Effects of semi-independent bedside rehabilitation of the upper limb assisted with a mobile wrist-hand exoneuromusculoskeleton (ENMS) for inpatients with subacute stroke: a randomized controlled trial (RCT)

Abstract

Background

Robotic-assisted rehabilitation has been used to release the labor burden in manual practice for early stroke rehabilitation. However, its broader application is constrained by high costs, limited accessibility, and high dependence on professional operation. This study aimed to investigate the feasibility and effectiveness of semi-independent, bedside upper limb rehabilitation assisted by a wrist-hand exoneuromusculoskeleton (WH-ENMS) for individuals with subacute stroke via a randomized controlled trial compared with the outcomes achieved by conventional therapy and training programs assisted by an interlimb-coordinated (IC) robot.

Methods

Fifty-four participants with subacute stroke were randomized into three groups: ENMS, IC, or conventional therapy. All groups underwent 21 rehabilitation sessions (60 min/day for 21 consecutive days), each including 30 min of standardized Bobath therapy. During the additional 30 min of training, the ENMS group engaged in semi-independent bedside training with minimal supervision, the IC group received supervised IC cycling, and the conventional group received proprioceptive neuromuscular facilitation. The rehabilitative effects were evaluated via clinical scores. The primary outcome was the Fugl-Meyer Assessment-Upper Extremity (FMA-UE), and the secondary outcomes included the action research arm test (ARAT), modified Ashworth scale (MAS), and modified Barthel index (MBI). The manpower consumption was quantified on the basis of the professional hours required during training.

Results

The ENMS group required significantly less total occupational therapist (OT) involvement (11.6 h) compared with the IC (22.9 h) and conventional (21.0 h) groups, representing a 49.4% reduction in professional manpower demand. All groups showed significant improvements in FMA-UE, ARAT, and MBI scores. Compared with the other groups, the ENMS group achieved superior gains in voluntary motor function (FMA-UE and ARAT, P < 0.05). The ENMS group also obtained significant reductions in muscle spasticity across multiple joints (MAS, P < 0.05). The IC group demonstrated the largest MBI improvements, while the conventional group showed greater gains in hand function than the IC group (P < 0.05).

Conclusions

Semi-independent bedside rehabilitation with WH-ENMS is feasible, safe, and effective for inpatients with subacute stroke. It reduced the demand for professional manpower while promoting upper limb recovery, particularly wrist–hand function, during the critical subacute neuroplasticity window. This approach could optimize rehabilitation resource allocation and facilitate more personalized interventions.

Trial registration Chinese Clinical Trial Registry Identifier: ChiCTR2300074469. Registered on 2023/08/08.

Background

Stroke remains one of the leading causes of mortality and long-term disability worldwide, with an annual incidence of 12.2 million new cases and over 5.7 million individuals developing permanent disabilities each year [1]. Global burden analyses revealed 93.8 million prevalent stroke cases [2]. Notably, more than 80% of these individuals experience moderate to severe upper limb motor impairments, particularly in wrist-hand function, which critically compromises poststroke independence and quality of life [3]. Neurorehabilitation research underscores that early intensive task-oriented therapy during the six-month subacute period is essential to leverage spontaneous neuroplasticity for maximal functional recovery and prevent learned nonuse [4, 5].

However, current inpatient rehabilitation systems face systemic resource constraints on both human resources and physical infrastructure, impeding optimal recovery delivery [6]. Conventional inpatient rehabilitation predominantly relies on labor-intensive, therapist-delivered, one-on-one manual interventions, and daily therapeutic doses for upper limb recovery often fail to meet evidence-based guideline thresholds, with systematic reviews documenting 30–70% shortfalls in treatment intensity [7–9]. This rehabilitation-delivery gap is further amplified by accelerating population aging and increasing stroke incidence, creating an urgent need for innovative automated rehabilitative technologies that can augment human resources while maintaining therapeutic efficacy. This crisis is particularly acute in China, where existing 17.8 million stroke survivors face critically limited rehabilitation capacity, with only 0.21 rehabilitation beds per 1,000 population (300,000 total rehabilitation beds), representing a 58% deficit relative to the national standard, and a severe shortage of 688,000 rehabilitation physicians is required to achieve the recommended practitioner-to-population ratio of 6:100,000 [10–13]. These resource limitations exacerbate fundamental inefficiencies in conventional upper limb rehabilitation paradigms and underscore the need for scalable rehabilitation solutions, particularly in China, but also in other healthcare systems with comparable resource constraints.

The integration of robotic technologies into poststroke rehabilitation has demonstrated benefits in enhancing motor recovery and daily life activities as a complement to conventional manual therapy through their capacity for highly intensive and repetitive training [14]. Several specific upper limb rehabilitation robots, such as end-effectors represented by MIT-Manus [15] and exoskeletons represented by the Armeo Power [16], as well as upper and lower limb robots that coordinate both limbs, such as cycle ergometers represented by MOTOmed [17], have become increasingly prevalent in hospital rehabilitation settings. Among them, the upper and lower limb robots that facilitate interlimb coordination show their unique benefits for reactivating neural pathways, strengthening cross-limb muscle synergies, and enhancing movement symmetry [18, 19]. Moreover, hybrid robots that integrate robotic devices with neuromuscular electrical stimulation (NMES) are widely adopted in clinical practice and have demonstrated more substantial improvements in motor function than robots that use either approach alone [20].

Despite these advances, current robotic rehabilitation systems remain difficult to integrate into early inpatient rehabilitation workflows. These systems are typically large-scale, expensive, and installed in a common therapy room within a hospital, frequently necessitating close professional supervision, specialized configurations, and sharing among multiple individuals undergoing rehabilitation, making access difficult for individuals who are bedridden. Individuals often experience delays due to limited system availability, as most hospitals may only have one system serving the entire unit due to expensiveness, resulting in suboptimal utilization of time and resources within rehabilitative care [21]. Moreover, one-on-one supervision and assistance by professional therapists are usually required when these systems are used in rehabilitation, implying that the demand for professional manpower in robotic-assisted rehabilitation remains comparable to that of traditional manual therapy. This undermines the cost-effectiveness of robotic service, aiming to conserve manpower resources while providing effective treatment to a wide population of individuals who have had a stroke. Additionally, centralized training environments with shared facilities may increase the risk of nosocomial infection and transmission among individuals recovering from stroke who have compromised immune systems, particularly during health crises such as the COVID-19 pandemic [22]. This challenge underscores the need for rehabilitation technologies that can be deployed safely within the ward environment and reduce the frequency of patient transport to communal therapy areas. Developing decentralized, bedside-compatible robotic solutions may help address both infection-control concerns and the practical limitations of conventional facility-based systems.

