This study is an unblinded, multicenter, prospective, pragmatic parallel 2-arm RCT comparing the efficacy and cost-effectiveness of a 12-week application-based telerehabilitation program to usual care across 3 CHs in Singapore. Patients recruited at each site will be randomly allocated in a 1:1 ratio to either an intervention or control group. The intervention group will be enrolled into ATLAS, consisting of a structured exercise program and Rebee, an AI-assisted telerehabilitation device, in addition to usual care. The control group will continue to receive usual care but will not have access to ATLAS (Figure 1). The study design is based on the Pragmatic Explanatory Continuum Indicator Summary Framework-2 (PRECIS-2) [27]. This study protocol is written according to the Standard Protocol Items: Recommendations for Interventional Trials reporting guidelines [28] (Figure 2). Study outcomes will be reported in accordance with the Consolidated Standards of Reporting Trials statement [28].

This study was conducted in 3 CHs (Outram Community Hospital, Seng Kang Community Hospital, and Jurong Community Hospital) in Singapore. CHs are purpose-built hospitals that provide medical, nursing, and rehabilitation care for patients who require a short period of care after their stay in the acute hospital before being discharged back to the community [26]. Patients admitted to CHs may come from any acute tertiary hospital in Singapore.

This study has been approved by the SingHealth Centralised Institutional Review Board (2024-2246) and is registered on ClinicalTrials.gov (NCT06683963). Written informed consent was obtained from all participants before study participation. Participants on each site are assigned a continuous study code to protect identifiable data (eg, OCH001-OCH150, SKCH001-SKCH150, and JCH001-JCH150). The assignment data are kept in a password-protected laptop, within a password-protected document, and only accessible by selected study team members. Participants are compensated for their time and inconvenience via grocery vouchers given upon completion of predefined time points in the study (Singapore $20 [Singapore $1=US $0.77 as of July 21, 2026] before discharge and Singapore $30 at the end of the study).

To ensure confidentiality of participants, only the study team will have access to the research data, both hard and soft copies, at all sites. Rebee Health will only receive deidentified and nonsensitive data. Hard copy data will be stored in designated locked locations (eg, cabinets and rooms) that are accessible to only authorized study personnel. Electronic data will be stored in an institution-approved, secured, and encrypted storage medium, such as databases, encrypted portable media (eg, USB drives, CD or DVD, and hard disks), and/or institution-approved online storage platforms. The electronic data will not contain research participant identifiers. Identification codes linking electronic data and research participants will be stored separately. The research data will be destroyed after they have been stored for the minimum duration of retention as specified by the institutional policy.

Patients will be recruited from 3 CHs in Singapore. The inclusion criteria are as follows: (1) age between 21 and 99 years; (2) admission primary rehabilitation diagnostic group (RDG) classified under hip fracture (3.11, 3.12, and 3.13), musculoskeletal conditions (5.11, 5.12, 5.13, and 5.21), or deconditioning (6.1, 6.2, and 6.3) [9]; (3) understand English or Mandarin; and (4) suitable and able to engage in Rebee exercises.

The exclusion criteria are as follows: (1) colonized by multi-drug-resistant organism; (2) presence of neuropsychiatric or neurocognitive disease affecting the ability to follow instructions; (3) unable to full weight bear, or restrictions to range of motion; and (4) refusal to participate or give informed consent.

Patients admitted to CHs undergo inpatient rehabilitation for around 3 to 4 weeks before being discharged home. During the CH admission, participants in both arms will receive standardized CH rehabilitation comprising both physiotherapy and occupational therapy sessions. Physiotherapy and occupational therapy sessions are each held for 30 to 60 minutes per session, up to 5 times a week, covering a range of items including warm-up, strength training, cardiovascular training, and balance training. All training may be conducted in either a group or one-to-one setting based on clinical needs and patient safety considerations [26]. Caregiver training is provided to patients and their respective family members or caregivers as needed prior to discharge.

All therapists delivering rehabilitation services are registered under the Allied Health Professions Council, a professional board that is governed and regulated by the Ministry of Health. This ensures standardization of professional qualifications across both study arms.

After discharge from a CH, participants will be referred to outpatient rehabilitation facilities as recommended by the CH team (eg, center-based therapy at DRCs or outpatient therapy clinic, or home-based therapy) as per usual care. They will be prescribed exercises communicated via printed materials, verbal instruction, and demonstration.

During their CH stay, patients in the intervention arm will be enrolled in the ATLAS program in addition to usual care (Figure 1). The ATLAS program is provided free of charge to the patients and their caregivers, with follow-up and support provided during their stay as well as upon discharge for up to 12 weeks. It consists of Rebee, an AI-assisted telerehabilitation device with a wearable sensor, and a structured rehabilitation program.

