Participants
We included in- and outpatients from the Swiss Children’s Rehab, University Children’s Hospital Zurich, Affoltern am Albis, Switzerland. In line with recommendations for clinical utility studies, which propose 8 to 25 participants [13], we aimed to include 15 to 20 participants. Inclusion criteria were neuro-orthopedic disorders, taller than 125 cm, able to understand simple instructions, able to walk ten meters with or without walking aid, younger than 18 years (a minimum age was not defined), and informed consent. Exclusion criteria were unconsolidated fractures or bone fragility of the lower extremities, skin lesions in the harness’ area which could not be protected, unstable hip, knee, and/or ankle joints, reduced head control or inability to maintain an upright position, inability to communicate discomfort or pain, surgery of the lower extremities in the last three months, recently implanted baclofen pump, implanted pacemakers, passive knee extension deficit > 30°, and self-selected walking speed in Andago > 3.2 km/h. According to the manufacturer, participants should not exceed a bodyweight of 135 kg and a body height of 2 m [14].
Patients were characterized by gender, age, height and –weight, diagnosis and severity, more affected side, and use of orthoses and other walking aids. Further, in line with the International Classification of Functioning, Disability and Health—Children and Youth (ICF-CY) version, we assessed on the body function level, leg muscle strength with the Manual Muscle Test (MMT), and selectivity using the Selective Control Assessment of the Lower Extremity (SCALE).
The MMT is a clinical standard to assess the strength of muscles (or muscle groups) against gravity and manual resistance [15]. The score ranges between zero (‘no contraction palpable’) and five (‘motion in the full range of motion against gravity and maximal resistance’). In this study, bilateral hip flexors and extensors as well as hip internal and external rotators, knee flexors and extensors, and ankle plantar- and dorsal-flexors were tested. We presented a sum score describing overall leg muscle strength, where the maximal score could vary between 0 and 80.
The SCALE is a valid and reliable clinical assessment tool used to evaluate selective voluntary motor control in children with cerebral palsy (CP) [16]. It tests the hip and knee flexion–extension, ankle dorsiflexion-plantarflexion, subtalar inversion-eversion, and toe flexion–extension. The score varies between zero (‘unable’) and two (‘normal’) and depends on the active range of motion and the occurrence of involuntary movements in other joints. The sum score varies between 0 and 20.
On the ICF-CY activity capacity level, patients performed the 10-m walk test (10MWT) at self-selected speed. The 10MWT is a valid and reliable tool to assess the walking speed over a short distance [17, 18]. Participants received standardized instructions to walk at their self-selected speed (with their normal walking aid and orthoses if used in daily life) along a 14-m walkway. The 10 m in the middle were stopped with a stopwatch (i.e., steady-state, so without acceleration and deceleration). Participants performed the 10MWT twice, and the average time of the two trials was reported.
On the ICF-CY activity performance level, we scored the patient’s walking performance in daily life with the Functional Mobility Scale (FMS) and the Gillette Functional Assessment Questionnaire Walking Scale (GFAQ). These are valid and reliable assessments in young patients with neuro-orthopedic disorders [19, 20]. The FMS scores the habitual mobility strategies considering the assistive devices the child uses for distances of 5, 50, and 500 m. The performance for each distance is rated on a 6-point ordinal level varying from 1 (‘uses wheelchair’) to 6 (’independent on all surfaces’). The GFAQ describes with a value between 1 (‘cannot take any steps at all’) and 10 (‘walks, runs, and climbs on level and uneven terrain and does stairs without difficulty or assistance. Is typically able to keep up with peers’) the daily functional mobility of the child.
Finally, we assessed the mobility and cognition subscales of the Functional Independence Measure for Children (WeeFIM) [21]. The WeeFIM contains 18 items covering the subscales self-care, mobility, and cognition, which can be assessed by observing a child’s daily life performance and score this according to criterion standards. For this study, we focused on the mobility subscale with 5 items (3 locomotion and 2 transfer items) and the cognition subscale with 5 items (2 communication and 3 social and cognitive items). Each item is scored from 1 (total assistance) to 7 points (complete independence). The WeeFIM is routinely assessed by trained and certified nurses in our center.
