This review was a systematic scoping review that followed the methodological steps first described by Arskey and O’Malley [16] and subsequently refined by Levac et al [17], Peters et al [18] and Pollock et al [19]. A scoping review was determined to be the most appropriate review methodology given the paucity of clinical trials investigating the effectiveness of SAR interventions in clinical physiotherapy, which precludes the conduct of a systematic review [20]. As such, our review question was “What is known about the use of SARs within physiotherapy interventions and activities with clinical populations?”

The advantages of a scoping review approach are the development of a systematic and reproducible search strategy, with clear inclusion and exclusion criteria to comprehensively identify and chart existing evidence related to our review question and identify gaps and potential avenues for future research. This scoping review report follows the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-analyses for Scoping Review) [21]. Given the nature of the scoping review being a type of systematic review, no ethical approval was needed.

The scoping review search terms were structured using the population, concept, context (PCC) framework [18]. For this review, the population was any clinical population, the concept of interest was SARs, and the context of interest was the conduct of physiotherapy interventions or activities. To ensure a balanced and comprehensive yield of existing publications related to robotics in physiotherapy, a combination of health and engineering literature databases was systematically searched. The engineering databases searched for this scoping review were the Association of Computing Machinery (ACM) and the Institute of Electrical and Electronics Engineers (IEEE) databases. The health databases searched were Ovid MEDLINE, Ovid Emcare, Ovid EMBASE, Ovid Cochrane Library, Scopus, and the CINAHL.

A primary database search string was developed in consultation with an academic librarian as recommended for scoping reviews [18]. The first search was constructed for execution in MEDLINE using Medical Subject Headings (MeSH), database-specific thesaurus equivalent words and keywords for SAR (eg, “social* assist* robot”) and physiotherapy (eg, “physiotherapy”). Details of complete syntax are listed in (Multimedia Appendix 1).

The primary search string was then adapted for use in all other electronic databases by modifying syntax and using database-specific subject headings (ie, MeSH). In cases where a database did not use subject headings (such as Scopus, ACM, and IEEE), only keywords were used. A study date limit filter was applied to restrict the search for publications since the foundational 2005 publication of Feil-Seifer and Mataric [1]. All databases were searched up to November 23, 2023. Complete search strings for all databases searched are presented in (Multimedia Appendix 2).

Identified records from all databases were uploaded to reference manager software Endnote (Clarivate; [22]) and imported into systematic review software Covidence (Veritas Health Innovation; [23]) for screening. Records were automatically deduplicated in Covidence before conducting title, abstract, and full-text screening. Further manual identification and removal of duplicates were conducted during the screening stages to ensure no duplicate records appeared in the final yield.

All references were imported into Covidence [23]. One reviewer (JH) screened all records at the title and abstract stage against the eligibility. To confirm reviewer accuracy in applying the eligibility criteria, a subset of 25 studies was randomly selected for independent evaluation by all 3 review team members. The outcome of the 3 independent reviews of the 25 studies was compared and agreement on the screening outcome was confirmed.

Two reviewers independently screened all papers at the full-text screening stage. One reviewer (JH) screened all papers while the second independent review was conducted by either MF or PCM. Meetings between all 3 reviewers occurred at regular intervals during all screening stages, ensuring close supervision of JH’s screening progress and a unified and consistent interpretation of the criteria. Conflicts identified at the full-text screening stage were resolved by an initial discussion between the reviewing pair (JH and MF or JH and PCM) until a consensus was reached. If a consensus could not be reached by the reviewing pair, the third reviewer was consulted and discussion continued until a consensus was reached. Included studies advanced to data extraction, while exclusion reasons were documented for excluded studies.

A data-charting form was developed in Covidence [23] to extract relevant data from the included studies [4,24-35] . The types of data extracted are shown in (Table 1). The review team piloted the data-charting form with the first 2 records and revised it before charting the data from the rest of the included studies [4,24-35].

One of the reviewers (JH) completed the remaining data extraction process to ensure data extraction consistency. Two reviewers (MF and PCM) audited the extracted data from 4 included studies [26,30,34,35] (approximately 30% of the final yield) and identified negligible errors in the data extracted. Extracted data were then exported from Covidence [23] to a spreadsheet for synthesis.