The primary outcome is functional status measured over 12 weeks. Functional status will be measured using modified Barthel index (MBI), a 100-point ordinal scale for the assessment of a patient’s functional state. The MBI assesses activities of daily living over 10 domains comprising transfers, ambulation, stair climbing, bathing, dressing, toileting, bowel control, bladder control, personal hygiene, and feeding [29].

Secondary outcomes include physical performance, quality of life, length of stay in CH, 30-day readmission rates, and cost-effectiveness over 12 weeks. Patients who are admitted under RDG total knee replacement will have an additional outcome measure of joint range. The outcome measures as well as the method of measurement are listed in Table 3. Health care cost will be calculated from electronic medical records and supplemented with survey data regarding the number of DRC or home therapy sessions attended, manpower hours to train patients to use Rebee, costs of Rebee software and hardware, and use of outpatient community services (eg, day care center, home nursing, housekeeping, and meals on wheels). The tools to be used in this study to measure the outcomes are validated and implemented in routine clinical care.

We calculated the sample size based on the primary outcome of MBI scores, which is evaluated at recruitment and at 4, 8, and 12 weeks. The historical mean of the MBI is 73 (SD 25). The study aims to detect a difference of 5 points between the control and intervention groups assuming a correlation of 0.35 between the repeated measurements. The sample size estimated to detect this difference with 80% power and a 5% significance level is 175 participants per group assuming equal allocation between arms. After accounting for 20% attrition, the total sample size required for the study is 450.

Participants will be recruited from the inpatient wards of 3 CHs, namely Outram Community Hospital, Sengkang Community Hospital, and Jurong Community Hospital. Each site will recruit 150 patients (75 in the intervention group and 75 in the control group), with a total of 450 participants recruited across all 3 sites.

The participants will first be screened by their ward therapists for eligibility, after which a member of the research team will approach them to provide information about the study as well as to take informed consent. Study team members will also provide reminders to ward therapists during their regular roll calls to maximize recruitment efforts.

Participants will be randomized to the intervention or control group based on a 1:1 allocation ratio. Stratified block randomization with randomly varying block sizes will be implemented with recruitment site as the stratification factor.

After taking informed consent, the research coordinator will contact a member of the team (who oversees the progress but is not involved in patient recruitment) to obtain the randomized assignment. The research coordinator will remain blinded to the allocation prior to contacting the team member. Due to the nature of the intervention, blinding of study participants will not be possible.

Participants will be analyzed via an intention-to-treat approach. Participant characteristics will be summarized using median (IQR) or mean (SD) for nonnormally distributed and normally distributed data, respectively, and count and percentage for categorical variables.

Linear mixed-effects model will be used to assess the pattern of change in the primary outcome of the MBI score and secondary outcomes of EQ-5D, 5-time chair stand test (5CST), and joint range over the 12-week follow-up period. A uniform correlation structure will be specified to model the within-subject correlation of these repeated measures. MBI, EQ-5D, 5CST, and joint range of motion will be treated as continuous outcomes. Normality of continuous outcome variables will be assessed through visual inspection methods. If normality assumptions are met, parametric methods will be used; otherwise, a suitable transformation may be performed to address the issue of nonnormality.

As individuals of different age groups have varying body composition, functional capacity, and physiological reserves, there may be differential influences on baseline functional ability and recovery trajectories. To ensure that any potential age-related differences are accounted for, we will first assess the interaction between the age category, treatment groups, and duration of time. We will stratify into age categories, and if statistically meaningful effect modification by age is observed, an age-stratified analysis will then be conducted.

Poisson regression will be used to compare length of stay between intervention and control groups. If there is evidence of overdispersion and/or zero inflation, negative binomial regression or corresponding zero-inflated models will be implemented instead.

Key prognostic factors such as baseline MBI, RDG grouping, and time will be adjusted a priori. The secondary outcome based on 30-day readmission will be compared between the intervention and control group using modified Poisson regression. The effect estimate will be quantified in terms of relative risk and its 95% CI.

For both the primary and secondary outcomes, linear mixed-effects model on the log-transformed scale will be applied if the model assumptions are violated.

Missing data are expected due to the nature of the study and will be handled by imputation.

The preferred method of imputation will be applied accordingly based on the degree of missingness, such as last observation carried forward.

The cost-effective analysis will evaluate both direct cost savings and cost-effectiveness in terms of quality-adjusted life years (QALYs). To assess direct cost savings, the analysis will evaluate the difference in length of stay at the hospital between the intervention and control group (Table 4). The results will be expressed as mean cost differences between groups. Cost per QALY gained will be evaluated for the cost-effectiveness of the intervention. QALY will be derived using EQ-5D-5L utility scores, and incremental cost-effectiveness ratios (ICERs) will be calculated. The ICERs will be compared to established willingness-to-pay thresholds to determine the cost-effectiveness of the intervention. Sensitivity analyses will be performed to account for uncertainties in cost and QALY estimates. Results will be presented using cost-effectiveness planes and acceptability curve.

