A Scoping Review on Technology-Based Approaches for Upper Limb Motor Rehabilitation after Stroke. Are We Really Targeting Severe Impairment?

1. Introduction

Most technology-based approaches for motor rehabilitation after stroke have been originally designed and developed with severely affected patients in mind [1,2]. As a matter of facts, plegic patients have little or no access to traditional approaches targeting upper limb function, which often rely on the presence of residual motor capabilities. Furthermore, technology can provide treatment with increased intensity, which is established as a crucial determinant of rehabilitation outcome [3]. An example of how technology can impact motor recovery of the upper limb can be found in Constraint-Induced Movement Therapy (CIMT), still one of the most effective approaches [4] that requires residual motor function on the affected limb. The lack or scarceness of such residual motor function can be at least partially overcome with technology, such as neuromuscular stimulation combined with CIMT [5]. Thus, technology can not only enhance the intensity of the treatment but also increase the number of patients that can access a given therapy, increasing the chances for all patients (including the more severely affected) to regain independence.

However, randomized controlled trials (RCT) testing these technology-based rehabilitation approaches often include patients with a wide range of motor impairment, and sometimes a small proportion of patients actually falls in the severe range as defined by clinical scales. Indeed, patients’ recruitment in rehabilitation is influenced by numerous factors and is still one of the most challenging steps for researchers aiming to bring technological devices to clinical practice [6]. Severity of motor deficit often comes with concomitant conditions, such as bed-confinement status (especially in the acute and subacute phases), cognitive impairment, pain, depression which altogether limit the ability of patients to participate in such trials. However, as the motor status at baseline is probably the most consistent prognostic factor of rehabilitation outcome [7], we believe that this aspect can be accounted for the hesitating translational success of many of these technological approaches, for several reasons. On one hand, success of clinical trials may be facilitated by the participation of less severe patients, who increase trial feasibility especially in the subacute phase, but are naturally destined to a more favorable outcome. As a consequence, the actual application of a given technology in severe patients may turn out disappointing in terms of outcome, regardless of the successful trial. On the other hand, some of these technological aids may provide little added benefit for less severe patients, who can take advantage of other, more traditional and less expensive rehabilitative exercises [4].

In this scoping review, we provide an overview on the actual severity of patients recruited in RCTs carried out to analyze the efficacy of technology-based rehabilitation approaches targeting severe upper limb impairment (explicitly mentioning patients severity in the title and/or abstract). These interventions are based on robotics, electrical or magnetic stimulation of the central nervous system (non-invasive brain stimulation, NIBS) or of peripheral structures (peripheral stimulation, PS), brain-computer interfaces (BCIs), virtual reality (VR) and, in general, advanced technological devices purposely developed for post-stroke motor rehabilitation. The identification of the most effective approaches is out of the scope of this review [8], where we intend to verify

i)

the actual severity of patients included in trials that explicitly declare to enroll severe subjects, to confirm or refute the anecdotal notion of extreme variability in baseline motor impairment which might be responsible for the hesitating translational success of such interventions [9]

ii)

whether some of these technological approaches have been more consistently tested on severe patients than others and eventually speculate on why they have.

Furthermore, we check whether the severity of patients was considered as an inclusion criterion and/or employed for stratification for primary or secondary analyses. We take into account whether each considered study achieved to confirm the efficacy of a given intervention or not, that is the superiority/not superiority of the target intervention vs the control condition (positive/negative results) also highlighting possible secondary analyses that were carried out by the authors to support their hypotheses.

The ultimate goal would be to provide useful hints to improve patients’ inclusion in RCTs, in order to favor the optimization of resources and efforts towards tailored, highly technological rehabilitation interventions that is instrumental to foster technology transfer into clinical practice.

2. Methods

2.1. Protocol and Eligibility Criteria

This scoping review was conducted according to the PRISMA standards. Our protocol was drafted using the Preferred Reporting Items for Systematic Reviews and Meta-analysis extension for Scoping Reviews (PRISMA—ScR, [10]) and revised by the research team. To be included in the review, papers needed to describe technology-based rehabilitative interventions targeting the recovery of motor function of the upper limb in stroke population.

