A prospective, pilot, feasibility trial of hemiparetic stroke patients with independent outcomes assessment and longitudinal follow up for 24 weeks was conducted in a single ambulatory centre, with affiliations to a tertiary inpatient rehabilitation unit. The goal of this study was to determine the feasibility, safety, and acceptability of implementing a clinic-to-home rehabilitation pathway using RAT, deploying a portable, 2D planar end-effector robot designed for upper-limb therapy, focused on training shoulder and elbow flexion and extension (Fig. 1a). The system provides smart physical human–robot interaction (haptics) in partial substitution of physical interaction with a therapist, coupled with remote telemonitoring via a web-based software (www.articares.com) (Fig. 1b).

Hardware

The upper-limb rehabilitation robot employed in this study is shown in Fig. 1a. H-Man is a portable, planar end-effector device designed to help train arm movements, and is essentially a powered, cable-driven differential mechanism [33].

The mechanism design provides the following advantages:

High back drivability: Back drivability refers to the ease with which the user can move the handle in the absence of motor actuation. Compared to other robot designs, the inertia and friction felt by the user’s hand when moving the handle are minimal. In this way, the user concentrates on performing the training tasks rather than overcoming the resistance of the mechanism. The device’s high back drivability also eliminates the need for feedback control, which in turn guarantees the robot’s contact stability. This makes H-Man inherently safe for manual interaction with the user.

Optimal workspace dimensions: The workspace of H-Man on the horizontal plane is 334.5 mm x 350 mm which defines all possible positions of the handle. The total footprint of the device is 665 mm x 620 mm x 105 mm.

H-Man can provide end-effector forces of up to 23 Newtons in any specified direction of the planar workspace to collaborate in the rehabilitation task. Previous clinical studies with Home-RAT can be found in [14, 20, 33].

Robotic intervention (exergames)

Therapy sessions with Home-RAT involved the participant performing a series of game-like training exercises or ‘exergames’ provided by the robot’s software. The exergame’s graphic user interface provides the user with a virtual manual task to execute, such as capturing fish in a pond, serving meals to customers, etc. The Home-RAT interacts physically with the user by exerting controlled forces on the handle. Depending on the type of task, these forces can either help the user in completing the required movements or create a challenge, such as adding resistance or introducing perturbations. In some games, the control software features an adaptive component that automatically adapts the intensity of the therapy to the patient’s current level of recovery. Tables 5 and 6 in Appendix 1 present a summary of the exergames employed in this study.

Exergames were prescribed by an OT and tailored to each participant’s needs; working towards prescribing exergames to improve arm coordination, strength and/or agility. Agility was defined as the average speed of point-to-point movements.

The exergame’s levels of assistance, resistance or perturbation are adjusted in the robot software based on the patient’s kinematic performance metrics. The metrics are computed from the sensor data (specifically handle position data) generated by the robot during the patient’s previous training exercises.

As the exergaming interface was in development during the pilot trial and initially commenced with 3 exergames, halfway into the study period, a further 5 exergames were added by the software developers. Consequently, we assigned participants to 2 groups for participation evaluation; Group 1 consisted of 6 participants (P01-P06) who trained with 3 different exergames at home. We refer to these as exergames E1. (Appendix 1, Table 5). Group 2 also consisting of 6 participants (P07-P012; group 2). For the trials with group 2 we incorporated a new set with an additional 5 exergames; we refer to these as exergames E2 (Appendix 1, Table 6). Thus, group 2 trained with a total of 8 exergames at home (Appendix 1, Table 6).

Remote monitoring software

The H-Man is controlled by a software application called the CARE Platform [34]. The software features a remote monitoring component capable of linking up the supervising clinician with one or several patients receiving robotic therapy in their homes (Fig. 2). In compliance with the institution’s Medical Devices and Operational Technology Security (MDOTS) [35], no personal identifiers (name, identity numbers, addresses) were stored in the robot or web-based platform which was not connected with the healthcare institution’s network and H-Man robot external USB ports were disabled.