Data synthesis and analysis was led by JH with regular review meetings with MF and PCM to confirm the accuracy of the synthesis process and devise a shared interpretation of the data synthesized. Charted data were first categorized into groups related to the scoping review questions: study and participant characteristics (population), features of physiotherapy (context), features of SARs (concept) used in interventions, robot design features, evaluation methods, key findings, limitations, and areas of application. Patterns were identified within each group and categorized, which aided the comparison and interpretation of data.

Subsequently, a more detailed analysis of the data extracted from included studies [4,24-35] by applying content analysis [36] was performed. Descriptive text extracted from studies relating to robot function and behavior was condensed, and patterns related to robot functions were coded and grouped into categories. Similarly, descriptive text concerning the role of physiotherapists, physiotherapists, and patient or client involvement in the protocol and robot design phase, key findings, limitations, and application areas were condensed, coded, and categorized. Throughout the data analysis process, all data were directly linked to the source material, and where necessary, the original text was revisited to confirm context or meaning if extracted data segments were ambiguous when being analyzed in isolation.

The database searches identified 9208 studies. After removing duplicates (3710 studies) and title and abstract screening, 215 studies were included for full-text screening. Of those, 202 studies were excluded with main reasons shown in Figure 1. Seven short papers related to our review question were identified at the full-text screening stage. A citation linkage search using Google Scholar was conducted to establish if any subsequent full-text reports had been published. No additional records meeting the inclusion criteria were identified by citation tracking of the studies included at the full-text stage. A final yield of 13 studies [4,24-35] met the criteria for inclusion in the review (Figure 1).

In this section, we begin by summarizing the general characteristics and demographic details of the selected studies. Following that, we provide an overview of the specific physiotherapy interventions, including prescribed exercises, dosage, and objectives.

To answer the question of how SARs are integrated into physiotherapy contexts, we categorized the information on SARs characteristics based on 4 subquestions.

1. What SAR platforms are most used in physiotherapy?

2. How do SARs and users interact with each other?

3. What roles do these robots serve within physiotherapy contexts?

4. What was the involvement of key stakeholders in the design and development process of SARs and intervention protocols, as well as the presence of technical support and the role of physiotherapists during interventions?

This information aims to provide insights into the relationship we seek to build between robots and users, and how to ensure that SAR interventions are designed with the needs of target users in mind, thereby promoting broader community adoption.

Table 4 summarizes the evaluation methods for treatment effectiveness reported in the included studies [4,24-35]. The outcomes are categorized into 2 groups. The first group includes clinical outcomes, user performance, or functional measures, which assess the direct impact of interventions on the user’s health and functional abilities, as well as their ability to perform tasks related to their functional goals. For example, Buitrago et al [24] used Goal Attainment Scaling [41] to evaluate the extent to which the patient with CP achieved his goals, with positive results indicating that the user successfully met the objectives during robotic training. Salas-Monedero et al [28] assessed upper extremity dexterity in patients with tetraplegia and paraplegia using metrics like the Box and Block Test [42]. Irfan et al [26] monitored physiological parameters, such as heart rate, the Borg scale [40], and blood pressure in a patient with a history of heart attack to observe the intervention’s effects.

The second group encompasses metrics that evaluate the acceptance, usability, and perception of robots. Among these, acceptance was the most frequently assessed construct, typically measured through subjective evaluations, such as interviews and questionnaires, with reports indicating a positive trend toward acceptance of SARs [4,26,29-31,33]. In addition, studies focusing on user perception reported positive feedback regarding the overall robot experience [35], usability [27], and increased user motivation and adherence to therapy [4,26,27,32]. However, 1 study found no improvement in adherence [33].

Although most of the included studies assessed the acceptance, usability, and perception of robots, 7 studies [4,29-32,34,35] did not report any clinical outcomes, user performance, or functional measures. This may limit the understanding of the actual therapeutic benefits of the robotic interventions.

Overall, 4/13 (31%) included studies [28,29,31,35] reported a limitation related to small sample sizes, thus limiting the generalizability of their findings. Selection bias was identified as a potential limitation in 1 study [31], where researchers may have been inclined to select participants they had previous positive interactions with. Other study limitations identified included slow recruitment rates due to strict inclusion criteria [33] and the homogeneity of participants from the same facility [29]. Technical issues and limitations in robot design were reported in 3/13 (23%) studies [33-35], such as system readiness issues and malfunctions leading to missed robot sessions [33], pelvic movement readiness and system heating concerns [34], and limited exercise variety due to robot size constraints [35].

