Telerehabilitation with ARC Intellicare to Cope with Motor and Respiratory Disabilities: Results about the Process, Usability, and Clinical Effect of the “Ricominciare” Pilot Study

Background: “Ricominciare” is a single-center, prospective, pre-/post-intervention pilot study aimed at verifying the feasibility and safety of the ARC Intellicare (ARC) system (an artificial intelligence-powered and inertial motion unit-based mobile platform) in the home rehabilitation of people with disabilities due to respiratory or neurological diseases. Methods. People with Parkinson’s disease (pwPD) or post-COVID-19 condition (COV19) and an indication for exercise or home rehabilitation to optimize motor and respiratory function were enrolled. They underwent training for ARC usage and received an ARC unit to be used independently at home for 4 weeks, for 45 min 5 days/week sessions of respiratory and motor patient-tailored rehabilitation. ARC allows for exercise monitoring thanks to data from five IMU sensors, processed by an AI proprietary library to provide (i) patients with real-time feedback and (ii) therapists with information on patient adherence to the prescribed therapy. Usability (System Usability Scale, SUS), adherence, and adverse events were primary study outcomes. Modified Barthel Index (mBI), Barthel Dyspnea Index (BaDI), 2-Minute Walking Test (2MWT), Brief Fatigue Inventory (BFI), Beck Depression or Anxiety Inventory (BDI, BAI), and quality of life (EQ-5D) were also monitored pre- and post-treatment. Results. A total of 21 out of 23 eligible patients were enrolled and completed the study: 11 COV19 and 10 pwPD. The mean total SUS score was 77/100. The median patients’ adherence to exercise prescriptions was 80%. Clinical outcome measures (BaDI, 2MWT distance, BFI; BAI, BDI, and EQ-5D) improved significantly; no side effects were reported. Conclusion. ARC is usable and safe for home rehabilitation. Preliminary data suggest promising results on the effectiveness in subjects with post-COVID condition or Parkinson’s disease.


Introduction
The Global Burden of Diseases, Injuries, and Risk Factors Study 2019 estimated that globally, in 2019, 2.41 billion individuals had conditions that would benefit from rehabilitation [1]. This number had increased by 63% from 1990 to 2019. The disease area that contributed most to prevalence was musculoskeletal disorders, followed by neurological disorders [1], at least until SARS-CoV-2-related pandemic disrupted life and good clinical practice worldwide.

Study Design and Objectives
'Ricominciare' is a pilot single-center, not controlled, prospective, pre-post-intervention study aimed at verifying the feasibility and safety of the ARC platform intended for the home rehabilitation of people suffering from mild to moderate disabilities due to respiratory or neurological conditions related to COVID-19 or PD.
In particular, the primary objectives of the study were to test the feasibility of integrating the ARC Intellicare solution into the care pathway for COVID-19 survivors and pwPD in terms of:

•
Adherence to the home rehabilitation program; • Safety of rehabilitation therapy.
Secondary objectives of the study were to investigate: • The usability and acceptability of the intervention; • The process to provide the new care pathway; • Clinical effectiveness: in fact, the participants will undergo pre-post-intervention monitoring of disability in basal activity of daily living (ADL), respiratory outcomes, endurance and fatigue, mood, and quality of life.

ARC Intellicare
ARC is a telerehabilitation solution based on the use of multiple wearable sensors, a mobile device, and algorithms of artificial intelligence (patent pending). ARC consists of a set of 5 inertial sensors (MetaMotionR+ from MbientLab, San Francisco, CA, USA) inserted in slap supports, a tablet with a dedicated application (App), and a charging station ( Figure 1). respiratory recovery for patients requiring continuous home rehabilitation. It has been conceived and developed to target post-stroke patients (MAGIC, grant agreement n. 687228), and then optimized to address the above-mentioned emerging needs coming from the COVID-19 pandemic (POR FESR 2014-2020-Asse I COVID-19 grant), following recommendations for people's recovery after COVID-19 [33,36].