Peer-reviewed journal papers were included if they were:

• published between the period of 2008–2024,
• written in English,
• involved human participants in the framework of a Randomised Controlled Trial.
• Papers were excluded if they
• did not fit into the conceptual framework of the study,
• were reviews, study protocols, and meta-analyses.

2.2. Information Sources and Search Strategy

Studies eligible for review were identified through electronic databases such as Scopus and PubMed from 2008 to 23 May 2024. Grey literature was excluded from the search process. The search strategies consisted of free text terms in the topic: “stroke” AND “severe” AND “rehabilitation” AND “Action Research Arm Test” OR “Fugl-Meyer Assessment” AND “randomised controlled trial” AND “upper limb” OR “hand”. The complete search terms and strategy are provided in the Supplementary materials (Table S1). The search strategies were developed and executed by a biomedical engineer (EC) and further refined through team discussion. The search was peer-reviewed by other expert researchers, i.e., a neurologist (FP) and a physiatrist (GM), using the Peer Review of Electronic Search Strategies checklist and modified as required [11]. The final search results were imported into the online systematic review software Rayyan [12]. Duplicates were identified by means of the Rayyan duplicates search algorithm and removed by a researcher.

2.3. Selection of Sources of Evidence

Two reviewers (EC, FP) screened titles and abstracts for inclusion, independently. For full-text screening, two reviewers (EC, FP) subsequently screened the full-text of potentially relevant articles to determine inclusion using similar inclusion and exclusion criteria. We resolved disagreements on study selection by consensus and discussion with other reviewers if needed or by a single arbitrator (GM). To ensure reliability between reviewers, a series of training exercises was conducted prior to commencing screening. Inter-rater agreement for study inclusion was calculated using percent agreement; when it reached > 80% across the team, we proceeded to the next stage. If lower agreement was observed, the inclusion and exclusion criteria were clarified, and another pilot test occurred.

2.4. Data Charting Process and Data Items

Data-charting form was jointly developed by three reviewers to determine which variables to extract. Three reviewers (EC, FP, GM) independently charted the data, discussed the results, and continuously updated the data-charting form in an iterative process.

For each article, data on the following characteristics were extracted:

• First Author Name
• Year of publication
• Source
• Population sample size (participants per group)
• Severity of the upper limb impairment, i.e., Upper Extremity Fugl-Meyer Assessment score, FMA [13], and/or Action Research Arm Test, ARAT [14], expressed as mean ± standard deviation (SD) or median and first and third quartile (Q1-Q3), per group, whenever available
• Inclusion Criteria in the RCT related to the upper limb impairment
• Availability of the dataset used (Yes/No)
• Time since injury (TSI), i.e., stroke event, classified as

  ◦

  ≤ 1 month (acute)

  ◦

  ≤ 3 months (early subacute)

  ◦

  ≤ 6 months (subacute)

  ◦

  > 6 months (chronic)

• Intervention type, classified as

  ◦

  Brain-Computer Interface (BCI)

  ◦

  Non-Invasive Brain Stimulation (NIBS)

  ◦

  Peripheral Stimulation (PS)

  ◦

  Robotic

  ◦

  Virtual Reality (VR) and Visual

• Comparator, i.e., control interventions and/or comparison conditions
• Active Motor Action required (Yes/Yes whenever possible/No):

  ◦

  Yes, if the intervention type requires participant’s residual motor ability (active motor exercise from the participant)

  ◦

  Yes whenever possible refers to conditions foreseeing active motor exercise when feasible, with the technology providing assistance as needed (e.g., robotics)

  ◦

  No otherwise

• Combination of technological interventions (Yes/No)
• Dose, expressed as minutes x number of sessions
• Primary and Secondary Outcomes
• Key Findings, classified as Positive, Positive on secondary analyses, Negative. We define Key Findings as

  ◦

  Positive if between-group statistical analyses evaluated for the primary outcomes statistically confirm the hypothesis investigated in the study.

  ◦

  Positive on secondary analyses if between-group statistical analyses evaluated for sub-items of the primary/secondary outcomes or considering sub-groups of the population under investigation confirm the hypothesis investigated in the study or if within-group statistical analyses evaluated for the primary/secondary outcomes reveal a statistical improvement only for the experimental group.

  ◦

  Negative if between- and within-group analyses do not reveal statistically significant differences among groups.