In addition, 3/13 (23%) studies [4,24,34] highlighted limitations in intervention design. For instance, Buitrago et al [24] noted that, although the objective measured by the Goal Attainment Scaling was achieved during the training, neither the robot nor the therapist adjusted the task or the environment to reflect the new goal. Carrillo et al [4] reported that the therapist performed tasks for the robot as they were not informed of the robot’s capabilities. Other reported limitations included a limited range of friendly interactions, the need for refreshment breaks, and limited flexibility in exercise sequences [34].

Other identified limitations included robotic technology expenses [35] and uncertainty regarding the impact of color blindness in users [30]. Identified limitations in evaluation methodologies included difficulty in measuring motivation in very young children [32], a lack of assessment of the robot’s impact on physical health [33,35], and the absence of cognitive ability data to indicate acceptability [31]. Three studies [25-27] did not report any limitations.

The areas of application proposed by the authors of the included studies spanned the domains of rehabilitation, such as cardiac rehabilitation [26], pediatric oncology, and neurorehabilitation [24,28,30,32], and applications in older adult care facilities [35]. Other identified applications included supporting exercise, progress monitoring, and maintaining motivation and adherence [26]. In addition, studies’ authors suggested integrating SARs into existing robot-assisted gait rehabilitation [25], using SARs to address gaps in telehealth care by enabling real-time program adjustment based on robot-patient interactions [34], and supporting long-term home exercise programs for people with chronic conditions who may not have access or resources for prolonged therapist contact to assist with managing their condition [33]. One study [29] did not propose any specific areas of application.

This scoping review aimed to explore the usage of SARs in physiotherapy contexts with clinical populations. Across the 13 studies [4,24-35] identified, SARs’ roles and functions were tailored to suit the specific needs of patients and the objectives of physiotherapy interventions. These findings indicate that, in current applications, the primary objective of integrating SARs into health care is therapeutic augmentation [43].

Numerous studies reported that the use of SARs led to repetitive exercises and sessions [44]. This issue may stem from simplistic programming or a lack of personalized adaptation, which can result in a loss of user interest [44]. To address this, efforts have increasingly focused on complementing SAR interventions with devices like Lokomat or incorporating more advanced gamified exercises [4,25,27,29,30]. This review identified several SAR capabilities, including coaching, demonstration, monitoring, and peer support. Currently, SARs seem to be prioritized for roles, such as time-extending coaching and monitoring, functions that are considered unfeasible for human therapists due to the mismatch between the number of available therapists and the size of clinical populations [43].

The dynamics of socially assistive interaction are elucidated through the roles that SARs take on and the communication channels used between robots and users. In all but 2 studies [28,32], where communication methods were not reported, verbal communication channels were used to deliver feedback, encouragement, or instructions (along with visual and motion-related communication). Verbal praise to users is considered positive feedback and thus has the potential to enhance users’ intrinsic motivation for task performance [43].

Of the SAR models discussed in the review, the humanoid robot NAO is the most frequently used. Although there is limited evidence guiding the selection of the most suitable SAR model for physiotherapy applications, NAO has been commonly adopted in aged care and pediatric rehabilitation contexts, noted for its good acceptance by users and physiotherapists and movement capabilities [13,45]. The prevalent use of NAO in clinical settings may stem from its perceived usefulness in prior feasibility tests [46], its commercial availability and its well-established functional and technical capabilities [47].

The involvement of domain experts (physiotherapists) and end users (clinical populations) in the design of the SAR system and the intervention protocols was investigated, with a focus on understanding whether the systems adequately meet the needs of important stakeholders. The findings show a lack of participation from key stakeholders in both processes. Among the 13 included studies [4,24-35], only 2 studies [4,35] appeared to have involved end users in the design and development phase, while the remaining studies either did not report or did not engage stakeholders during these stages. In Back et al’s study [35], participants provided feedback on exercise programs to improve their effectiveness. In Carrillo et al’s study [4], the authors emphasized the early engagement of patients and physiotherapists in the design and development of SARs, highlighting the benefits of overcoming technological limitations, fostering effective engagement with end users, and facilitating the development of the robot’s demonstration function. A previous study has identified the value of stakeholder input in robot development, as it aids in identifying potential uses for developers, pinpointing flaws in current robot designs to avoid, and suggesting improvements to ensure usability, particularly in health care settings [48].