Fig. 2

Schematic illustration of H-Man and web platform architecture. Motion and performance data are generated from each training session with the robot. Data collection is performed by the software application (CARE Platform installed in the robot’s PC). Only non-identifiable (non-PII) data are collected from the user. The bulk of the data consist of robot motion data and performance data generated during training. Data are uploaded in encrypted form to a secure cloud-based server. Data can be accessed remotely by registered users (for example the supervising clinician) by means of a web-based software application

The software’s communication framework featured encrypted transmission of training data from Home-RAT to a secure database, and generation of data analysis and progress reports, allowing remote access by clinicians with secure log in passwords to view and manage participants’ therapy schedules and generate reports remotely.

Study setting

The study was conducted from 3 March 2022 to 1 September 2023 at the Tan Tock Seng Hospital, Clinic for Advanced Rehabilitation Therapeutics (TTSH-CART) in Singapore, an ambulatory rehabilitation facility providing comprehensive medical rehabilitation consultations and multi-disciplinary rehabilitation therapies, incorporating various rehabilitation technologies (e.g., robot-aided therapies, virtual reality training, neuromuscular electrical stimulation etc.). TTSH CART is directly linked to Tan Tock Seng Hospital (TTSH) Rehabilitation Centre, a 95-bed inpatient tertiary rehabilitation unit providing acute inpatient neurorehabilitation programs.

Study participants

The majority of participants had completed inpatient rehabilitation at TTSH Rehabilitation Centre and were recruited consecutively according to the following study inclusion criteria; first-ever clinical stroke (ischaemic or hemorrhagic) confirmed by admitting doctors and CT, CT angiography or MRI brain imaging, aged 21–90 years, duration of > 28 days post-stroke, upper limb motor impairment measured with Fugl-Meyer Motor Assessment scale (FMA) scale between 10 and 60/66 [25], presence of stable home situation and a carer to supervise home-based RAT, Montreal Cognitive Assessment (MoCA) score > 21/30 and ability to understand purpose of research [36].

The study’s exclusion criteria were: non-stroke related causes of arm motor impairment, severe aphasia, medical conditions incompatible with research participation such as uncontrolled medical illnesses (hypertension or diabetes, ischaemic heart disease, congestive heart failure, bronchial asthma, severe / untreated depression, agitation, end stage renal/liver/heart/lung failure, unresolved cancers, anticipated life expectancy of < 6 months, inability to tolerate sitting continuously for 60 min, local factors potentially worsened by intensive robot-aided arm therapy and computer-based training: active seizures within 3 months, spasticity of Modified Ashworth Scale grades > 2 skin wounds, shoulder, arm pain visual analogue scale > 5/10, active upper limb fractures, arthritis, fixed upper limb flexion contractures, hemi anesthesia of affected limb, severe visual impairment or visual neglect affecting ability to interact with the H-Man user interface, history of dementia, severe depression or behavioural problems, absence of a reliable carer to provide supervision during home training. Pregnant and or lactating females were also excluded.

Study protocol and ethics statement

Institutional ethical approvals were obtained by the National Healthcare Group, Domain Specific Review Boards (NHG-DSRB 2021/00156) prior to participant recruitment and study procedures. The study was conducted in accordance with the Declaration of Helsinki, which governs ethical principles for medical research involving human subjects. All participants signed written informed consent prior to enrolment. The study was registered with www.clinicaltrials.gov (NCT: 05212181) [37].

Retrospective data related to participants’ demographic, acute stroke characteristics and individualised billed cost data were extracted from institutional electronic medical records. All other clinical or robotic metric data were prospectively collected.

The protocol for the home-based training and follow-up is shown schematically in Fig. 3. Following eligibility screening and signed informed consent, 2 clinic onboarding (Visits V1, V2 occurred before and after, T0 respectively, for baseline outcomes) sessions of 90 min each were conducted within a week by an OT for both the participants and their appointed carer. This was followed by a single home visit (Visit V3) by the vendor to deliver and set up the H-Man at the participants’ homes. Simultaneously, an OT was present at this home visit for appropriate interfacing of the participants to Home-RAT, reinforcement of Home-RAT training, safe operations, and handling of the robot. From the next day, Home-RAT was commenced for 30 consecutive days. The Home-RAT was then retrieved from the participants’ homes.

Fig. 3

At week 5 (Visit V4, T1), participants returned to the clinic for 1 session of clinic-based OT. Follow-up assessment sessions using standardised outcome measures were conducted in the clinic on weeks 5 (T1), 12 (T2) and 24 (T3) (Fig. 3). All T0-T4 assessments and up to 10 remote telemonitoring sessions were conducted by an independent experienced OT, not involved in V1-2 interventions.