Threshold for statistical significance was set to 0.05.

• Stratification for secondary analyses according to an upper limb impairment criterium
• Follow-up (Yes/No), i.e., if Yes, we reported the number of months after the end of the intervention
• Setting: Inpatient/Outpatient.

The extracted data were collected in a table in which rows represent the included articles, columns represent variables (data items). The spreadsheet software, Microsoft Excel, was used to create our custom extraction form. The choice was based on its ease of use, high customizability and worldwide diffusion. Before extracting the data from all papers included in the scoping review, the extraction form was tested for further refinements and underwent a calibration phase. This entailed three reviewers independently extracting data from 5 papers each and meeting afterward to discuss any discrepancies, with further refinement of the form if a high level of agreement between reviewers was not obtained.

2.5. Synthesis of Results

To analyse the database the following explanatory approaches are used: descriptive and frequency analysis and association analysis.

2.5.1. Descriptive and Frequency Analysis

Descriptive statistics concern the population of participants that are overall included in the selected studies in terms of: number of participants, sample size of intervention and control groups, dose of intervention, severity of upper limb motor impairment upon enrolment, as described by FMA and/or ARAT. Data have been summarized, according to their distribution (modality and dispersion), by means of mean and standard deviation (SD) or median and interquartile range (IQR), presented as difference between the first quartile (Q1) and the third quartile (Q3), i.e., Q1-Q3.

Frequency analyses concern the following variables: time since injury (at least 4 classes, i.e., acute, early subacute, subacute, chronic participants), setting (at least 3 classes, i.e., inpatient/outpatient/ inpatient and outpatient), availability of the dataset used (2 classes, i.e., Yes or No), technological rehabilitative intervention type (at least 5 classes, i.e., BCI, NIBS, PS, Robotic, VR and Visual) and whether it did or did not require an active motor exercise from the participants (3 classes, i.e., Yes/Yes whenever possible/No), comparison conditions, primary and secondary outcome measures, presence/absence of follow-up evaluations (2 classes, i.e., Yes or No), Key findings (3 classes, i.e., positive, negative and positive on secondary analyses), severity of upper limb deficit employed as inclusion criteria for participant enrolment and/or stratification for secondary statistical analyses. In frequency analysis, the counts and percentages of articles in each cluster are calculated. Studies that share a similar approach towards a specific variable are clustered together and those following different approaches are assigned to different groups. Clustering can be carried out based on values of a single variable on the entire data set or on a subset of articles which already belong to a cluster on a higher level.

All results are presented both narratively and by means of plot and pie charts when relevant.

2.5.2. Association Analysis

Association analysis explores the relationships between the variables. Since the number of possible combinations of variables are relatively large, the results are focused on those regarding the research questions. We hypothesize that the characteristics of the technological rehabilitation interventions would determine a difficulty in recruiting severely affected participants. Therefore, we have separately analyzed the severity upper limb motor impairment at baseline (as assessed via FMA) according to the

• type of intervention (e.g., Robotic, BCI, PS, …)
• required active upper limb motor actions from the participant by the intervention itself.

For each analysis, studies that share a similar approach, i.e., type of intervention or required active motor action, are clustered together. For each level of the analyzed variable, i.e., 5 levels for the variable TYPE OF INTERVENTION and 3 levels for the variable MOTOR ACTION, FMA data are pooled together. If needed, mean and SD data are estimated from data reported in the paper as median and IQR by means of the formula in [15].

All results are expressed as mean ± SD and presented both narratively and by means of plot charts.

3. Results

3.1. Selection of Sources of Evidence

The search returned a total of 189 papers. After the duplicate removal (96), 93 articles were screened. After screening of titles and abstracts, 41 papers were excluded, due to the following reasons:

• the rehabilitative intervention under investigation does not include a technology-based approach; videos/instruction displayed on screens or other devices of everyday use (personal computers, tablet, smartphones) were not included
• the design of the study does not follow the Randomised Controlled Trial design (wrong study design)
• the effectiveness of the rehabilitative intervention under investigation was not assessed in term of motor function improvement (wrong outcome)
• the paper presents a study protocol, a review or meta-analysis.