Another critical gap in current clinical practice is the lack of stratified strategies for the allocation of professional resources in robotic rehabilitation tailored to the varying medical conditions and functional levels of individuals after stroke. Individuals with better conditions may be capable of engaging in robot-assisted rehabilitation more independently, thereby allowing therapists to allocate more time and attention to more complicated cases. The adoption of self-rehabilitation strategies with remote professional supervision has proven its feasibility and effectiveness in individuals with chronic stroke, suggesting that independent physical training with necessary supervision could serve as a viable alternative to traditional outpatient rehabilitation methods [23]. Nevertheless, self-help robot-assisted training has seldom been utilized in individuals in the early stages after stroke. One of the barriers is that the majority of current robots either necessitate intricate operations or fail to offer intuitive feedback to users when used alone, limiting their independent usage by nonprofessionals. Another concern is that physical training alone without close professional monitoring may pose safety risks for individuals in the acute and subacute phases after stroke who are being transported to a common therapy room from their wards, as their health conditions remain weak in the early stage poststroke. This creates an unmet need for a compromise approach, e.g., a semi-independent training paradigm, where robotic systems can provide core therapeutic exercises with intermittent therapist supervision and assistance, enabling individuals to initiate early and intensive practice without overwhelming demands of professional manpower.

In our previous study, a hybrid soft robotic technique, namely, the mobile exoneuromusculoskeleton (ENMS) [24], was developed as a lightweight and portable system suitable for bedside rehabilitation. The ENMS includes wearable elbow and wrist-hand modules and integrates the advantages of neuromuscular electrical stimulation (NMES), pneumatic muscles, and exoskeletons in a single platform [24]. The system has been commercially available on the Hong Kong market. Peripheral voluntary motor effort (VME), as captured by EMGs, is used to control multichannel NMES and provide mechanical assistance to elbow, wrist, and finger motions [25]. We successfully implemented a self-help telerehabilitation program using the wrist-hand ENMS (WH-ENMS) module in a public rehabilitation center [26]. Individuals with chronic stroke demonstrated significant long-term improvements in upper limb function through home-based training with this assistive technology under intermittent remote professional supervision [26–28]. However, its application for individuals in the subacute stage after stroke receiving inpatient care remains unexplored in a semi-independent clinical setting without close assistance from healthcare professionals. Therefore, this study aimed to investigate the feasibility and effectiveness of semi-independent upper limb bedside rehabilitation assisted by the WH-ENMS for individuals with subacute stroke receiving inpatient care through a randomized controlled trial (RCT) compared with rehabilitation assisted by an upper and lower limb robot and conventional therapy.

Methods

EMG-driven ENMS for wrist-hand training

The WH-ENMS (Mobilexo Arm, Thecon Technology, Hong Kong, China, shown in Fig. 1A and detailed in Table 1) is a lightweight and wearable hybrid system that combines pneumatic actuation with NMES driven by residual voluntary electromyography (EMG) signals from the paretic upper limb following stroke. It consists of a glove embedded with pneumatic fingers, two EMG-NMES channels, a control box (9 × 16 × 5 cm (length × width × height)), and a mobile application (App) [25, 26, 28], with a weight of 150 g for the wearable unit (glove and wires) and 650 g for the control box powered by a 12 V rechargeable battery inside the control box, enabling portable usage in a ward environment, e.g., bedside usage. The entire device can be compactly stored in a container with an A4 paper box. The ENMS can aid individuals after stroke in performing repetitive, phased, and coordinated movements of the wrist and hand, i.e., wrist extension with the hand open and wrist flexion with the hand closed. It employs EMG-triggered control, using WH extensors (Extensor Carpi Ulnaris (ECU) and Extensor Digitorum (ED)) and WH flexors (Flexor Digitorum (FD) and Flexor Carpi Radialis (FCR)) as voluntary neuromuscular inputs to initiate mechanical and NMES support (square pulse with adjustable pulse width of 0–200 µs, 70 V, 40 Hz) during each movement effort [25]. The selected muscles (ECU, ED, FD, FCR) were chosen as the primary agonists for wrist–hand extension and flexion, the fundamental movements targeted in ENMS-assisted training, and are conventionally adopted in EMG-driven robotic wrist–hand rehabilitation systems with validation [27–29]. When the EMG activation level of the driving muscle exceeds 10% of maximal voluntary contraction (MVC), the pneumatic fingers inflate to provide extension torque to each finger or deflate passively due to voluntary finger flexion of the affected limb, and NMES is delivered to the driving muscle. This 10% MVC threshold ensures signals explicitly reflect voluntary activation while representing a feasible contraction level that activates motor pathways and minimizes fatigue [30]. Implemented in our previous EMG‑driven robotic systems, it successfully identified VME from impaired muscles in both subacute and chronic stroke patients [27, 29, 31, 32]. When individuals in the early stage after stroke exhibit insufficient voluntary EMG, a passive mode could be provided as an alternative solution, i.e., the system provides mechanical and NMES assistance cyclically without EMG initiation. Wrist-hand training assisted by ENMS has previously been proven to significantly improve voluntary motor function with less compensatory motion and release muscle spasticity in chronic stroke [25, 27, 28].

Fig. 1.

Experimental setup of the subacute stroke upper limb training. A ENMS-assisted wrist-hand training in the ward; B Interlimb-coordinated training in the therapy room

Table 1.

Comparison of ENMS-assisted training and interlimb-coordinated training

	ENMS-assisted training	Interlimb-coordinated training
Robot specifications
Robot device	ENMS wrist-hand module	Upper and lower limb robot
Mobility	Portable	Stationary
Occupied area	Approximately 0.3 m × 0.5 m	Approximately 1 m × 1.5 m
Power consumption	12 VA	250 VA
Robotic control	EMG-driven/passive	Unaffected upper limb-driven
Electrical stimulation	EMG-driven wrist-hand extension/flexion (40 Hz, 0-300 µs pulse width)	Nil
Cost	80,000 HKD	120,000 RMB
Training regimen details
Training tasks	Wrist-hand extension/flexion	Interlimb-coordinated cycling
Training intensity	30 min/session over 21 consecutive days
Training venue	The ward	Therapy room
Therapist Dependency (per person)
Pre-training tutorial	10 min
Total training preparation	25 min	105 min
Total supervised training	31.18 ± 2.81 min	630 min

