Powered knee-assisting exoskeletons for gait rehabilitation: a systematic review and meta-analysis on clinical outcomes

Abstract

Background

This systematic review and meta-analysis aim to evaluate the clinical outcomes of powered knee-assisting exoskeletons (PKAEs) in individuals with gait disorders. The primary goal is to compare functional outcomes and spatiotemporal gait parameters between PKAEs and conventional physiotherapy (CPT).

Methods

A comprehensive literature search was conducted in PubMed®, Embase®, Scopus®, and Web of Science™ for articles published up to October 2024. Controlled clinical trials involving PKAE-based rehabilitation in patients with gait disorders were included. Data extraction and bias assessment were independently performed by two reviewers using established tools (RoB-I and ROBINS-I). Effect sizes were calculated using random-effects models.

Results

Fifteen studies (9 randomized and 6 non-randomized clinical trials) met the inclusion criteria. Pooled analyses showed significant improvements in gait speed for PKAEs compared to CPT at post-treatment (SMD = 0.40; 95% CI 0.01—0.79), but no significant differences were found in other outcomes, including the 6-min walk test, 10-m walk test, cadence, timed up and go, timed stair test, and functional independence measure, between the PKAE and CPT groups. Other outcomes, such as balance, gait symmetry, and metabolic efficiency, were explored qualitatively due to the insufficient number of studies for meta-analysis.

Conclusion

PKAEs demonstrated superior results in gait speed but showed comparable findings to CPT in other outcomes, suggesting their potential in gait rehabilitation. However, further research is required to assess their long-term clinical value and to identify the specific patient populations that would benefit most from their use.

Study registration

Registered in the International Prospective Register of Systematic Reviews (PROSPERO) with the code CRD42024626418.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13018-026-06895-0.

Introduction

Gait disorders are prevalent, impacting approximately 35% of individuals aged 70 and older and nearly 60% of hospitalized neurology patients [1]. In older adults and individuals with conditions like stroke, multiple sclerosis (MS), and Parkinson's disease (PD), gait impairments strongly correlate with cognitive decline and substantially elevate the risks of cardiovascular disease, disability, falls, institutionalization, and mortality [2–4].

Treatment for gait impairment depends on its underlying causes, often involving management of medical conditions to reduce long-term debility. Surgical interventions may be indicated for specific cases, including severe osteoarthritis or spinal disorders. A comprehensive approach, including balance exercises, gait training, and physical and occupational therapy, can enhance gait once reversible causes are addressed [3]. Conventional gait rehabilitation methods provide essential mobility but often fall short in fully restoring function or preventing complications. Traditional orthotic devices, like knee-ankle–foot orthoses (KAFOs), provide support for walking but demand substantial physical effort, limiting their extended use [5]. Wearable robotic exoskeletons, in contrast, provide structural support and enable control over gait parameters, such as velocity, cadence, and step length, fostering motor skill re-learning [6]. This technology may reduce health complications linked to prolonged sitting, thereby improving autonomy, quality of life, and social engagement [5, 7].

Regarding knee exoskeletons, several devices have been developed to enhance gait function, including Welwalk (or GEAR) by Toyota Inc. [8], Keeogo by B-Temia Inc. [9], ABLE by ABLE Human Motion Inc. [10], AlterG by AlterG Inc. [11], REFLEX by Ekso Bionics Inc. [12], MAK by Marsi Bionics Inc. [13], as well as Orthobot by Integra Inc. [14], P.REX [15], and KAEGO [16]. These robotic systems vary in design and functionality, providing users with targeted support to improve mobility and aid in knee rehabilitation.

Recent studies highlight the positive impact of robotic knee exoskeletons on gait rehabilitation across various patient populations. The ABLE exoskeleton has demonstrated significant enhancements in gait parameters, including walking time, step count, and speed for spinal cord injury (SCI) patients, with a strong safety profile [17]. The REFLEX exoskeleton has shown promising results in hemiparetic patients by reducing compensatory movements in the nonparetic limb, promoting a more symmetrical gait [12]. For stroke rehabilitation, the Welwalk exoskeleton has achieved notable improvements in step length, width, and single support phase, leading to better gait symmetry compared to traditional orthoses [18, 19]. The Keeogo exoskeleton has also provided significant benefits to individuals with MS, improving daily mobility and both physical and emotional well-being [20].

Research findings on robot-assisted gait training (RAGT) have been mixed. For instance, a before-and-after trial reported that Welwalk training improved gait speed in hemiparetic stroke patients [21]. However, two other studies—a before-and-after trial and a randomized controlled trial (RCT) on hemiparetic stroke patients—found that neither Welwalk nor RAGT training produced significant benefits for gait speed [18, 22].

In terms of gait phases, a before-and-after trial on hemiparetic stroke patients using Welwalk and an RCT on SCI patients using ABLE both found no beneficial effect of knee exoskeletons on the swing and double support phases. However, the findings differed for the single support phase: the former showed a beneficial effect from pre- to post-intervention, while the latter reported no difference between control and intervention groups [10, 18].

Regarding abnormal gait patterns, an RCT on hemiparetic stroke patients reported no significant differences between the Welwalk and control groups at discharge in the distribution of two-point and three-point gait patterns or in the Wisconsin Gait Scale (WGS) scores [23]. Conversely, three other studies reported beneficial effects of robot-assisted gait training (RAGT) on compensatory gait movements [10, 18, 21].

Additionally, one clinical trial with matched controls found that Welwalk increased Functional Independence Measure (FIM) scores in hemiparetic stroke patients compared to matched controls [24]. However, other studies—a clinical trial with matched controls using GEAR and an RCT using Welwalk—reported no significant differences in FIM scores between intervention and control groups of stroke patients [21, 23].

While knee exoskeletons offer promising benefits for rehabilitation, the evidence supporting their superior effectiveness compared to conventional physiotherapy remains limited and inconsistent. Powered knee-assisting exoskeletons (PKAEs) are wearable robots that provide active actuation exclusively at the knee joint (s), without powered hip or ankle components, and can be unilateral or bilateral. To date, no comprehensive review has specifically examined the clinical gait outcomes of PKAEs compared with conventional rehabilitation. This systematic review and meta-analysis aim to fill that gap by providing a detailed comparison of clinical outcomes between PKAEs and conventional physiotherapy for gait disorder rehabilitation. The outcomes considered include spatiotemporal gait parameters such as gait speed, cadence, stride length, step length, and the duration of gait cycle phases; performance-based measures including the 6-min walk test (6MWT), 10-m walk test (10MWT), Timed Up and Go test (TUG), and Timed Stair Test (TST); the Functional Independence Measure (FIM); balance assessments; and gait quality measures such as abnormal gait patterns.

Materials and methods

Protocol

This systematic review and meta-analysis were registered on the Prospective Register of Systematic Reviews (PROSPERO) website with the code CRD42024626418 and followed the guidelines outlined in the PRISMA Statement (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) [25].