Study Design and Objectives
'Ricominciare' is a pilot single-center, not controlled, prospective, pre-post-intervention study aimed at verifying the feasibility and safety of the ARC platform intended for the home rehabilitation of people suffering from mild to moderate disabilities due to respiratory or neurological conditions related to COVID-19 or PD.
In particular, the primary objectives of the study were to test the feasibility of integrating the ARC Intellicare solution into the care pathway for COVID-19 survivors and pwPD in terms of: • Adherence to the home rehabilitation program; • Safety of rehabilitation therapy.
Secondary objectives of the study were to investigate: • The usability and acceptability of the intervention; • The process to provide the new care pathway; • Clinical effectiveness: in fact, the participants will undergo pre-post-intervention monitoring of disability in basal activity of daily living (ADL), respiratory outcomes, endurance and fatigue, mood, and quality of life.

ARC Intellicare
ARC is a telerehabilitation solution based on the use of multiple wearable sensors, a mobile device, and algorithms of artificial intelligence (patent pending). ARC consists of a set of 5 inertial sensors (MetaMotionR+ from MbientLab, San Francisco, CA, USA) inserted in slap supports, a tablet with a dedicated application (App), and a charging station ( Figure 1). Figure 1. ARC Intellicare device for patient use (Home Unit): a portable and lightweight charging station hosts 5 wearable inertial sensors and their biocompatible supports (2 blue-colored, longer for ankle placement; 2 grey-colored, shorter for wrist positioning; and a longer and more flexible support to be placed on back-side of the neck). The tablet, provided with an anti-shock cover, comes with a preinstalled application with a user-friendly interface and functionalities dedicated to the patient (Home version).
The mobile App comes with a Home version, with specific features for the patient, and a Clinical version, designed instead for the healthcare professional. The device allows rehabilitation professionals to prescribe exercises from the available library according to Figure 1. ARC Intellicare device for patient use (Home Unit): a portable and lightweight charging station hosts 5 wearable inertial sensors and their biocompatible supports (2 blue-colored, longer for ankle placement; 2 grey-colored, shorter for wrist positioning; and a longer and more flexible support to be placed on back-side of the neck). The tablet, provided with an anti-shock cover, comes with a preinstalled application with a user-friendly interface and functionalities dedicated to the patient (Home version).
The mobile App comes with a Home version, with specific features for the patient, and a Clinical version, designed instead for the healthcare professional. The device allows rehabilitation professionals to prescribe exercises from the available library according to specific therapeutic needs and/or according to the rehabilitation protocols adopted by the center, and to monitor patients' performances and progresses remotely. With the Home version, ARC guides patients in the unsupervised execution of the rehabilitation project through simple instructions, video tutorials, and the automatic counting of the correct number of repetitions performed. The counting of the number of exercise repetitions correctly performed is the output of the developed AI algorithm, which is the innovative core of this device. It runs real-time on the ARC backend deployed in a dedicated cloud server. Input of this algorithm are the tri-axial accelerations and angular velocities. These data are streamed simultaneously via Bluetooth to the tablet from 3 out of 5 IMU sensors worn by the patients during the exercise session. The sensors involved in the monitoring are selected based on the exercise type and the target body area. For upper body exercises and for those mainly involving upper arm movements, the upper limb-trunk sensor configuration is used (i.e., ULT, two on both wrists and one back neck sensors). For all exercises requiring movement mainly on the lower limbs, instead, a lower limb-trunk sensor configuration is used (LLT). The positioning and orientation of the sensors on the body, with their relative axes, are shown in Figure 2.