The majority of studies reported limitations within their findings, particularly regarding small sample size [28,29,31,35], challenges in patient recruitment [31,33], the robot’s design and technical capabilities [33-35], intervention protocol design [4,24], and evaluation methods [24,32,33]. Meanwhile, despite significant efforts in the application of SARs in physiotherapy, many of the studies are primarily pilot studies intended to pave the way to clinical trials with end users. However, they often fail to progress into larger studies due to the various limitations highlighted in Lo’s [49] paper, such as the inappropriate selection of control groups to assess therapeutic benefits, the absence of power calculations to detect minimally clinically important differences, and insufficient detail in robot rehabilitation protocols, hindering reproducibility. The current included studies have flaws, such as slow patient recruitment processes [31,33] and small sample size [28,29,35]. This could potentially compromise the validity of these trials and their impact on the clinical evaluation of the effect of SARs on patient outcomes.

Other limitations stem from the shortcomings of the current SAR designs and technological limitations, such as pelvic movement robustness, system overheating [34], robot size that affects exercise variety [35], and system readiness and malfunction [33]. This aligns with a study that investigated NAO-assisted pediatric CP rehabilitation, where most physiotherapists considered system or hardware failures during sessions as crucial features needing improvement [50].

In addition, the personalization of the robots to adapt to the user’s mood, fatigue, and performance variations during long-term involvement, such as in rehabilitation, remains a major challenge. In a review analyzing patients’ and clinicians’ opinions on the use of SARs in rehabilitation, the loss of user interest frequently arose as a concern [51]. This could pose a challenge in applying SARs when working with vulnerable user populations, supporting self-rehabilitation, or long-term exercise programs for chronic conditions [33].

Previous papers have described several barriers to the integration of SARs in clinical settings [44,49]. When recruiting end users (clinical populations) in clinical trials, several ethical issues must be carefully considered regarding patient physical and psychological safety, especially when involving vulnerable clinical populations (eg, children or people with cognitive impairments) [52]. Another significant hurdle is justifying the investment in clinical SAR research without clear economic or clinical evidence of its cost-effectiveness, which may impede the adoption of SARs in clinical settings. In addition, conducting research in the clinical setting can be time-consuming, as it requires multidisciplinary collaboration between robotics and health science experts who are rarely co-located [43]. These factors may indeed contribute to the high budget requirements of robot research in clinical settings.

A strength of this scoping review is the interdisciplinary review team with expertise in the fields of physiotherapy (JH and MF), robotics (PCM), and scoping review methodologies (MF). This interdisciplinary perspective was important for identifying relevant locations to search, terms to use, and knowledge of disciplinary publication conventions. While numerous medical and engineering databases were comprehensively searched, citation tracking was limited to short reports that otherwise met the inclusion criteria for pragmatic reasons. This may have resulted in the omission of potential studies suitable for inclusion, but this risk is considered minimal given that the citation tracking found no further eligible studies for inclusion.

Furthermore, one team member led the majority of title and abstract screening, data extraction, synthesis, and analysis. This approach may introduce bias into the study exclusion, data extraction, and synthesis processes, potentially leading to incomplete data capture due to the risk of overlooking relevant information and nuances [53]. The primary reviewer’s limited experience in scoping review methodology may further compound these potential biases [53]. However, this risk was counterbalanced in this review through the auditing of processes and data, the use of collaborative review tools with clear audit trails, and regular meetings for confirmation of decisions and team endorsement to progress after each review stage.

This scoping review identified what is currently known about the usage of SARs in physiotherapy interventions and activities with clinical populations. Despite the limited publications on this topic, the review offers insights into SAR design and application areas in physiotherapy interventions. This review has highlighted the future potential of SARs to be increasingly applied in health care settings. Future research should prioritize addressing the limitations in intervention design and evaluation methods to enhance the quality of evidence in this field. There is a critical need to resolve technical issues associated with SARs, such as system readiness, malfunction, size restriction, and overheating, to enhance the quality of interventions. Engaging key stakeholders, including physiotherapists and clinical populations, at all stages of SAR design and development is essential to overcoming technical challenges and promoting user acceptance. In addition, conducting more clinical trials and using reliable and valid clinical outcome measures are essential steps to investigate the effectiveness of SARs in promoting health outcomes.