Search strategy

An independent reviewer (M.MA.) conducted an electronic search in the PubMed®, Embase®, Scopus®, and Web of Science™ databases from their inception until October 31, 2024. The search strategy utilized the terms (Robotics OR Exoskeleton OR Bionics OR Orthosis) AND (Knee OR “Patellofemoral Joint” OR “Tibiofemoral Joint” OR “Tibiofibular Joint”) AND (Gait OR Locomotion OR Walking OR Ambulation OR Mobility OR Rehabilitation OR “Physical Therapy”), including synonyms and alternative spellings. The search string was tailored to meet the specific syntax requirements of each database (Appendix 1). No restrictions were applied to the language or type of study in the search strategy. Additionally, the bibliographic references of the selected articles were reviewed for related studies, and relevant articles were added manually. Grey literature was not included in this search.

Eligibility criteria and study selection

This systematic review screened all studies that met the PICOD protocol requirements: P (Problem, optional): gait disorder; I (Intervention): rehabilitation with powered knee-assisting exoskeleton (PKAE); C (Comparison): conventional rehabilitation; O (Outcomes): gait spatiotemporal parameters and functional scores; and D (Design): clinical trials. All search results were imported into EndNote X20 citation manager, and duplicate entries were removed. Two independent reviewers (S.Ebh. and M.MA.) screened the titles and abstracts to identify relevant studies. Subsequently, the same two reviewers (S.Ebh. and M.MA.) conducted a full-text review of the selected references. Any disagreements during the screening process were resolved by consensus, with the involvement of a supervisor (M.D. or A.M.) at each stage. Studies were eligible if they evaluated the outcomes of PKAE robots on gait rehabilitation in individuals with gait disorders (e.g., stroke, joint arthroplasty, multiple sclerosis, cerebral palsy, elderly individuals) or healthy participants. Any robotic system that provided powered actuation at the knee joint was included. Systems were eligible regardless of whether they were unilateral or bilateral and regardless of the presence of passive structural components at the hip or ankle joints. However, systems that provided powered actuation at the hip or ankle were excluded. Excluded from the review were abstracts, review articles, textbook chapters, letters to the editor, prototypes, animal studies, and articles without full text availability. Additional exclusion criteria included studies involving powered exoskeletons for joints other than the knee, full lower-body robots, passive knee exoskeletons, artificial limbs, and non-clinical studies. In cases where multiple publications originated from the same clinical trial, only data from distinct outcome measures were included to prevent duplication. Each study's unique contributions were considered based on the specific parameters reported.

Data extraction

Two reviewers (S.Ebh. and M.MA.) independently collected data from the full texts of the included studies using a pre-designed Excel form. The collected results were compared and cross-checked by the same reviewers to ensure accuracy. Standardized, pilot-tested forms were used to ensure consistency and methodological rigor across all stages of the review process. Extracted data included the following: first author, year of publication, country, study design, study population, sample size, demographic data, rehabilitation method, device used in each study group, study time-points, and outcome measures. The pre-defined outcome measures for this systematic review were related to gait rehabilitation and included spatiotemporal gait parameters such as gait speed, cadence, stride length, step length, and the duration of gait cycle phases; performance-based measures including the 6-min walk test (6MWT), 10-m walk test (10MWT), Timed Up and Go test (TUG), and Timed Stair Test (TST); the Functional Independence Measure (FIM); balance assessments; and gait quality measures such as abnormal gait patterns. In addition, other outcome measures that were commonly reported across included studies were also extracted and analyzed, either through quantitative meta-analysis or qualitative synthesis.

Risk of bias assessment

Two independent authors (M.Sh. and S.Ebh.) assessed the methodological quality of the included RCTs and non-randomized studies of interventions (NRSIs), respectively, using the Cochrane Risk of Bias tool for randomized trials (RoB-I) and the Cochrane Risk of Bias tool for non-randomized studies of interventions (ROBINS-I), depending on the study design [26, 27]. Any disagreements were resolved in consultation with a third reviewer (M.S.).

Discrepancies between reviewers were documented and resolved through structured consensus discussions. When agreement could not be reached through direct deliberation, additional members of the research team were consulted to adjudicate and reach a final decision. This consensus-based approach was applied systematically to ensure accuracy, transparency, and reproducibility of study inclusion and data synthesis. Agreement between reviewers was continuously monitored throughout the review process. Disagreements were infrequent and primarily related to the interpretation of eligibility criteria or outcome categorization rather than data abstraction inconsistencies. Final decisions were reached through multidisciplinary consensus, thereby maintaining methodological robustness in accordance with PRISMA 2020 recommendations.

Rating the quality of outcomes

The overall quality of the evidence was assessed by the GRADE (Grading of Recommendations Assessment, Development, and Evaluation) approach for each outcome [28]. The overall quality of the evidence based on specific domains was then either High (+ +  + +), Moderate (+ + +), Low (+ +), or Very low ( +). It may be rated down for reasons such as risk of bias, imprecision, indirectness, inconsistency, and publication bias, or rated up based on plausible confounding. Reasons for downgrading the quality of evidence are transparently reported in front of each outcome.

Statistical analysis

The eligibility of the studies included in the meta-analysis was determined by evaluating the outcome measures, the methodologies employed to assess these measures, and the reporting units for the outcomes. The primary analysis focused on spatiotemporal gait parameters and performance-based metrics, while secondary analyses examined functional independence, balance assessments, and gait quality measures, whether evaluated quantitatively or qualitatively. For the quantitative analysis, forest plots were generated to visually assess heterogeneity and to present the pooled weighted mean difference (WMD) and standardized mean difference (SMD) with corresponding 95% confidence intervals across studies. The selection of effect size metrics (WMD versus SMD) was determined based on methodological, statistical, and clinical considerations. The Weighted mean difference was applied when studies reported outcomes using identical measurement scales and units, and when the underlying clinical populations, intervention characteristics, and assessment conditions were sufficiently comparable to support direct pooling of absolute mean differences. In contrast, the SMD was used when studies assessed the same underlying construct but employed different instruments, scales, or scoring systems, or when variability in study design, population characteristics, or intervention protocols was expected to influence outcome dispersion. Therefore, the choice between WMD and SMD was guided not solely by unit homogeneity, but also by the degree of clinical and methodological heterogeneity to ensure both statistical validity and meaningful interpretation of pooled estimates. Given the anticipated conceptual, clinical, and methodological heterogeneity among studies, pooled effect size estimates were calculated using a random-effects model with the DerSimonian–Laird method [29]. Statistical heterogeneity among included studies was evaluated using Cochran’s Q test and quantified with the I² statistic (with I² ≥ 50% indicating substantial heterogeneity). Given the limited statistical power of Cochran’s Q when the number of included studies is small, statistical significance for heterogeneity was defined a priori as p < 0.10. In addition to statistical metrics, heterogeneity was interpreted in light of clinical and methodological diversity across studies, including differences in patient populations, intervention protocols involving PKAEs, outcome measurement tools, and follow-up durations. Sensitivity analysis was performed to examine the influence of each included study on the pooled estimate and to assess the robustness of the pooled estimates. All statistical tests were two-tailed, and the significance level was set at less than 0.05 for all, except for the heterogeneity test. Statistical analyses were performed using Stata version 17.0 (Stata Corp., College Station, Texas, USA).