specific therapeutic needs and/or according to the rehabilitation protocols adopted b center, and to monitor patients' performances and progresses remotely. With the H version, ARC guides patients in the unsupervised execution of the rehabilitation pr through simple instructions, video tutorials, and the automatic counting of the co number of repetitions performed. The counting of the number of exercise repetitions rectly performed is the output of the developed AI algorithm, which is the innovative of this device. It runs real-time on the ARC backend deployed in a dedicated cloud se Input of this algorithm are the tri-axial accelerations and angular velocities. These dat streamed simultaneously via Bluetooth to the tablet from 3 out of 5 IMU sensors wor the patients during the exercise session. The sensors involved in the monitoring ar lected based on the exercise type and the target body area. For upper body exercises for those mainly involving upper arm movements, the upper limb-trunk sensor con ration is used (i.e., ULT, two on both wrists and one back neck sensors). For all exer requiring movement mainly on the lower limbs, instead, a lower limb-trunk sensor figuration is used (LLT). The positioning and orientation of the sensors on the body, their relative axes, are shown in Figure 2. Accelerations and angular velocities coming from the three pre-defined inertial sors are buffered in a temporally ordered manner and normalized. A neural netwo used to recognize which exercise is being performed and to segment the data into si repetitions. When a repetition is identified by the algorithm, it is considered correctly formed, and real-time feedback is provided to the patient through the App user inter At the end of each exercise execution, a counting unit outputs the total number of re tions performed, which is then used to compute patient adherence to the prescribed apy (refer to Section 2.4). This information, together with other feedback collected dir from the patient on his/her health status during and at the end of each session (pa reported outcomes), is collected and displayed on a results dashboard in the Clinical sion of the ARC App.
The accuracy and reliability of the commercial IMU sensor part of ARC Intelli i.e., MetaMotionR+, was previously assessed in the literature for both sensor fusion bedded algorithm [49] and raw data (i.e., accelerations and angular velocities) [50]. AI algorithm was validated on a group of 30 healthy subjects and for all the 41 m exercises available in the ARC exercise library. The validation consisted of the compar between the number of exercise repetitions identified by an expert operator (gro Accelerations and angular velocities coming from the three pre-defined inertial sensors are buffered in a temporally ordered manner and normalized. A neural network is used to recognize which exercise is being performed and to segment the data into single repetitions. When a repetition is identified by the algorithm, it is considered correctly performed, and real-time feedback is provided to the patient through the App user interface. At the end of each exercise execution, a counting unit outputs the total number of repetitions performed, which is then used to compute patient adherence to the prescribed therapy (refer to Section 2.4). This information, together with other feedback collected directly from the patient on his/her health status during and at the end of each session (patient reported outcomes), is collected and displayed on a results dashboard in the Clinical version of the ARC App.
The accuracy and reliability of the commercial IMU sensor part of ARC Intellicare, i.e., MetaMotionR+, was previously assessed in the literature for both sensor fusion embedded algorithm [49] and raw data (i.e., accelerations and angular velocities) [50]. The AI algorithm was validated on a group of 30 healthy subjects and for all the 41 motor exercises available in the ARC exercise library. The validation consisted of the comparison between the number of exercise repetitions identified by an expert operator (ground truth), N operator , and the number of exercise repetitions counted by the algorithm for the same exercise execution (model predictions), N ARC . Assuming a normal distribution, a paired t-test between the couples N operator and N ARC (one per exercise) was used to verify the null hypothesis, i.e., no statistical difference between the assessment made by an expert operator and the AI algorithm. In 35 of 41 exercises (86%), no significant difference in relative repetition count between the ground truth and the model predictions (p ≥ alpha = 0.05), was obtained, confirming the functional performance of the device on 35 different exercises (validation data are available upon request). Only exercises for which validation was successful have been considered in this study.
The novelty of the ARC device with respect to other commercial solutions is the use of a validated AI algorithm able to recognize the high number of exercises included in the library, varying in typology and complexity level. In the ARC exercise library, indeed, different types of rehabilitation exercises (e.g., for mobility, coordination, balance, core stability, strengthening, etc.) are available, and all of them are already widely used in clinical practice [51]. These exercises are characterized by a periodic movement, which can be composed of several phases and can comprehend the use of different limbs. Five levels of difficulty have been defined based on exercise pattern (i.e., bilateral symmetric-BS, bilateral alternated-BA, mono-lateral-ML, static-ST), exercise starting position (comfortable or uncomfortable), and the number of limbs and joints involved.