The ENMS has a mobile app that wirelessly communicates with the control unit via Bluetooth and serves as the user interface, offering visual feedback to users, including individuals after stroke and their caregivers, as well as professionals (e.g., occupational therapists (OTs), nurses, nursing assistants, etc.). Figure 2A shows the calibration interfaces for the training parameters, which include the NMES intensity level, EMG level, air pressure, and training interval durations. The intensity level of NMES is typically set to induce maximal muscle contraction without causing immediate pain, with a pulse width ranging from 10 to 24 µs for the WH extensors and 10 to 16 µs for the WH flexors. The EMG level is defined by recording the EMG during baseline and wrist-hand extension/flexion, as guided by the visual instructions on the screen. The inner air pressure of the pneumatic fingers should be calibrated to a maximum of 40 kPa to ensure adequate mechanical support. In the passive mode, the precise durations for the training interval of each phase of extension, rest, and flexion should be initially defined. In addition, the app incorporates visual instructions (Fig. 2B) to guide users through a standardized step-by-step wrist-hand training cycle. Each cycle follows a structured sequence of starting with extending the wrist-hand, followed by a relaxation phase, then flexing the wrist-hand, and finally concluding with another relaxation phase. The interface simultaneously displays the performance metrics, including the cumulative count of completed training cycles, elapsed session duration, and real-time inner pressure value of the pneumatic fingers. There is a termination button positioned in the app training interface to quit training at any time. In addition, a physical stop button is located on the control box, enabling the system to be powered off immediately when necessary. Moreover, the app features an automated data archival and history tracking system (Fig. 2C) that systematically documents quantitative parameters from each training session, including training duration and completion of WH task cycles.

Fig. 2.

User interfaces of the ENMS group and IC group. A ENMS interfaces for calibrating training parameters, i.e., NMES intensity, EMG level, air pressure, and training intervals; B ENMS interfaces during wrist-hand training with step-by-step instructions; C ENMS interfaces for training records and history; D Upper and lower limb robot interfaces, i.e., NMES calibration and training interfaces

Upper and lower limb robot for interlimb-coordinated (IC) cycling training

The upper and lower limb rehabilitation robot (SL4-3 S, Guangzhou Yikang Medical Equipment Industrial Co., Ltd., China, Fig. 1B; Table 1) adopted in this study is a system designed for cycling the upper and lower limbs and is actuated with four motors, which independently, or jointly, regulate assistive/resistive torque outputs to a handlebar for an upper limb and pedal for a lower limb. The robotic system was installed in a therapy room within an area of 1.0 m×1.5 m, requiring a 220 V AC power supply. The IC cycling system can operate in three distinct modes, i.e., passive (no resistance/assistance), active (adjustable resistance), and assistive (adjustable motor support) modes. We adopted the active mode in this work; meanwhile, the movements of the pedals were mechanically locked with the upper limb handles, with the purpose of using upper limb movements to drive the follow-up motions in the lower limb. The user was required to perform VME in the upper limb as much as possible during IC cycling, whereas VME was not mandatory in the lower limb. Resistive levels from the motors could be adjusted across a calibrated range (0–20; 0 = no resistance, 20 = maximal resistance) for individuals. This interlimb coordination has been posited to enhance neuromuscular control efficacy via augmented corticospinal connectivity between the upper and lower extremities [33, 34], and IC robot-assisted training has been a routine intervention for individuals after stroke receiving inpatient care in China [33]. The IC robot provides user interfaces (Fig. 2D) for the training parameter configuration, including the motor resistive levels and completion progress in a training session, i.e., training duration, speed, cycling distance, consumed energy, etc.

Participants

This study was approved by the Human Participants Ethics Sub-Committee of the Seventh Affiliated Hospital of Sun Yat-sen University (Ethics number: KY-2023-033-03). A total of 59 participants receiving inpatient care from the hospital were screened for training. Fifty-six participants satisfied the following inclusion criteria: (1) had a singular and unilateral brain lesion due to a stroke acquired within 14–180 days after onset (subacute stage); (2) had standard medical care and were in stable condition; (3) were aged from 20 to 80 years; (4) were able to sit for 30 min without assistance; (5) had enough cognition to understand the content or purpose of the study and follow simple instructions as assessed by the Mini-Mental State Examination (MMSE > 23) [35]; (6) both the metacarpophalangeal (MCP) and proximal interphalangeal (PIP) joints could be extended to 180° passively; and (7) muscle spasticity during extension at the finger joints and the wrist joint was less than 3, as measured by the Modified Ashworth Scale (MAS) [36].

This study employed a non-blinded randomized controlled trial design. This trial was registered at the Chinese Clinical Trial Registry (Identifier: ChiCTR2300074469). Eligible participants were informed that the assigned training program could involve ENMS-assisted wrist-hand training, upper and lower limb robot-assisted IC cycling training, or conventional occupational therapy. Following this disclosure, 54 participants provided written informed consent before randomization. Stratified randomization with respect to Fugl-Meyer Assessment-Upper Extremity (FMA-UE) scores during recruitment, i.e., distributed into two strata (FMA-UE: 0–33, 34–66), was performed to minimize heterogeneity in the initial impairments of the baseline between the groups, as practiced in our previous RCTs with small sample sizes [29, 31, 32]. FMA-UE was selected as the primary outcome of this study because it is an effective and widely used indicator for the clinical assessment of voluntary limb function. The participants were then randomly assigned to one of three groups, i.e., the ENMS group, the IC group, or the conventional group, via a computer-based random number generator (MATLAB, MathWorks, Inc., 2024), which allocated individuals with equal probability (1:1:1 ratio) by generating values of “1,” “2,” or “3” corresponding to each group. The Consolidated Standards of Reporting Trials (CONSORT) flowchart outlining participant progression through the training program is presented in Fig. 3.

Fig. 3.

Consolidated standards of reporting trial flowchart of the experimental design

Training protocols

All the participants engaged in a 21-session upper limb rehabilitation program, with an intensity of 60 min per session and 1 session per day for 21 consecutive days in the hospital. In each training session, the robotic groups (i.e., the ENMS and IC groups) first received 30 min of conventional intervention on the basis of the Bobath concept [37], followed by 30 min of robot-assisted intervention. Bobath therapy was administered via an OT in a one-to-one manner, with a focus on the motor function of key anatomical points, e.g., radiocarpal, carpometacarpal, PIP, and MCP joints [37], including thumb mobilization, shoulder forward flexion at 90° for multiplanar movement training, forearm pronation and supination exercises, and wrist flexion and extension exercises. After the 30-minute OT intervention, participants in the robot groups accessed robot-assisted training programs with the assigned robotic devices.

Semi-independent wrist-hand training assisted by the ENMS

In the ENMS group, participants primarily completed ENMS-assisted upper limb training in a semi-independent manner in their wards, without close monitoring by healthcare professionals for the entire duration. The ENMS training program consisted of 21 training sessions, comprising one supervised session followed by 20 semi-independent sessions, with each session divided into a 5-minute assisted-preparation period and a 30-minute independent training segment for performing wrist-hand extension and flexion movements (Fig. 4A).