Andago robotic device
Andago (V2.0) is a patient-following robotic BWU system for over-ground gait rehabilitation with few spatial limitations (Fig. 1a). Patients are secured with a harness connected to a dynamic BWU system ensuring the safe practice of walking and balance exercises. The Class IIa device weighs 185 kg and measures 107 cm (front-back) by 195 cm (height) by 85 cm (width; inner width is 67 cm), which makes it slender enough to pass through standard doors. Between the harness and the device, sensors are mounted that register the force and direction in which the patient is walking (Fig. 1a). The sensors, in combination with the motorized wheels, allow the robotic system to follow the patient. The dynamic bodyweight support system can be set from 0 to 55 kg by the therapist while the patient is standing upright. It remains constant during the entire gait cycle unless the therapist decides to increase or decrease the support by a few kilograms, which can be done without stopping. However, a stop is needed to readjust the dynamic unloading range. The system protects against and registers potential falls. The robot has two detachable handrails, and sensors integrated in the front of the device to detect collisions, and a patient lift for the transfer from and to the wheelchair (Fig. 1a).
Andago adapts its speed to the patient who can accelerate, decelerate, stop, turn, and even walk backward at any time during the therapy. The system is limited to a maximal speed of 3.2 km/h. It has three different control modes: (i) ‘patient-following mode’: the device follows the movements of the patient in any direction, (ii) ‘straight-line mode’: walking is limited to a straight line (backward or forward), and (iii) ‘manual mode’: the therapist can steer the device with a remote controller. In this modus, the sensors are deactivated, and the system does not follow the patient’s movements [22].
Study protocol
Children and adolescents with various diagnoses and gait impairments participated in three sessions (lasting 60, 45, and 90 min, respectively) in our center (Fig. 2). The first two appointments were provided, amongst others, to become familiarized with walking in the Andago and on the treadmill.
In session 1, we assessed the patients’ characteristics and various functional tests (MMT, SCALE, etc.) during the first 25 min. Patients then walked in the Andago (20 min) and on the treadmill (10 min). When training in the Andago, participants practiced the following tasks: 2 × 30 m walking in a straight line (once in the straight line and once in the patient-following mode), 2 × 20 m backward walking in the straight-line mode, 1 × 30 m walking in the straight-line mode forward without using the bars of the Andago, and finally, passing 5 times through a door (width 110 cm) in the patient-following mode. On the treadmill, children walked forward (60 steps) while holding the parallel bars with both hands, followed by 60 steps with one hand on a parallel bar and finally walking forward without holding the parallel bars. The session was closed by a short questionnaire where the children responded to acceptability questions about the system (5 min). In session 2, we investigated the accuracy of the Andago unloading system (10 min), again followed by 30 min of training in the Andago and on the treadmill with the same protocol as in session 1. The acceptability questions of session 1 were repeated at the end of this session (5 min). In session 3, participants were equipped with surface electromyography (sEMG) electrodes, accelerometers, and goniometers (25 min), after which they walked 15 min in Andago and 10 min on the treadmill. This was followed by 20 min of walking over-ground in the Andago at various BWU conditions. At the onset and end of this session, participants performed the 10MWT at their maximum speed twice to control for factors such as fatigue.
Furthermore, all therapists performing therapeutic training sessions with the Andago filled out the System Usability Scale (SUS) questionnaire, which reports on the practicability of the system from the therapists’ perspective [23].
Practicability
We investigated the following practicability issues:
-
Time needed to get the patient in and out of Andago. During sessions 1 to 3, we stopped the time needed to position each participant in the Andago harness, and, at the end of the session, we recorded the time to get the patient out of the device. We calculated the average time for each participant over the three sessions.
-
Accuracy of the bodyweight support system. As the bodyweight of some of the children is relatively low, and the Andago was not primarily developed for children, we considered it important to determine the accuracy of the weight support system. We tested three conditions, 25%, 50%, and 75% of BWU. Participants were standing in the Andago in the dynamic BWU range on a digital weighing scale. The Andago weight support was set at 25%, 50%, or 75% (randomized order) of their bodyweight as accurately as possible. We compared the weight based on the Andago settings with the weight displayed on the digital scale [kg]. Besides the absolute differences, we calculated also relative deviations [(Weighing scale—Andago BWU setting)/Andago BWU setting × 100%].
-
During the first two sessions, we recorded the number of device deficiencies, the number of prevented falls, the number of safety stops (caused by too rapid movements of the participant), and the number of bumps when passing 5 times through the door in patient-following mode.
Acceptability
We quantified the young patients’ acceptability of Andago by asking (1) whether they felt safer walking in Andago or walking normally, (2) whether they would prefer to train walking in Andago or on the treadmill, and (3) whether they would like to walk again in Andago. Furthermore, we asked how cool it was to train in Andago (Likert scale 0, i.e., not cool at all, to 10, i.e., very cool). We asked these questions at the end of sessions 1 and 2 (Fig. 2).