Results

Study selection

A total of 10,165 records were identified through database searches. After removing 2773 duplicates, 7392 studies underwent title and abstract screening, and 811 full-texts were assessed for eligibility criteria. Fifteen studies met the inclusion criteria, including 9 RCTs and 6 NRSIs. Of these, all were eligible for inclusion in the meta-analysis. The PRISMA 2020 flow diagram detailing the selection process is presented in Fig. 1.

Fig. 1.

PRISMA flow diagram for reference selection process

Study characteristics

The included studies investigated PKAE in various clinical populations, including patients with stroke (N = 11), Parkinson’s disease (PD, N = 1), osteoarthritis (OA, N = 1), multiple sclerosis (MS, N = 1), and spinal cord injury (SCI, N = 1). The devices used in the interventions included Welwalk/GEAR (N = 9), Keeogo (N = 4), AlterG (N = 1), and ABLE (N = 1) (Fig. 2). Study durations, treatment intensities, and outcome measures varied across the trials. Most interventions combined robotic gait training with conventional physical therapy (CPT), while the control groups received only CPT or other non-robotic exercise regimens. In total, 212 individuals (62% male) were in the exoskeleton group, and 299 individuals (63% male) were in the control group. The included studies applied interventions ranging from 2 to 8 weeks, with 6 to 36 sessions and 40 to 180 min per session (Tables 1 and 2).

Fig. 2.

Illustrations of (A) Keeogo, (B) GEAR, (C) Bionic Leg, and (D) ABLE Devices. Respectively, Reproduced and with permissions from McGibbon et al. (2022) [55], Hirano et al. [23], Byl et al. [56], and Rodríguez‑Fernández et al. [10] licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/)

Table 1.

The characteristics of randomized controlled trials (RCTs)

First Author, Year [ref]	Country	Injury	Groups		Device		Number (Total: M/F)		Age (M (SD))		BMI		Intensity of treatment					Reported outcomes
			Intervention (I)	Control (C)	I	C	I	C	I	C	I	C	w	s	s/w	min/w	min/d
McGibbon [1]	Canada	PD	Exercise with Exo	Exercise without Exo	Keeogo (B-Temia Inc., Quebec City, Canada)	–	13: 4/9	14: 7/7	67.6 (5.9)	70.7 (7.3)	26.0 (5.8)	27.2 (5.9)	8	16	2	120	60	6MWT, FoG-Q, Gait speed, Dual-task gait cost index, B-BESTest, ABC, PDQ-39, SCOPA-COG, UPDRS, HADS, ACPM, Log intensity index, AEs
McGibbon [2]	Canada	OA	Keeogo	Physical activity without Exo	Keeogo, B-Temia Inc., Canada	–	11: 9/2	13: 10/3	59.9 (7.4)	58.5 (10.1)	30 (7)	28.3 (2.9)	2	–	–	–	–	TUG, 6MWT, TST, WOMAC, FSS, ABC, NPRS, SF36, AEs
McGibbon [3]	USA, Canada	MS	Keeogo	Physical activity without Exo	Keeogo (B-Temia Inc., Quebec City, Canada)	–	15: 7/8	14: 5/9	48.4 (11.4)	50.1 (10.0)	25.5 (5.1)	24.9 (6.0)	2	–	–	–	–	Step count, TUG, 6MWT, TSTup, TSTdn, FSS, ABC, NPRS, MSWS
Hirano [4]	Japan	Stroke	RAGT + CPT	CPT	Welwalk WW-1000 (Welwalk, Toyota Motor Corporation, Aichi, Japan)	–	42: 27/15 (34 at discharge)	43: 31/12 (41 at discharge)	58.9 (11.6)	62.9 (9.8)	23.1 (5.1)	22.6 (4.3)	4	28	7	1260 (RAGT: 560)	180 (RAGT: 80)	6MWT, 10MWT, WGS, Gait pattern, FIM, SIAS
Ogino [5]	Japan	Stroke	GEAR	Treadmill	GEAR by Welwalk WW-1000, /Japan	Treadmill	8: 6/2	11: 9/2	66.1 (9.6)	65 (7.7)	–	–	4	20	5	300 (GEAR: 200)	60 (GEAR: 40)	Step length, Stride length, Gait speed, Stance phase, Swing phase, Gait pattern, Circumduction
Ogino [6]	Japan	Stroke	GEAR	Treadmill	GEAR by Welwalk WW-1000, Japan	Treadmill	8: 6/2	11: 9/2	66.1 (9.6)	65 (7.7)	–	–	4	20	5	300 (GEAR / Treadmill: 200)	60 (GEAR / Treadmill: 40)	Stride length, Cadence, Gait speed, TUG, 6MWT, SF8, GRC
Tomida [7]	Japan	Stroke	GEAR	CPT	GEAR by Welwalk WW-1000, Japan	–	13: 11/2	13: 5/8	55.0 (9.0)	61.2 (12.9)	–	–	4	28	7	1120—1260 (GEAR / CPT: 280)	160—180 (GEAR / CPT: 40)	Gait pattern, SIAS-L/E, FIM-walk
Stein [8]	USA	Stroke	RAGT	Group exercise without Exo	Bionic Leg, AlterG (formerly Tibion), USA	–	12: 83%/17%	12: 58%/42%	57.6 (10.7)	56.6 (15.1)	-	-	6	18	3	180	60	TUG, 6MWT, 10MWT, Gait speed, Five-times-sit-to-stand test, CAFÉ 40, EFAP BBS, Romberg test
Rodríguez-Fernández [9]	Spain	SCI	ABLE	KAFO	ABLE, ABLE Human Motion, Spain	KAFO	5: 5/0	5: 4/1	46.6 (5.5)	41.6 (5.7)	26.1 (1.9)	25.3 (2.5)	5	8	2	180	90	Knee ROM, COM, Step length, Stance phase, Swing phase, Cadence, Gait speed, 6MWT, 10MWT, TUG, MCoT, VO2peak, FIM, PIADS, Circumduction, Circuity index, QUEST 2.0., AEs

