Effects of Robot-Assisted Gait Training in Individuals with Spinal Cord Injury: A Meta-analysis

2.1. Types of Participants

This study included only SCI subjects, regardless of traumatic or nontraumatic lesion, the time since injury, age, and sex.

2.2. Types of Interventions

Any kind of RAGT compared with other training modalities or no training for lower limbs was included.

2.3. Outcome Measures

Primary outcome measures were spasticity and pain. Spasticity was assessed by modified Ashworth scale (MAS) or Ashworth scale (AS) for lower limbs. Pain was assessed by the visual analog scale (VAS). The VAS is widely used to assess self-perceived pain [29]. It is a 10-centimeter line in which 0 represents no pain and 10 at the right edge means intolerable pain. Participants subjectively reported their pain condition on the VAS scale.

Secondary outcome measures were LEMS and walking ability assessed by 6MWT, 10MWT, TUG, WISCI, and FIM-L. LEMS assessed motor score for lower limbs according to ASIA standard. 6MWT measured the walking distance in 6 minutes. 10MWT assessed the walking speed measuring the time necessary to walk 10 meters. 6MWT and 10MWT were reliable and responsive tools in assessing walking ability in incomplete SCI [30]. TUG assessed the time that a subject took to rise from a chair, walk three meters, turn around, walk back to the chair, and sit down. WISCI measured improvements in ambulation in persons with SCI by evaluating the amount of physical assistance, braces, or devices required to walk 10 meters. WISCI I scored from 1 to 19 and WISCI II from 1 to 20 [31]. FIM was an 18-item assessment of physical, psychological, and social function. The assessor graded the functional status of a person based on the need of assistance [32].

2.4. Type of Studies

RCT, non-RCT, and crossover trials (only the RAGT period) were included in analysis.

2.5. Searching Criteria

The searching criteria were limited to human studies and English language.

2.6. Data Sources

Four electronic databases (PubMed, Scopus, Medline (Proquest), and Cochrane Central Register of Controlled Trials (CENTRAL)) before November 2019 using Medical Subject Heading terms combined with keywords, such as robotics, spinal cord injury, pain, and spasticity, were processed in title, abstract, and keywords. Appendix 1 shows the combinations used.

2.7. Study Selection

Two authors independently searched and screened the titles, abstracts, and literatures to identify potentially relevant studies. Then, full texts of relevant studies were obtained and assessed to determine whether the articles met the inclusion criteria. Any disagreement was discussed and solved with the third author to reach a consensus in every relevant detail.

2.8. Data Extraction and Management

Two authors extracted data independently from included studies and filled into an extraction form. The following data were extracted: (1) authors; (2) year of publication; (3) study design; (4) inclusion/exclusion criteria; (5) subject demographics (age, gender, number of subjects, level of lesion, classification of ASIA, duration of injury); (6) intervention; (7) outcome measures; and (8) summary of the results.

Data at baseline and at the end of the intervention were extracted for the analysis of the effect of training. Measurements during the intervention or at follow-up were excluded due to inconsistent measuring time points used across different studies. Studies were excluded if necessary outcome measures were missing or not measured.

2.9. Quality Assessment

The methodological quality of the selected RCTs was independently assessed by two authors using the Cochrane risk of bias assessment tool [33]. For the assessment of the methodological quality of the selected cohort studies and clinical trials, the Newcastle Ottawa Scale [34, 35] was employed and done by two authors independently. Any disagreement was resolved through discussion and consensus with the third author.

2.10. Statistical Analysis

RCTs and non-RCTs were grouped and analyzed separately. Statistical analysis was performed using Comprehensive Meta-Analysis (CMA) version 3 to analyze the treatment effect. Mean differences and 95% confidence interval (CI) were calculated for each primary and secondary outcome. Random effect models were used to calculate the pooled mean difference estimates if heterogeneity occurred. Fixed effects models were used to calculate the pooled mean difference estimates if no heterogeneity occurred.

3.1. Studies Included

A total of 223 studies from electronic databases and two studies from the reference lists of included studies were identified. In these, 105 of the selected studies were duplicates and thus were removed from analysis. Out of the retained studies, 18 studies were retained for quality synthesis which included 7 RCTs and 11 non-RCTs. The flow of studies through the review process is shown in Figure 1. Six studies were included after review for quantitative synthesis of which the characteristics are shown in Table 1 and Table 2.

3.2. Interventions

The intervention of RAGT was 3 to 5 sessions per week, 30 min to one hour for 4 to 12 weeks in RCTs. The training protocol of non-RCTs was 2 to 5 sessions, 30 min to 90 min for one week to 90 days.

The apparatus used for RAGT in these studies included Lokomat, hybrid assistive limb (HAL), Indego Exoskeleton, ReWalk, ARKE 2.0, and Ekso GT in which all included RCTs used Lokomat for training.