Fig. 4.

Training protocol of the upper limb training program. A Timeline for the 21-session semi-independent ENMS-assisted upper limb training program. B Total accumulated occupational therapist involvement hours per participant across the training program, with data presented as the mean values for each group

In the first session, the OT conducted a 10-minute bedside pre-training tutorial covering proper training posture, mobile app operation, and training protocols. A participant was instructed to maintain a seated position with their affected forearm positioned on an overbed table (shown in Fig. 1A) in the ward while raising the backrest of the bed to sit in a comfortable position. The participant was trained to perform repetitive wrist-hand extension and flexion tasks according to the instructions displayed on the App interface (Fig. 2B). The OT then set the system parameters of the maximal air pressure in the pneumatic fingers and the NMES intensities to the targeted muscles. The OT supervised the entire training onsite in the first session, with the purpose of familiarizing participants with the ENMS-assisted training tasks and the system usage for later independent training alone. After that, the OT conducted a competency assessment to determine the participant’s readiness for independent training with the following criteria: (1) independent use of the training function of the app and (2) ability to follow the training instructions in the app interface to perform WH tasks. Additional supervised sessions would be provided by the OT if a participant was not ready for independent practice. In this study, all participants could perform ENMS-assisted WH training independently after the first supervised session.

In the subsequent sessions, the participant independently conducted the training tasks in the ward after a certified nursing assistant helped mount the ENMS system and attach the electrodes to the paretic limb. The OT inspected at predetermined checkpoints (i.e., sessions 6, 11, 16, and 21) to make necessary adjustments to the training parameters in the ENMS for an individual to achieve rehabilitative progress. After the completion of the desired training tasks, the participant informed the nursing assistant to help with device doffing via a bedside call.

All the training sessions were scheduled during the working hours of the OT, who was working near the participant’s ward and could arrive onsite in 5 min. If any difficulties arise in the independent training sessions, e.g., misuse of the system, the participant could reach out for help from the nursing assistant first. Then, more professional assistance would be provided if deemed necessary.

Supervised interlimb-coordinated training in the IC group

The participants in the IC group underwent arm-initiated and leg-followed cycling tasks assisted by the upper and lower limb robot in a common therapy room (Fig. 1B). Before the start of the training in the first session, a 10-minute tutorial was provided by an OT to prepare a participant to become familiar with the IC training tasks, appropriate postures in cycling, training protocols, and safety guidelines. The OT helps the participant sit on the robot in a seated-to-stand or semi-reclined posture to ensure spinal alignment and joint safety. The affected hand was firmly attached to the handlebar via bandages to stabilize the handlebar-grasping position, whereas the lower limb was stabilized with hook-and-loop fasteners to ensure ankle joint stability during cycling (Fig. 1B). This preparation period lasts approximately 5 min.

During the 30-minute IC cycling training, the participant was required to exert the initial effort from the nonparetic upper limb first, coupled with the voluntary effort from the paretic upper limb. The resistance level of the robot was gradually increased by the OT starting from level 1 and then adjusted to levels 3–4 according to the participants’ tolerance and performance metrics. Continuous supervision from the OT was provided to ensure adherence to appropriate postures, provide corrective feedback, and prevent fall risks. The participants were instructed to report fatigue or cardiovascular strain (e.g., elevated heart rate) for immediate training suspension. Training parameters (speed, distance, energy expenditure) were monitored in real time through the system’s user interface (Fig. 2D).

Traditional occupational therapy in the conventional group

The participants in the conventional group received 60 min of conventional training matched with the robotic interventions in total duration and intensity. It comprises sequential phases beginning with a 30-minute Bobath-based intervention identical to robotic groups, followed by a 30-minute intervention of proprioceptive neuromuscular facilitation (PNF) [38]. The Bobath component emphasized fine motor control and joint mobility, including exercises for the thumb, wrist, forearm rotation, and shoulder movements to promote multiplanar motion. The subsequent PNF systematically implemented multiplanar upper extremity diagonal patterns, specifically the flexion-adduction-external rotation (flexion pattern) and extension–abduction–internal rotation (extension pattern) sequences, designed to elicit synergistic muscle activation across multiple joints. These multi-joint movement sequences could enhance proprioceptive input, neuromuscular coordination, and dynamic stability. The entire conventional training process was manually conducted by one OT in a one-to-one manner.

Outcome evaluations

The motor improvements in participants after the interventions were assessed by a physical therapist blinded to group allocations via standardized clinical scales before and after the intervention. Evaluations included the FMA-UE [39] to measure motor impairment in upper limb movements; the Modified Ashworth Scale (MAS) [36] for quantifying resistance during passive stretching of the finger, wrist, and elbow flexors (indicating spasticity); the Action Research Arm Test (ARAT) [40] to evaluate voluntary upper limb function with emphasis on fine motor skills and grasping; and the Modified Barthel Index (MBI) [41], which rates independence and performance in activities of daily living (ADLs).

The total accumulated time of OT involvement per participant was quantified across the three groups, including conventional Bobath therapy, PNF therapy, pre-training tutorials, preparatory procedures conducted over the 21 sessions, IC training, and ENMS-assisted training. Additionally, during the ENMS semi-independent sessions, the supplementary assistance time provided by the OT, along with the corresponding rationales, was recorded.

Statistics

The varying demographic characteristics of the participants in the three groups were assessed via one-way analysis of variance (ANOVA) or Fisher’s exact test. The normality of the clinical scores was evaluated via the Lilliefors method [42]. Parameter tests were employed for the clinical scores of the FMA-UE and MBI, which followed a normal distribution, whereas non-parametric tests were utilized for the clinical scores of the MAS and ARAT, as these scores did not conform to a normal distribution. The baselines of the clinical scores for the three groups were compared via one-way ANOVA with Bonferroni post hoc correction or the Kruskal‒Wallis test with the Mann‒Whitney U test as a post hoc test. One-way analysis of covariance (ANCOVA) or the Quade test was then used to evaluate the group differences in the post-training clinical assessments, with the pre-assessment as a covariate. A paired t test or Wilcoxon signed-rank test was subsequently performed to examine the intragroup differences among the three groups before and after training. Furthermore, the changes in each clinical assessment after training were compared among the groups via the Kruskal‒Wallis test with the Mann‒Whitney U test as a post hoc test. Statistical significance was defined as a p value less than 0.05, and significance levels of 0.01 and 0.001 were also indicated.

Results

A total of fifty-two participants finished the assigned training. Two participants, one from the ENMS group and one from the conventional group, withdrew from the study because of unexpected hospital discharge for personal reasons. Table 2 presents the demographic details of the participants ultimately recruited into the three groups. There were no statistically significant differences among the three groups in terms of age, sex, type, affected side, or onset time.

