PubMed
(n = 725)
Scopus
(n = 53)
Web of Science
(n = 32)
Cochrane
(n = 22)
Ebsco
(n = 36)
Engineering Village
(n = 32)
Records screened
(n = 900) Duplicates
(n = 58)
Titles & Abstracts screened
(n = 842) Records excluded
(n = 803)
Full-text articles assessed
for eligibility
(n = 39)
Included
(n = 17)
Additional articles –
manual reference
screening
(n = 12)
Excluded; reasons:
(n = 22)
Lokomat/BWST (4)
No robotic device (2)
Cross-design (1)
No SCI-specific data (3)
No intervention (1)
Conference proceeding
(6)
FES (1)
Case report (3)
Proposal (1)
study was conducted at multiple centres across the
UK, Europe and Australia (11), and 2 others were
conducted across 5 centres within the USA (12) and
9 centres within Europe (13). Sample sizes ranged
from 3 to 44 participants.
Participant characteristics
Mean participant age was 40 years (SD 7) , with
males accounting for 78% of the sample. The ma-
jority of participants presented with complete SCI
between T1 and T12 (Fig. S1 1 ). Time since injury
varied considerably between studies, with the major-
ity of participants presenting with chronic injuries
(> 1 year).
Training protocols
ment and Evaluation (GRADE) system. The quality of the studies
was extracted independently by 2 reviewers using a standardized
protocol and data collection table according to STROBE (Table
SII 1 ) guidelines adapted from the Equator system. Data analysis Ambulatory outcomes
Meta-analyses were completed for the walking performance
tests. Random effects models were reported based on the high
variability noted between studies. The standardized mean differ-
ences and 95% confidence interval (95 % CI) were calculated. To
ensure positive estimates for all meta-analyses, the study means
were multiplied by –1 according to the statistical correction
described in the Cochrane Handbook. Forest plots were used to
illustrate the individual study findings and the pooled estimate
results. Statistical significance level was set at p < 0.05. The I 2
statistic was used to estimate heterogeneity of effects across
studies, with values of < 25%, 50% and > 75% representing
low, moderate and high inconsistency, respectively. Statistical
analyses were performed using RevMan Review Manger 5.3.
Participant and intervention characteristics were summarized
using means and standard deviations (SD) for continuous data,
and counts and percentages for categorical data. Walking performance. Meta-analyses were performed
on the 7 studies that assessed walking performance
tests, including the 6MWT (Fig. 2), the 10MWT (Fig.
3) and the TUG (Fig. 4). Five studies reported a posi-
tive pooled effect of –0.94 (95% CI –1.53, –0.36) with
moderate heterogeneity (I 2 = 27%, p = 0.002) for the
distance achieved during the 6MWT. Six studies re-
ported a positive pooled effect of –1.22 (95% CI –1.87,
–0.57) with high heterogeneity (I 2 = 60%, p = 0.0002)
for the speed achieved during the 10MWT. Five studies
reported a positive pooled effect of 0.74 (95% CI 0.36,
1.11) with no heterogeneity (I 2 = 0%, p = 0.0001) for the
time required to complete the TUG. Importantly, 4 of
the 7 studies included in the meta-analyses involved
chronic recovery phases (7, 14–16), one involved
both acute and chronic phases (13) and one study did
not report on this recovery period (12). Variations in
other participant characteristics, including age, level
of injury, and intervention protocols differences, could
also act as cofounders to the results observed between
studies. This variability between studies is evident in
the moderate-high heterogeneity scores observed in
Figs 2 and 3.
Excluded
(n = 2)
Argo (1)
Lokomat (1)
Total included in review
(n=27)
Fig. 1. Outline of the literature search procedure and article selection.
725
The ReWalk™ (ReWalk Robotics Inc., Marlborough,
MA, USA) powered exoskeleton was evaluated in
11 studies, Ekso® (Ekso Bionics, Richmond, CA,
USA) in 10 studies, Indego™ (Parker Hannifin Corp.,
Cleveland, OH, USA) in 3 studies, WPAL (Fujita
Health University, Japan) in 2 and REX (Rex Bionics
plc, London, UK) in one study. The mean interven-
tion length was 12.1±19.6 weeks, with a wide range
of 1–24 weeks, 8 weeks being the most common.
Typically, training was conducted 3 times per week
for 60 min per session (Table I).
Robotic locomotor training in rehabilitation
RESULTS
Study characteristics
All studies included in this review were prospective
non-randomized, uncontrolled trials, of which all,
except 3, were single-centre studies conducted in the
USA (12), Italy (3), Canada (3), Japan (2), Germany
(1), Netherlands (1), UK (1) and Israel (1). A single
J Rehabil Med 51, 2019