Parkinson’s disease (PD), Osteoarthritis (OA), Multiple sclerosis (MS), Spinal cord injury (SCI), Exoskeleton (Exo), Robot-assisted gait therapy (RAGT), Conventional physical therapy (CPT), Gait exercise assist robot (GEAR), Treadmill gait training (TGT), Knee ankle foot orthosis (KAFO), Weeks (w), Sessions (s), Sessions per week (s/w), Minutes per week (min/w), Minutes per day (min/d), 6-Minute walk test (6MWT), Freezing of gait questionnaire (FoG-Q), Brief–Balance evaluation system test (B-BESTest), Activity-related balance confidence scale (ABC), The 39 item Parkinson's disease questionnaire (PDQ‐39), Scales for Outcomes in Parkinson’s-Cognition (SCOPA-COG), Unified Parkinson's disease rating scale (UPDRS), Hospital anxiety and depression scale (HADS), Activity counts per minute (ACPM), Adverse events (AEs), Timed up and go test (TUG), Timed stair test (TST) up and down (dn), MS Walking Scale (MSWS), Western Ontario and McMaster Universities Arthritis Index (WOMAC), Fatigue severity scale (FSS), Numeric Pain Rating Scale (NPRS), 36-item Short form health survey 36 (SF36), 8-item Short form health survey (SF8), 10-Meter walk test (10MWT), Wisconsin Gait Scale (WGS), Functional Independence Measure (FIM), Stroke Impairment Assessment Set (SIAS), SIAS-Lower extremities (SIAS-L/E), Global rating of change (GRC), Stance-swing Ratio (SSR), Gait Quality Index (GQI), Functional Ambulatory Category (FAC), California functional evaluation 40 (CAFÉ 40), Emory functional ambulation profile (EFAP), Berg balance scale (BBS), Range of motion (ROM), Center of mass (COM), Metabolic cost of transport (MCoT), Peak oxygen uptake (VO2peak), Psychosocial Impact of Assistive Devices Scale (PIADS), he Quebec User Evaluation of Satisfaction with Assistive Technology (QUEST 2.0)

Table 2.

The characteristics of non-randomized studies of interventions (NRSIs)

First Author, Year [ref]	Country	Injury	Groups		Device		Number (total: M/F)		Age (M (SD))		BMI		Intensity of Treatment					Reported Outcomes
			Intervention (I)	Control (C)	I	C	I	C	I	C	I	C	w	s	s/w	min/w	min/d
Huang [10]	Taiwan	Stroke	Keeogo + CPT	CPT	Keeogo (Wistron Medical Technology, Taiwan)	–	17: 11/6	81: 61/20	56.4 ± 11.4	57.1 ± 9.1	–	–	C: - I: 2—3	C: - I: 6	C: 5 I: 5 (Keeogo: 2—3)	C: 250 I: 330—370 (Keeogo: 80—120)	C: 50 I: 90 (Keeogo: 40)	C: 5MWT, 6MWT, Gait speed, BBS, BI, FuglUL I: ROM, TUG, 5MWT, 6MWT, Gait speed, BBS, BI, MMSE, FuglUL, FuglLL, FuglSE
Katoh [11]	Japan	Stroke	GEAR + CPT	CPT	GEAR (Welwalk WW-1000, Japan)	–	15: 8/7	15: 10/5	54.5 (11.6)	58.8 (7.6)	24.2 (4.2)	25.2 (5.5)	–	–	5	C: 5400 I: 5400 (GEAR: 200)	C: 180 I: 180 (GEAR: 40)	First double support, Second double support, Single support, Stance phase, Swing phase, Side step length, Cadence, Stride length, Gait speed, Gait pattern, Circumduction
Ii [12]	Japan	Stroke	Welwalk + CPT	CPT	Welwalk WW-1000 (Welwalk, Toyota Motor Corp., Aichi, Japan)	–	36: 22/14	36: 20/16	59 (9)	60 (8)	–	–	C: - I: 5.2 (3.1)	C: - I: 26—36	C: 7 I: 6—7 (Welwalk: 5—7)	C: 1260 I: 1080—1260 (Welwalk: 200—280)	C: 180 I: 180 (Welwalk: 40)	FIM-walk, SIAS-L/E, Days from stroke onset to discharge
Wang [13]	Japan	Stroke	GEAR + CPT	CPT	GEAR (Welwalk WW-1000, Japan)	–	10: 5/5	10: 5/5	60 (10)	59 (11)	–	–	C: 5 (3) I: 4 (3)	C: 25—35 I: 20—28	5—7	C: 900—1260 I: 900—1260 (GEAR: 200—280)	C: max 180 I: max 180 (GEAR: 40)	Step length, Stride length, Stance phase, Swing phase, Double support time, Cadence, Gait Speed, FIM-walk, SIAS-L/E, Gait pattern, Circumduction
Hirano [14]	Japan	Stroke	GEAR	CPT	GEAR (Welwalk WW-1000, Japan)	–	1: 1/0	15: 8/7	70	57.6 (10.7)	–	–	C: 4 I: 4	C: 28 I: 24 (20)	C: 7 I: 6 (GEAR: 5)	C: 1260 I: 1080 (GEAR: 200)	C: max 180 I: max 180 (GEAR: 40)	C: FIM-walk I: Gait speed, Dependence level in GEAR training, Knee extension assist level, Swing assist level, SIAS, Gait pattern
Hirano [15]	Japan	Stroke	GEAR + CPT	CPT	GEAR (Welwalk WW-1000, Japan)	Non-powered Orthosis	6: -	6: -	61 (11)	54 (12)	–	–	3.8 (0.8)	–	5	200	40	10MWT, FIM-walk, SIAS-L/E

Gait exercise assist robot (GEAR), Conventional physical therapy (CPT), Weeks (w), Sessions (s), Sessions per week (s/w), Minutes per week (min/w), Minutes per day (min/d), 5-Meter walk test (5MWT), 6-Minute walk test (6MWT), Berg balance scale (BBS), Barthel index (BI), Mini-Mental State Examination (MMSE), Fugl–Meyer assessment-Upper limb (FuglUL), Fugl–Meyer assessment-Lower limb (FuglLL), Fugl–Meyer assessment-Sensory (FuglSE), Range of motion (ROM), Timed up and go test (TUG), Functional Independence Measure (FIM), Stroke Impairment Assessment Set (SIAS), SIAS-Lower extremities (SIAS-L/E), 10-Meter walk test (10MWT)

Risk of bias assessment

Among the 9 RCTs included, the risk of bias varied across domains. While most trials demonstrated low or unclear risk in key aspects such as randomization, blinding, and outcome reporting, certain studies, particularly those by Ogino et al. [30], and McGibbon et al. [31], showed potential concerns due to insufficient details regarding participant selection and blinding procedures (Fig. 3A). These methodological gaps may affect the internal validity of their findings. Nevertheless, the majority of RCTs were judged to provide generally reliable data on robotic exoskeleton interventions.

Fig. 3.

Quality assessment of included studies. (A) RCTs’ quality assessment using cochrane risk of bias tool for randomized trials (ROB-I). (B) NRSIs’ quality assessment using cochrane risk of bias tool for non-randomized studies of intervention (ROBINS-I)

Regarding the 6 included NRSIs, quality assessment with the ROBINS-I tool revealed that most studies were at low to moderate risk of bias, primarily due to confounding, selection bias, and deviations from the intended interventions. In particular, challenges in adequately controlling for baseline differences and limited information on how outcomes were measured contributed to potential bias in some studies. As shown in Fig. 3B, only a few NRSIs did not reach a low risk in most domains. However, it is advisable to approach these findings with a degree of caution, given the quality of evidence derived from the NRSIs.