A total of 52 full-text papers have been examined. Four studies have been excluded during full-text search and check; thus 48 articles have been included in the scoping review.

Flow-chart in Figure 1 presents the detailed search and selection process.

3.2. Results of Individual Sources of Evidence

Results of individual sources of evidence are shown in Table 1. Table 1 reports the following subset of the items among those described in the section Data charting process and data items:

• First Author Name and Year of Publication
• Population sample size (participants per group)
• Severity of the impairment (FMA and ARAT per group, whenever available)
• Inclusion Criteria for the RCT related to the upper limb impairment
• Time since Injury
• Intervention Type
• Active Motor Action required
• Comparator
• Primary Outcome
• Key Findings

3.3. Synthesis of Results

3.3.1. Descriptive and Frequency Results

The included articles report data from 3000 adult participants. Three articles [18,23,52] include participants with a diagnosis of stroke and traumatic brain injury (TBI), respectively 5 and 3 TBI participants in [18] and [23], for a total of 8 participants. No detailed information on stroke/TBI ratio is reported in [52].

The average sample size of the groups (target intervention and control groups) is 29 ± 39 (mean ± SD). The RCT in [39] is the only study in which a very large number of participants, i.e., 770 participants, was enrolled. Conversely, there is more than one study in which a very small number of participants, i.e., less than 10 per group, is analysed [21,27,34,44,45,49,58]. Therefore, median and quartiles values, i.e., 20 (median) and 11-32 (Q1-Q3), provide more accurate estimates on the target intervention and control group sample size.

Enrolled participants include stroke in acute, early subacute, subacute and chronic phase. Most studies (62.53%) consider a homogeneous group of participants: 6.25% acute stroke participants, 16.70% early subacute stroke participants, 2.08% subacute stroke participants, 37.50% chronic stroke participants. The remaining studies (37.47%) include more than one group of stroke participants who differ in terms of time from the cerebral lesion to enrolment, e.g., early subacute and subacute. The distribution of studies across TSI classes is reported in Figure 2.

Most participants enrolled in the studies are inpatient (60.40%). Outpatient studies and those considering both inpatient and outpatient participation are 20.85% and 6.25%, respectively. For the remaining 12.5% there are no clear indications referring to the setting.

As for the severity of upper limb motor impairment upon enrolment, as described by FMA and/or ARAT, Figure 3 reports respectively the FMA (panel a) and ARAT (panel b) score at baseline of participants enrolled. Forty-two studies assess motor impairment at baseline in stroke participants by means of the FMA score; fifteen studies by means of the ARAT score. As shown in Figure 3, both FMA and ARAT scores at baseline extend throughout the whole range, i.e., FMA: 21.71 ± 11.41 (mean ± SD across 42 studies) and ARAT: 12.28 ± 10.85 (mean ± SD across 15 studies).

The complete dataset about characteristics of each participant enrolled in the study (e.g., individual FMA or ARAT scores) is available in 15% of studies.

As for the type of technological rehabilitative intervention, the pie chart in Figure 4 shows the distributions of studies across intervention types. Rehabilitative interventions administered by means of robotic devices and peripheral stimulation and their combination cover more than half of the tested interventions: 27.08%, 29.17% and 6.25% respectively. Fewer studies focus on BCI-based and NIBS interventions, both combined with other intervention types, and Virtual Reality and Visual rehabilitative interventions: 16.66% (8 articles), 14.58% (7 articles) and 6.25% (3 articles) respectively.

Most rehabilitative intervention approaches require participant’s residual motor ability (active motor exercise from the participant, 52.10%), 10.40% are categorised as “Yes whenever possible” referring to conditions foreseeing active motor exercise when feasible, with the technology providing assistance as needed (e.g., robotics), and 37.50% do not require any active motor action from the participants.

As for the dose of rehabilitative intervention, studies differ among them in terms of both minutes of each training session, 40 minutes (median) and 30-60 minutes (Q1-Q3), and number of training sessions (21.78 ± 14.39, mean ± SD) ranging from 12 sessions (Q1) to 28 sessions (Q3). The overall dose of intervention, resulting from the multiplication of minutes per session and number of sessions, is 13h (median) and 8-27h (Q1-Q3).