Subjects
People (men and women > 18 years) following COVID-19 (COV19) or diagnosed with PD were enrolled between 26 March 2021 and 30 May 2021 if received by a physiatrist an indication for rehabilitation to cope with motor or respiratory disorders in order to optimize the independence in activity of daily living (ADL). Inclusion criteria were (i) mild-moderate dyspnea (Barthel dyspnea Index ≤ 95) [52] or (ii) Walking Handicap Scale (WHS) [53] ≤5; and (iii) signed informed consent. The study exclusion criteria were (1) at the enrollment, presence of fever (TC ≥ 37 • C), cough, cold, sore throat, diarrhea, or pneumonia signs and diagnosis of moderate to severe cognitive impairment; (2) formal rehabilitation performed in the last month; (3) pre-existing disability related to dementia, epilepsy, seizures, and a history of severe dizziness and falls; (4) severe non-stabilized comorbidities, such as oncological diseases in the active phase, New York Heart Association (NYHA) Functional Classification stage IV congestive heart failure [54], or severe respiratory failure requiring cough and breath support; (5); Rankin mod. Score ≥ 4 [55], i.e., moderate or severe dependence in activities of daily living for any medical reason (but we accepted an inconstant need for help to use technology or a supervision to prevent falls during motor training); (6) for women of potential childbearing, not using suitable valid methods of contraception; (7) pregnancy.

Intervention Protocol
Upon obtaining informed consent, each participant underwent clinical-functional evaluation, usability tests, and training in ARC use, receiving an ARC unit, to be used independently at home in the next 4 weeks.
Regarding the training, at the enrollment, each subject received from the site investigator a 30 min explanation on the use of both software (SW) and hardware (HW) components. During this training, the patient also performed some of the exercises in the rehabilitation protocol that had been explained to him/her. Once completed, each subject carried out a usability test, in which he/she was asked to perform 15 tasks (6 related to HW and 9 to SW components). For each task, the patient was asked to indicate if he/she needed support from the investigator and the degree of difficulty encountered in carrying it out on a scale from 0 to 10 (0 = no difficulty and 10 = impossible to perform). Each subject then completed the System Usability Scale (SUS) [56]. After enrollment, each subject underwent the clinical assessment and received an ARC unit to perform 45 min exercise sessions at home, 5 days/week for 4 weeks and an at least 30 min/week video call with the investigator using the ARC app. The use of ARC during unsupervised sessions allows for exercise and patient adherence monitoring.

Study Endpoints and Outcome Measures
Primary study endpoints and outcome measures (Table 1) were:

•
To reach an effective adherence to the home rehabilitation program (at least 80%), measured as the rate of performed/prescribed sessions [57]; • Safety of rehabilitation therapy, based on the number/type of adverse events [58]. Considering motor and respiratory data separately, we proposed two different ways to calculate adherence to home-based exercises prescribed with ARC: 1.
Adherence-Days = Total number of days the patient accessed the platform for training versus the total number of days the exercises were prescribed (1); where Adh thresholded is computed using a threshold on the number of daily exercise executions (N executions ), considering together all exercises prescribed for that day. In particular: • Adh thresholded (d) = 0, when the patient never tried to access into ARC device to perform one of the exercises prescribed for day d; • Adh thresholded (d) = 1, when N executions (d) ≥ 1.

