Participants
Participants were people with iSCI that had been referred to the GRAIL (gait real-time analysis interactive lab) training by a rehabilitation physician to improve their gait capacity and dynamic balance. Inclusion criteria were: (1) a motor incomplete spinal cord injury with a traumatic or non-traumatic cause (American spinal injury association impairment scale (AIS) C or D), (2) 6 months post injury, (3) ability to walk in a self-paced mode on the GRAIL without using the handrails, and (4) age ≥ 18 years. Subjects were excluded if they had pre-injury impairments of the nervous system, or lower limbs, or any other impairment that might affect balance control. Healthy controls were included if they were 18 years or older without a history of neurological or musculoskeletal problems. The study was approved by the regional medical ethics committee of Arnhem-Nijmegen (2019–5255). All participants provided written informed consent under the Declaration of Helsinki.
Data collection
Participants were tested on an instrumented split-belt treadmill (GRAIL, Motek Medical BV, The Netherlands). Kinematic data were acquired using an eight-camera motion capture system (VICON, Oxford, United Kingdom). Reflective markers were placed on 19 anatomical landmarks: 7th cervical vertebra and left and right acromion process, humeral lateral epicondyle, ulnar styloid process, anterior superior iliac spine (ASIS), posterior superior iliac spine (PSIS), femoral lateral epicondyle, lateral malleolus, metatarsal II, and calcaneus. Marker data were sampled at 100 Hz.
Protocol
All participants performed a two-minute walk test (2MWT) at a self-paced speed on the treadmill. The participants with iSCI performed the 2MWT at the start of their first training session. The speed of the belt was adjusted in real-time to the anterior–posterior position and velocity of the pelvis to allow participants to walk at a self-selected walking speed (self-paced mode), which is a suitable alternative to fixed-speed treadmill walking in gait analysis [21, 22]. In the self-paced mode, walking on the front part of the treadmill results in acceleration proportional to the difference between the pelvis position and middle of the belt, and to the velocity of the pelvis. Likewise, walking on the back part of the treadmill results in deceleration. The participants were instructed to walk at a comfortable walking speed. The healthy controls performed one extra 2MWT at a fixed speed equal to 50% of their mean self-paced walking speed (preferred speed) to analyze the effects of walking speed on their ML foot placement strategy, and because this speed was presumed to be similar to the preferred walking speed of the participants with iSCI [23]. A fixed speed was selected because walking in the self-paced mode at 50% of the preferred speed is challenging, and previous research found no significant differences between self-paced and fixed-speed walking [21, 22]. Before the 2MWTs, participants performed one to four one-minute practice rounds to familiarize themselves with walking on the treadmill. To ensure safety, all participants wore a safety harness attached to a rail on the ceiling, without body weight support.
Data analysis
Data were processed using MATLAB (R2019b, MathWorks). The first 20 and last 5 seconds of each 2MWT were excluded from the analysis to remove the start and stop phases. Gaps in the ASIS and PSIS marker data were automatically filled using the rigid body method as previously described [24]. Cubic spline fill was used for the remaining markers when a gap was no more than 10 samples. Marker data were filtered with a 4th order zero-phase low-pass Butterworth filter with a cut-off frequency of 20 Hz.
Hip joint centers were estimated using the regression method reported by Dumas et al. [25]. Marker data and hip joint centers were used to estimate the COM location of nine segments (torso and head, upper leg, lower leg and foot, upper arm, forearm and hand) as described by Tisserand et al. [26]. The whole-body COM location was computed using a weighted sum of the segment COM locations. Gaps in the whole-body COM, resulting from gaps in the marker data, were filled using the pattern fill method as described by Camargo et al. [24]. The average location of the ASIS and PSIS markers was used as the donor pattern.
Marker data of the feet were used to detect gait instances [27]. Heel strike was defined as the instant at which the anterior–posterior velocity of the calcaneus marker reversed with respect to the walking direction. Toe-off was defined as the instant at which the velocity of the metatarsal II marker reversed to the positive walking direction. Step width was defined as the distance between the left and right calcaneus marker at the instant of midstance.
To study the foot placement strategy of a participant, linear regression was used to predict the ML foot placement (FP) based on ML COM position and velocity at heel strike [19, 28,29,30]. We used the following regression equation:
$$FP= {\beta }_{pos}\cdot COM+ {\beta }_{vel}\cdot \dot{COM}+ \varepsilon$$
in which \({\beta }_{pos}\) and \({\beta }_{vel}\) are the regression coefficients of the COM position and velocity, respectively, and \(\varepsilon\) the model error. Foot placement was defined as the demeaned ML distance between the left and right calcaneus markers at midstance. The COM position was defined with respect to the calcaneus marker of the stance foot at mid stance, and both predictors were demeaned.
Outcome measures
The accuracy of the foot placement strategy was evaluated by the root mean square error (RMSE) between the predicted and actual foot placements. The RMSE was selected as primary outcome measure and referred to as foot placement deviation.
To confirm adherence to the foot placement strategy, the goodness of the fit of the linear regression model was evaluated with the coefficient of determination (R2), here referred to as foot placement adherence. Substantial adherence to the foot placement strategy was considered when the coefficient of determination was larger than 0.26 [31]. In addition, the within-subject standard deviation (SD) of actual foot placement was determined, because foot placement adherence is influenced by the dispersion of the actual ML foot placement.
Step width was selected as a secondary outcome measure, because wider steps have previously been linked to instability during walking [32, 33] and a reduced foot placement strategy [34].
Statistical analysis
Participant characteristics of both groups were compared with independent t-tests for continuous variables and Chi-square tests for nominal variables. Foot placement deviation of people with iSCI was compared with values obtained from healthy controls at different walking speeds using independent t-tests, whereas a difference in foot placement deviation between different walking speeds in healthy controls was tested with a dependent t-test. Likewise, group differences in foot placement adherence, in the SD of actual foot placement, and in step width were tested with independent t-tests, whereas differences between different walking speeds within the healthy control group were tested with dependent t-tests. We performed the Student's independent t-test when the assumption of homogeneity of variance was met and the Welch's independent t-test when this assumption was not met (resulting in fractional degrees of freedom). When the assumption of normality was violated, non-parametric equivalent tests were performed. The level of significance (α) was adjusted for the number of tests performed (3) and set at 0.017.