Table 2 shows for each surveyed primary and secondary outcome, the number of studies in which each outcome is used as primary or secondary. As for the primary outcome, the Upper Extremity Fugl-Meyer Assessment results the most frequent scale administered for the assessment of the efficacy of the rehabilitative interventions (38 on 48 studies, 79.17%). Conversely, as for the secondary outcomes, several evaluations, clinical/functional as well as instrumental, are considered. Among them the most frequent are the Stroke Impact Scale [64] (22.92%), the Barthel Index [65] (22.92%), the Wolf Motor Function Test [66] (18.75%), the Action Research Arm Test [14] (16.67%), and even also transcranial magnetic stimulation, electroencephalographic, electromyographic, kinematic and kinetic parameters (brain outcome and motor outcome 31.25%).

In 37.50% studies, participants are followed-up from 2 weeks to 9 months after the end of the rehabilitation, i.e., 3 months (median) and 3-5 months (Q1-Q3). For the other studies (62.50%) no follow-up evaluations are reported.

As for the comparison conditions, most studies (81.25%) are 2-arm RCTs. The remaining 18.75% compare more than 2 groups, up to 4 groups in [53]. For the 2-arm studies, the control conditions that are most commonly observed can be categorised as follows:

• Sham Stimulation/Control (applies to NIBS/PS and BCI studies, referring to conditions where the participants are induced to believe they are receiving stimulation or controlling a BCI system while they are not): 28.20%
• Similar intervention “without technology” (e.g., mirror therapy in contrast to VR-based mirror therapy): 17.95%
• Usual care: 17.95%
• Dose equivalent upper limb training (dose equivalent therapy session focused on the upper limb, considered in add-on to usual care): 10.26%
• Different combinations of technology-based approaches: 10.26%
• Different technology: 7.69%
• Different parameters of the same technology (e.g., different robotic assistance, anodal vs cathodal transcranial direct-current stimulation): 7.69%.

For studies comparing more than 2 groups, different interventions, e.g., as for technology or parameters, are mostly compared to the usual care control condition.

As shown in Table 1, most studies (89.58%) employ inclusion criteria related to the upper limb impairment for the participant enrolment. Figure 5 shows the inclusion criteria and their frequency across studies (43 studies include inclusion criteria). More than half of the studies define inclusion criteria by means of a FMA score (53.50%) or ARAT (9.30%). Nevertheless, evaluation based on motor outcome, i.e., range of motion or ability/inability to perform a specific task, such as that involved in the study protocol, are taken into account in 16.3% studies, as well as spasticity-related scales, i.e., Brunnstrom stages and Tardieu scale, in 11.63% studies. Specifically, for the FMA inclusion criteria, Figure 6 shows the reduction gained by defining inclusion criteria for the participant enrolment. That reduction, expressed as percentage, is computed as the one’s complement of the ratio between the FMA range defined as in the inclusion criteria and the whole FMA range (66-0), according to the following formula:

$$FMArangereduction=100\mathrm{*}(1-{\displaystyle \frac{FMArangedefinedasinclusioncriteria}{66}})$$

Seven studies make a reduction between 70% and 80%; among them, the study by Schrader et al. [52] in which inclusion criteria have been defined on the basis of the hand section of the FMA scale (maximum value 14). Most of the studies achieve a 50% reduction, thus including in the same analyses participants who differ in FMA score by 33, i.e., 50% of (66-0), for example in the range [8 38] in [61].

As for the severity of the upper limb impairment at baseline for the stratification of participants in secondary analyses, 25.00% studies exploit such evaluation, defining two or more levels of the variable: FMA (9 studies on 12) and ARAT, motor evoked potentials and range of motion (one per study).

As for the key findings of the trials, i) 41.67% report positive results, i.e., the studies confirm the hypothesis of efficacy of a given intervention via between group analysis, ii) 31.25% report positive results on secondary analyses, i.e., between-group statistical analyses evaluated for sub-items of the primary/secondary outcomes or considering sub-groups of the population under investigation confirm the original hypothesis or if within-group statistical analyses evaluated for the primary/secondary outcomes reveal a statistical improvement only for the experimental group. Negative results are reported from 25.00% of studies, i.e., the target intervention is not superior to the control condition.

