Assessing Gait Adaptability in People With a Unilateral Amputation on an Instrumented Treadmill With a Projected Visual Context

Condition 1: anticipatory and reactive obstacle avoidance.

This condition comprised two 4-minute trials during which participants were confronted with a series of 12 visual obstacles each. Obstacles were projected onto the treadmill belt as a white patch of light moving toward the participant at a speed corresponding to the belt speed. The obstacle dimensions were set according to the width of the belt and the size of the participant's shoe. The starting position of the obstacle was such that it would arrive exactly beneath either the left foot or the right foot (6 each, in random order) if the participant continued walking in the way he or she had been walking, as indicated by the foot position determined online²⁰ and with the average length of the 8 preceding steps.

In the anticipatory obstacle avoidance trial, the 12 obstacles were presented at the moment of foot contact, 4 steps in advance of the anticipated contact (Fig. 1A), allowing for an anticipatory gait adjustment to avoid the obstacle. In the reactive obstacle avoidance trial, the 12 obstacles were presented under time pressure, at the moment of foot-off, 1 step in advance of the anticipated contact (Fig. 1B), requiring a reactive gait adjustment during the swing phase of the ongoing step to avoid the obstacle. Sufficient time was available between obstacles for participants to reach a steady-state walking pattern again. Anticipatory and reactive obstacle avoidance trials were presented in counterbalanced order, each followed by a 5-minute seated rest period.

Success rates (as percentages) in both obstacle avoidance trials were scored manually by one of the experimenters through visual inspection. To be classified as successful avoidance, both feet had to be positioned outside the area of the obstacle (ie, no overlap of shoe and obstacle). This visual inspection was deemed superior to using available COP data because COP data do not reveal the real margins of the foot and hence cannot reveal whether part of the foot area was within the area of the obstacle. Sagittal video recordings were made for verification of uncertain obstacle crossings.

Condition 2: walking on regular or irregular patterns of stepping targets.

In this condition, rectangular stepping targets (length: the participant's shoe length; width: half the belt width) were projected onto the treadmill belt for both the left foot and the right foot. Participants were instructed to walk on the presented step pattern, aligning their feet as much as possible in the anterior and posterior directions on the stepping targets. Approximately 4 or 5 stepping targets were visible at each moment in time (Fig. 1C). This condition consisted of 3 trials with different levels of regularity in the patterns of stepping targets (ie, 0%, 20%, and 30% variations). The 0% trial consisted of a regular pattern of stepping targets corresponding to a participant's mean gait pattern, as registered in the preceding 20 seconds of unconstrained treadmill walking.^17,18 In the irregular trials, a random variation of 20% or 30% of a participant's mean stride length was added to the aforementioned regular pattern of stepping targets. Each trial consisted of a sequence of 80 stepping targets. The 3 trials were offered in a counterbalanced order and were followed by a period of seated rest.

Stepping accuracy (in millimeters) was quantified in terms of the variation in foot placement relative to the stepping targets. The anterior-posterior COP position at mid-stance (ie, the COP position halfway between the subsequent instances of foot-off and initial contact of the contralateral foot) were compared with the anterior-posterior position of the center of the stepping targets; the former position was used as a reference for the foot placement location on the treadmill. Note that although the COP position at mid-stance is close to the center of the foot, it is not necessarily perfectly so; in the pilot study on people who were able bodied (n=12; mean age=28.6 years, SD=4.4), we found a systematic error between the true center of the foot and the COP position at midstance of ±1 cm (95% confidence interval), depending on the manner in which people walked. As a consequence, a systematic offset between the center of the stepping target and the COP position at mid-stance may be observed, even when the foot is perfectly aligned on the stepping target. To circumvent such bias, we used the standard deviations of the distances between the COP position at mid-stance and the center of the stepping target (instead of the mean) as a measure of the accuracy of foot placement. The smaller this standard deviation, the more accurate the foot placement positions are relative to the stepping targets.

Supplementary experiment: effect of walking speed on gait adaptability scores.

Walking speed may affect gait adaptability performance. In a supplementary experiment, we therefore assessed the effect of walking speed per se on the gait adaptability scores of participants in the CO group. These participants performed both conditions at their self-selected walking speed and at a slower speed (expected to represent the lower range of walking speeds observed in participants with an amputation), in randomized order, as further described in the eAppendix.

General group characteristics and self-selected treadmill walking speed.

The 3 groups were compared in terms of age, body weight, height, and self-selected treadmill walking speed by use of separate 1-way analyses of variance (ANOVAs) (factor: group; levels: CO, TT, and TF) followed by post hoc independent t tests for significant group effects. The TT group and the TF group were compared in terms of the side and the cause of amputation by use of chi-square tests and in terms of the time since amputation and the time using the current prosthesis by use of independent t tests.

Condition 1.

Nonparametric statistics were used to examine group effects on obstacle avoidance success rates. First, Wilcoxon signed rank tests were performed to compare anticipatory and reactive obstacle avoidance success rates separately for the 3 groups. Differences in success rates between groups were evaluated with Kruskal-Wallis tests; Mann-Whitney U tests were used for post hoc follow-up in case of significant group effects.

Condition 2.

Differences in stepping accuracy between groups were tested by use of a repeated-measures ANOVA with the between-subject factor group (3 levels: TF, TT, and CO) and the within-subject factor pattern regularity (3 levels: 0%, 20%, and 30%). Post hoc comparisons were performed with independent t tests (group effects) or paired-sample t tests (pattern regularity effects).

Gait adaptability performance in relation to clinical test scores.

Gait adaptability scores were related to 10MWT_comf, 10MWT_max, TUG, Emory-FAP, and ABC scores by use of Spearman (success rates) or Pearson (stepping accuracy) correlation coefficients. To examine the relationships between gait adaptability scores and self-reported incidence of falls and fear of falling, we stratified the groups into subgroups (“fallers”/“nonfallers” and fearful/nonfearful) and then used Mann-Whitney U tests to compare obstacle avoidance success rates for these subgroups. Likewise, to compare stepping accuracies for the stratified subgroups, we used repeated-measures ANOVAs for subgroup (2 levels: fallers versus nonfallers and fearful versus nonfearful) × pattern regularity (3 levels: 0%, 20%, and 30%).

All statistical tests were performed with SPSS version 17.0 (SPSS Inc, Chicago, Illinois). Level of significance (α) was set at .05 for testing main and interaction effects. For post hoc tests, α was corrected for multiple comparisons by dividing .05 (α) by the number of post hoc comparisons performed. In general, for post hoc tests among the 3 subgroups, α was .017.