2.
Adherence-Repetitions = total number of repetitions performed versus total number of repetitions prescribed, considering all exercises included in the rehabilitation plan (e, from 1 to n, where n is the total number of exercises prescribed to a subject) and all days (d) of treatment (from d = first, i.e. first day of treatment to d = last, i.e. last day for which an individual rehabilitation program was prescribed) (2).
In the first analysis, we evaluated the number of days each patient used ARC compared to the total days they had a prescription, despite the number and type of exercises prescribed by the rehabilitation professional, and without considering if the patient completed them or not.
In the second analysis, we evaluated the number of repetitions that each patient performed with ARC compared to the total repetitions assigned by the rehabilitation professional. All sets/exercises prescribed are considered in this analysis. Both adherence calculations were expressed as a percentage.
Secondary endpoints and outcome measures ( Table 2) were: • usability and acceptability of the intervention studied through the System Usability Scale (SUS) [56] and a semi-structured ad hoc-prepared questionnaire; • The process to provide the new care pathway, measured by the percentage of subjects resulting eligible to the study. In addition, pre-and post-treatment clinical outcomes were monitored as follows: Quality of Life: Euro-Quality of life Questionnaire self-assessment-5 Dimension (EQ-5D) and EQ-5D-Visual Analogic Scale (EQ-5D-VAS) [64].
In Tables 1 and 2, we also reported the cut-off scores of the outcome measures of the different end-points based on the availability of the minimum clinically important difference (MCID) score in the literature.
The assessment was performed at the enrollment (T0) and at the end of the 30-day intervention period (T1), when monitoring of primary and secondary outcome measures on adherence, acceptability, and safety was completed.
All the secondary endpoint outcome measures were assessed at the University Hospital outpatient facilities, at each assessment time point, by blinded clinicians who used standardized questionnaires sheets and devices (finger pulse oximeter, OXY Watch ChoiceMMed, Pikdare SpA; Hensych Blood Pressure Cuff Kit with Manual Sphygmomanometer and Stethoscope) independent from ARC. Participant selection and enrollment were performed by physiatrists MGC and MC while the blinded assessment of study outcomes was performed by physiatrists FAB and MEL. The physiatrist RC and the physiotherapists MH, PC, and RI built the exercise library. RC and PC supervised patients during the rehabilitation period, selecting and updating the training protocol, if needed, during the weekly synchronized training sessions. All raters involved in the study underwent a preliminary course to harmonize methods and increase inter-rater reliability to a Cronbach alpha 0.8.

Ethical Procedures
The study was performed according to the Declaration of Helsinki and approved by the Local Institutional Committee (protocol number: CERM1781). All participants signed informed consent forms prior to participation in the study. Trial registration: the trial ClinicalTrials.gov Identifier is NCT05074771.

Statistical Analysis
At least 20 subjects (10 subjects per group: i.e., COV19 group and pwPD) were expected to be enrolled in this exploratory study. The sample size was calculated by the confidence interval method (IC95%), using the Clopper-Pearson exact method, and defining a 95% confidence level. A sample size of 10 subjects per group was needed to estimate an adherence (calculated as the percentage of day accesses) of at least 80%, with an interval width of 0.8 (IC95% 0.44-0.97). Such a sample of 20 subjects was deemed appropriate to achieve results characterized by an error of the estimate compatible with the exploratory nature of the study.
The distribution of variables in the whole sample as well as in the 2 sub-groups (COV19 and pwPD) was determined using mean and standard deviation, and number/percentage for categorical variables.
Pre-post-treatment within-group changes were checked by using the Wilcoxon rank test for the whole sample as well as for each study group. The Mann-Whitney U test with Z analysis was applied for comparing data across the 2 groups, after verifying the normal distribution through the skewness and kurtosis tests.
Finally, to control the influence exerted by a patient's personal and clinical characteristics on the treatment effect, firstly we calculated a change index delta = ([t1 score − t0 score]/t0 score) × 100 for each clinical outcome, whereas we applied the Spearman rank correlation analysis to study the effect of age, MTUAS, MoCA, and years of education on delta score.
Statistical significance was set at the 0.05 level. The analysis was performed using Statview Statistics, version 5.0.
Of the 23 subjects receiving training sessions (12 COV19 and 11 pwPD), 2 were defined as unsuitable due to limitations in the use of technology because of limitations in using the technology and a lack of caregivers who could help them use it; therefore, 21 (mean age 61 ± 10 years [range: 29-72], 8 women) were enrolled, and all completed the study: 11 were COV19 (mean age 57 ± 13 years [range: 29-72], 5 women) and 10 were pwPD (mean age 65 ± 4 years [range: 59-70], 3 women). Figure 3 shows the flow diagram of the Ricominciare study. Table 3 shows the detailed demographic and clinical data of the enrolled subjects.