3.3.2. Association Results

Figure 7 shows for each type of intervention (top panel in the figure) and motor action required to the participant by the intervention itself (bottom panel in the figure), the upper limb impairment severity (UE-FMA) of participants recruited in that group. Each study has been categorised both according to the intervention type (analysis presented in the top panel) and motor action required (analysis presented in the bottom panel). Data from studies belonging to the same category, e.g., intervention type PS, are pooled together and summarised by means of box charts.

As for the intervention type, most seem to cover almost half of the UE-FMA scale. On average PS-based interventions concern participants with a slightly lower UE-FMA value (18.46 ± 11.55) than other interventions (BCI: 19.52 ± 9.72, NIBS: 25.47 ± 10.67, Robotic: 23.46 ± 10.74; VR and Visual: 29.05 ± 10.36). However, robotic interventions, as shown from the distribution outliers, include studies with UE-FMA values both lower than 5 and higher than 40. As for the motor action required to the participant, studies in which experimental protocol does not require action to participant result to enrol participants with UE-FMA on average lower (19.44 ± 10.62) than studies requiring active motor action (22.41 ± 9.94), even whenever possible (25.14 ± 14.09). However, studies that require active motor actions enrol participants with reduced motor action ability (FMA < 5) as well as participants with moderate motor impairment (FMA=43).

4. Discussion

In this scoping review we provide a portrait of the current evidence derived from RCTs investigating the efficacy of technology-based interventions targeting upper limb motor recovery in patients with severe impairment after stroke. Our main aim is to verify the actual severity of the included patients enrolled in such trials, to confirm or refute the anecdotal notion of extreme variability in baseline motor impairment which might be responsible for the lack of strong evidence supporting efficacy of such interventions [9].

We included papers reporting at least one between FMA and ARAT to assess upper limb motor impairment (Table S1, Supplementary Materials), as they are commonly employed to investigate efficacy rehabilitative treatment [67]. Our results (Table 2) show that FMA is by far the most commonly employed as primary outcome measure, followed by ARAT; other measures that are specific for upper limb function are more commonly employed as secondary outcome, such as BBT, WMFT, MAL.

According to our descriptive analyses on FMA and ARAT values upon enrolment, we verified that the included patients altogether virtually cover the whole range of both scales (Figure 3a,b). That is, these RCTs that were originally implemented to investigate the efficacy of rehabilitative interventions designed for severe patients often include patients with mild to moderate deficits as well. This occurs despite the fact that almost 90% of the studies actually defined inclusion criteria based on severity. As a matter of facts, there is no unique definition of severity, even for the same assessment scale. Again, among those studies in which the inclusion criteria were based on severity, about 50% employed FMA for such a definition. However, the references for the proposed stratification did not converge on a unique subdivision. For example, several papers refer to Fugl-Meyer et al. [13] or Gladstone et al., [68] to justify the use of cut-off values for their inclusion criteria or stratification analyses, but no subdivision is provided in neither of these papers. Woodbury et al., [69] suggest a cut-off below 19 for severe patients and of 47 for moderately impaired patients, which is applied as an inclusion criterion by Carrico et al., [36]. Ding et al. [37] apply the clustering suggested in Woytowicz et al. [70] to define severity in patients with FMA < 35. Conroy et al., [38] apply a cut-off of 25 as suggested by Luft et al., [71]. As a possible explanation for such a wide range of severity among the enrolled patients, we hypothesized that the technology in study could play a role. For example, some robotic devices cannot be proposed to patients with severe spasticity, or an electromyographically triggered orthosis cannot be activated if patients have no residual movement in the target muscles. Thus, we categorized the papers according to the proposed technology. Electrical or magnetic stimulation of peripheral structures (PS) are the most represented technology in our review, followed by robotics. These two or their combinations represent approximately 63% of the studies. The BCI and NIBS (often in combination with other devices) are the topic of about 31% of the studies, while the less represented technology is VR, with just about 6% of the papers. In an attempt to further characterize the interventions according to the fact that they required (or not) active motor actions from the patients, we found out that more than 50% of the interventions required some residual motor ability from the patients. This could be an explanation for the trend to include “not-so-severe” patients in order to increase trial feasibility. Nevertheless, we did not observe a clear difference in the severity of the included patients when we divided the studies according to the technology or to the presence/absence of an active motor exercise. The distributions of patients’ severity (Figure 7) qualitatively show a tendency towards the more severe range for the interventions based on PS and for interventions not requiring active motor tasks. We also highlighted the heterogeneity of included patients in terms of time since injury. While the majority of studies targeted chronic patients exclusively (37.5%), the studies including mixed groups altogether reached a similar percentage (37.47%). It is well known that brain plasticity that underlies motor recovery is time dependent and the recovery potential is different according to the post-stroke phase [72].

