The Improvement of Walking Ability Following Stroke
A systematic review and network meta-analysis of randomized controlled trials
; ; ;
Background: Gait velocity and maximum walking distance are central parameters for measuring the success of rehabilitation of gait after a stroke. The goal of this study was to provide an overview of current evidence on the rehabilitation of gait after a stroke.
Methods: A systematic review of randomized, controlled trials was carried out using network meta-analysis. The primary endpoint was gait velocity; secondary endpoints were the ability to walk, maximum walking distance, and gait stability. The following interventions were analyzed: no gait training, conventional gait training (reference category), training on a treadmill with or without body weight support, training on a treadmill with or without a speed paradigm, and electromechanically assisted gait training with end-effector or exoskeleton apparatus.
Results: The systematic search yielded 40 567 hits. 95 randomized, controlled trials involving a total of 4458 post-stroke patients were included in the meta-analysis. With respect to the primary endpoint of gait velocity, gait training assisted by end-effector apparatus led to significant improvement (mean difference [MD] = 0.16 m/s; 95% confidence interval [0.04; 0.28]). None of the other interventions improved gait velocity to any significant extent. With respect to one of the secondary endpoints, maximum walking distance, both gait training assisted by end-effector apparatus and treadmill training with body weight support led to significant improvement (MD = 47 m, [4; 90], and MD = 38 m, [4; 72], respectively). A network meta-analysis could not be performed with respect to the ability to walk (a different secondary endpoint) because of substantial inconsistencies in the data. The interventions did not differ significantly with respect to safety.
Conclusion: In comparison to conventional gait rehabilitation, gait training assisted by end-effector apparatus leads to a statistically significant and clinically relevant improvement in gait velocity and maximum walking distance after stroke, while treadmill training with body weight support leads to a statistically significant and clinically relevant improvement in maximum walking distance.
Stroke is one of the most frequently occurring diseases worldwide and leads to permanent disability, diminished quality of life, and thus to a heavy burden of illness. A high proportion of stroke patients have impaired walking ability and can only walk in their own home. Their reduced mobility often means they are unable to go outdoors at all. Approximately 70% of those who retain the ability to walk cannot move at a normal speed and are therefore limited in daily activities such as crossing the road at a stop light (1). Regaining the ability to walk at a speed approaching normal is thus one of the principal goals for stroke patients and their family members.
In recent years interventions such as treadmill training and electromechanical-assisted training have been introduced to help improve walking after stroke (2). During treadmill training the patient is secured by a belt system that bears part of the body weight (3, 4). Another approach is treadmill training with systematic increase of the walking speed (5). In electromechanical-assisted training the patient’s gait cycle is partly automated, which eases the work of the therapist (6). This method increases the number of steps that can be taken during treatment sessions and enables severely affected patients to practice walking earlier and more intensively than was possible previously (7). The GT-1 walking trainer is an example of the end-effector type (1), while exoskeleton models are represented by the Lokomat and LOPES trainers (6, 8). Moreover, studies published particularly in the past few years have described mobile exoskeletons (9–11) and special “limb robots” (12–14).
The exoskeleton system consists of a treadmill and an exoskeleton, i.e., an orthosis with rods and joints designed to imitate the skeleton of the lower extremities that is adapted to the dimensions of each individual patient (1). Integrated into the exoskeleton are programmable power units that move the hip and knee joints during ambulation. The feet are also led or controlled by the device (1). In the end-effector system the patient, secured by straps, stands on two footplates that simulate walking (1). The device moves only the feet, fixed to the footplates; The knee and hip joints follow and are not controlled by the device but have to be actively moved by the patient (1).
Although the evidence on training stroke patients to walk seems robust, no review has yet been compiled that summarizes and evaluates the results of all studies and interventions regarding the improvement of walking ability after a stroke. The existing Cochrane Reviews, for example, have a narrow focus such as the efficacy of treadmill training or the efficacy of electromechanical-assisted rehabilitation of walking (4, 15). However, there are hardly any comparisons of two or more interventions to improve walking ability, although in practice it is crucial to know which device performs more effectively than others in a given situation. The treating physician also encounters difficulties in deciding which specific form of treatment to prescribe for a stroke patient.
An approach to solving this problem is offered by network meta-analyses. These enable quantitative synopsis of the “evidence network” by combining direct and indirect comparisons of three or more interventions in randomized, controlled trials on the basis of a common comparative intervention (16).
We set out to gain an overview of the evidence from randomized, controlled trials on the improvement of walking speed, walking distance, walking ability, and safety in stroke patients. A further aim was to estimate the relative efficacy of the various interventions, taking effect modifiers into account.
