The Effects of Robotic Training on Walking and Functional Independence of People with Spinal Cord Injury: A Systematic Review, Meta-analysis and Meta-regression

We performed screening for this review in two phases. The first phase served at larger project with a wider scope[21] and included studies using the PICOS (patient, intervention, comparison, outcome, study design) framework as follows: P) adults or children requiring medical rehabilitation; I) any type of robotic device designed for rehabilitation purposes; C) conventional rehabilitation, wait-list-control, or other training modalities different from the experimental group; O) body functions and structures, activities, or participation according to International Classification of Functioning, Disability and Health (ICF), or quality of life; and S) RCT or cross-over RCT. The second phase was carried out after the updated search with more specified PICOS criteria to identify eligible studies of interest in this particular review: P) adults with both SCI and walking impairments; I) robot-assisted lower extremity or walking training intervention; C) a different type of exercise (active control) or no exercise (inactive control) or placebo as comparator; O) validated and standardized measures of walking or functional independence, and S) RCTs and cross-over RCTs. No language or publication date restrictions were imposed.

The titles, abstracts and full texts of the included studies were independently assessed by two researchers (AK, SH, RY, MK, OI, and EA) according to the eligibility criteria using Covidence software.[22] Disagreements were resolved by discussion or consultation with a third review team member (EA). All eligible RCTs were included in the systematic review. Meta-analyses excluded passive and other type of robot control interventions to control clinical heterogeneity.

A customized template was designed in Covidence[22] to extract information on participants, interventions, outcomes, and adverse events of the included studies, and to perform quality assessment according to the Cochrane Risk of Bias 2 tool [23]. Two review team members independently extracted data and assessed the quality of the studies (AK, MK, SH, and RY). Disagreements were resolved by discussion or consultation with a third review team member (EA). Researchers of the RCTs were contacted when necessary to acquire missing data or to clarify ambiguities. If adequate data were not received despite three requests, the study or some of the outcomes of the study were excluded from the quantitative analyses. RCTs eligible for this review were included in the meta-analysis, regardless of the risk of bias judgement.

All outcomes measuring walking ability or functional independence in individuals with SCI were extracted from the included studies. A combination of the 10-m walk test (10MWT) measuring walking speed and the Walking Index for Spinal Cord Injury (WISCI), measuring walking independence or the change in the need for a walking aid, is suggested to provide the most valid measure of improvement in gait and ambulation [7, 8]. To provide the most comprehensive battery, a measure of endurance, such as the 6-min walk test (6MWT), is recommended.[7] For the walking ability meta-analysis, all walking outcomes in the included studies were prioritized in accordance[7] the following order: walking speed, walking independence, and walking endurance (Supplementary material).

Both the Spinal Cord Independence Measure (SCIM III) and the Functional Independence Measure (FIM) have been used to measure the broader functioning and independence in everyday life of individuals with SCI. The SCIM was chosen as the primary measure because it was specifically developed for persons with SCI [14, 24].

To assess the treatment effect after the intervention, the meta-analysis was conduct-ed using R software with the Metafor package for R.[25] Postintervention mean and standard deviation (SD) values were used in the analyses. Data reported as median or interquartile range (IQR) were converted to mean and SD assuming a normal distribution. A correlated effects model with robust variance estimation (RVE) using the Robumeta package on R and small-sample corrections was used, as it considers the possible dependent effect of the studies used multiple times in the same meta-analysis [26]. This was the case when a study had multiple control groups [27‐29] or the study population was divided according to the level of injury [30]. It is also considered to be a more reliable analysis model for studies with small number of participants [26]. The intervention effect size (Hedges’ g), 95% confidence interval (CI), and statistical heterogeneity (I2) were estimated using a forest plot. The scale of Hedges’ g was evaluated as small (0.20–0.49), medium (0.50–0.79), or large (0.80 or more) effect [31]. Statistical heterogeneity was assessed as follows:0–40% might not be important heterogeneity, 30–60% may represent moderate heterogeneity, 50–90% may represent substantial heterogeneity, and 75–100% represents considerable het-erogeneity [32]. If a crossover-RCT did not have a washout period, only the first intervention period was included in the meta-analysis.

Meta-regression analysis was performed using the Metafor package for R. We computed the Univariate Mixed effects model with intercept and restricted maxi-mum-likelihood estimation to determine whether covariates related to intervention content (duration of intervention, number of training sessions per week, time of one training session, weekly total volume of training), characteristics of rehabilitees (age, time since injury in months, the baseline WISCI score), and quality of the study (domains of risk of bias) could have an impact on the results. Sensitivity analysis was conducted excluding the studies with a high risk of bias in the domains that were found to be significant in the meta-regression. The certainty of evidence was graded at the outcome level according to the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) guidelines [33‐35].