Quantitative meta-analysis outcomes

The results of the quantitative assessment of the included studies are summarized in Table 3.

Table 3.

Meta-analysis of the difference between RAGT and CPT, regarding gait clinical outcomes after physiotherapy

Gait outcome	Number of studies	SMD (95% CI)	I2 (%)	I2 P-value	P-value
6MWT (meters), Total	7	0.00 (− 0.32 to 0.32)	5.11	0.51	0.99
6MWT, subgroup (Excluding McGibbon (2021))	6	0.14 (− 0.21 to 0.49)	0.00	0.95	0.42
6MWT, subgroup by FU (1-month)	2	0.00 (− 0.73 to 0.74)	0.00	0.83	0.99
6MWT, subgroup by FU (3 months)	2	− 0.13 (− 0.88 to 0.63)	0.00	0.97	0.74
10MWT (second)	2	0.28 (− 0.39 to 0.96)	0.00	0.69	0.40
Gait Speed (meter/second)	5	0.40 (0.01 to 0.79)	0.00	0.52	0.04
Cadence (step/min)	2	0.32 (− 0.42 to 1.06)*	0.00	0.57	0.40
TUG (second)	5	0.36 (− 0.16 to 0.88)	41.36	0.15	0.17
TST (second)	2	0.94 (− 0.20 to 2.07)*	0.00	0.92	0.11
FIM walk (score)	3	0.08 (− 0.51 to 0.68)*	79.19	0.01	0.37
FIM motor (score)	2	2.16 (− 3.03 to 7.36)*	0.00	0.88	0.41

^* Weighted mean difference (WMD) was calculated

Robot-assisted gait therapy (RAGT), Conventional physical therapy (CPT), Standardized mean difference (SMD), Weighted mean difference (WMD), 6-Minute walk test (6MWT), Follow-up (FU), 10-Meter walk test (10MWT), Timed up and go test (TUG), Timed stair test (TST), Functional Independence Measure (FIM)

6-min walk test (6MWT), meter

A meta-analysis of seven eligible studies assessing the 6MWT revealed no statistically significant difference in walking distance improvements between participants undergoing exoskeleton-assisted gait training and those in control groups. The pooled SMD was 0.00 (95% CI –0.32 to 0.32), indicating no difference between groups (Fig. 4A). To address potential bias from outlier data, a meta-analysis was conducted by excluding the study by McGibbon et al. [32]. The resulting pooled SMD was 0.14 (95% CI –0.21 to 0.49), indicating a favorable but statistically no significant trend toward improved 6MWT performance in the intervention group (Fig. 4B). Sensitivity analysis showed that the SMD in 6MWT scores remained consistent across analyses (range of summary SMDs: 0.09–0.17), indicating the robustness of the meta-analysis model. A subgroup analysis by follow-up duration also showed no significant difference in 6MWT outcomes between the intervention and control groups, with a pooled SMD of 0.00 (95% CI –0.73 to 0.74) at 1-month follow-up (Fig. 4C).

Fig. 4.

Forest plots of standardized mean difference (SMD) in 6-Minute Walking Test (6MWT, meter), after Robot-assisted Gait Training as Intervention. Diamond represents the summary (pooled SMD) estimate, and its width indicates the corresponding 95% CI for the random-effects estimate. I2 test and Cochran’s Q statistic were used to assess the statistical heterogeneity (p < 0.10) across studies. (A) Meta-analysis of all seven eligible studies assessing the 6MWT. (B) Meta-analysis after excluding McGibbon (2021) Study. (C) Subgroup meta-analysis by Follow-up Time

10-m walk test (10MWT), second

The meta-analysis of two studies evaluating the 10MWT showed a pooled SMD of 0.28 (95% CI –0.39 to 0.96), indicating a medium effect favoring the exoskeleton group over the control group. However, this improvement was not statistically significant from pre-intervention to post-intervention. The consistency of a positive SMD across both studies supports the reliability of this finding. In the individual studies, Stein et al. (2021) observed a favorable improvement for the exoskeleton group compared to controls from pre-intervention to 1-month follow-up (SMD = 0.28; 95% CI –0.57 to 1.12). However, at the 3-month post-intervention follow-up, the results reversed, showing a small favor for the control group over the exoskeleton group (SMD = –0.03; 95% CI –0.91 to 0.84) (Fig. 5).

Fig. 5.

Forest plots of standardized mean difference (SMD) in 10-Meter Walking Test (10MWT, second), after Robot-assisted Gait Training as Intervention. Diamond represents the summary (pooled SMD) estimate, and its width shows the corresponding 95% CI for the random-effects estimate. I2 test and Cochran’s Q statistic were used to assess the statistical heterogeneity (p < 0.10) across studies

Gait speed, meters/second

The meta-analysis of five studies evaluating gait speed demonstrated a significant increase in gait speed following robotic knee exoskeleton interventions compared to controls at post-treatment (SMD = 0.40; 95% CI 0.01 to 0.79). Sensitivity analysis showed the SMD in gait speed (m/s) was consistent (range of summary SMDs: 0.24, 0.52), indicating that the meta-analysis model was robust (Fig. 6A). However, subgroup analysis stratified by follow-up duration revealed inconsistent benefits over time. A non-significant improvement in gait speed was observed at 1-month follow-up compared to baseline (SMD = 0.22; 95% CI –0.39 to 0.82). However, at 3-month follow-up, no improvement was noted, with the SMD indicating a slight favor for controls (SMD = –0.20; 95% CI –0.82 to 0.41) (Fig. 6B).

Fig. 6.

Forest plots of standardized mean difference (SMD) in Gait Speed (meters/second), after Robot-assisted Gait Training as Intervention. Diamond represents the summary (pooled SMD) estimate, and its width indicates the corresponding 95% CI for the random-effects estimate. I2 test and Cochran’s Q statistic were used to assess the statistical heterogeneity (p < 0.10) across studies. (A) Meta-analysis of all five eligible studies assessing the gait speed. (B) Subgroup meta-analysis by follow-up time

Cadence, steps/minute

The meta-analysis of two studies evaluating cadence revealed a non-significant improvement in cadence following robotic exoskeleton interventions (WMD = 0.32; 95% CI − 0.42 to 1.06), as shown in Fig. 7A. Subgroup analysis further corroborated this finding. Similar to the 10MWT outcomes, the cadence results exhibited a comparable pattern. In the individual study by Ogino et al. (2020), a favorable improvement was observed for the exoskeleton group compared 95% CI to controls from pre-intervention to 1-month follow-up (SMD = 0.31; 95% CI –0.60 to 1.23). However, at the 3-month post-intervention follow-up, the results reversed, with the control group showing a favorable change in cadence over the exoskeleton group (SMD = –0.62; 95% CI –1.59 to 0.35) (Fig. 7B). Certainty of evidence was low due to imprecision, given the small sample size and a wide confidence interval that included both possible harm and meaningful benefit (Table 4).

