Spontaneous recovery after spinal cord injury (SCI) at the chronic stage is limited, and individuals with severe SCI are less likely to recover motor function than individuals with moderate SCI [1]. Over the past 25 years, since the first publication showing that humans possess a central pattern generator [2], body weight-supported treadmill training (BWSTT) has been reported to be effective for restoring gait capacity in SCI patients with varying levels of severity [3]. More recently, robot-assisted gait training (RAGT) devices have been used clinically, including Ekso [4], Lokomat [5], and ReWalk [6] devices. In general, a combination of BWSTT and RAGT offers advantages including reducing the effort required of physical therapists, longer durations of training for patients, and more accurate and reproducible gait patterns. In addition, the sensorimotor cortical regions of the cerebrum can be activated after BWSTT with the Lokomat [7]. It is important to understand who best responds to what type of treatment, and some reports have indicated that individuals with chronic SCI are responsive to each therapy included RAGT [8,9,10].

The hybrid assistive limb (HAL; Cyberdyne Inc., Ibaraki, Japan), one of a voluntary driven exoskeleton (VDE), is also effective in improving gait function for chronic SCI patients when used in conjunction with BWSTT [11,12,13,14,15]. However, most of these previous studies included participants with some residual lower-limb motor function and walking ability [11, 13] and only showed the effectiveness of VDE for chronic SCI patients with the ability to walk. In the report by Jansen et al., a participant with more severe gait disability at baseline (Walking Index for Spinal Cord Injury II score of one) was included and showed an improvement in gait function after BWSTT with a VDE [12]. Nevertheless, his injury level and zone of partial preservation were enough to walk with orthosis clinically (Th12 level, above L1). Based on this, it is not fully clear if BWSTT with VDE is effective for all severities of SCI.

In order to optimize the use of VDE, it is necessary to understand the treatment effect that can be expected for individuals with severe SCI. The aim of this study was to quantify the effect of BWSTT with VDE in SCI patients according to the severity of SCI, evaluated using baseline walking ability.


Study participants

Twenty participants with chronic SCI were recruited from the patients of Keio University Hospital by individual referral regardless gait ability between November 2016 and November 2018. The inclusion criteria were as follows: (1) age between 20 and 75 years, (2) more than 6 months post injury, (3) self-reported disturbed gait and plateau in recovery from paralysis symptoms. The exclusion criteria were as follows: (1) disease or skin disorder that would make training impossible or be made worse by training, (2) received interferon-alpha or Botox injection within the past 6 months, (3) participated in other neurorehabilitation training (such as other BWSTT or functional electrical stimulation) within the past 3 months, and (4) underwent walking training with VDE within the past 12 months. This study was registered to UMIN-CTR (UMIN000021907).

Study design

This study was conducted as a non-randomized open-label single-arm study.

Before training, neurological classification was determined using the American Spinal Injury Association Impairment Scale (AIS), and all outcome measures were evaluated. Two physical therapists evaluated all outcomes measures throughout the research. In this investigation, every participant underwent 20 sessions of BWSTT with VDE (2–5 sessions per week), at Keio University Hospital referring to previous report [11] and investigator-initiated clinical in Japan [16]. After 20 training sessions, all outcomes were evaluated and feedback was collected from the participants. Written informed consent was obtained from the study participants, including consent to participate and to publish the findings before enrollment. No incentive was given to the study participants.

Treadmill training with VDE

As VDE, we used HAL which was designed for this purpose as it provides voluntary motion assistance to the lower limbs and is controlled by detecting the weak bio-electrical signal generated by active muscle contractions. Electrodes were placed on the skin surface according to a previous report to capture the voluntary bio-electrical signals from extensor and flexor muscles of hip and knee joints [15]. These signals were used to provide motion support.