An initial 1 405 abstracts were identified from the electronic databases after duplicates were removed (Fig. 1). After removal of studies considered ineligible according to the PICOS criteria, 23 RCTs were included in this review and 14 in the meta-analyses with all of them studying walking ability and 9 also functional independence. The remaining studies compared two types of robot-assisted walking training [36‐40], had the same patient population as another included study [41, 42], had insufficient reporting of results [43], or the comparison group included no exercise [44]. Detailed characteristics of the included studies, justification for full-text exclusions and the information requested from RCTs are provided (Supplementary material).

Most commonly, the injuries of the participants in both meta-analyses were at the cervical or thoracic level, but there were also participants with lumbar-level injuries. Consequently, the meta-analyses included both paraplegic and tetraplegic participants. Most studies included participants who were grade C or D on the ASIA Impairment Scale (AIS) [45], with the majority being grade D. One study divided the participants into complete and incomplete injuries, without naming the AIS grade [46].

The overall risk of bias was assessed as unclear [36‐39, 46‐48, 52, 53, 55] or high [27‐30, 40‐44, 49‐51, 54] in each study (Supplementary material). No studies with a low overall risk of bias were found. High risk originated mainly from deviations from intended interventions but also from missing outcome data. An unclear risk of bias was found in the randomization process, deviations from the intended interventions, and selection of the reported results. Visual inspection of funnel plots suggests that some degree of publication bias is possible, smaller studies seem to favor the comparator (Supplementary material).

Statistically significant improvements were observed in functional independence in favor of the robot-assisted walking training group compared to the control group (Hedges’ g 0.31, 95% CI 0.02 to 0.59; I2 = 19.7%, 9 studies, 419 participants), whereas there were no statistically significant differences between groups in walking ability (Hedges’ g 0.02, 95% CI –0.27 to 0.31; I2 = 35.5%, 14 studies, 498 participants), walking speed (Hedges’ g –0.09, 95% CI –0.51 to 0.33; I2 = 32.8%, 10 studies, 290 participants), walking endurance (Hedges’ g -0.03, 95% CI –0.65 to 0.58; I2 = 63.1%, 8 studies, 259 participants) or walking independence (Hedges’ g 0.25, 95% CI –0.14 to 0.64; I2 = 51.3%, 9 studies, 419 participants) (Figs. 2, 3, 4, 5 and 6). Certainty of evidence proved to be low for functional independence and walking independence and very low for walking ability, speed, and endurance (Supplementary material).

In the meta-regression analyses, no relationships were found between the effects of robot-assisted walking training and intervention content or characteristics of rehabilitees. A high risk of bias in selection of the reported results was associated with the effect in functional independence. When excluding the high risk of bias study [29] from the meta-analyses, robot-assisted walking training remained statistically significant in improving functional independence compared to the control group (Hedges’ g 0.35, 95% CI 0.05 to 0.64; I2 = 15.5%, 8 studies, 389 participants) (Supplementary material).

Adverse events were examined in 11 of 23 studies included. Reported adverse events were mostly mild and infrequent and four studies reported no adverse events (Supplementary material).

This study provides new information on the effects of robot-assisted walking exercise on functional independence and various aspects of walking ability in individuals with SCI. It is the first to assess the association between personal, clinical, and intervention characteristics and intervention effects. Meta-analyses excluded passive control interventions to control clinical heterogeneity. Meta-regression clarified the results, and GRADE guidelines graded the evidence certainty at the outcome level [33, 34]. To our knowledge, no recent review has provided graded clinical recommendations on this topic.

This review has limitations to consider when interpreting the results and generalizing evidence. The meta-analyses mainly included individuals with AIS grade C or D, and all but one study used an exoskeleton robot (Lokomat or Ekso), so the findings may not be generalizable. More data is needed on the effects of different robots. RCTs’ methodological quality limits the reliability of the results. However, sensitivity analyses excluding studies based on the risk of bias did not alter the results. Future studies should pay particular attention to the methodological quality to ensure unbiased results.

Low level evidence suggests that robot-assisted walking training results in a slight improvement in functional independence, but little to no difference in walking independence in persons with SCI when compared to other exercises. The evidence is very uncertain regarding the effects of robot-assisted walking training on walking ability, walking speed, and walking endurance in persons with SCI when compared to other exercises. Heterogeneity between studies was substantial, and there is no clear evidence if positive effects were associated with age, time since injury, baseline walking independence, intervention programming, or quality of the study. Robot-assisted walking training appears to be a safe rehabilitation method for individuals with SCI. However, additional high-quality RCTs with larger sample sizes, similar outcome measures and differentiation of results between paraplegic and tetraplegic individuals are needed to further evaluate the effects and safety of robot-assisted walking training on functional independence and walking ability in individuals with SCI. When seeking to improve the functional independence of persons with SCI, robot-assisted gait training may be considered as a potential training option.

Low-level evidence suggests that in people with SCI robot-assisted walking training may slightly improve functional independence but has little to no effect in walking independence compared to other exercises. Evidence is very uncertain on walking ability, speed, and endurance. Intervention or rehabilitee characteristics, and risk of bias didn’t affect results.