Fig. 7.

Forest plots of weighted mean difference (WMD) in Cadence (steps/minute), after robot-assisted gait training as intervention. Diamond represents the summary (pooled WMD) estimate, and its width indicates the corresponding 95% CI for the random-effects estimate. I² test and Cochran’s Q statistic were used to assess the statistical heterogeneity (p < 0.10) across studies. (A) Meta-analysis of Two Eligible Studies Assessing Cadence. (B) Subgroup Meta-analysis by Follow-up Time

Table 4.

GRADE assessment for outcomes of PKAE in gait rehabilitation

Papulation	Patients with a gait disorder
Intervention	Rehabilitation with PKAE
Comparison	Conventional physical therapy (CPT)
Outcomes	N of participants (studies)	Certainty of the evidence (GRADE)	Pooled estimates (95% CI)	Rationale
6MWT (meters)	154 (7 studies)	⨁⨁⨁⨁ High	SMD: 0.00 (− 0.32 to 0.32)	–
10MWT (second)	34 (2 studies)	⨁⨁⨁◯ Moderate	SMD: 0.28 (− 0.39 to 0.96)	Limited number of participants and contributing studies
Gait Speed (meter/second)	107 (5 studies)	⨁⨁⨁⨁ High	SMD: 0.40 (0.01 to 0.79)	–
Cadence (step/min)	29 (2 studies)	⨁⨁◯◯ Low	SMD: 0.32 (− 0.42 to 1.06)	The estimate was based on only two small studies (n = 29), and the confidence interval was wide, encompassing both potential harm and clinically important benefit
TUG (second)	106 (5 studies)	⨁⨁⨁⨁ High	SMD: 0.36 (− 0.16 to 0.88)	–
TST (second)	53 (2 studies)	⨁⨁◯◯ Low	WMD: 0.94 (0.20 to 2.07)	Downgraded by two levels due to imprecision and potential for risk of bias. As only two studies provided data, resulting in a wide confidence interval
FIM motor (score)	111 (2 studies)	⨁⨁◯◯ Low	WMD: 2.16 (− 3.03 to 7.36)	Limited number of studies. The pooled effect estimate covers a wide range of confidence intervals
FIM walk (score)	111 (2 studies)	⨁⨁⨁◯ Moderate	WMD: 0.32 (0.36 to 1.01)	Imprecision because the pooled estimate was based on only two studies, and heterogeneity among the studies

Powered knee-assisting exoskeleton (PKAE), Grading of Recommendations Assessment, Development and Evaluation (GRADE), 6-Minute walk test (6MWT), 10-Meter walk test (10MWT), Timed up and go test (TUG), Timed stair test (TST), Functional Independence Measure (FIM), Standardized mean difference (SMD), Weighted mean difference (WMD)

Timed up and go (TUG) test, second

The meta-analysis of five studies assessing the TUG test revealed favorable but non-significant effects of exoskeleton-assisted rehabilitation (SMD = 0.36; 95% CI –0.16 to 0.88). (Fig. 8). Sensitivity analysis showed the SMD in TUG time (seconds) was consistent (range of summary SMDs: 0.18, 0.56), indicating that the meta-analysis model was robust.

Fig. 8.

Forest plots of standardized mean difference (SMD) in timed up and go (TUG) test, (second), after robot-assisted gait training as intervention. Diamond represents the summary (pooled SMD) estimate, and its width indicates the corresponding 95% CI for the random-effects estimate. I2 test and Cochran’s Q statistic were used to assess the statistical heterogeneity (p < 0.10) across studies

Timed stair test (TST), second

Pooled results from two studies assessing the Keeogo exoskeleton in patients with OA and MS showed no significant difference in improvement of stair-climbing ability between the exoskeleton and control 95% CI groups (WMD = 0.94; 95% CI –0.20 to 2.07) (Fig. 9). However, our confidence in this finding was low due to the possible effects of inconsistency and imprecision on the calculated effect estimate (Table 4).

Fig. 9.

Forest plots of weighted mean difference (WMD) in timed stair test (TST, second), after Keeogo Gait Training as Intervention. Diamond represents the summary (pooled WMD) estimate, and its width shows the corresponding 95% CI for the random-effects estimate. I2 test and Cochran’s Q statistic were used to assess the statistical heterogeneity (p < 0.10) across studies

Functional independence measure (FIM), score

The meta-analysis of three studies assessing levels of independence showed mixed results. No significant improvement in the FIM-walk (WMD = 0.08; 95% CI –0.51 to 0.68) and FIM-motor (WMD = 2.16; 95% CI –3.03 to 7.36) scores were observed in the exoskeleton group compared to the control group (Figs. 10 and 11); overall certainty of evidence was low due to inconsistency and imprecision (Table 4). Regarding the FIM-walk score, a higher pooled WMD was observed when including only the two RCTs in the meta-analysis (WMD = 0.32; 95% CI –0.36 to 1.01). A lower WMD was found when the analysis included an NRSI, yielding a WMD of 0.08 (95% CI –0.51 to 0.68) (Fig. 10).

Fig. 10.

Forest plot of the effect of gait exercise assist robot (GEAR) as the Robotic Exoskeleton Intervention on Functional Independence Measure-walk (FIM-walk, score) in Subjects with Stroke. Diamond represents the summary weighted mean difference (pooled WMD) estimate, and its width shows the corresponding 95% CI for the random-effects estimate. Subgroup meta-analysis was conducted by study design. I2 test and Cochran’s Q statistic were used to assess the statistical heterogeneity (p < 0.10) across studies

Fig. 11.

Forest plot of the effect of gait exercise assist robot (GEAR) as the Robotic Exoskeleton Intervention on Functional Independence Measure-motor (FIM-motor, score) in Subjects with Stroke. Diamond represents the summary weighted mean difference (pooled WMD) estimate, and its width shows the corresponding 95% CI for the random-effects estimate. Subgroup meta-analysis was conducted by follow-up time. I2 test and Cochran’s Q statistic were used to assess the statistical heterogeneity (p < 0.10) across studies

Qualitative outcomes

Several outcomes could not be synthesized quantitatively due to limited pre–post data or methodological heterogeneity. These included durations of stance phase and swing phase, stride length, step length, joint range of motion, gait patterns, balance outcomes, metabolic cost of transport (MCoT), pain, fatigue, quality of life (QoL), psychosocial functioning, user satisfaction, adverse events (AEs), and the stroke impairment assessment set (SIAS). Across these measures, robotic interventions generally showed mixed results. Improvements were observed in specific gait parameters (e.g., stride length, abnormal gait components) and selected QoL domains, but these effects were not consistent across studies or populations. AEs were generally mild and infrequent, and user satisfaction was high, particularly regarding safety for certain devices (See Supplementary File 1 for full details of the qualitative outcomes).