Training with VDE was performed on a treadmill (Aeromill STM-1250, Nihon Kohden Corporation, Tokyo, Japan) with half of their body weight supported (VDE device weight not included), by a weight-support device (PneuWeight, Pneumex, Idaho, USA). If necessary, participants were allowed to hold the handrails of the treadmill. Assistance for shifting the center of gravity and swinging of the lower limbs was provided by therapists. The velocity of the treadmill was individually set to the participant’s comfortable walking speed (between 0.5 and 2.5 km/h), and there was no incline. The duration of each training session was 60 min, which included time to rest that did not exceed 20 min. Five participants had difficulty walking with VDE and performed weight-shift training or stepping training with VDE, instead of gait training, for the initial one to three sessions.

The treadmill training with VDE was supervised by physical therapists with more than 6 months of experience using VDE, all of whom had attended the safety training courses provided by Cyberdyne, Inc. and were certified to use the device.

Outcome measures

Gait performance on the treadmill with VDE

The speed, distance, and duration the participant walked in one session with VDE was recorded in every training session. On the same days, participants reported the perceived training intensity using the Borg scale, which quantifies the perception of effort during exercise. The scale ranges from 6 to 20 points, where 6 means very very light and 19 means very very hard [17].

Overground walking ability without VDE

Overground walking ability was evaluated without VDE, but with the use of self-selected aids and braces. Walking ability was quantified using the Walking Index for Spinal Cord Injury II Scale (WISCI-II) [18]. Gait speed and number of steps needed to walk 10 m were assessed using the 10 meters walk test (10MWT) [19] performed at maximum speed. The 2 minutes walk test (2MWT) [20] was used to measure the distance walked in 2 min at a self-selected speed. Mobility was evaluated using the timed up and go (TUG) test [19], which is the time needed to stand up from a wheelchair, walk 3 m, return to the chair and sit down.

Balance and performance in activities of daily living (ADL)

Balance was evaluated using the Berg Balance Scale (BBS) [21]. The result of BBS was also evaluated in three categories—sitting balance, standing balance, and dynamic balance (position change) according to the previous study [22]. Lower extremity motor score (LEMS) was evaluated as function of lower limb [23]. Performance of ADL was evaluated using the Barthel Index (BI) [24] and Functional Independence Measure (FIM) [25].

Subjective assessment

The degree of subjective improvement and degree of satisfaction were rated by the participants on the evaluation day set after the last training session. Improvement was rated on a seven-point scale (completely improved, much improved, slightly improved, no change, slightly worse, much worse, very much worse), and satisfaction was rated on a four-point scale (satisfied, slightly satisfied, slightly dissatisfied, dissatisfied).


The change in each outcome measure from pre to post training was assessed using a Wilcoxon signed-rank test due to the heterogeneity of the study sample. In addition, participants were categorized into two groups according to the baseline WISCI-II score: low walking ability (low group; unable to walk 10 m or could walk only in parallel bars; WISCI-II score 0–5), and high walking ability (high group; able to walk 10 m with a walker or canes/crutches, braces, or physical assistance, or without any device or assistance; WISCI-II score 6–20). We compare each parameter of participant characteristics between low and high group in using a chi-square test and Mann–Whitney U test. Within each group, the changes in each outcome measure from pre to post training was assessed using a Wilcoxon signed-rank test. Differences were considered statistically significant at p < 0.05. Data were analyzed using IBM SPSS Statistics version 25.0 (IBM Japan, Japan).


Participant characteristics

The characteristics of 20 participants completed the study protocol and were left for the final analysis are shown in Table 1 (see Supplementary Table 11 for participant-level data). The mean (SD) age at the time of enrollment was 43.3 (16.6) years. The mean (SD) time since injury was 80.4 (128.8) months. Two participants were categorized as AIS grade A, four as grade B, eight as grade C, and six as grade D. The level of the injury was cervical in ten participants, thoracic in nine, and lumbar in one. The median (range) WISCI-II score was 9 (0–20), including seven participants with a score of zero. Based on baseline WISCI-II score, eight participants were categorized into the low walking ability group, and 12 participants were categorized into the high walking ability group. The mean (SD) frequency of intervention was 2.6 (1.1) days per week. All participants achieved 20 training sessions with VDE. There were no adverse events.

