Searches were conducted from six databases: MEDLINE, Ovid Healthstar, Cochrane Database of Systematic Reviews, Embase, Web of Science and CINAHL. The following search strings and keywords were used to carry out the search: multiple sclerosis OR (multiple sclerosis AND chronic progressive OR relapsing-remitting) AND robotic rehabilitation OR (robotics AND exercise therapy OR resistance training OR endurance training OR neurological rehabilitation) AND muscle strength OR (muscle contraction OR isometric contraction OR isotonic contraction). Search strategies were completed with the assistance of the institutional Librarian services and all database searches were completed on the same day (March 31, 2021).

Eligibility criteria included peer-reviewed journal publications or conference proceedings that have been published in the last 20 years. Randomized controlled trials, controlled studies, cohort studies, case control studies, pilot studies and case series/case reports were included in the search. Studies were considered if they included participants with MS (any level/progression of disease) and used robotic rehabilitation (upper or lower extremity) for a minimum of 3 training sessions/days. Articles were excluded that did not directly measure muscular strength. For the purpose of this search, muscular strength was defined as a measurement of maximal voluntary isometric contraction (MVC) and the maximum force produced against an external resistance (4). Exclusion criteria comprised of books, editorials, dissertations, or if the article was not published in English. Studies were further excluded if the following were not included: outcomes (strength measures), intervention (robotic training), or based on study design (minimum of 3 training days).

All database searches were completed by a single researcher, followed by a double-blinded screening from two researchers. Articles that met the search terms within each database were extracted and collated into Covidence (Veritas Health Innovation, Melbourne, Australia). After removing duplicates, title/abstract screening was completed using the inclusion/exclusion criteria independently by two researchers. Following this initial screen, all qualifying articles underwent full-text screening using the same inclusion/exclusion criteria by the same researchers as the primary screening. Conflicts regarding inclusion/exclusion were resolved by a meeting between these two researchers until all disagreements were settled. A PRISMA flow diagram outlines the search strategy in Figure 1.

Methodological quality of the included studies was assessed with the Downs and Black Quality Checklist (16) and risk of bias assessment using Cochrane Handbook for Systematic Reviews of Interventions (17). The modifications to the original Downs and Black Quality Checklist included 16 items from the checklist that were relevant to this specific review (Table 1). Risk of bias assessment was based on bias arising from the randomization process, bias due to deviations from intended interventions, bias due to missing outcome data, bias in measurement of the outcome and bias in selection of the reported result (Table 2).

The literature search resulted in a total of 31 articles: 6 (MEDLINE), 6 (Ovid Healthstar), 2 (Cochrane Database of Systematic Reviews), 17 (Embase). After duplications were removed and articles were screened for inclusion criteria, 5 articles were included for extraction and included pilot studies (n = 2), randomized controlled trials (n = 2) and controlled trials (n = 1). Data extraction included study subject size, duration of interventions, outcome measures of strength during baseline, post intervention and follow-up. Robotic training targeted the upper limb/reaching tasks (n = 2) and lower body/gait training (n = 3); upper limb focused training was unilateral training whereas the lower limb studies were primarily gait training interventions with bilateral exercises. Of the studies included, 3 reported MS phenotype which included participants with relapsing-remitting, primary progressive and secondary progressive MS, the remaining 2 studies did not report disease phenotype. Sample sizes ranged from 6 to 20 participants with a mean sample size of 12.7 ± 5.2. Expanded Disability Status Scale (EDSS) scores ranged from 6.0 to 8.0 and a disease duration ranging from 15 to 25 years. Number of training sessions ranged from 5 to 24 total active sessions with an average duration of 38.3 ± 11.0 minutes per session. Table 3 provides an overview of each study included in this review and Table 4 represents an article summary from the data extraction.

The Haptic Master was used for robot-mediated upper limb rehabilitation training. In combination with I-TRAVL it allows for reaching tasks within a 3-dimensional workspace. Gravity compensation at the hand and elbow are individualized and allow for stabilized end-positions to aid the participant to a successful completion of the task. In both studies using the Haptic Master, participants underwent 30-minutes of task-oriented and dynamic movement training (10, 11). Specific outcomes and interventions of each study are reported in the article summary (Table 4). Handgrip force (kg) significantly increased in one of the two upper limb studies (p < 0.05), following a robotic training intervention between baseline and post training, and perceived strength (visual analog scale—p < 0.05) also significantly increased (22).

The Lokomat system was used for the objective of robotic gait training. The Lokomat system was used as body weight supported orthosis and required extension and flexion of muscles of the hip and knee joints. This robotic rehabilitation also includes an interactive component inducing feedback-controlled training. In both studies using the Lokomat, leg movements were assisted by the robotic device with a preprogrammed physiological gait pattern. Body-weight support assistance was reduced as a progression throughout the training. Mean knee-extensor force increased by 3.5 and 3.3 kg for right and left side, respectively (right: p < 0.05 left: p < 0.05) (18); knee and hip flexion and extension torque increased by 7.1 and 3.7 Nm, respectively (p < 0.05) (21). The IntelliStretch was implemented to investigate ankle sensorimotor function in individuals with MS. The IntelliStretch is an ankle rehabilitative robot with controlled passive stretching and active movement. This robotic device also incorporates a computer game interface for feedback-control and participant engagement. The training program consisted of concentric and eccentric contraction of the dorsi and plantar flexors with a combination of passive and active stretching. Ankle dorsiflexion and plantarflexion torque increased by 11.4 and 10.7 Nm (all p = 0.028) (20).