Table 2.

Demographic characteristics of the participants

Characteristics	ENMS group (n = 17)	IC group (n = 18)	Conventional group (n = 17)	P value
Sexa (male/female)	16/1	14/4	15/2	0.472
Affected sidea (right/left)	13/4	16/2	13/4	0.377
Type of strokea (ischemic/hemorrhagic)	15/2	18/0	16/1	0.837
Ageb in years (mean ± SD)	54.47 ± 8.47	57.47 ± 7.37	53.64 ± 7.65	0.944
Time since strokeb in days (mean ± SD)	92.82 ± 24.56	86.94 ± 24.20	90.17 ± 24.26	0.774

No significant differences were found among the groups (P > 0.05)

^aFisher’s exact test

^bOne-way analysis of variance (ANOVA)

Figure 4B compares the total OT involvement time per participant across the 21-session training program among the three groups. Both the conventional group and the IC group required continuous OT engagement, which lasted 21.00 and 22.92 h, respectively. The OT time distribution in the IC group included 10.5 h of conventional Bobath therapy, 10 min of pre-training tutorial, 105 min of preparation time (5 min per session across all 21 sessions), and 10.5 h dedicated to IC training. In contrast, the ENMS group with semi-independent training significantly reduced OT involvement to 11.6 h per participant, which comprised 10.5 h of Bobath therapy (equivalent to other groups), 10 min for pre-training tutorials (equivalent to the IC group), 25 min of preparation (5 min for the 5 critical sessions), and 31.18 ± 2.81 min of ENMS-assisted training (30-minute mandatory supervised session and additional OT assistance), demonstrating a 49.4% decrease in OT-occupied time compared with the IC group. Three participants after stroke requested additional OT assistance during their semi-independent training sessions, with an accumulated 20 min. One participant experienced pain from NMES during the training, and the OT decreased the NMES intensity level by 10% to relieve the pain (5 min). Another participant requested a personal consultation with the OT and received a five-minute consultation. The third participant was unable to trigger the ENMS system due to muscle fatigue, so the OT allowed the participant to rest and provided a massage to enhance muscle excitability (10 min). Other system malfunctions were addressed by nursing assistants, including one instance of electrode displacement and one unintended cable disconnection initiated by the participant.

In terms of training control mode, more than half of the participants in the ENMS group were able to generate sufficient voluntary EMG to operate the system in the EMG‑driven active mode. Specifically, 9 participants completed all training sessions using the active mode, whereas 8 participants required the passive mode due to insufficient voluntary EMG activation at baseline. Among them, 3 participants subsequently transitioned from passive to active mode as their voluntary neuromuscular activation improved during the training period.

Figure 5 illustrates the pre- and post-intervention clinical assessments (i.e., FMA-UE, ARAT, MAS, and MBI) of participants across the three groups. Table 3 summarizes these clinical scores as the means with 95% confidence intervals, alongside probabilities (P values) and effect sizes (EFs) for intergroup and intragroup comparisons. A statistically significant baseline difference was detected in the MAS finger scores (P = 0.026, EF = 0.109, Kruskal‒Wallis test with the Mann‒Whitney U test as a post hoc test; Table 3). Specifically, the ENMS group demonstrated higher baseline MAS finger scores compared to the conventional group, indicating a greater degree of muscle spasticity prior to the intervention. No significant baseline differences were observed among the groups for FMA-UE, ARAT, MAS elbow, MAS wrist, or MBI (P > 0.05, one-way ANOVA with Bonferroni correction or the Kruskal‒Wallis test with Mann‒Whitney U post hoc analysis; Table 3). Intragroup comparisons revealed significant improvements in FMA-UE, ARAT, and MBI scores across all groups (P < 0.05, paired t test or Wilcoxon signed-rank test; Table 3). Both robotic groups (ENMS and IC) exhibited significant reductions in MAS elbow scores (P < 0.05, Wilcoxon signed-rank test, Table 3). Notably, the ENMS group achieved additional reductions in MAS finger and wrist scores (P < 0.05, Wilcoxon signed-rank test, Table 3). According to the results of the post-assessment intergroup analyses, the ENMS group presented superior FMA-UE scores than both the IC and conventional groups (P < 0.001, EF = 0.259, one-way ANCOVA with the Bonferroni post hoc correction; Table 3). Furthermore, the ENMS group demonstrated significantly higher ARAT scores than the IC group post-intervention (P = 0.041, EF = 0.122, Quade test with Tukey’s post hoc test; Table 3).

Fig. 5.

Clinical scores before and after training in the three groups. A Fugl-Meyer Assessment-Upper Extremity (FMA-UE); B action research arm test (ARAT); C modified Barthel index (MBI); D–F modified Ashworth scale (MAS) scores at the fingers D, wrist E, and elbow F, presented as the means ± SEs (error bars) in each evaluation session. The solid red lines represent the ENMS group, the dashed blue lines represent the IC group, and the dotted black lines represent the control group. Significant intergroup differences are indicated by “#” (P < 0.05, one-way ANOVA with the Bonferroni post hoc correction or the Kruskal–Wallis test with the Mann–Whitney U post hoc test) and “△” (P < 0.05, one-way ANCOVA with the Bonferroni post hoc correction or the Quade test with the Tukey post hoc test). “*” indicates significant intragroup differences (P < 0.05, paired t test or Wilcoxon signed-rank test)

Table 3.