Table 1 Baseline characteristics of participants in Low group and High group with chronic spinal cord injury.

Gait performance on the treadmill with VDE

The speed, distance, and total duration participants walked in one training session increased significantly from the first to the last training session in all participants (Table 2 and also see Supplementary Tables 1–2 for participant-level data). Five participants who initially had difficulty walking with VDE were analyzed afterwards, when they were able to walk with VDE for the first time. Mean ratings of perceived exertion across the 20 sessions were significantly higher in the low group compared with the high group. This provides evidence that the training intensity was adequate in the low group, even though the other three training parameters were significantly lower in this group (see Supplementary Table 2 for training intensity in two groups).

Table 2 Training parameters in the first and last training session with HAL in 20 participants.

Overground walking ability without VDE

Overground walking ability before and after 20 sessions of VDE training is shown in Table 3 (see Supplementary Tables 13 for participant-level data). The WISCI-II score tended to increase. No participants in the low group were able to complete the tests of overground walking ability (10MWT and TUG) at either time point. In the high group, there was a significant improvement in 10MWT time (134.0 to 88.3 s, p = 0.01) and speed (0.26 to 0.34 s/m, p < 0.01), 10MWT number of steps (44.8 to 36.5 steps, p = 0.05), and TUG time (83.5 to 68.5 s, p = 0.01). The decrease in 10MWT number of steps indicates an extension of step length. The WISCI score was not significantly improved in the high group (10.5 to 11.5, p = 0.11). In terms of the relation between relative change in 10MWT time and WISCI-II score at baseline, the relative change in 10MWT time varied among participants (Supplementary Figure).

Table 3 Change in overground walking ability, balance, and performance of ADL from baseline to post-20 sessions of BWSTT with HAL.

Balance and performance of ADL

Balance and performance of ADL before and after 20 sessions of VDE training are shown in Table 3 (see Supplementary Tables 14 for participant-level data). When the group was considered as a whole, there was a significant improvement in BBS score, but not BI or FIM (Table 3). Although the BBS score improved following training in the high group (p = 0.02, Table 3), there was no change in the low group (p = 0.06, Table 3). When BBS was evaluated by three categories, participants in the low group showed little improvement in dynamic balance and sitting balance but no improvement in standing balance (Supplementary Table 3). Neither the BI nor the FIM changed from pre to post training in high group (p = 0.32, p = 1.00) and low group (p = 1.00, p = 1.00), indicating that there was no effect of training on performance of ADL regardless of baseline walking ability (Table 3).

Subjective assessment

Subjective improvement was reported by 19 participants (two answered “much well” and 17 answered “slight well”), and all participants were either fully or slightly satisfied with the training (eight were satisfied and 12 were slightly satisfied).


This study is novel in the point that the performance in balance and gait of chronic SCI patients was improved after as few as 20 sessions of BWSTT with VDE. The degree of improvement depended on baseline walking ability, with patients with a baseline WISCI-II score of less than six hardly benefitting from the training. This result was against previous study that reported the improvement of walking ability even in the patients with severe motor dysfunction with a baseline WISCI-II score of one [12]. Although the overall change in BBS score was not significant for our participants with a baseline WISCI-II score of less than six, some of them did show an improvement in sitting balance.

Jansen et al. showed improvement in gait ability after 60 sessions with VDE in chronic SCI participants who included a participant with severe gait disability at baseline (WISCI-II score of one) [12]. He was classified as AIS grade A, however, his injury was at Th12 and zone of partial preservation was above L1. In addition, his baseline Janda muscle function testing grade was more than two for the hip joint flexors and knee joint extensors, which meant that he could move his hip and knee joints when supported against gravity. In contrast, the low group in the present study had more severe motor dysfunction than the patient reported by Jansen et al. Other reports also showed that gait function was not or poorly improved in the participant without baseline gait ability by locomotion training [26] or RAGT [27] too. Moreover, the above intervention by Jansen et al. included regular physical therapy and exercises in addition to BWSTT with VDE [12]. By contrast, participants in the present study performed only BWSTT with VDE. This may require additional interventions for chronic SCI patient with severe motor dysfunction and gait disability.