Study protocol and registration
The study protocol for this systematic review is registered in the PROSPERO database under the ID CRD42017056820 and meets the PRISMA criteria (17).
Inclusion and exclusion criteria
Our analysis embraced all published and unpublished studies on adults following stroke (clinically defined). We compared all types of training designed to improve the walking speed, walking distance, and walking ability of stroke patients. All randomized, controlled trials of parallel-group design and randomized crossover studies that compared walking training with other interventions were included. We combined comparable interventions and approaches into treatment categories.
The primary endpoint was walking speed, while the secondary endpoints were walking ability, walking distance, and walking safety.
We defined the following categorization of study interventions in advance:
- No walking training
- Conventional training (walking on the floor, preparatory exercises in sitting position, balance training etc. without technical aids and without treadmill training or electromechanical-assisted training) (reference category)
- Treadmill training without or with body-weight support
- Treadmill training with or without walking speed paradigm
- Electromechanical-assisted training with end-effector devices or exoskeletons
Our systematic survey yielded 44 567 records. After exclusion of irrelevant records, 95 randomized controlled trials with a total of 4458 patients were included for quantitative analysis (Figure 1).
Of the 95 publications included, 80% were randomized controlled trials and the remaining 20% were randomized crossover studies. The trial size ranged from five to 282 patients (mean: 26 patients). The patients’ mean age ranged from 43 to 76 years (eTable 1). The mean time elapsed since stroke was 3 days to 8 years. Altogether, 92 of the 95 trials compared an active experimental group with an active control group (eTables 2–4).
Ninety-two (97%) of the 95 publications included reported proper generation of the randomization sequence, 72 (76%) stated adequate concealment of the randomization sequence, and 77 (81%) confirmed satisfactory blinding of the investigators. The methodological quality of the trials, depicted in eFigures 1–3 and eTables 2–4, was included as a covariable in the calculations (adjusted effect mass). SUCRA (surface under the cumulative ranking curve) presentation of the endpoints can be found in eTables 5–7.
Summary of network geometry
Walking speed was used as an endpoint in 75 studies with a total of 3614 patients. Most of the trials compared treadmill training against walking rehabilitation without treadmill training (Figure 2 and eFigures 1–5).
Walking distance was the secondary endpoint in 44 trials with a total of 2509 patients. In these studies too, the majority compared treadmill training against walking rehabilitation without treadmill training (Figure 3 and eFigure 6).
Achievement of walking ability was a secondary endpoint in 22 studies with a total of 1517 patients. Most of these trials compared electromechanical-assisted walking training with walking training that did not involve electromechanical assistance (eFigure 3 and eTable 3).
The secondary endpoint safety was reported in 57 trials with a total of 2889 patients, most of which compared electromechanical-assisted walking training with walking training that did not involve electromechanical assistance (Figure 4 and eFigure 7).
The network structure and geometry are described in more detail in the eMethods.
For the primary endpoint of walking speed, end-effector-assisted training achieved significantly greater improvements than conventional walking rehabilitation (mean difference [MD] = 0.16 m/s, 95% confidence interval [CI]: [0.04; 0.28]). None of the other interventions improved walking speed significantly (Figure 2).
With regard to the secondary endpoint of walking distance, both end-effector-assisted training and treadmill training with body-weight support increased the distance walked significantly more than conventional walking rehabilitation (MD = 47 m, 95% CI: [4; 90] and MD = 38 m, 95% CI: [4; 72], respectively). No other interventions improved walking distance significantly in comparison with conventional walking rehabilitation (Figure 3).
No network analysis was carried out for the secondary endpoint of walking ability owing to statistically relevant inconsistency; the central precondition of transitivity was infringed. No approach was statistically significantly superior to any other approach.
Altogether 42 studies with a total of 2207 patients were included for analysis. At the end of treatment 639 patients (29%) were able to walk. Seventy study arms with a total of 1572 patients investigated the efficacy of conventional walking rehabilitation, while 21 study arms with 415 patients examined the efficacy of treadmill training. A detailed account of all trials with regard to patient and study characteristics, age, interventions, and walking ability can be found in eTables 1–4.
As for the secondary endpoint of safety, we found no systematic differences among the various interventions for walking rehabilitation following stroke.
Our sensitivity analysis revealed no significant difference in study effects with regard to the methodological quality of the trials included.