Description of in-clinic phase

Following screening and informed consent, each participant was assigned a unique research identifier code, which was used in data collection forms, the clinic and home robots and a web-based platform to identify participants. Participants were then assessed at baseline by an OT using the above outcome measures (T0, visit 1), followed by 2 × 90-min clinic onboarding sessions (V2-V3) at TTSH-CART. The main purposes were to introduce participants to Home-RAT training, familiarise participants to the various exergames, training schedules and progression and to train their carers on proper operational handling, safety aspects and progression of training on the Home-RAT. Particular attention was paid to proper trunk posture and positioning in height-adjustable chairs with appropriate hemiplegic shoulder positioning and hand straps to the robotic handle as needed.

Subsequently, visit 3 occurred at the participants’ homes with the concurrent delivery and installation of the Home-RAT by the vendor and training set up by CART OT over 90 min (Fig. 3). The goal of this visit was to ensure continuity of ergonomic positioning of the participant, which was previously established during the prior 2 clinic onboarding sessions; also, supervision or manual assistance from carers or next of kin as needed for proper positioning at the Home-RAT or for turning on /off and adjustment of controls; and revision of safety and trouble-shooting protocols by participants and carers. Participants were given contact numbers to short message or contact OTs or vendor in case of physical or technical difficulties respectively. A paper record was also provided for manual logging of dates, start and end times of each of the training sessions as a consistency countercheck against the web-based cloud data.

Home training phase

Participants were instructed to perform daily home-based Home-RAT training for the next 30 days, starting at 20–30 min per session daily and progressing with rest breaks as needed to 60 min/day at the end of the first week and further increasing to 120 min daily in distributed sessions by the end of the second week. OTs did not perform synchronous tele-monitoring facing the participants during the 30-day home training phase.

Remote asynchronous tele-monitoring via the web-based cloud platform was performed by OTs in the clinic for 10 min each, up to 10 sessions over 30 days (i.e., 2–3 times per week). This involved accessing the cloud data and participants’ performance (log-in duration, dates, times via a graphical interface). The first remote monitoring session occurred 24 h after visit 3 (delivery and set-up of Home-RAT) and proceeded as per protocol at 2-3x/week up to 10 sessions/30 days. Telephone calls or short messaging from OTs to participants/carers were on an as-needed basis, when the following situations were encountered: absence of web-based cloud activity noted for > 24 h initially, intermittent, or poor compliance (i.e., irregular, or infrequent log-in < 20 min each time) or failure to progress training duration to 60 min/day by day 14/30 days.

At the end of 30 days, the Home-RAT was retrieved from participants’ homes by the vendor.

Follow-up phase

These consisted of 3 clinic visits of 60–90 min each (visits 4–6, or T1,2,3). These included 1 session of independently rated outcome measures and functional retraining by OT at week 5 (T1, visit 4), and 2 further follow up outcome measures, assessed by OT at weeks 12 (T2, visit 5), and weeks 24 (T3, visit 6). At T1, visit 4, functional retraining was performed prior to T1 outcome assessment, this consisted of limb ranging and mobilisation followed by guided practice of reach coordination and grasp/release functions utilising neuro-facilitatory handling techniques such as the Bobath Concept and Neurodevelopmental Treatment, with Task-oriented Training [38, 39]. At T3 and T4 follow-up points, we documented which participants had concomitant rehabilitation interventions, however, the exact amounts or intensity of upper limb training or interventions, were not documented in line with institutional ethical regulations due to electronic medical records outside of study protocol and site.

Participants were discharged from the study at week 24 upon completion of all study interventions and outcome measures.

Outcome measuresTherapy plan: adherence

Primary outcomes of participants’ adherence with the therapy plan, were defined in two ways. Firstly, we defined as an "active day" any day within the 30-day therapy programme in which a participant training was logged into the robot’s software for at least 20 min. Secondly, we defined “active hours/30 days” or “active minutes/day”, the total time spent, removing idling time of the robotic handle. These were counter checked against participants ‘manual logs filled out during home RAT training.

Patterns of participant usage per day according to date and time stamped on the web application.