Usability and Acceptability
None of the COV19 subjects needed support for using the device after training; 4 out of the 10 pwPD needed a caregiver during home sessions. In the 1-month period of ARC use at home, forty-two technical support requests were raised from the twenty-one patients that completed it (i.e., an average of two technical support tickets per patient). No  The MTUAS score is slightly related to age (Spearman corr: Z = −2.0; p = 0.048) and highly with MoCA score (Z = 2.8; p = 0.005).

Usability and Acceptability
None of the COV19 subjects needed support for using the device after training; 4 out of the 10 pwPD needed a caregiver during home sessions. In the 1-month period of ARC use at home, forty-two technical support requests were raised from the twenty-one patients that completed it (i.e., an average of two technical support tickets per patient). No issues prevented the study from being completed successfully. Indeed, those related to hardware components, leading to the replacement of a sensor or charging station, occurred in 3 out of 420 ARC usage sessions (i.e., 0.7% of the overall system usage). These issues caused training discontinuation for 2 days on average (corresponding to the time needed for component substitution). All other issues were generally solved by restarting the device/app, and thus not hindering patients from completing the prescribed rehabilitation program. Twenty-four issues (5.7%) were related to the management of the Bluetooth connection between the tablet and sensors. Four issues (1%) were caused by 3G/4G or WiFi network connectivity, while four issues (1%) were linked to the real-time feedback provided by the device during exercise execution. Finally, the remaining seven tickets (1.7%) were not triggered by actual technical problems and were solved by clarifying to the user how to properly use the system and where to find the requested functionality.

Clinical Data Evolution
No correlation between demographic (age, education) or clinical data (MoCA, modified Rankin disability score, MTUAS, SUS) with respect to clinical effects (delta score of BaDI, mBI, BFI, 2MWT, and EQ.5D) was found using the Spearman rank test (p > 0.05); nor between adherence measures and clinical effects.
No adverse events were reported except for fatigue in 10% of pwPD.             Stratifying by pathology, at the baseline, no significant differences between groups emerged. COV19 improved in all outcome measures, while pwPD improved in all measures, except for 2MWT, whose performance improved without reaching statistical significance.
No correlation between demographic (age, education) or clinical data (MoCA, modified Rankin disability score, MTUAS, SUS) with respect to clinical effects (delta score of BaDI, mBI, BFI, 2MWT, and EQ.5D) was found using the Spearman rank test (p > 0.05); nor between adherence measures and clinical effects.
No adverse events were reported except for fatigue in 10% of pwPD.

Discussion
The analysis of the flow of subjects through the study showed that sixty-seven out of seventy-eight (83%) examined people with COVID-19 or pwPD outcomes needed rehabilitation. Among them, twenty-nine (37%) were already undergoing rehabilitation, while thirty-six (46%) were eligible for the study. However, seven subjects (9% of the whole sample, most pwPD) showed difficulty in using or accepting the technology, while five (6%) preferred to undergo outpatient in-person rehabilitation and did not sign the informed consent. Firstly, those numbers reflect the high frequency of outcomes needing rehabilitation in both post-COVID-19 condition (80% of participants, irrespective of the results of the pulmonary function tests according to Ceravolo et al., 2021) [73] and PD [19][20][21][22][23].
Secondarily, the observed prevalence of people (46%) who were not undergoing rehabilitation despite needing it may reflect a persistent difficulty that had been observed in 2020, during the first phase of the COVID-19 pandemic, when up to 80 percent of pwPD had to discontinue or were unable to access physiotherapy treatments due to social distancing, fear of contagion, or hospital difficulties [28,29].
The last result that emerges from the study enrollment phase is the level of difficulty in using or accepting the technology and remote rehabilitation: this was manifest in less than 10% of the whole sample and less than 5% among people receiving training.
For effective use, technologies in healthcare require rigorous validation to prove their acceptability and usability in addition to clinical benefits [74]. The "acceptability" of technology is a tradeoff among all those factors that influence the adoption of new technologies [75]. Two key dimensions may characterize the users' acceptability: the "perceived ease of use" and the "perceived usefulness" according to the Technology Acceptance Model (TAM) [76].