Clinical assessments before and after training in each group

Assessment		PRE		POST	Intragroup comparison	Intergroup comparison
						Pre- assessments	Post- assessments
		Mean Values(95% Confidence Interval)			P (Cohen’s d/r)	P(Partial η2/η2)	P(Partial η2)
FMA-UEa	ENMS	23.41 (15.57 ~ 31.25)		33.35 (25.27 ~ 41.44)	< 0.001*** (7.529)	0.466 (0.031)	< 0.001### (0.259)
	IC	19.89 (14.90 ~ 24.88)		22.83 (16.83 ~ 28.84)	0.037* (5.525)
	Conventional	26.47 (16.04 ~ 36.90)		29.65 (19.14 ~ 40.15)	< 0.001*** (2.942)
ARATb	ENMS	9.06 (0.75 ~ 17.37)		16.59 (6.53 ~ 26.65)	< 0.001*** (-0.802)	0.071 (0.067)	0.041# (0.122)
	IC	9.11 (6.51 ~ 11.71)		10.61 (7.45 ~ 13.77)	0.018* (-0.575)
	Conventional	11.94 (1.34 ~ 22.54)		15.47 (4.68 ~ 26.26)	0.003** (-0.715)
MBIa	ENMS	70.53 (60.01 ~ 81.05)		77.35 (67.23 ~ 87.48)	0.013* (10.020)	0.449 (0.032)	0.802 (0.009)
	IC	63.61 (57.36 ~ 69.87)		71.11 (64.05 ~ 78.17)	< 0.001*** (7.717)
	Conventional	63.53 (52.56 ~ 74.50)		69.41 (58.40 ~ 80.43)	0.002** (6.431)
MAS Fingerb	ENMS	1.00 (0.52 ~ 1.48)		0.41 (0.15 ~ 0.67)	0.015* (-0.589)	0.026△ (0.109)	0.187 (0.066)
	IC	0.61 (0.19 ~ 1.03)		0.56 (0.13 ~ 0.98)	0.317 (-0.243)
	Conventional	0.24 (0.01 ~ 0.46)		0.18 (-0.03 ~ 0.38)	0.317 (-0.243)
MAS Wristb	ENMS	0.76 (0.34 ~ 1.19)		0.41 (0.09 ~ 0.73)	0.014* (-0.594)	0.390 (-0.002)	0.130 (0.080)
	IC	0.72 (0.31 ~ 1.13)		0.61 (0.26 ~ 0.96)	0.157 (-0.343)
	Conventional	0.41 (0.09 ~ 0.73)		0.24 (0.01 ~ 0.46)	0.180 (-0.325)
MAS Elbowb	ENMS	0.82 (0.41 ~ 1.24)		0.53 (0.21 ~ 0.85)	0.025* (-0.542)	0.090 (0.058)	0.397 (0.037)
	IC	0.83 (0.34 ~ 1.32)		0.61 (0.22 ~ 1.00)	0.046* (-0.485)
	Conventional	0.29 (-0.01 ~ 0.60)		0.29 (0.05 ~ 0.54)	1 (0)

The means and 95% confidence intervals of the pre- and post-clinical assessments and the probabilities with the estimated effect sizes of the statistical analyses. Differences with statistical significance are marked with superscripts next to the P values (“*” for intragroup comparison, “^△” for pre-assessment intergroup comparison, “^#” for post-assessment intergroup comparison). Significant levels are indicated as 1 superscript for < 0.05, 2 superscripts for ≤ 0.01, and 3 superscripts for ≤ 0.001

^aParametric test: Intragroup comparison, paired t test; Pre-assessment intergroup comparison, one-way ANOVA with the Bonferroni post hoc correction; Post-assessment intergroup comparison, one-way ANCOVA with the Bonferroni post hoc correction

^bNon-parametric tests: Intragroup comparison, Wilcoxon signed-rank test; Pre-assessment intergroup comparison, Kruskal‒Wallis test with the Mann‒Whitney U test as a post hoc test; Post-assessment intergroup comparison, Quade test with Tukey’s post hoc test

Figure 6 presents the longitudinal changes in the clinical assessment outcomes across the three groups following their respective interventions. The detailed numerical data and comparative statistical results are summarized in Table 4. Compared with both the IC and conventional groups, the ENMS group demonstrated significantly greater improvements in FMA-UE, ARAT, and MAS finger scores (P < 0.05; Kruskal‒Wallis test with Mann‒Whitney U post hoc comparisons; Table 4). Moreover, the conventional group demonstrated a significantly greater improvement in ARAT scores compared to the IC group. However, no statistically significant intergroup differences were detected in the MBI, MAS wrist, or MAS elbow scores (P > 0.05; Kruskal‒Wallis test with Mann‒Whitney U post hoc comparisons; Table 4).

Fig. 6.

Changes in each clinical assessment after treatment in the three groups. Fugl-Meyer Assessment-Upper Extremity (FMA-UE), action research arm test (ARAT), modified Barthel index (MBI), and modified Ashworth scale (MAS) scores at the fingers, wrist, and elbow, presented as the means ± SEs (error bars). Significant differences are indicated by “▲” (P < 0.05, Kruskal–Wallis test with Mann–Whitney U post hoc test)

Table 4.

Changes in each clinical assessment of each group

Assessment	ENMS group	IC group	Conventional group	P value (η2)
FMA-UE	9.94 ± 7.52	2.94 ± 5.52	3.17 ± 2.94	< 0.001▲▲▲ (0.327)
ARAT	7.52 ± 8.63	1.50 ± 2.38	3.54 ± 4.24	0.012▲ (0.138)
MBI	6.82 ± 10.02	7.50 ± 7.72	5.88 ± 6.43	0.830 (-0.033)
MAS Finger	-0.58 ± 0.79	-0.05 ± 0.23	-0.05 ± 0.24	0.006▲▲ (0.170)
MAS Wrist	-0.35 ± 0.49	-0.11 ± 0.32	-0.17 ± 0.52	0.151 (0.036)
MAS Elbow	-0.29 ± 0.46	-0.22 ± 0.42	0.00 ± 0.35	0.124 (0.044)

The means and standard deviations (SDs) for the changes in each clinical assessment and the probabilities with the estimated effect sizes of the statistical analyses. Statistically significant intergroup differences are marked with “▲” (the Kruskal–Wallis test with the Mann–Whitney U test as a post hoc test). Significant levels are indicated as 1 superscript for < 0.05, 2 superscripts for ≤ 0.01, and 3 superscripts for ≤ 0.001

Discussion

In this work, we investigated the feasibility of semi-independent upper limb bedside rehabilitation assisted by the WH-ENMS and compared its rehabilitative effectiveness and professional manpower demand with those of routine rehabilitation approaches, including IC training assisted by an upper and lower limb robot and conventional occupational therapy, in participants with subacute stroke receiving inpatient care.