Participant subjective ratings

Patient reported outcome measures (via standard questionnaire), where participants rated on a Likert scale [40] of 1–5, with 1 being strongly disagree and 5 being strongly agree on their home-based experience with Home-RAT. The questions (1–7) were as follows:

1.

It is easy to learn how to use the system.

2.

The set-up was comfortable.

3.

The training was easy to complete at home.

4.

The training was not boring.

5.

The training was useful for exercising my arm.

6.

The home robot training should be part of standard therapy.

7.

I am overall satisfied with the performance of the robotic system.

Standardised clinical outcomes

The following secondary efficacy and health-related quality of life (Hr-QOL) outcomes were measured T0,1,2,3 by an independent, experienced OT assessor not involved in training visits 1–3. These were done to assess the durability of any gains over 24 weeks (follow-up period of 19 weeks).

Upper extremity (UE) Fugl-Meyer Motor Assessment (FMA) is a widely used quantitative measure of motor impairment to evaluate upper-limb recovery [25]. Its scores range from 0 being the minimum to the maximum score of 66 points and is divided into UE including shoulder-elbow, and coordination and speed (0–42) and distal wrist-hand scores (0–24).

Action Research Arm Test (ARAT) is a 19-item observational measure of upper-extremity performance score and score ranges from 0 being the minimum to the maximum score of 57 points. [41, 42]. It consists of 4 sub-tests (grasp, grip, pinch, and gross movement). Each task performance is rated on a 4-point scale ranging from 0 (no movement) to 3 (normal movement). The subscale ranges for each subtest are; grasp (6 items, 0–18), grip (4 items, 0–12), pinch (6 items, 0–18) and gross movement (3 items, 0–9). Scores from each task are summed, with a minimum total score of 0 to a maximum score of 57 [41, 42].

Affected hand grip strength was measured using Jamar Dynamometer (kg) using the mean reading of 3 attempts [43].

The Stroke Specific Quality of Life Scale (SSQOL), an instrument intended to measure the quality of life specific to stroke patients [44]. The instrument consists of 49 items within 12 domains such as family roles, self-care, and mobility. Each item is scored on a 5-point Likert scale [40] from 1–5, with a minimum total score of 49 and a maximum of 245. Higher scores imply higher QOL.

Safety data

In terms of participant safety monitoring, these included clinical measures of hemiparetic limb spasticity of shoulder adductors, elbow flexors, wrist and finger flexors using the Modified Ashworth Scale scores (MAS) [45] and shoulder/arm visual analogue scale pain scale (VAS 0–10) at rest for T0,1,2,3. [46] (Appendix 3, Tables 8, 9).

All participant demographic and clinical data were collected and managed on the REDCap electronic tool hosted at the National Healthcare Group [35].

Statistical analysis

As this was a pilot feasibility trial, formal statistical power calculation was not performed. A minimum sample size of 10 was planned and factoring in a ~ 20% drop out rate (~ 2 subjects), the total sample size was 12. All eligible patients were consecutively screened and recruited. Modified intention to treat analyses was performed [47]. A normality test (Shapiro Wilk p value was > 0.05) was performed for all the main outcome variables. All variables were found to be normally distributed, except for ARAT at week 24, whereby, skewness and kurtosis tests conducted for ARAT at week 24 showed a skewness of -0.0242 indicating a nearly symmetric distribution and kurtosis value of 1.253 suggesting a less peaked distribution and has lighter tails, indicating that data points are more evenly spread around the mean compared to a normal distribution. Overall, these values suggested that the distribution is approximately normal, which is generally acceptable for conducting parametric tests. We then analysed differences for main outcome variables between T0-T1, T0-T2 and T0-T3.

In our analysis, we employed a mixed random effect modelling procedure for all pre-specified outcomes (FMA total, ARAT Total and SSQOL), following the standard univariate and multivariate stepwise backwards regression analysis technique.

In this standard procedure, the first step was to conduct a univariate analysis to identify putative predictor variables associated with the outcome measures. A significance threshold p ≤ 0.1 was used in this stage in order not to miss any potentially clinically important predictors.

In the multivariate analysis, the final fitted models were constructed using multivariate backward stepwise regression procedures. The known clinically important variables are forcefully adjusted in the model. The final statistical significance remained conventionally defined as p ≤ 0.05.