Our systematic review and network meta-analysis embraced a total of 95 trials with 4458 patients. The special feature of this network meta-analysis is that for the first time, competing methods for improvement in walking following stroke are evaluated together and rendered directly statistically comparable with one another, thus enabling nuanced assessment of their effect. Our work can be viewed as complementing the existing Cochrane Reviews. Evaluation of the network meta-analysis showed that electromechanical control of the leg from distal (the end-effector principle) improves walking speed significantly more than conventional walking rehabilitation. The mean increase of 0.16 m/s (corresponding to 0.58 km/h) achieved by end-effector-assisted training is clinically meaningful (27).
For walking distance, it emerged that both an end-effector method and treadmill training with body-weight support can be expected to be superior to conventional walking rehabilitation in increasing the distance walked. According to Flansbjer the smallest clinical improvement was 0.15 to 0.25 m/s in walking speed and 37 to 66 m in walking distance in the 6-minute walking test (27).
The mean improvement over conventional walking rehabilitation of 38 m and 47 m, respectively, in the 6-minute walking test lies in the lower range of clinical relevance but can still be regarded as meaningful (27).
No statements were made with regard to achievement of walking ability. We refrained from statistical evaluation because of the clear statistical inconsistency in the evidence network (26). The individual studies, the interventions used, and the patient characteristics were therefore described qualitatively instead (eTable 3).
Overall, the number of adverse events was relatively low in all studies and the safety level therefore high. No systematic differences were found among the various interventions for walking training following stroke (eTable 4).
Comparison of results with previously published data
Previous reviews of walking rehabilitation after stroke have had a much narrower focus, e.g., the efficacy of treadmill training (15), electromechanical-assisted training (4), or repetitive conventional training (28). The advantage and novelty of the network analysis presented here lie in its inclusion of randomized controlled trials on various methods of walking rehabilitation in one common statistical analysis.
It is well known that treadmill training is appropriate for stroke patients who can already walk (15), and electromechanical-assisted training above all for non-ambulant patients (15, 28). Our network analysis shows that distally supportive electromechanical-assisted training is best for increasing walking speed following stroke and treadmill training with body-weight support best for improving the walking distance. This analysis supplements the existing evidence with the confirmation that the walking training for stroke patients should be highly repetitive with (distal) partial support, rather than relying on complete assistance systems.
In agreement with earlier publications, our analysis points to superiority of walking training with end-effector devices over conventional walking rehabilitation (4, 6). However, there are no controlled trials directly comparing the efficacy of the various devices available.
We applied a systematic, comprehensive strategy to search various databases for published and ongoing trials. Nevertheless, publication bias cannot be entirely ruled out because negative results may not have been submitted for publication.
Inconsistent description of treatments by different authors could possibly have resulted in excessively heterogeneous intervention categories, which would limit the generalizability of the findings. However, prior to statistical evaluation we discussed how best to define the intervention groups and then compare them statistically.
One could argue that the treatments within both the control group and the experimental group were heterogeneous. However, on the basis of the information provided in the studies included we strove to categorize all treatments to the best of our ability.
The described effects of some individual interventions—for both walking speed and walking distance—were not only statistically significant but also clinically meaningful. However, no conclusions could be drawn for walking ability in general. We selected a conservative approach and did not perform a network analysis for this parameter; rather, we described the studies in qualitative terms.
It could be reasoned that the initial degree of disability following stroke was a source of bias in the joint analysis of all patients. In this network analysis we used walking ability as one aspect of disability following stroke and employed it as a covariable in the statistical evaluation. However, the fact that no account was taken of other variables, such as stroke site, may have distorted the results—although it is not clear in which direction.
A further potential criticism lies in our categorization of the selected interventions. It could be that certain assisted interventions were used particularly in more severely affected patients (e.g., those who could not walk), as recommended in the current guidelines. However, closer inspection of the studies shows that not all study authors adhered to the latest guideline recommendations. A glance at the tabulated presentation of the interventions in the individual trials (eTables 1 and 2) reveals that sometimes mildly affected patients were treated with robotic systems and severely affected patients with treadmill systems, contrary to the recommendations in the guidelines. The effect and the direction of such a distortion on the basis of the study data cannot be assessed with any accuracy.
One can also voice the criticism that we used only the mean values from each trial, not the data from every individual patient. Undoubtedly much more precise estimates of the different effects could have been made on the basis of individual patient data, but this exceeded the remit of our study.
One limitation of our systematic review and network meta-analysis is that we did not include mobility, falls, and quality of life as endpoints. We chose to concentrate on endpoints clinically relevant to walking ability, i.e., walking speed and distance, that are also very important for patients in their recovery from stroke. Nevertheless, further studies should focus particularly on other endpoints such as activities of daily life, mobility, social participation, and also falls.