Therapists who applied the Andago during regular clinical therapies in our center filled out the SUS. It comprises ten questions that are scored on a 5-point Likert-scale (from strongly agree to strongly disagree) and reflects the therapist’s impression of the practicability of the device. We calculated the SUS scores for each therapist and the average SUS score over all therapists. We transferred the average SUS score also to a percentile score, where a value of 68 or higher is considered practical. Therapists further responded to the general question ‘How did you find Andago in general?’ and were asked to provide specific comments on their experiences with the Andago.
Appropriateness
We investigated appropriateness in session 3 (Fig. 2). As we investigated differences in stride-to-stride variability of the stride time and lower limb inter-joint coordination between the Andago and treadmill trials, we needed to account for the influence of walking speed [24]. As it is not possible to walk in Andago at a pre-defined speed, we determined the self-selected walking speed during the Andago condition first and set the treadmill speed in the subsequent trial accordingly. This excluded the possibility to randomize the order of the conditions. After that, participants performed the other Andago trials with different levels of BWU in a randomized order. To investigate whether fatigue might have affected the results, participants performed the 10MWT at maximal speed (i.e., the instruction included now ‘… as fast as possible’) twice at onset and twice at the end of this session.
Technical equipment
Participants walked on the treadmill of our Lokomat® Pro V6 system. The BWU system shares similar characteristics as the Andago weight support system. Prior to the trials, patients were equipped with DTS 3D accelerometers (Noraxon U.S.A. Inc., Scottsdale/USA), self-adhesive hydrogel electrodes (Covidien, Mansfield/USA) to record sEMG, and DTS 2D flexible electrical goniometers (Biometrics Ltd, Newport, UK). Signals were transmitted with a sampling rate of 1500 Hz from the sensors to the MyoResearchXp software (Noraxon U.S.A. Inc., Scottsdale/USA) by the Wireless TELEmyo DTS (Noraxon U.S.A. Inc., Scottsdale/USA). Accelerometers were mounted on the patient’s left and right shoe over the Achilles-tendon region to detect initial contact and toe-off, which was needed to determine gait phases and cycles. EMG electrodes were positioned in line with the SENIAM guidelines [25] on the M. Gluteus Medius, M. Biceps Femoris, M. Gastrocnemius Medialis, M. Vastus Medialis, and M. Tibialis Anterior of the more affected leg. Still, we present only data of the antigravity muscles. The location was marked with a skin pencil, the skin was shaved, lightly rubbed with an abrasive gel to decrease skin impedance, and the electrodes were fixed. The participants were asked to contract each muscle to test the quality of the signal [25]. The three goniometers were mounted so that the middle of the goniometer was placed over the center of rotation of the hip, knee, and ankle joint of the more affected leg and calibrated.
We further used a Logitech HD Webcam C270 (Logitech Europe S.A., Nijmegen, Netherlands) to record the participants’ feet from behind during walking. The signal was transmitted with a sampling rate of 30 Hz via USB cable to the MyoResearchXp software. These synchronized video recordings of the feet enabled us to determine initial contact and toe-off for each Andago and treadmill trial in case of unclear accelerometer signals.
Differences between Andago and treadmill walking
The participants performed all measurements wearing their regular orthoses. The Andago trials were performed in the straight-line mode. We first measured the walking pattern in the Andago at the reference level of BWU (BWUref), i.e., the minimal amount of BWU needed for the participant to keep a physiological knee position (or as physiological as possible) during stance. This is also the BWU level used during therapeutic training sessions. We determined BWUref as follows. Participants were secured in the Andago harness and stood on the digital scale. Weight was unloaded until they could hold a physiological knee position.
Participants walked twice along a 17 m-long walkway at their self-selected pace. The time needed to cover ten meters was manually stopped with a stopwatch to calculate the mean speed. Between the trials, participants had one minute of rest. We asked the participants to walk without holding the handrails if this was physically possible.
Then, participants performed the treadmill trial (also at BWUref). We increased the treadmill speed until we reached the mean walking speed of the previous Andago BWUref trial. After six steps, we recorded 15 strides for data analysis. The treadmill was stopped, the participant had one minute of rest, and then the procedure was repeated.
Differences between walking at various bodyweight unloading levels
To investigate the influence of different levels of unloading, participants walked in the Andago at BWUref plus 15% additional BWU (BWUref + 15%) and BWUref plus 30% additional BWU (BWUref + 30%). The order of these conditions was randomized (www.randomizer.org/). The procedure was otherwise similar to the first Andago BWUref trial.