3.3. Risk of Bias of the Included Studies

Figure 2 summarized the risk of bias judgements related to all RCTs. In all included RCTs, only one study [36] had high risk of bias level. Five studies [37, 38, 44–46] reported randomization. Two studies did not mention randomization [36] or were unclear [47]. Allocation concealment was fulfilled by two studies [44, 45]. Three of the included studies [44, 45, 47] had blinding and the other three studies [37, 38, 46] did not report the methodology of allocation concealment. Two studies [45, 47] did intention to treat analysis and the other three studies [37, 38, 44] were unclear on the information about attrition bias. Two studies [36, 46] had high risk of attrition bias for not reporting their management on the drop-out data.

Table 3 summarized the risk of bias judgements related to non-RCTs. All studies had general to good quality. All non-RCTs recruited representative SCI subjects and no control group. All studies had secure record on training protocol but one study [39] did not. All studies assessed outcomes independently. The duration of three non-RCTs [43, 50, 52] was from 40 min to 2 weeks, most trials with 8 weeks.

3.4. Effects of the Interventions

Pain and walking ability were analyzed in RCTs. Spasticity, pain, and walking ability were analyzed in non-RCTs. Summarization on spasticity and TUG in RCTs and FIM-L in non-RCTs were done without meta-analysis due to insufficient data information.

3.5. Results of Primary Outcomes: Spasticity

Four RCTs [36, 38, 44, 46] assessed spasticity. However, different muscle groups were assessed in these studies; therefore, the data could not be pooled together. In these studies, all participants’ spasticity was mild (MAS 0-2) and none of them changed significantly after RAGT.

Six eligible non-RCTs were included but only 4 studies’ data were retained to pool for analysis. One trial was excluded for analysis because it assessed spasticity of 36 joints together [51]. The other one trial was excluded for analysis because the participants in this trial had no spasticity [52]. Out of the four non-RCTs analyzed, 2 studies [49, 50] use MAS as their outcome measure on 28 participants. The other two studies [42, 43] used AS to assess 23 participants for spasticity (Figure 3). The robotic group showed significant decrease in MAS (95%CI = −2.886 to -1.412, p ≤ 0.001) and AS (95%CI = −0.202 to -0.068, p ≤ 0.001) measures. The pooled mean difference using MAS and AS (fixed effects model) were -2.149 and -0.135, respectively.

3.6. Results of Primary Outcomes: Pain

Two RCTs [44, 45] and 3 non-RCTs [43, 49, 50] were included for analysis. Eighty-four and 31 participants were involved in RCT and non-RCT studies, respectively. Figure 3 showed the results on the analysis of the primary outcomes of pain after RAGT. Although the trend for pain reduction was in favor of robotic group, there was no significant difference between robotic and control group, regardless of RCTs (p = 0.427) or non-RCTs (p = 0.239). The pooled mean difference (random effects model) of RCTs and non-RCTs were -0.890 and -1.676, respectively. The level of pain ranged from painless [44] to moderate [49, 50] in all included studies.

3.7. Results of Secondary Outcomes: LEMS and Walking Ability

3.7.1. LEMS

Three RCTs [36, 44, 45] included 104 participants, and three non-RCTs [42, 52, 53] with 30 participants were pooled for analysis. Significant positive effect in favor of robotic group in both RCTs (95%CI = 1.143 to 2.732, p ≤ 0.001) and non-RCTs (95%CI = 1.508 to 4.839, p ≤ 0.001) were shown in the results of LEMS. The pooled mean differences (fixed effects model) were 1.938 and 3.173 for RCTs and non-RCTs, respectively (Figure 4).

3.8. Publication Bias

Appendix 2 demonstrated the funnel plots of VAS, 6MWT, 10MWT, TUG, WISCI, and LEMS. There were no funnel plots of MAS and AS due to only two studies of each measurement. It seemed a symmetrical funnel plot of VAS, but small study bias was identified in Egger’s test (p = 0.03289). No small study bias was found in other measurements.

4.1. Spasticity

The current meta-analysis revealed that spasticity decreased after RAGT in non-RCTs. Several possible mechanisms could explain the reduction of spasticity after RAGT. Spasticity is defined as a velocity-dependent increase in tonic stretch reflexes with exaggerated tendon jerks [54]. However, spasticity also involves nonreflex component such as intrinsic muscular properties [54]. Mirbagheri et al. reported that RAGT reduced reflex and intrinsic stiffness of ankle in individuals with SCI [37].

RAGT produces rhythmic movements of lower limbs and provides sensory inputs. Previous studies reported that rhythmic passive exercise could induce spinal circuitry reorganization and decrease spasticity in patients with SCI [27, 55]. Improving spasticity and locomotor function by the activation of spinal locomotor centers might also be influenced by the repetitive elements of the therapeutic program [56]. RAGT is a type of repetitive functional task training. These above mechanisms might possibly explain the finding that RAGT reduces spasticity.

The reasons that the RCTs did not show significant reduction in spasticity in the RAGT group might be due to the floor effect [44, 46] (MAS 0 to 1) and the measurements done on different joints [38, 44, 46]. In the included studies, the subjects’ initial spasticity level was not high enough to show change after RAGT. It is suggested that subjects with more severe spasticity could be recruited for further investigation of RAGT.