Our findings show that highly repetitive electromechanical-assisted training is probably the best intervention for improving the walking speed of stroke patients. Walking distance is most likely to be increased by end-effector-assisted training and treadmill training with body-weight support. These results have important consequences for the neurological rehabilitation of stroke patients with impaired walking ability, in that device-supported training must be universally integrated into rehabilitation practice. Furthermore, the findings have considerable implications for the practice of community and inpatient physiotherapy and for the financing of such treatment in the out-of-hospital setting. A change of direction is required—away from special physiotherapy employing neurophysiological techniques (29) towards device-supported walking rehabilitation.
Future studies should investigate both the number of repetitions and the intensity and escalation of treatment in walking rehabilitation for stroke patients. Forthcoming systematic reviews should include individual patient data to enhance the accuracy of description of the effects of walking training.
Conflict of interest statement
The authors declare that no conflict of interest exists.
Manuscript submitted on 5 December 2017, revised version accepted on 29 May 2018
Translated from the original German by David Roseveare
Prof. Dr. rer. medic. Jan Mehrholz
Medizinische Fakultät der Technischen Universität Dresden
Fiedlerstr. 27, 01307 Dresden, Germany
eMethods, eFigures, eTables:
Prof. Jan Mehrholz, Prof. Joachim Kugler, Prof. Bernhard Elsner
SRH Hochschule für Gesundheit, University of Applied Health Sciences: Prof. Jan Mehrholz,
Prof. Bernhard Elsner
Helios Klinik Schloss Pulsnitz: Prof. Marcus Pohl
|1.||Hesse S, Mehrholz J, Werner C: Robot-assisted upper and lower limb rehabilitation after stroke. Dtsch Arztebl Int 2008; 105: 330–6 VOLLTEXT|
|2.||Dohle C, Tholen R, Wittenberg H, Saal S, Quintern J, Stephan K: S2e-Leitlinie: Rehabilitation der Mobilität nach Schlaganfall (ReMoS). Neurologie und Rehabilitation 2015; 21: 355–494.|
|3.||Mehrholz J, Thomas S, Elsner B: Treadmill training and body weight support for walking after stroke. Cochrane Database Syst Rev 2017: CD002840 MEDLINE|
|4.||Mehrholz J, Thomas S, Pohl M, Werner C, Kugler J, Elsner B: Electromechanical-assisted training for walking after stroke. Cochrane Database Syst Rev 2017: CD006185 MEDLINE|
|5.||Pohl M, Mehrholz J, Ritschel C, Ruckriem S: Speed-dependent treadmill training in ambulatory hemiparetic stroke patients: a randomized controlled trial. Stroke 2002; 33: 553–8 CrossRef|
|6.||Mehrholz J, Pohl M: Electromechanical-assisted gait training after stroke. A systematic review comparing endeffector and exoskeleton devices. J Rehabil Med 2012; 44: 193–9 CrossRef MEDLINE|
|7.||Werner C, von Frankenberg S, Treig T, Konrad M, Hesse S: Treadmill training with partial body weight support and an electromechanical gait trainer for restoration of gait in subacute stroke patients: a randomized crossover study. Stroke 2002; 33: 2895–901 CrossRef|
|8.||Colombo G, Joerg M, Schreier R, Dietz V: Treadmill training of paraplegic patients using a robotic orthosis. J Rehabil Res Dev 2000; 37: 693–700 MEDLINE|
|9.||Buesing C, Fisch G, O‘Donnell M, et al.: Effects of a wearable exoskeleton stride management assist system (SMA) on spatiotemporal gait characteristics in individuals after stroke: a randomized controlled trial. J Neuroeng Rehabil 2015; 12: 69 CrossRef MEDLINE PubMed Central|
|10.||Stein J, Bishop L, Stein DJ, Wong CK: Gait training with a robotic leg brace after stroke. Am J Phys Med Rehabil 2014; 93: 987–94 CrossRef MEDLINE|
|11.||Watanabe H, Tanaka N, Inuta T, Saitou H, Yanagi H: Locomotion improvement using a hybrid assistive limb in recovery phase stroke patients: a randomized controlled pilot study. Arch Phys Med Rehabil 2014; 95: 2006–212 CrossRef MEDLINE|
|12.||Forrester LW, Roy A, Hafer-Macko C, Krebs HI, Macko RF: Task-specific ankle robotics gait training after stroke: a randomized pilot study. J Neuroeng Rehabil 2016; 13: 51 CrossRef MEDLINE PubMed Central|