Discussion
The analysis of the flow of subjects through the study showed that sixty-seven out of seventy-eight (83%) examined people with COVID-19 or pwPD outcomes needed rehabilitation. Among them, twenty-nine (37%) were already undergoing rehabilitation, while thirty-six (46%) were eligible for the study. However, seven subjects (9% of the whole sample, most pwPD) showed difficulty in using or accepting the technology, while five (6%) preferred to undergo outpatient in-person rehabilitation and did not sign the informed consent. Firstly, those numbers reflect the high frequency of outcomes needing rehabilitation in both post-COVID-19 condition (80% of participants, irrespective of the results of the pulmonary function tests according to Ceravolo et al., 2021) [73] and PD [19][20][21][22][23].
Secondarily, the observed prevalence of people (46%) who were not undergoing rehabilitation despite needing it may reflect a persistent difficulty that had been observed in 2020, during the first phase of the COVID-19 pandemic, when up to 80 percent of pwPD had to discontinue or were unable to access physiotherapy treatments due to social distancing, fear of contagion, or hospital difficulties [28,29].
The last result that emerges from the study enrollment phase is the level of difficulty in using or accepting the technology and remote rehabilitation: this was manifest in less than 10% of the whole sample and less than 5% among people receiving training.
For effective use, technologies in healthcare require rigorous validation to prove their acceptability and usability in addition to clinical benefits [74]. The "acceptability" of technology is a tradeoff among all those factors that influence the adoption of new technologies [75]. Two key dimensions may characterize the users' acceptability: the "perceived ease of use" and the "perceived usefulness" according to the Technology Acceptance Model (TAM) [76].
Moreover, additional external factors may influence the user acceptance of technology, including demographic and clinical characteristics (i.e., age, education, and disability level), technological usage and attitude, and social or cultural influences, usability, availability, assurance of privacy and security [77,78]. Assurance of privacy seems a critical concern for older adults, followed by the functionality of the system [79]. Moreover costs, usability (both as ease of use and suitability for daily use), stigma and technical support are critical barriers to digital health adoption [79,80]. Considering the usability issue, we selected the SUS score as an important outcome of the study: poor usability is one of the main causes of technological system abandonment [75,76] and was shown to influence people's acceptance of digital solutions and adherence to the treatment [79]. In our study, the level of usability of the ARC solution, as measured by the SUS [56], is in line with that reported by Rossetto and colleagues (2023) [74], who studied a telerehabilitation system in chronic disabilities, including chronic obstructive pulmonary disease and PD. On average, the SUS score was above 70/100, recognized as an appropriate cut-off of good acceptance [56]. In our sample, pwPD, who were older and more disabled than COV19, presented slightly lower acceptability and usability of telerehabilitation at the time of enrollment, but after the study, the level of reported ARC usability improved slightly. It is possible that ARC met some of the requirements that several authors [58,75,[80][81][82][83][84][85][86] believe are important for improving usability and adherence, such as the establishment of a good therapeutic alliance with patients through initial training, the ability to choose and adapt the physiotherapy program at any time tailored to the patient, and to reach goals important to the patient. These characteristics may also be useful in the case of COVID-19, which showed a good baseline value that was slightly lowered. It is possible that some technical problems encountered during rehabilitation may affect the SUS value at follow-up.
The quality and efficacy of healthcare depends on patient adherence to recommended treatment regimens. Poor adherence to treatment may lead to worse outcomes across many healthcare disciplines including physiotherapy [87,88].
Poor adherence has implications for subjects' health and well-being, as well as treatment cost and effectiveness [65,87]. The clinical effectiveness of training depends on repetition as well as salience, specificity, difficulty, shaping, motivation, and feedback [8,95,96].
In our study, adherence is the primary outcome, and was set at a high score of 80% to obtain the theoretical maximum effect based on daily exercise [57,58]. ARC Intellicare allows for the monitoring of both daily access to the platform and the number of repetitions of the gesture. By monitoring the number of repetitions and-through the developed algorithm-the correct execution of the gesture, it allows us to give patients real-time feedback on goals and performance, which are crucial for motor learning [95,96].