Although muscular strength was not the targeted outcome of the training in the included studies, increases in maximal contractions were present in most of the studies and illuminate the benefits of using robotics as a rehabilitation modality. Included studies with a focus on the upper limb showed increases in handgrip force through sustained end-effector manipulation. These muscular strength adaptations were supplementary to the goal of the intervention and occurred as a result of the training. Regardless of the mechanism behind the improvements in grip force, handgrip strength can be an important predictor and component of basic activities of daily living (22). For training interventions focused on the lower limb, voluntary eccentric muscle contraction of the quadriceps, and increased knee joint moment was required that, in turn, improved neuromuscular control and gait patterns, resulting in increased muscular strength compared to conventional gait-support training (18, 21).

None of the selected studies reported on training volume (i.e., number of repetitions, sets or training load) and only reported on the duration and frequency of the training sessions (Table 4). There is basis for improvement for future studies to report such information as training volume is critical when participating in conventional resistance training and is suggested to be of importance for rehabilitation as well (12). Due to low load requirements for MS patients, increased volume is necessary to increase strength. As such, the premise behind robotic therapy is to use a high volume of repetitions with minimal resistance added to the robotic device or acting against the participant's voluntary movements. It is plausible that the low loads exerted during gripping of the device or the eccentric movement of the quadriceps at a high dosage elicited adaptations in muscular strength in the studies included in this review. This theoretical speculation is based on evidence in healthy populations, proving that resistance training at low loads and a high dosage can still provoke strength gains (16). The work investigated here did not report changes in muscle size (hypertrophy), so it can only be speculated that the mechanisms of strength changes are from neural adaptations. This could be a deconditioning effect; all studies included were acute resistance trained (<12 weeks in duration) and likely, participants were not highly resistance trained prior to the intervention. With acute resistance training, there is often little to no change in muscle mass but still a greater max torque is produced (21). This is caused by neural adaptations as opposed to chronic resistance training whereby there is a larger increase in muscle mass and a smaller change in neural adaptations over time (24).

In general, a consensus on proper therapy dosage and a link between total therapy time and strength outcomes is lacking in the MS literature. The studies included had therapy session frequencies ranging from 2 to 5 days, with total sessions completed ranging from 3 to 24 and duration times of 30–60 min. Typically, 8-week studies are chosen and are based on previous literature of non-robotic exercise focused studies (exercise/resistance training) (20, 23) and previous robotic literature (24, 25). However, only one study included in this review utilized an 8-week duration (19). All studies included in this review had time-dependent sessions as compared to repetition-dependent and omitted any resistance progressions throughout the therapy protocol despite knowledge that progression guidelines exist in the literature for resistance and cardiovascular exercise (6). Future work is necessary to develop guidelines for robotic therapy that are based on individual characteristics such as severity of disease, age, joint angle range of motion and progressions that clinicians can refer to like the physical activity guidelines that exist (26). If these general fitness principles are incorporated, the changes noted in strength following robotic rehabilitation may be even more apparent.

There is clinical importance in individualizing rehabilitative treatment according to MS phenotype as symptoms, impairments and rehabilitative responses may differ between relapsing remittent, primary progressive and secondary progressive MS. Three of the five included studies (18, 21, 22) reported MS phenotype of their participants however, all neglected to categorize the results according to disease impairment. For example, it is unknown based on these findings if individuals with differing phenotypes or EDSS scores benefit from this rehabilitative modality. Beer et al. (18) included participants with severe walking impairments only (EDSS 6.0-7.5) and report improvements in knee-extensor strength post-intervention (Table 4). It is suggested that future studies consider MS phenotype as response to treatment can be affected by disease progression. It is important to note that no studies reported adverse effects to the training protocols. Thus, this therapy modality is well tolerated by individuals with MS.

Robotics used in the included studies have examined clinical uses for both upper and lower limb rehabilitation. Lower limb robotic devices focused primarily on improving gait and mobility, and upper extremity robotics focused on tasks that require reaching and grasping, both are common movements affected by symptoms of MS and critical in everyday lives. In this review, three studies examined robotics for the lower limb and two studies examined robotics for the upper limb. One out of the two studies focusing on the upper limb did not show significant improvements in muscular strength and because of the lack of focus on handgrip force during the training, perceived fatigue and the inability to complete the task overpowered any potential improvements in strength outcomes (19, 22). Conversely, all studies of the lower limb reported improvements in muscular force and torque production following the intervention/training program. All studies with an upper or lower limb focus are limited by lack of progression overload and protocol individualization. Though, all studies showed significant improvements in most of their primary measures aside from muscular strength, including motor performance and accuracy.

There are a few limitations to this review that should be considered. Firstly, only five studies met the criteria for full-text analysis. Therefore, results are limited, and further work is necessary in this field. However, given the documented important link between muscle strength and daily function, this should highlight a call for more work in the area. Secondly, the included studies failed to report any rationale for the prescribed intervention including therapy dosage, frequency or load. This is a gap in robotic rehabilitative literature whereby intervention guidelines should be developed so clinicians know how long and how often they should prescribe robotic therapy to their patients. Lastly, none of the included studies considered MS phenotype or how response to treatment differs depending on level of disability or disease progression. There is not enough research in this field to yet identify this. For future work, this is necessary to investigate given that MS has a wide range of debilitating symptoms according to level of disability.