4.2. Pain

Pain and spasticity are intricate consequences of spinal cord injury [57]. Researchers suggested that pain and spasticity are closely linked [57]. In addition, pain and spasticity might share similar pathophysiological mechanisms [5]. Hence, it is reasonable to expect a reduction of pain with spasticity reduction after RAGT. However, the result of this meta-analysis did not show significant decrease of pain accompanying reduced spasticity following RAGT. This might be that the pain suffered by the participants in these included trials was not mainly from spasticity. Some other potential sources, such as muscle soreness due to excess exercise, joint pain due to malposture, or poor biomechanics, might be the cause of pain. One should also note that the neuropathic pain, more than 50% prevalence in spinal cord injured persons [58], was not reported in the included studies. Therefore, they lacked source of data for meta-analysis. It is suggested to be investigated in future studies.

Although, the included RCTs and non-RCTs did not show significant change in pain, the trend favored RAGT group. Past studies revealed that physical activities could relieve musculoskeletal and neuropathic pain [59, 60]. However, participants in the current meta-analysis did not subjectively feel significant alteration in pain level with VAS assessment. The baseline floor effect of mild [44] to moderate [50] intensity of pain felt by the participants might account for the nonsignificant result. Future studies that include participants with higher level of pain at baseline are suggested.

4.3. LEMS

This study showed that LEMS significantly improved after RAGT. As discussed previously, rhythmic muscle activations could be detected during RAGT. In addition, weight bearing may be an important factor. RAGT provides support which allows users to load their weight on lower limbs during training. Lower limbs weight bearing and the enhancement of muscle activation may contribute to the improvement of LEMS.

Decreasing guiding force as RAGT progresses might increase the muscle strength of lower extremities. Subjects needed greater engagement to activate muscles and participate in the training program. Since the guiding force has not been quantified in the included studies, investigation of the relationship between guiding force and the improvement of LEMS is suggested in future studies.

4.4. 6MWT

This meta-analysis showed that 6MWT increased significantly in favor of the RAGT group. 6MWT is an indicator of endurance. Clinically meaningful change (CMC) of 6MWT was 19-22 m in healthy older adults [61]. The 95% CI of the current meta-analysis includes the range 4.394-106.628 m for included RCTs and 7.218-52.586 m for included non-RCTs. Hence, RAGT can be clinically practical for endurance training.

In physiological point of view, endurance could be improved by multiple sessions of submaximal voluntary exercises [62]. RAGT, due to its lack of active participation from the users during training, was doubted to increase cardiopulmonary fitness in subjects [63]. However, increased 6MWT in this meta-analysis indicates it could improve endurance without emphasizing voluntary muscle contraction. According to Mazzoleni et al. [64], bilateral muscular activity increased after RAGT in people with SCI. Thus, the activation of muscles might increase the challenge to cardiopulmonary systems and, thus, increase the endurance of participants with SCI.

4.5. Walking Speed (10MWT and TUG)

TUG is commonly used to assess functional mobility. It was correlated with muscle strength of the lower extremities and gait speed [18, 65]. The current meta-analysis showed that both parameters improved after RAGT in SCI. The results of 10MWT also supported that RAGT increased walking speed in SCI. The CMC of 10MWT was 0.04-0.06 m/s in healthy older adults [61]. The pooled mean difference (0.123 m/s) was above CMC in included non-RCTs in this meta-analysis.

Kim et al. [66] reported that muscle strength of the lower extremities was correlated with walking speed in chronic incomplete SCI. As shown with the result of LEMS’s improvement, lower extremity strength might be the cause of improved walking speed.

Another explanation for improved walking speed was the strengthening of central pattern generator (CPG). Previous studies supported that RAGT, which involved rhythmic activations of lower extremities, could strengthen CPG [67]. CPG is an essential neural mechanism of walking. Enhanced CPG would lead to increase walking ability. The other explanation was the reduction of spasticity. Spasticity could increase the resistance of movement and interfere with gait. As shown in the above result, spasticity reduced after RAGT and, thus, resulted in less resistance during walking.

4.6. WISCI and FIM-L

WISCI and FIM-L showed no significant difference after RAGT in this meta-analysis. The reason might be that studies investigated that WISCI and FIM-L were few in the current meta-analysis. However, the CMC for the WISCI was 1 point [68]. The pooled mean differences of included studies all exceed CMC (3.383 for RCTs and 5.945 for non-RCTs).

FIM was graded according to the assistance required by a subject. People with SCI might experience fear of falling that impeded transferring the improved walking abilities to functional tasks after RAGT.

4.7. Study Limitations

The first limitation of this current review is no classification of subgroups according to the level and severity of SCI. More trials and subjects are needed for subgroup analysis. The second limitation is the risk of bias exists in all studies. Compared with the non-RCTs, the number of RCTs studies is relatively few. The third limitation is that the training protocol used in each study is not identical. Further reviews are suggested to compare effects of different protocols with increased trials.