symmetric stepping. IA was significantly associated with cost of transport during preferred walking (r=0.74, n=9, p=0.02) and symmetric stepping (r=0.82, n=9, p<0.01; Figure 2D, right), suggesting that kinematic asymmetries are correlated with cost of transport regardless of step length asymmetry.
We next considered that IA could remain similar across conditions while individual segment asymmetries could be reorganized. We did not find this to be the case. We compared the individual segment asymmetries (e.g., ) across segments and between conditions. ANOVA revealed a significant main effect of segment (F(2.312,18.497)=4.87, p=0.017). Post hoc analyses revealed that segment asymmetry was significantly larger in pelvic rotation (d+e) than both leading (f; p=0.022) and trailing pelvis translation (c; p=0.023). We did not observe a significant main effect of condition (F(1,8)=4.49, p=0.067; Figure 3A) or segment x condition interaction (F(2.03,16.25)=0.6, p=0.49). Figures 3B and 3C show how the segment asymmetries contribute to IA for each participant during each condition. When we compared segment asymmetries after ordering them by which contributed most-to-least strongly to IA (during preferred walking) between conditions, we also did not observe a significant main effect of condition (F(1,8)=4.49, p=0.067; Figure 3D) or segment x condition interaction (F(2.5,19.98)=2.153, p=0.134). As expected, we observed a significant main effect of segment (F(1.5,12.06)=28.38, p<0.001).
Asymmetries in AP GRFs, ML GRFs, and vertical COM velocities observed during preferred walking persist during symmetric stepping
We then aimed to identify the features of these asymmetric walking patterns that may influence the elevated cost of transport regardless of step length asymmetry. We investigated whether these features were similar in both preferred walking and symmetric stepping, or whether the costs of transport were similarly high in these conditions but resulted from different underlying mechanics. Asymmetric kinematics at heel-strike should result in asymmetric mechanical work done on the COM by each leg, and previous studies demonstrated that mechanical work done on the COM is related to cost of transport in healthy adults26,30. Furthermore, prior studies identified periods of the gait cycle where excessive positive work is often observed post-stroke, contributing to an elevated mechanical energetic cost7,8,31.
We investigated GRF and COM velocity profiles between legs and conditions, as these contribute to the work done over the gait cycle. ANOVA revealed a main effect of leg on the AP GRF peak (Figures 4A and 4B, top; F(1,8)=10.29, p=0.01), ML GRF peak (Figures 4A and 4B, middle; F(1,8)=7.55, p=0.03), AP COM velocity peak (Figures 4C and 4D, top; F(1,8)=8.53, p=0.02), and vertical COM velocity peak (Figures 4C and 4D, bottom; F(1,8)=6.63, p=0.03). Post hoc analyses revealed that the AP GRF peak was significantly larger in the nonparetic leg than the paretic leg (p=0.01), the ML GRF peak was significantly larger in the paretic leg than the nonparetic leg (p=0.03), the AP COM velocity peak was significantly larger during nonparetic late stance as compared to paretic late stance (p=0.02), and the vertical COM velocity peak was significantly larger during paretic late stance as compared to nonparetic late stance (p=0.03). There were no significant effects of leg on the vertical GRF peak (F(1,8)=0.43, p=0.53) or ML COM velocity peak (F(1,8)=2.80, p=0.13). We did not observe significant effects of condition on GRF or COM velocity variables (all p>0.17) or leg x condition interactions (all p>0.31).
The nonparetic leg does more positive work than the paretic leg during preferred walking and symmetric stepping
We next investigated the work done on the COM by each leg across conditions. We first calculated COM power for each leg during preferred walking and symmetric stepping (Figure 5A). We calculated COM work by integrating COM power over each of the four time periods described in the methods (Figure 5B and 5C). ANOVA revealed a significant main effect of leg on positive work done 1) by the paretic leg during the first period (step-to-step transition, nonparetic leg trailing) vs. the nonparetic leg during the third period (step-to-step transition, paretic leg trailing; F(1,8)=10.96, p=0.01), and 2) by the paretic leg during the second period (paretic single support) vs. the nonparetic leg during the fourth period (nonparetic single support; F(1,8)=11.85, p<0.01). Post hoc analyses revealed that the nonparetic leg did significantly more positive work during the third period than the paretic leg did during the first period (p=0.01). The nonparetic leg also did significantly more positive work during the fourth period than the paretic leg did during the second period (p<0.01).
ANOVA also revealed a significant main effect of leg on negative work done 1) by the paretic leg during the first period (step-to-step transition, nonparetic leg trailing) vs. the nonparetic leg during the third period (step-to-step transition, paretic leg trailing; F(1,8)=6.35, p=0.04), and 2) by the paretic leg during the third period vs. the nonparetic leg during the first period (F(1,8)=7.06, p=0.03). Post hoc analyses revealed that the paretic leg did significantly more negative work during the first period than the nonparetic leg did during the third period (p=0.04). However, the nonparetic leg did significantly more negative work during the first period than the paretic leg did during the third period (p=0.03). Note on Figure 5A that the first transition period begins prior to paretic heel-strike at approximately 95% of the prior gait cycle.
We did not observe a significant main effect of condition on work done over any of the time periods (all p>0.13). We did observe a significant leg x condition interaction for the positive work done during the fourth period (nonparetic single support; F(1,8)=7.43, p=0.03); however, post hoc analyses did not reach statistical significance.
A separate ANOVA revealed a significant main effect of leg on positive (but not negative; F(1,8)=0.03, p=0.87) work done across all time periods (F(1,8)=7.25, p=0.03). We did not observe a significant main effect of condition on positive or negative work done across all time periods (both p>0.57) nor did we observe a significant leg x condition interaction on positive or negative work done across all periods (both p>0.10).
Less positive work done by the paretic leg is correlated with higher cost of transport and slower walking
We then assessed whether the positive and negative work done by each leg across the gait cycle were correlated with cost of transport, gait speed, or IA during preferred walking and symmetric stepping. Positive paretic work was significantly correlated with decreased cost of transport during both conditions (preferred walking: r=-0.84, p<0.01; symmetric stepping: r=-0.80, p<0.01; Figure 6A, left); positive nonparetic work was not (both p>0.63, Figure 6A, right). Positive paretic work was also significantly correlated with increased walking speed (preferred walking: r=0.90, p<0.01; symmetric stepping: r=0.87, p<0.01; Figure 6B, left) whereas positive nonparetic work was not (both p>0.66, Figure 6B, right). Finally, positive paretic work was negatively correlated with decreased IA during preferred walking and symmetric stepping, though these trends did not reach statistical significance (preferred walking: r=-0.62, p=0.07; symmetric stepping: r=-0.65, p=0.06; Figure 6C, left). Positive nonparetic work was not significantly correlated with IA during either condition (both p>0.50; Figure 6C, right). We did not observe significant correlations between negative paretic or nonparetic work and cost of transport, walking speed, or IA during either condition (all p>0.10).