Discussion

This systematic review and meta-analysis assessed the effects of powered knee-assisting exoskeletons (PKAEs) on gait rehabilitation. Human gait is a result of coordinated multi-joint interactions involving the hip, knee, and ankle. Among these, the knee joint contributes substantially to both the stance and swing phases and plays a key role in load acceptance, limb advancement, and shock absorption. From a design perspective, targeting the knee joint may offer a practical balance between mechanical simplicity and functional assistance, as it allows meaningful gait support while limiting device weight and structural complexity [33]. In total, 15 studies were included in this review, examining the comparative effects of PKAEs and CPT on gait-specific parameters (e.g., gait speed and cadence), performance-based walking and mobility tests (e.g., 6MWT, 10MWT, TUG, and TST), and functional independence measures (e.g., FIM). The meta-analysis revealed a significant pooled improvement in gait speed for PKAEs compared to CPT; however, all other outcomes were comparable between PKAEs and CPT. These findings suggest that knee exoskeletons yield better results than CPT in gait speed and produce comparable results in other aspects of recovery, emphasizing the need for further research to optimize their use across diverse clinical populations.

The 6MWT, a common measure of aerobic capacity and endurance [34], showed no significant difference in walking distance between groups, consistent with the meta-analysis by Yang et al. [35]. Excluding McGibbon et al., which studied individuals with OA [32], yielded slightly more favorable—but still non-significant—results. This highlights the need to consider patient population and disability type, as knee exoskeletons may offer greater benefits for neurological than orthopedic conditions, potentially influenced by disability characteristics and patient expectations. For OA patients, rehabilitation typically aims to restore natural joint function and independence without assistive devices, as they generally retain joint control and have fewer complications in adjacent joints [36]. In contrast, neurological conditions like stroke often involve multi-joint impairments and cognitive or sensory challenges [37], making the stabilizing and assistive functions of knee exoskeletons more critical for movement and compensation. This review included patients with stroke, PD, SCI, OA, and MS, encompassing diverse medical and physical challenges. OA patients typically have joint pain and stiffness but intact neurological function [38], whereas MS patients may face muscle weakness, spasticity, balance issues, and fatigue from nervous system damage [39]. Such differences influence walking strategies and the benefits gained from exoskeletons, making consistent conclusions difficult. The heterogeneous participant mix likely contributed to more variable and less interpretable findings.

The 10MWT, similar to the 6MWT, showed a modest advantage for the exoskeleton group, though the difference was not statistically significant. These results suggest some benefits, but they are not conclusive. Postol et al.'s systematic review on lower limb robotic exoskeletons for acquired brain injury rehabilitation also reported similar findings for the 10MWT [40]. Cadence, which refers to the number of steps taken per second [41], showed a modest, non-significant improvement with robotic exoskeletons, similar to the 6MWT and 10MWT. Interestingly, Ogino et al. (2020) found that after one month, the exoskeleton group had a higher step frequency, but by three months, the control group outperformed them [42]. This suggests that initial benefits may diminish over time, highlighting a challenge with robotic exoskeletons: short-term gains may not be sustained without additional support, potentially due to user adaptation difficulties or the need for further device customization.

The gait speed outcome, unlike the other assessed parameters, showed a significant advantage for the exoskeleton group in our analysis. Notably, among the studies included in the gait speed meta-analysis, Huang et al. (2024) reported a greater improvement in the intervention group compared with the control group and substantially influenced the overall pooled effect in favor of the exoskeleton group [43]. However, this study was an NRSI, whereas the other studies included in this analysis were RCTs, which provide a higher level of evidence. Therefore, the impact of this NRSI should be carefully considered when interpreting the observed superiority of the exoskeleton group in gait speed. In contrast, Hunt et al.’s systematic review reported inconsistent findings regarding gait speed across studies evaluating different types of robotic exoskeletons in children with CP. They attributed this variability to factors such as adaptation time, training duration, stride characteristics, and optimization of device torque, noting that limited familiarization and suboptimal assistance levels may reduce improvements in walking speed [44]. Methodological differences between the two reviews should also be acknowledged. Hunt et al. examined hip, knee, and ankle exoskeletons exclusively in children with CP, whereas our review focused solely on powered knee exoskeletons and included a broader patient population, with direct comparisons between intervention and control groups over time.

The TUG test, assessing functional mobility and balance [45], showed a favorable trend for the exoskeleton group, suggesting potential improvements in dynamic mobility, including standing and initiating walking. While these improvements may vary across individuals and lack statistical significance at the group level, they indicate potential clinical relevance, especially for populations with balance issues or movement initiation difficulties, where even small gains could enhance independence and safety. The TST, used to assess strength and coordination [46], showed no significant difference between the exoskeleton and control groups. Stair climbing involves technique, balance, strength, and timing, making it more complex than level walking. It also requires more power and energy [47], which knee exoskeletons may not fully provide, especially for individuals with significant impairments.

The results on functional independence were mixed. The FIM-walk score, measuring walking ability [48], showed a modest improvement in the exoskeleton group, but with a statistically nonsignificant difference compared to the control groups. This effect became more pronounced when only high-quality studies (RCTs) were included, suggesting that the design of the study is important and that RCTs may help minimize potential biases. On the other hand, the FIM-motor score, assessing overall daily movement, showed no significant improvement. This difference may be due to the distinct nature of walking versus daily activities. Knee exoskeletons are designed to support walking, not the full range of movements needed for independence. This explains the improvement in the FIM-walk score but not in the FIM-motor score, which covers activities like bathing, dressing, and transitions between sitting and standing [49], not directly aided by knee exoskeletons. Thus, while walking ability improves, broader motor function may not. Limited improvement in independence may also stem from patients not fully relying on the exoskeleton. Many individuals remain cautious, preferring to use their own strength or seek therapist support instead of depending on the device. This reluctance, due to factors like lack of trust, fear of falling, or discomfort, may prevent the exoskeleton from reaching its full potential [50].

The findings from this meta-analysis on gait speed indicate that PKAEs significantly improve gait speed compared to controls. However, the results from the 6MWT, 10MWT, cadence, TUG, TST, and FIM consistently show that, although robotic knee exoskeletons lead to improvements, these changes do not reach statistical significance. This suggests that, beyond clinical data, human-centered factors may limit their full potential. Much like toddlers learning to walk, users require an extended adaptation period to develop coordination, balance, muscle engagement, and trust in the device. The initial learning phase often involves awkward movements and a mismatch between intended actions and device responses [51], which may explain why early gains, such as the cadence improvement reported by Ogino et al. (2020) at one month, tend to diminish in later follow-ups, with the control group sometimes outperforming [42].

