A convenience sample of 25 participants with isolated ankle pathology took part in the experiment. All participants were recruited through the orthopaedic department of a local hospital and the electronic mailing list of employees and students at Université Laval. The ethics committee of the Centres intégrés universitaires de santé et de services sociaux de la Capitale Nationale (CIUSSS-CN) (rehabilitation and social integration section) and of the Centre hospitalier universitaire de Québec granted the ethical approval. All participants provided their written informed consent.
Inclusion criteria were: 1) to be aged over 18; 2) to be living with one of these two isolated ankle pathologies: ankle fracture or ankle osteoarthrosis for at least 3 months; and 3) to be able to walk for at least 20 min without a walking aid. Exclusion criteria were: 1) to have a history of chronic pain or presence of pain unrelated to the ankle condition and 2) to have a neurological disorder that could affect task performance.
Experiment
Baseline characteristics such as age, injury type and anthropometric characteristics were first collected on day one. Then, maximal ankle dorsiflexion strength (using a dynamometer) [21] and maximal weight-bearing dorsiflexion range of motion [22] were measured and participants filled three validated self-reported questionnaires: the Lower Extremity Functional Scale (LEFS), the Tegner Activity Scale (current level of activity) and the Pain Interference Subscale of the Brief Pain Inventory (BPI). The LEFS is a 20-item questionnaire assessing the impairment of the lower-extremity musculoskeletal system in everyday activities with a score ranging from 0 (minimal impairment) to 80 (maximal impairment) [23]. It has been validated in individuals with ankle pathologies [23]. The Tegner Activity Scale evaluates work and sports activities using a score ranging from 0 (maximal disability) to 10 (full participation in sports) [24, 25]. The BPI includes 11 items scored on a numeric 0 to 10 scale where 0 = no interference and 10 = total interference. The mean score on the 11 items was reported [24, 25]. The BPI was also completed on Day 2, to characterize the stability of pain experienced on Day 1 and Day 2 [26].
Thereafter, all participants performed a gait adaptation task on two consecutive days, to assess both motor acquisition (Day 1) and retention (Day 2). On each day, they walked on a treadmill at 1 m/s [27, 28] while wearing the robotized ankle foot orthosis (rAFO) on their injured side [29]. The rAFO is a custom-designed electrohydraulic ankle-foot orthosis that can produce several types of force fields during walking [29]. It has been used in several studies evaluating force-field adaptation paradigm during walking [16, 30]. Detailed information on the rAFO can be found in Noel et al. [29] During the gait task, the level of ankle pain was rated verbally every minute on a 0-10 numerical rating scale (0 = no pain and 10 = worst imaginable pain) and the mean level of pain during the task was reported. A familiarization period (5 min) preceded data collection.
Motor learning test
Details of the experimental procedures have been previously described by Bouffard et al. [10, 15, 16] For 5 min, participants walked on the treadmill with the rAFO while no force field was applied to quantify baseline gait. For the next 5 min, the rAFO applied a force field resisting ankle dorsiflexion during the midswing phase of each stride (adaptation phase). The torque magnitude of the perturbation was constant during the entire adaptation phase. Participants were not told about the exact time at which the force field would be turned on. They were instructed to “overcome the perturbation in order to walk as normally as possible.” For the last 5 min, participants walked again without the force field to recover their normal walking pattern (washout). The rAFO actively cancelled torques produced by its mechanical components to minimise interference with gait pattern during baseline and washout periods (i.e. force cancellation mode) [31].
During the experiment, the ankle angle in the sagittal plane was recorded by an optical encoder located on the rAFO (encoder resolution is < 1°) [29]. The torque applied by the rAFO was measured by a load cell and the heel contact (custom made foot switch placed under the shoe) was recorded to calculate stride cycle duration. The tibialis anterior (TA) muscle activity was recorded on the trained lower limb using surface electromyography (EMG). The electrodes were placed just below the calf band of the rAFO, at 1/3 on the line between the tip of the fibula and the tip of the medial malleolus as recommended by the Surface Electromyography for the Non-Invasive of Muscles (SENIAM) guidelines [32].
Variables of interest
Participants’ global performance (i.e., how much the participant can cancel the effect of the force field) was characterized using the Mean absolute ankle angle error. This measure represents the difference in the relative ankle angle curves between the baseline and the adaptation phases. The motor strategy used by the participant to overcome the force field during the adaptation phase was characterized by 1) the Relative timing of error (a measure of the temporal center of error distribution relative to the peak force command) [10] and by 2) the Tibialis anterior EMG activity change before and after the perturbation (peak force command [PFC]) (i.e., TAratioBeforePFC and TAratioAfterPFC, indicators of feedforward and feedback control, respectively) [15].