On the other hand, our result in high group (WISCI-II score of 6–20) was consistent with previous studies [11, 13]. More specifically, they improved gait performances without VDE at less times of training sessions using VDE than that of them. Considering our result and previous studies that included participants with some residual lower limb motor function and walking ability (i.e., WISCI-II score ≥ 6), BWSTT with VDE is effective for chronic SCI patients with the ability to walk with some aid at baseline. In other word, the patients who are most likely to benefit from BWSTT with VDE have certain stability of the lower limbs, which means enough to move one’s leg under the condition without gravity. Some previous studies reported that showed baseline strength of lower muscle was useful for person with SCI to predict responsiveness in BWSTT and activity-based therapy [8, 26]. Meanwhile, in this study, a small improvement in WISCI-II scores was shown and this was not statistically significant in the high group, which was consistent with two previous reports that enrolled 8 patients [15] and 21 patients [12] with chronic SCI using VDE. These results may indicate that the sample is too small sample to achieve a statistically significant change in WISCI-II scores.

In the present study, there was significant improvement in the BBS score in all patients and high group following BWSTT with VDE, which is novel finding. Although there was no significant change in the low group, some of them showed an improvement in sitting balance or dynamic balance (Supplementary Table 3). This is in accordance with a previous study that demonstrated RAGT improved trunk muscular activity during gait in high-thoracic, motor-completed SCI patients [28]. Thus, similar to previous research [29], RAGT may improve trunk muscular function and sitting balance in SCI patients with severe motor impairment.

The present study has several limitations. First, the duration of the intervention (20 sessions) may have been insufficient to improve the ability of the low group. We chose this duration because we considered it worthwhile to prove information on the minimum training duration required to elicit an effect, regarding many chronic SCI patients have returned to social life and are difficult to spend long time on training. The selection of duration was based on studies that adopted 9 sessions [16] and 30 sessions [11]. While a previous report showed that RAGT was ineffective for re-establishing gait in chronic SCI patients without baseline gait function after 60 sessions [10], longitudinal research utilizing >60 sessions is required. Second, the frequency of the intervention differed across participants, and this could have led to dispersion of the results. However, as there was no significant difference in intervention frequency between the high and low group, it is likely our main finding was not severely affected. Third, this was a heterogeneous study sample that included a wide range of time since injury, a wide range of injury level and type (nontraumatic and traumatic), a mixture of complete and incomplete SCI, and was also no control group. For this reason, we could not confirm significant difference of the treatment effect of our patients by analysis by the level of injury or functional severity, as was shown in previous studies [2, 30]. Although this baseline heterogeneity may lead effect modification, baseline data showed no significant difference between groups except for AIS grade and WISCI score in Table 1, which may be related at baseline gait ability. Fourth, we did not perform a detailed evaluation of spinal circuitry (e.g., electromyography, somatosensory, or motor evoked potentials) after RAGT in our study. Sczesny-Kaiser et al. reported that BWSTT with VDE improved cortical excitability in the primary somatosensory cortex for people with chronic SCI [14]. Accordingly, it is possible that the difference in the recovery of gait function between the low and high group of our study was also influenced by differences in spinal circuitry, especially sensory function. Further research is needed to clarify the sensory functional influence on effectiveness in BWSTT using VDE. Lastly, it was not possible to collect data in a blinded manner.

We found that chronic SCI patients with high walking ability at baseline showed significant improvement in walking function and balance function by 20 sessions of BWSTT with VDE. On the other hand, patients with a baseline WISCI-II score of less than six did not show significant improvement in either of these functions. In the near future, BWSTT with VDE combined with new therapies under development, such as cell transplantation or medication, may further promote functional recovery even in individuals with severe chronic SCI.