|13.||Forrester LW, Roy A, Krywonis A, Kehs G, Krebs HI, Macko RF: Modular ankle robotics training in early subacute stroke: a randomized controlled pilot study. Neurorehabil Neural Repair 2014; 28: 678–87 CrossRef MEDLINE PubMed Central|
|14.||Waldman G, Yang CY, Ren Y, et al.: Effects of robot-guided passive stretching and active movement training of ankle and mobility impairments in stroke. NeuroRehabilitation 2013; 32: 625–34 MEDLINE|
|15.||Mehrholz J, Thomas S, Elsner B: Treadmill training and body weight support for walking after stroke. Cochrane Database Syst Rev 2017: CD002840 MEDLINE|
|16.||Mills EJ, Bansback N, Ghement I, et al.: Multiple treatment comparison meta-analyses: a step forward into complexity. Clin Epidemiol 2011; 3: 193–202 CrossRef MEDLINE PubMed Central|
|17.||Hutton B, Salanti G, Caldwell DM, et al.: The prisma extension statement for reporting of systematic reviews incorporating network meta-analyses of health care interventions: checklist and explanations. Ann Intern Med 2015; 162: 777–84 CrossRef MEDLINE|
|18.||StataCorp LLC: Stata Statistical Software: Release 15. College Station, TX, 2017. www.stata.com/features/documentation/ (last accessed on 1 March 2018).|
|19.||Higgins JPT, Altman DG, Sterne JAC: Assessing risk of bias in included studies. In: Higgins JPT, Green S (editors). Cochrane handbook for systematic reviews of interventions version 5.1.0 [updated March 2011]. The Cochrane Collaboration, 2011. www.cochrane-handbook.org (last accessed on 1 March 2018).|
|20.||Chaimani A, Higgins JP, Mavridis D, Spyridonos P, Salanti G: Graphical tools for network meta-analysis in STATA. PLoS One 2013; 8: e76654 CrossRef MEDLINE PubMed Central|
|21.||Chaimani A, Salanti G: Visualizing assumptions and results in network meta-analysis: the network graphs package. Stata J 2015; 15: 905–50.|
|22.||Dias S, Welton NJ, Caldwell DM, Ades AE: Checking consistency in mixed treatment comparison meta-analysis. Stat Med 2010; 29: 932–44 CrossRef MEDLINE|
|23.||Hutton B, Salanti G, Caldwell DM, et al.: The PRISMA extension statement for reporting of systematic reviews incorporating network meta-analyses of health care interventions: checklist and explanations. Ann Intern Med 2015; 162: 777–84 CrossRef MEDLINE|
|24.||Kiefer C, Sturtz S, Bender R: Indirect comparisons and network meta-analyses: estimation of effects in the absence of head-to-head trials—part 22 of a series on evaluation of scientific publications. Dtsch Arztebl Int 2015; 112: 803–8 VOLLTEXT|
|25.||Salanti G, Ades AE, Ioannidis JP: Graphical methods and numerical summaries for presenting results from multiple-treatment meta-analysis: an overview and tutorial. J Clin Epidemiol 2011; 64: 163–71 CrossRef MEDLINE|
|26.||White IR, Barrett JK, Jackson D, Higgins JP: Consistency and inconsistency in network meta-analysis: model estimation using multivariate meta-regression. Res Synth Methods 2012; 3: 111–25 CrossRef MEDLINE PubMed Central|
|27.||Flansbjer UB, Holmback AM, Downham D, Patten C, Lexell J: Reliability of gait performance tests in men and women with hemiparesis after stroke. J Rehabil Med 2005; 37: 75–82 CrossRef MEDLINE|
|28.||French B, Thomas LH, Coupe J, et al.: Repetitive task training for improving functional ability after stroke. Cochrane Database Syst Rev 2016; 11: CD006073 MEDLINE|
|29.||Heilmittelkatalog 2017. www.heilmittelkatalog.de/files/luxe/hmkonline/physio/zn2.htm (last accessed on 1 May 2018).|
Walking Instability in Constant Speed and Variable Speed Treadmill Walking Evaluated by the Maximum Lyapunov Exponent of Chaos AnalysisRigakuryoho Kagaku, 201910.1589/rika.34.607
An Exploratory Clinical Study on an Automated, Speed-Sensing Treadmill Prototype With Partial Body Weight Support for Hemiparetic Gait Rehabilitation in Subacute and Chronic Stroke PatientsFrontiers in Neurology, 202010.3389/fneur.2020.00747
Cochrane Database of Systematic Reviews, 202010.1002/14651858.CD006185.pub5
Applied Sciences, 202010.3390/app10217626
Deutsches Aerzteblatt Online, 201910.3238/arztebl.2019.0174b
Deutsches Aerzteblatt Online, 201910.3238/arztebl.2019.0174a