Feasibility and acceptability of WH-ENMS-assisted training

Participants in the ENMS group successfully completed the prescribed semi‑independent bedside training with minimal professional assistance, and most tolerated the intervention without adverse effects. The reported adverse effects experienced by the participants were muscle fatigue (n = 1) in session 6 and pain (n = 1) in session 2, which were caused by the NMES delivered to the extensors of the wrist-hand by the ENMS system. In the stimulation configuration, the initial intensity for the extensors set before training was usually greater than that for the flexors in assisting muscle contraction, as most individuals after stroke have better motor return in the WH flexors than in the extensors even in the early stage [43]. However, the pain report was not immediately received in session 1 but in session 2, which could be related to delayed somatosensory perception in the early stage poststroke [44]. Late reports on pain caused by NMES were absent in previous ENMS-assisted upper limb training for chronic stroke, whose somatosensory functions were relatively stable [27, 28]. This finding also implies that ENMS-assisted training for participants with subacute stroke should be more frequently checked on the somatosensory status in the early training sessions. The pain could be relieved immediately by lowering the NMES intensity, but this operation reduced the evoked muscular contraction level in the extensor at the same time. The participant could still complete the WH extension cycle with combined mechanical assistance from the pneumatic fingers and sub-threshold NMES assistance. One participant complained about muscle fatigue in the extensor in the middle session of the training, which was related to the common side effects of motor-level NMES clinically, which could be relieved by more rest and massage of the fatiguing muscles [45]. On the other hand, the variability in control modes observed in the ENMS group further illustrates the heterogeneous neuromuscular status of subacute inpatients and indicates the adaptability of the ENMS system to different levels of voluntary motor engagement. Participants who were capable of sustaining EMG‑driven control demonstrated preserved or rapidly recovering corticospinal excitability, allowing them to engage more directly in voluntary wrist–hand activation. In contrast, those who initially relied on passive assistance represented individuals with weaker or more inconsistent voluntary motor output early in the recovery process. Notably, the fact that several individuals were later able to transition from passive to active control suggests that ENMS-assisted practice may help strengthen voluntary drive through repeated task‑specific activation and sensorimotor feedback. The ability of the system to support both passive initiation and voluntary engagement, and to facilitate transitions between the two, highlights its suitability for early-stage inpatient rehabilitation, where motor capabilities fluctuate substantially. Such flexibility may be particularly important for maximizing individualized rehabilitation dosage while ensuring safety and usability at the bedside. These findings indicate that the ENMS-assisted training is acceptable and feasible for participants with medical conditions similar to those in this study, representing approximately 80% of poststroke inpatients in the hospital department, with an annual intake exceeding 300.

In addition, the designated supervision is illustrated in Fig. 4A, the OT only provided requested assistance or supervision on the basis of the participants’ needs or requests. The large portion of the hands-on operation, e.g., attaching electrodes, mounting the device, etc., could be handled by a nursing assistant. Furthermore, once the training setup was completed, the participants could practice the training tasks in the ward without professional companions during most of the session. This practice was feasible in this study and could be applied to similar clinical applications. It was also observed that the rehabilitative professionals in the hospital of this study are open to new technologies that may enhance their practice efficiency. Neither the OT nor the nursing assistant expressed any hesitation or concern about integrating ENMS-assisted training into their routine clinical practice. Prior to the commencement of the study, their primary concerns centered on the actual therapeutic efficacy of the technology and the ease with which clinical staff could operate the device.

Moreover, only one participant discontinued the ENMS-assisted training for reasons unrelated to the intervention itself, with a dropout rate comparable to that observed in the other two professionally supervised groups. When considered alongside the reduced therapist involvement required, this low attrition provides further evidence supporting the feasibility of ENMS-assisted upper limb rehabilitation for bedridden inpatients in the early poststroke stage.

Professional manpower occupation

The groups in this study represented three different levels of professional manpower occupation. As demonstrated in Fig. 4B, the ENMS group required an average of 11.6 h of total OT involvement per participant across the 21 sessions, representing 49.4% and 44.8% decreases compared with the IC group (22.92 h) and the conventional (21.00 h) groups, respectively. The conventional group required the full occupation of the OT’s presence when serving a participant, including labor repetitions, concentration, and one-to-one manual operation. The IC group released the OT from the majority of the labor work in a training session. However, continuous supervision was still essential in the IC group due to safety risks associated with the use of large, rigid robotic cycling device powered by alternating current. As a shared device with limited ergonomic adaptability, the system was prone to misalignment between patient anthropometry and mechanical configuration, thereby increasing the likelihood of joint strain or inadvertent injury under improper positioning. Moreover, the device was capable of generating considerable output torque to facilitate target motions, and without precise modulation this could exceed participants’ tolerance thresholds. Accordingly, sustained therapist oversight was required to mitigate these risks by stabilizing multiple joints, correcting posture, and dynamically adjusting resistance levels in real time, which ensured both the safe execution of training tasks and the therapeutic efficacy of the intervention. In contrast, the WH-ENMS exhibited comparatively higher wearable comfort and ergonomic conformity through flexible textile-based glove, and its output torque generated by its pneumatic artificial muscles was relatively modest, which may contribute to an inherent safety profile. Training was conducted at the bedside with participants seated in a stable posture, and the affected forearm supported on an overbed table to further reduce potential risks. The ENMS system additionally provided standardized visual instructions and real-time feedback via a mobile application, facilitating semi-independent adherence to the training protocol. Therapist involvement was limited to periodic checkpoints, thereby reducing continuous clinical oversight and stratifying the workload into professional and administrative levels, with certain tasks shared by a nursing assistant. This arrangement enabled the OT professional manpower to be distributed to more participants and/or other professional duties.

Although the semi‑independent design of the ENMS demonstrated manpower‑saving benefits, the ENMS group generated a higher frequency of participant‑initiated requests for OT consultation and clarification of unfamiliar experiences when training alone, e.g., pain and muscle fatigue. The participants in the other two groups seldom reported similar alertness about the training experiences when the OT was present throughout the training sessions. This study incorporated predetermined OT checkpoints at sessions 1, 6, 11, 16, and 21, whereas these fixed intervals might be insufficient to adequately address participants’ training‑related questions. Therefore, more detailed investigations are needed to determine the optimized checking distance, which can further reduce professional occupation while maintaining a high quality of service.

Rehabilitation effectiveness

Clinical assessments demonstrated significant improvements in upper limb motor functions (FMA-UE, ARAT) and independence of ADLs (MBI) across all three groups following training (Figs. 5A–C). Compared with both the IC and conventional groups, the ENMS group achieved the most pronounced gains in voluntary motor function, reflected by significantly greater longitudinal increases in FMA-UE and ARAT scores (Fig. 6). Additionally, participants receiving ENMS-assisted training showed a reduction in muscle spasticity at the distal wrist and finger joints (Figs. 5D–F and 6). These enhanced outcomes in the ENMS group were attributable to the unique neurophysiological mechanism of the NMES-robot hybrid system, which integrates voluntary effort with NMES and soft exoskeletal support to deliver high-intensity, repetitive wrist-hand exercises that specifically modulate corticospinal pathways [25]. This approach establishes a bidirectional closed-loop sensorimotor system in which EMG-driven assistance facilitates the execution of descending motor commands (top-down control), whereas localized NMES delivers targeted sensorimotor afferent input to distal muscles (bottom-up input), inducing muscle contractions and supporting the integrity of sensorimotor pathways critical for functional recovery, particularly in wrist and hand functions. Compared with conventional physical therapy and IC training, the ENMS system offered enhanced sensorimotor integration and more precise neuromodulation. Conventional therapy relies primarily on therapist-guided voluntary movements, which might lack the intensity and muscular precision achievable by robotic or NMES-based interventions. The contralateral-driven paradigm in IC training inadequately quantifies or engages the voluntary motor effort of the impaired limb, thereby limiting its effectiveness in promoting neuromuscular pathway restoration.