Both the Mean absolute ankle angle error and the Relative timing of error were derived from generated error curves. Using the heel contact and rAFO control signals, data were separated into individual gait cycles and tagged as perturbed or non-perturbed strides. Ankle angle data were low-pass filtered with a second-order zero-lag Butterworth filter at 15 Hz. With the use of the ankle angle obtained from the optical encoder, the swing phase was identified as described by Bouffard et al. [16] and time normalized to 1000 points. A baseline swing phase ankle angle template was then created by averaging 45 of the last 50 baseline strides for each day (after removing the 5 most different strides from the mean). The ankle angle error was then calculated by subtracting point-by-point the baseline template values from each stride of the adaptation phase. The absolute value of ankle error of all 1000 swing phase points was averaged to define the Mean absolute error. An increased Mean absolute error represents a lower motor learning performance. In addition, changes leading to smaller (i.e., earlier) Relative timing of error during the adaptation phase represent switching to a more anticipatory strategy, while larger (i.e., later) values represent a more reactive strategy.
As for the Tibialis anterior EMG activity gains, EMG data were digitally filtered with a second-order zero-lag butterworth filter (bandpass 20–450 Hz) and rectified, and the envelope was extracted using a nine-point moving average [33]. As EMG activity precedes movement onset, the time window used for EMG analysis was extended by 30% of the identified swing phase, starting earlier to include the onset of TA stance-to-swing burst.
To quantify changes in TA activity during adaptation, an EMG gain was calculated, consisting of a point-by-point ratio between the TA activity of adaptation divided by baseline (TAratio) (see Fig. 1 for an example). EMG gains were then linearized using a log2 transformation. Mean gains before (TAratioBeforePFC) and after (TAratioAfterPFC) PFC were computed. For more details on data analysis, see Bouffard et al. [10] All data were analysed using custom-made software written in MATLAB (The MathWorks Inc., Natick, USA).
Statistical analysis
Participants who experienced pain constantly during the adaptation phase on both Day 1 and 2 were assigned to the Pain group (minimum pain at each time point ≥1/10). Participants who did not experience pain during the adaptation phase on both days were assigned to the No Pain group. Participants who had intermittent pain (e.g., ≥1/10 on Day 1 and < 1/10 on Day 2) were excluded from statistical analyses. Personal and clinical characteristics were compared between the Pain and No Pain groups using Mann Whiney tests and χ2 (e.g., age, gender, anthropometric characteristics, functional performance, strength, range of motion and pain during the task). The stability between the perceived levels of pain during the task on Day 1 and Day 2 was also evaluated by comparing participant’s scores between the days for the Pain group, using an Intraclass Coefficient Correlation (ICC; Two-way mixed effects) and a paired t-test [34]. The number (%) of participants excluded from analyses was reported as an indicator of protocol feasibility.
For the second aim, data from the Pain and No Pain groups were compared using a three-way non-parametric ANOVA for repeated measures (NparLD; Time [within subject]: Early vs. Late; Day [within subject]: Day 1 vs. Day 2 and Group [between subjects]: No Pain vs. Pain) on the following dependent variables: Mean absolute error, Relative timing of error and TAratios during the adaptation period. Time was characterized as Early adaptation (mean of strides 2–11 of the adaptation phase) and Late adaptation (mean of strides 151–200 of the adaptation phase). NparLD analyses are particularly relevant for small sample sizes and do not require normality of the data [35]. Effect sizes were reported as relative treatment effect (RTE). RTE is used to compare causal effect of a treatment on outcome; the distribution of the two groups is compared based on mean ranks and can thus be related to each other (≥.71 or ≤ .29: high effect; ≥.64 or ≤ .36: medium effect; ≥.56 or ≤ .44: low effect) [36].
Statistical analyses were conducted using the nparLD and AOV packages of the R software, respectively (version R.2.7.2.; R Foundation for Statistical Computing, Vienna, Austria). Mann Whiney test and χ2 were conducted in IBM SPSS Statistics (IBM SPSS Statistics 26, IBM Corp., NY, USA). Results are presented as means ± standard errors of the mean (SEM). Considering the exploratory design of this study and the statistical power limitation due to the small sample size, we decided to not apply correction for multiple comparisons for post hoc analyses. Level of statistical significance was set at p < 0.05.