Age, nature of the stroke (Haemorrhagic and Infarct/ Ischemic), duration of stroke, affected side of stroke (right, left & both), frequencies of home-based exercise (captured via cloud data) were forcefully adjusted in the multivariate model. The inclusion of these covariates, despite the lack of statistical significance in the univariate analysis, was motivated by their known clinical relevance and potential to influence the outcomes.

Initial exploratory analysis employed the paired t-test to determine the changes in the mean scores of outcome variables over time using paired t-test.

Final adjusted clinical effect sizes for FMA referring to intervention of COT, RAT at clinic and RAT at home (using Home-RAT) were calculated using multivariate mixed random effect models with unstructured covariances and sandwich regressor (Robust Variance) option to take into account for quantifying heterogeneity within subject variability for repeatedly measured FMA scores over time; unstructured covariance matrix which provide a flexible framework for modelling the correlation structure of the data, while the sandwich estimator helped to correct for any potential misspecification of the covariance matrix.

The final independent variables included in the multivariate model were: (1) nature of the stroke, classified as infarct, haemorrhagic, (2) recurrence of stroke (classified as yes or no), (3) side of the stroke (classified as right, left, or both), (4) intensity of training (measured as total days trained per month). (Refer to Appendix 5 Table 15).

The level of statistical significance for all tests was set at two-sided p < 0.05.

Cost-effectiveness analysis (CEA)

The CEA [48] was conducted based on societal (stroke survivors’ and hospital’s) perspectives. The time horizon of the analysis was set at ~ 24 weeks in line with the study duration. In this initial analysis, we quantified and compared the costs of each intervention at the follow up (week 24), COT, RAT at clinic and RAT at home (using Home-RAT) and their corresponding effectiveness measures, such as clinical FMA outcomes. Cost data for COT, RAT at clinic were retrospective, billed data for each participant where available. RAT at home billed data was collected prospectively for all 12 participants. All costs were estimated in Singapore dollars (S$). Details on the estimation of healthcare and non-healthcare related medical costs for all interventions are provided in Appendix 2.

FMA scores for COT and RAT at clinic were obtained from an earlier conducted randomised clinical trial (RCT) in 2018 at the same clinic as the current study [14]. With reference to this study by Budhota et al., we provide a summary of the comparison groups’ baseline characteristics, intervention and follow up, which were used for CEA. The non-inferiority RCT design was used in the first laboratory prototype of Home-RAT to compare 2 intervention groups: (i)18 × 90-min sessions over 6 weeks, thrice weekly, of in-clinic supervised RAT (60 min of H-Man and 30 min of COT per session) with (ii) duration-matched in-clinic 90 min per session, thrice weekly, supervised COT by OT.

The baseline FMA and ARAT of both intervention groups were similar; FMA (RAT group) 40.3 (SD 9.3) vs FMA (COT group) 35.9 (SD 11.7), p > 0.05; and ARAT (RAT group) 26.6 (SD 16.6) vs ARAT (COT group) 18.9 (SD 15.6), p > 0.05. The baseline FMA and ARAT values (RAT intervention group) were comparable with the current study’s baseline values of FMA 42.1 (SD 13.2) and baseline ARAT 25.4 (SD 19.5) [14]. The duration of follow up was similar for Budhota et al. and current study, at 24 weeks after baseline measurements.

Adjusted clinical effect sizes for FMA referring to COT, RAT at clinic and RAT at home were calculated using multivariate mixed random effect models and clinically important variables were adjusted in the models (more details in Sect. 2.8.4). CEA was carried out using model-based, estimated individual predicted clinical effect sizes, and healthcare, non-healthcare costs, and total costs for 3 unique treatment pathways.

Incremental Cost-Effectiveness Ratio (ICER) was calculated using the following formula (1):

$$ICER = \frac \right)}} \right) }}$$

(1)

The ICER indicates the additional cost incurred to attain an additional unit of effectiveness with the new intervention when compared to the alternate choice or comparator.

Budget impact analysis (BIA)

The BIA [49] aimed to estimate the potential impact of increased uptake of a new intervention (RAT at home) compared to the current model—only COT. BIA used Singapore's national perspective and a five-year time horizon. To estimate the annual number of stroke survivors eligible for post-stroke rehabilitation, we used national statistics reported by the Singapore Stroke Registry [50] and the Ministry of Health data [51].