Short-term interventions may therefore be insufficient for sustained improvements in outcomes such as the 6MWT, TUG, or cadence, as adaptation to a new motor strategy requires time, repetition, and continued support [52]. Psychological and behavioral factors also play a role; some patients perceive exoskeletons as rigid or impersonal, limiting motivation and engagement, which could explain the absence of significant improvements in measures like the TUG and 10MWT [53]. Furthermore, as with traditional orthotics, overreliance on the device may unintentionally reduce muscle activation if users expect it to carry most of the load. This could potentially contribute to the modest gains in the FIM-walk score and the lack of significant changes in the FIM-motor score, as patients may not be sufficiently engaging their muscles or retraining their neuromuscular systems [54].

Strengths and limitations

This systematic review and meta-analysis had several strengths. The inclusion of both RCTs and NRSIs enabled a comprehensive examination of PKAEs across diverse clinical populations. Additionally, the use of the GRADE framework to assess evidence quality provides valuable insight into the reliability of the findings. However, there were some limitations. First, the inclusion of NRSIs, while at low-to-moderate risk of bias, introduces potential confounding and selection bias. Second, clinical heterogeneity among the patient populations (neurological vs. orthopedic) adds variability to the results. Third, small sample sizes and few studies for some outcomes limit statistical power, especially for measures like TST and FIM. Finally, the absence of long-term follow-up data prevents conclusions about the sustained effects of PKAE interventions. Future research should address these limitations with larger, more standardized studies and longer follow-ups.

Clinical implications

This review suggests that PKAEs are comparable to CPT for gait rehabilitation, but they are not yet recommended for routine clinical practice due to the lack of significant improvements in several outcomes. PKAEs may be most beneficial for neurological conditions such as stroke, MS, and SCI, where gait assistance is more critical. For orthopedic conditions like OA, the benefits may be less pronounced. Implementing PKAEs in practice requires resources such as specialized training for clinicians, time for patient adaptation, and addressing the high cost of the devices. Further research is needed to refine these devices and improve accessibility for clinical use.

Recommendations for future research

Future studies on PKAEs should focus on more homogeneous populations, such as individuals with chronic stroke, to reduce variability in treatment response and improve the consistency of results. Adopting standardized core outcome sets, including the gait speed, 6MWT, 10MWT, and TUG, will allow for easier comparison across studies and more robust conclusions. Additionally, future research should incorporate follow-up durations of at least 3 months to capture the long-term impact of PKAEs, as short-term outcomes may not fully reflect the sustained effects. Measuring user experience, adherence, and neuromuscular engagement (e.g., using electromyography, EMG) will provide valuable insights into how users interact with the devices and the feasibility of long-term use. Finally, studies should report specific device parameters such as the level of assistance provided, session duration, and adjustments made to the devices. This will help standardize interventions and improve the reproducibility of future trials.

Conclusions

The findings of this review suggest that powered knee-assisting exoskeletons (PKAEs) show significant improvement over conventional physiotherapy (CPT) only in gait speed, with no differences observed in other gait parameters, such as cadence, 6MWT, 10MWT, TUG, TST, and FIM. Although both groups showed improvements from baseline, the only statistically significant advantage of PKAEs over CPT was in gait speed. Overall, PKAEs perform better than CPT in gait speed and are comparable in other gait rehabilitation parameters. However, further studies with longer treatment durations, a wider range of patient populations, and more diverse clinical profiles are necessary to fully assess their long-term potential.

Data availability statement

All data analyzed in this review are derived from previously published studies, which are available in the public domain. Correspondence and requests for materials should be addressed to mahladalirib96@gmail.com.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

MS

Multiple sclerosis

PD

Parkinson's disease

KAFOs

Knee-ankle–foot orthoses

SCI

Spinal cord injury

RAGT

Robot-assisted gait training

RCT

Randomized controlled trial

FIM

Functional independence measure

PKAEs

Powered knee-assisting exoskeletons

6MWT

6-Minute walk test

10MWT

10-Meter walk test

TUG

Timed up and go test

PROSPERO

Prospective register of systematic reviews

PRISMA

Preferred reporting items for systematic reviews and meta-analyses

TST

Timed stair test

SIAS

Stroke impairment assessment set

NRSIs

Non-randomized studies of interventions

RoB-I

Cochrane risk of bias tool for randomized trials

ROBINS-I

Cochrane risk of bias tool for non-randomized studies of interventions

GRADE

Grading of recommendations assessment, development, and evaluation

WMD

Weighted mean difference

SMD

Standardized mean difference

OA

Osteoarthritis

CPT

Conventional physical therapy

MCoT

Metabolic cost of transport

QoL

Quality of life

AEs

Adverse events

%GC

% Of gait cycle

cm

Centimeter

%BH

% Of body height

m

Meter

ROM

Range of motion

3D

3-Dimensional

PCOM

Projected center of mass

BBS

Berg Balance Scale

s

Second

ABC

Activities-specific balance confidence

B-BESTest

Brief–balance evaluation system test

ml/kg/m

Milliliters of oxygen/kilogram/meter

%VO₂peak

Peak oxygen uptake

FSS

Fatigue severity scale

NPRS

Numeric pain rating scale

WOMAC

Western Ontario and McMaster Universities Osteoarthritis Index

SF-36

Short Form

HADS

Hospital anxiety and depression scale

PIADS

Psychosocial impact of assistive devices scale

QUEST 20

Quebec user evaluation of satisfaction with assistive technology

SIAS-L/E

SIAS-lower extremity

Authors' contributions

M.MA.: Data collection, Data curation, Writing – original draft, Writing – review & editing, Literature review. S.Ebh.: Data collection, Data curation, Writing – original draft, Writing – review & editing, Literature review. A.M.: Conceptualization, Investigation, Resources, Supervision. M.S.: Data analysis, Methodology, Investigation. A.A.: Conceptualization, Investigation, Writing – review & editing. M.Sh.: Data analysis, Methodology. I.K.: Investigation, Writing – review & editing. MH.G.: Data collection. M.F.: Data collection. S.DA.: Data collection. N.J.: Investigation, Conceptualization M.D.: Conceptualization, Resources, Writing – review & editing, Supervision, Project administration. All authors read and approved the final manuscript.

Funding

The authors did not receive support from any organization for the submitted work.

Declarations

Competing interests

The authors declare no competing interests.

Consent for publication

Not applicable.

Ethics approval

Not Applicable. Ethics approval from the Ethics Committee of Mashhad University of Medical Sciences was waived since no new data were collected, and this was a review study.

Study registration

Registered in the International Prospective Register of Systematic Reviews (PROSPERO) with the code CRD42024626418.

Statement of the location

Orthopedics Research Center, Ghaem Hospital, Mashhad University of Medical Sciences, Mashhad, Iran.

Declaration of Generative AI and AI-assisted technologies in the writing process

ChatGPT (OpenAI, GPT-4) was used only to improve the grammar and language clarity of the manuscript. No AI tool was used for generating scientific content, data analysis, interpretation of results, or drawing conclusions. The authors take full responsibility for the content of this work.