Data analysis
For each condition, we analyzed 30 ‘steady-state’ strides for each participant. A stride was defined as the time between two consecutive initial contacts of the measured leg and identified using the accelerometer signals and video recordings. We always analyzed an equal number of strides, as, for example, the stride-to-stride variability decreases with increasing numbers of strides [26]. Although a previous study in healthy adults recommended that 50 gait cycles are needed to evaluate the variability of spatiotemporal gait parameters reliably [26], we considered this too much for most of our young patients, and we agreed on 30 gait cycles, as this might be sufficient to achieve a more or less normal data distribution. Data were further processed in MATLAB R2016A.
Andago vs. treadmill walking
Stride-to-stride variability: We quantified the stride-to-stride variability by calculating the coefficients of variation (CV [%] = 100% x SD/mean) over the 30 stride times for each participant per condition.
Inter-joint coordination: To quantify the variability in inter-joint coordination, we calculated a normalized inter-joint coordination coefficient. For the joint signals, each of the 30 strides was normalized to 1500 data points. In line with Chiu et al., [27] the angular positions (x) were then normalized between 1 and −1 with the following formula to minimize differences in movement amplitude between conditions:
$${x}_{i, norm}= \frac{2 \times ({x}_{i}-\mathrm{min}({x}_{i}))}{\mathrm{max}\left({x}_{i}\right)-\mathrm{min}({x}_{i})}-1$$
Then, we averaged the normalized kinematics over the 30 strides to achieve an averaged stride for each joint. To obtain the inter-joint coordination variability, the magnitudes of the vectors pointing from the average stride to each of the other 30 strides on the knee-hip and knee-ankle angular position plots were calculated for each data point. The variability is represented by the mean of the magnitude of all vectors [normalized root mean square (RMS)].
Different BWU levels in Andago: For the EMG signals, we determined the stance-phase (time from initial contact to consecutive toe-off of the measured leg) for each of the 30 strides. Each stance phase was normalized to 900 data points. EMG data were rectified, filtered with a 20 Hz Butterworth high-pass filter [28], and smoothened by RMS with a time window of 50 ms. We calculated the mean EMG amplitude for each stride [μV] and then calculated the average amplitude over the 30 stance phases for each participant per condition.
Possible confounding of holding parallel bars
As some patients required the use of the parallel bars of the Andago or the treadmill, and different levels of support could influence specific parameters, we recorded the amount of pressure applied by the patients on the parallel bars. We mounted per handrail three 40 cm long (Andago) or 30 cm long (treadmill) FSR 408 pressure sensor stripes (Interlink Electronics, Westlake Village, USA). To ensure that the force applied by the patients was distributed entirely over the measurement area, we fixated long stripes over the active force-detecting part of the sensor stripes (Fig. 1b and c). We covered this with tape and inserted a tube of slightly larger diameter around the bar to focus the force distribution on the measurement area and to decouple the vertical support force (we were interested in) from circular grasping forces. The signal was transmitted via cable from the sensors to an Arduino MEGA 2560 microcontroller board and then to MATLAB R2016A (MathWorks Inc., Massachusetts/USA) with a frequency of approximately 5 Hz. Children who were not able to walk hands-free in a condition (also based on the previous two familiarization sessions) were asked to hold the handrails during all Andago and treadmill trials to ensure comparability between the conditions. We calculated the mean values of the forces produced with the right and left hand on the handrails over the two trials for each condition with MATLAB. We summed these values of the left and right hand and presented them as a percentage of the participant’s bodyweight.
Statistical analyses
Statistical analyses were performed with the software R (Version 3.4.3, R Core Team, Vienna, Austria, 2017) and SPSS (Version 24, IBM).
We tested the distribution of the data with the Shapiro–Wilk test and Q–Q plots. Based on their outcomes, we performed merely non-parametric statistics. Differences between two dependent conditions were tested with two-sided Wilcoxon signed-rank tests (i.e., Practicability: the number of bumps when passing a door between session 1 and 2; Acceptability: differences between the first and second session; Appropriateness: stride-to-stride variability of stride time and inter-joint coordination). Data reflecting the accuracy of the BWU system compared to the digital scale (for 25%, 50%, and 75% of BWU) were normally distributed, so we performed paired t-tests. Alpha was set at 0.05.
Differences between the three Andago BWU conditions (Appropriateness) were tested with a Friedman’s test followed by pairwise Wilcoxon signed-rank tests. Here, we applied a Bonferroni’s correction for the pairwise comparisons and set alpha at 0.017.