Feedback on the Rebee device and application will be collected from the intervention group through 6 close-ended questions and 1 open-ended question at the end of the study (Multimedia Appendix 1). Specifically, feedback will be collected on the ease of use of the Rebee app and exercises, the ability to complete Rebee exercises independently, any technical issues encountered, the overall user experience, the suggested improvements, and whether they would recommend this technology to others. A simple descriptive analysis will be conducted to summarize the quantitative findings using frequencies and percentages. A content analysis will be performed on the qualitative responses to identify keywords and phrases. If there is substantial variation, we will conduct a thematic analysis to summarize the common themes.

Rebee use data for each participant will be retrieved and analyzed via the Rebee software [25]. Use data will be broken down in detail to evaluate daily compliance rate to use of Rebee and its exercises, as well as overall use and uptake over the 12-week period.

To the best of our knowledge, this study is the first to determine the clinical effectiveness and cost-effectiveness of an AI-assisted telerehabilitation device with a structured training program starting from inpatient stay and continued upon discharge via an RCT study design among patients requiring postdischarge rehabilitation care in Singapore. We anticipate that patients receiving the telerehabilitation device and structured training program will demonstrate greater improvement in their functional status, as measured by the MBI, compared to those receiving standard rehabilitation alone. We also expect improvements in joint range, physical performance, and quality of life, with reduction in length of CH stay, lower 30-day readmission rates, and increased cost-effectiveness as compared to usual care alone.

Previous studies have demonstrated that telerehabilitation devices can achieve comparable outcomes compared to conventional rehabilitation services [14,15,22]. However, studies are heterogeneous in terms of the type of intervention and technology used, with varying combinations of in-person or digital personnel required for supervision and execution of the program.

This trial will contribute in several ways to existing knowledge and research regarding telerehabilitation devices and rehabilitation programs. First, it will test user acceptability and compliance with telerehabilitation devices and systems to guide the practicability of application of such devices to routine clinical practice. Second, this will evaluate the cost-effectiveness of such systems versus usual care alone to provide an economic evaluation encompassing resource use, patient-reported quality of life, and physical function improvement. Finally, this will also test whether the addition of such a telerehabilitation system as compared to usual care alone will improve outcomes in terms of functional improvement, length of stay in a CH, and 30-day readmission rates.

There are some limitations to this study. First, it was not possible to blind participants as the intervention involves receiving a physical Rebee device as well as physical sessions to teach the participant about the use of the Rebee device. This might change the behavior and response of participants in the intervention group, causing them to provide more positive outcomes. The difficulty in blinding participants may also increase patient dropout in the control arm as they do not perceive any immediate benefits to them being in the study. Second, the total health care costs calculated in the cost-effectiveness analysis may be underestimated due to omission of certain costings such as postdischarge service use costs and unplanned readmission costs. These costs were excluded because the data on postdischarge service use were collected through patient self-report and could not be verified. On the other hand, there is insufficient information on readmission costs. To minimize inaccuracy, it was decided that these costings will be excluded from the cost-effectiveness analysis. Another limitation is the language barrier, as the Rebee application is only available in English and Mandarin. Considering that Singapore is a multiracial country, this may deter participants who can only read Malay or Tamil from participating in this study. Finally, there is also a potential confounder, which is the digital literacy of participants, which was not assessed in this study. To address this issue, the study team will provide support to participants and caregivers, ensuring that they are able to operate Rebee independently.

Results from our study will be disseminated through peer-reviewed journals, including publications on the trial protocol, main outcomes, and secondary analyses. Results will also be presented at local and international conferences centered on rehabilitation, DH, and implementation science. Relevant findings will be provided to participating clinical sites and collaborating institutions to inform decisions regarding future adoption and scalability of such telerehabilitation technologies. Dissemination will adhere to ethical, regulatory, and data-sharing requirements.

Populations around the world are rapidly aging, and health systems are attempting to bend the cost curve by shifting care from hospital-based acute care to community-based preventive health and rehabilitation. The insights on clinical and cost-effectiveness from this study shall also assist policymakers to make better decisions regarding the use of AI-assisted telerehabilitation systems to plug the present gaps in this care transition.

By incorporating both objective clinical measurements as well as health use and cost-effectiveness analyses, this trial aims to extend prior work and provide more evidence and guidance for future scaling of such telerehabilitation models.