Our study's small sample performed very consistently and with high compliance in both parameters monitored by ARC Intellicare. The measure of repetitions was related to the trend in adherence measured by daily accesses to the platform, although it was lower in percentage value. Adherence measured by repetitions might be a more accurate variable for measuring learning progress: in any case, no differences emerged between the two variables with respect to effectiveness.
A recent scoping review by Yen and colleagues [97] found that the effectiveness of telerehabilitation in promoting independence in activities of daily living (ADLs) is supported by the integration of sensing tools coupled with processing algorithms, which enable remote monitoring and quantifiable measurements. Versatile smart sensors can generate clinically relevant data and recognize user gestures in real time. Machine learning algorithms should be included to promote the learning process. Although remote rehabilitation was conceived with the intention of reducing the healthcare burden and reaching the most distant people, it is still important to ensure-during telerehabilitation as well as in presence-discussion with rehabilitation professionals and adaptability/progression of training in the context of selecting new goals in parallel with learning [97]. Therefore, it is useful to promote mixed models of synchronous and asynchronous rehabilitation and to allow for video consultation with the practitioner. Finally, it might be useful to provide the user with a digital tool literacy program.
ARC allowed for a synchronous session of rehabilitation, that in the study was provided weekly to verify progression and acquire patients' feedback. Treatment efficacy was demonstrated by a homogeneous and statistically significant improvement, which was observed at follow-up in all selected clinical outcome measures except mBI. The mBI score, although improved, did not reach statistical significance, particularly in the COV19 group, where it already started from very high values (99 on average) at baseline. The chronic condition, typical of both enrolled patient groups, determines the mBI results in a measure that is not sensitive to change by a ceiling effect.
This study has two main limitations. The first is the small sample size, justified by the pilot scope of the study, and the latter is the absence of a control group.
To reduce the bias due to the uncontrolled design and to increase the reliability of interpretation, we measured the clinical outcome through internationally validated scales (Barthel Dyspnea Index [52], Brief Fatigue Inventory [62], etc.) or tests (i.e., 2MWT) [61], and considered the minimum clinically important difference (MCID) value, when available, as a comparison. At the end of the study protocol, wearable sensor-assisted home telerehabilitation, performed through a patient-tailored schedule and with asynchronous sessions at 80%, was effective in increasing patients' scores beyond the MCID of Barthel's Dyspnea Index [69], modified Barthel's Index [68], 2MWT [61,71] and Borg Scale [71,72].
No MCID values are available for the other outcome measures. The pwPD did not achieve a statistically significant improvement in the 2MWT, despite having on average walked 8.1 m more at the end of treatment and reporting less fatigue and a better quality of life. However, walking tests may be widely dependent on the effect of medication rather than cardio-fitness level in moderate-advanced pwPD, such as those enrolled in our study (Hoehn and Yahr score = 3, which means moderate-advanced PD phase) [97]. On the other hand, the BFI score improved significantly in COV19 subjects as well as in pwPD. In pwPD, fatigue could predict the progression of motor dysfunction severity over a longitudinal duration in subjects with disease progression, having a decline in physical and mental fatigue [98].

Conclusions
Telerehabilitation through the ARC Intellicare system is usable and safe for home rehabilitation in subjects with chronic motor and respiratory disabilities. Preliminary data suggest promising results on the effectiveness in subjects with post-COVID-19 condition or PD. Devices as ARC Intellicare, based on wearable sensors and ML processing algorithms, allow for real-time monitoring of several aspects of adherence: both daily adherence and repetition based on exercise recognition. The results obtained should be confirmed through a larger-sample randomized controlled trial.
Finally, these preliminary results suggest that after rigorous testing, ARC could be useful in the treatment of several conditions causing disability that require motor and respiratory rehabilitation that can be performed remotely.