Regarding the study design we also pointed out extreme variability in terms of dose/intensity of treatment, type of comparator and sample size. In almost 20% of the two-arm studies “usual care” is the only control condition, while an active comparator would be most desirable. Different dosage and very different sample sizes (ranging from 10 to almost 800 participants) limit the potential impact of our attempt to interpret the results of this scoping review that takes into account such a wide scenario of clinical studies. An indirect result of our scoping review derives from the papers that we excluded for reporting study protocols (n=13). These papers were all published between 2017 and 2023 [73,74,75,76,77,78,79,80,81,82,83,84,85], with ten of them being subsequent to 2020 testifying a progressive increase in the efforts dedicated to rigorous clinical trial design in this field of clinical research.

The ultimate goal of our review work is to improve the design of RCTs to boost the translation of rehabilitative technologies into clinical practice. Indeed, the majority of the selected studies report positive results, indicating efficacy of the proposed technology. However, about 30% of the studies required secondary analyses to support the benefit derived from the intervention in study. Some of them report positive results on secondary outcomes e.g., SIS and WMFT [17,38] and/or outcome sub-items, e.g., ARAT, FMA, MAS items [21,41]. Some required subgroup analyses e.g., proving efficacy only in severe patients [32,41,43,50]. The remaining 25% of openly negative studies (showing non-superiority of the proposed intervention) is likely underestimated, as it is known that publishing a negative result is harder and often occur only in the case of very well designed and well conducted, large clinical trials.

A further observation derived from our work is that along with established clinical and functional scales employed as primary or secondary outcomes (Table 2), 16 papers applied objective measurements of brain and motor activity as means to verify the effects of the intervention on the motor system performance. Advanced analyses on electroencephalographic [26,49] and transcranial magnetic stimulation recordings [43,49,58,60], electromyographic, kinematic and kinetic data [20,24,40,45,49,51,56,57,60,61,62] were performed in these studies to identify modifications in the motor system performance subserving a favorable motor outcome derived from the intervention in study. This suggests that technology is not only employed for the design of rehabilitative interventions but may play a crucial role in improving the outcome assessment with more objective, measurable and reproducible parameters that may serve as biomarkers of motor recovery.

This scoping review has several limitations, mostly related to the heterogeneity of the included studies in terms of numerosity, characteristics of enrolled patients, trial design, implementation, and reporting. We considered papers published between 2008 and 2024 observing a progressive improvement in all of these aspects which will hopefully be fruitful in the upcoming years to properly address some of the issues that we raised here and that remain unanswered. Only 7 studies provide a complete dataset documenting the individual severity of enrolled patients, that would allow a statistical analysis on how effective the interventions were according to a homogeneous stratification of patients based on severity. The policies for data availability are also progressively pointing towards safe data sharing according to internationally agreed General Data Protection Regulations.

5. Conclusions

We represented the current evidence derived from RCTs investigating the efficacy of technology-based interventions targeting upper limb motor recovery in patients with severe impairment after stroke. Our aim was to shed light on the problems of the current research in rehabilitation technologies, to ultimately boost the translational success of such approaches. There is undoubtedly a need for patient stratification upon enrolment, in order to direct resources to patients who will benefit the most from a given approach. The correct taxonomy of patient severity and the related correct reporting in clinical trials could significantly improve the transnationality and contextualization of the results obtained, avoiding biases that could affect potential effectiveness. Only severe patients should be recruited for the clinical validation of devices that are designed specifically for them, while design and development of technologies with adequate sensorimotor and cognitive stimulation would probably increase their salience (and effectiveness) for less affected subjects. As clinical trial design, implementation and reporting improves in this research field, subsequent systematic reviews will probably help in identifying strong evidence and thus clear indications for clinicians.