The beneficial effects of ENMS-assisted training might also be related to increased neuroplasticity during the subacute “golden window”, a period characterized by increased synaptic remodeling, axonal sprouting, and increased responsiveness to sensorimotor input [46]. In contrast, the chronic phase results in diminished plasticity and consolidation of maladaptive compensatory patterns, restricting functional gains from similar interventions. This interpretation was supported by our prior studies in the chronic stage, where FMA-UE improvements in the present subacute cohort (+ 9.94 points) were more pronounced than those observed in chronic cohorts using similar ENMS systems (+ 5.67 and + 6.5 points, respectively) in clinical settings [26, 28], underscoring the critical role of the subacute recovery period.

Although IC training did not achieve the same degree of fine motor improvements as ENMS training did, it remains valuable for enhancing overall motor function, which is consistent with previous reports [47–49]. By leveraging neural coupling between the upper and lower limbs, IC training activates both cortical and subcortical motor pathways [50], promoting motor unit recruitment and gross motor coordination, as reflected by the highest MBI increase among the groups (Fig. 6). Furthermore, IC training may provide additional aerobic exercise benefits relevant to cardiopulmonary function, survival, reduction in recurrence risk, and comprehensive rehabilitation.

Conventional PNF techniques incorporate task-specific neurorehabilitation principles and actively engage proprioceptive feedback. Diagonal spiral movement patterns (e.g., D1/D2 flexion-extension) resemble functional upper limb tasks such as grasping and reaching, directly activating sensorimotor networks [51]. Techniques such as rhythmic stabilization, stretch reflex facilitation, and resistance applied to agonist–antagonist muscle chains further increase proprioceptive input and promote distal muscle activation [52], which is critical for restoring isolated finger movements. Although overall post-training outcomes did not differ significantly between the IC and conventional groups (Fig. 5), the conventional group demonstrated greater improvements in hand function, as evidenced by a significantly greater increase in the ARAT score (Fig. 6). This might reflect the strong relevance of PNF techniques to ARAT subtest demands, whereas the repetitive and stereotyped movements in interlimb coordination cycling may lack sufficient task-specific neural activation to effectively enhance fine motor control.

Limitations

In this study, most participants in the IC and conventional groups exhibited slight to no increase in muscle tone across joints after training, and no intergroup differences were observed. This limited sensitivity of the MAS might be partly attributable to baseline differences. Prior to training, MAS finger scores in the ENMS group were significantly higher than those in the conventional group, with no difference compared with the IC group, suggesting poorer muscle tone in the ENMS group, whereas participants in the IC and conventional groups presented relatively low spasticity and thus limited potential for further reduction. Notably, participants in the ENMS group achieved significant intragroup improvement, and analysis of change scores revealed a significantly greater reduction in muscle tone compared with both the IC and conventional groups. Although these findings require cautious interpretation given the baseline disparities in spasticity, they nonetheless suggest that ENMS-assisted training can effectively reduce finger spasticity. Future studies with balanced baseline characteristics and additional quantitative outcome measures, such as joint range of motion and instrumented spasticity assessments, are warranted to more sensitively evaluate therapeutic efficacy and to clarify whether the observed improvements are attributable to baseline differences or to the intervention itself.

Meanwhile, the comparison across the three groups did not constitute a fully balanced head-to-head design: the ENMS group received semi-independent training, whereas the IC and conventional groups underwent fully supervised training. In principle, continuous therapist supervision could be incorporated into the ENMS group to enhance the quality of movement execution. Nonetheless, the main objective of this study was to investigate the feasibility of ENMS-assisted semi-independent training with minimized therapist involvement, rather than to evaluate training effectiveness under identical training conditions. A commonly adopted upper and lower limb robot for inpatient rehabilitation used in the IC group was selected as a comparator to provide a meaningful clinical context, situating the feasibility assessment of ENMS within a relevant and practical reference framework. Future studies focusing on therapeutic efficacy could be designed with more balanced experimental conditions to enable direct comparison across groups.

Conclusion

This study demonstrated that semi-independent upper limb rehabilitation assisted by the WH-ENMS system is both feasible and well accepted among inpatients with subacute stroke, requiring significantly less professional OT involvement compared to routine IC training and conventional therapy. The ENMS-assisted approach effectively reduced the OT workload by enabling participants to perform training with minimal supervision while maintaining safety and addressing side effects through timely monitoring. Clinically, the ENMS group achieved improvements in voluntary motor function and reduced muscle spasticity in distal joints, attributable to the unique neurophysiological mechanism of the system that integrates voluntary effort with NMES and robotic assistance to enhance sensorimotor integration. These advantages underscore the capacity of this semi-independent ENMS-assisted training to maximize functional recovery during the critical subacute neuroplasticity window, optimize the use of available resources, and support the development of more personalized interventions aimed at improving functional independence and quality of life after stroke.

Abbreviations

ADLs

Activities of daily living

ANCOVA

Analysis of covariance

ANOVA

Analysis of variance

App

Application (mobile application)

ARAT

Action research arm test

CONSORT

Consolidated standards of reporting trials

ECU

Extensor Carpi Ulnaris

ED

Extensor digitorum

EF

Effect size

EMG

Electromyography

ENMS

Exoneuromusculoskeleton

FCR

Flexor carpi radialis

FMA-UE

Fugl-meyer assessment-upper extremity

IC

Interlimb-coordinated (training)

MAS

Modified ashworth scale

MBI

Modified barthel index

MMSE

Mini-mental state examination

MVC

Maximal voluntary contraction

NMES

Neuromuscular electrical stimulation

OT

Occupational therapist

PNF

Proprioceptive neuromuscular facilitation

RCT

Randomized controlled trial

VME

Voluntary motor effort

WH

Wrist-Hand

Author contributions

YWW contributed to intervention implementation and manuscript drafting. YHH was responsible for the data analysis and manuscript drafting. YRM carried out participant recruitment and clinical assessments. HJC and YJH assisted to intervention implementation and data acquisition. DFH conceived the study and provided critical revisions to the manuscript. XLH supervised the study and contributed to critical manuscript revision. All the authors read and approved the final manuscript.

Funding

This work was supported by ZSQY202373519, GRF 15218324, ITT/012/23GP, ITS/011/23, and STG1/M-401/24-N.

Data availability

The datasets generated and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the Human Participants Ethics Sub-Committee of the Seventh Affiliated Hospital of Sun Yat-sen University (Ethics number: KY-2023-033-03). Written informed consent was obtained from all participants before enrollment.

Consent for publication

Consent for publication was obtained from participants prior to enrollment in the study.

Competing interests

The authors declare no competing interests.