Non-Immersive Virtual Reality Telerehabilitation System Improves Postural Balance in People with Chronic Neurological Diseases

Background: People with chronic neurological diseases, such as Parkinson’s Disease (PD) and Multiple Sclerosis (MS), often present postural disorders and a high risk of falling. When difficulties in achieving outpatient rehabilitation services occur, a solution to guarantee the continuity of care may be telerehabilitation. This study intends to expand the scope of our previously published research on the impact of telerehabilitation on quality of life in an MS sample, testing the impact of this type of intervention in a larger sample of neurological patients also including PD individuals on postural balance. Methods: We included 60 participants with MS and 72 with PD. All enrolled subjects were randomized into two groups: 65 in the intervention group and 67 in the control group. Both treatments lasted 30–40 sessions (5 days/week, 6–8 weeks). Motor, cognitive, and participation outcomes were registered before and after the treatments. Results: All participants improved the outcomes at the end of the treatments. The study’s primary outcome (Mini-BESTest) registered a greater significant improvement in the telerehabilitation group than in the control group. Conclusions: Our results demonstrated that non-immersive virtual reality telerehabilitation is well tolerated and positively affects static and dynamic balance and gait in people with PD and MS.


Introduction
Over the past 30 years, the burden of chronic neurological diseases is continuously augmenting due to population increase and aging, with dramatic medical and economic implications [1]. Chronic neurological diseases such as Parkinson's Disease (PD) and Multiple Sclerosis (MS), have a significant long-term impact on Activities of Daily Living progressively reducing the Quality of Life (QoL) [2][3][4]. For these reasons, the management of PD and MS is a health priority of the WHO considering the increasing demand for treatment, rehabilitation, and support services [5].

Participants
A sample of individuals diagnosed with PD (according to the United Kingdom PD Society Brain Bank criteria [42]) or MS (according to McDonald's criteria [43]) was recruited between 2017 and 2020 within two multicenter randomized controlled trials that involved five Italian rehabilitation hospitals (Istituti di Ricovero e Cura a Carattere Scientifico-IRCCS) of the Italian Neuroscience and Rehabilitation Network (https://www.reteneuroscienze.it/en; accessed on 7 February 2023). The subjects were enrolled if they met the following eligibility (inclusion and exclusion) criteria: • Between 25 and 70 years of age; • Stage of disease: mild to moderate as documented by Hoehn and Yahr (H&Y score range between 2 and 3-PD group) or Expanded Disability Status Scale (EDSS score ≤ 6.5-MS group); • Absence of cognitive impairment measured by the MoCA total score ≥ 18 [44] and sufficient cognitive and linguistic level to understand and comply with study procedures; • Stabilized drug treatment for at least 3 months before starting this study; • Absence of moderate and severe dyskinesia and freezing episodes as documented by MDS-UPDRS (PD group); • No other neurologic conditions different from MS or PD; • No psychiatric complications or personality disorders, as indicated in the medical documentation; • Absence of severe primary sensory deficits such as blurring or low vision, severe hearing loss and speech disorder Written and informed consent was obtained for all participants. None of the participants were involved in other experimental trials during the entire duration of the present study.

Rehabilitation Procedures
All enrolled subjects were randomized into two groups: (1) the Intervention Group (IG) which received a non-immersive virtual reality telerehabilitation system and (2) the Control Group (CG) which received at-home conventional rehabilitation. The randomization procedure was applied to people with PD and to MS separately.

Intervention Group (IG)
The IG underwent 30-40 sessions (5 days/week, 6-8 weeks; 45 min each session) of motor, and cognitive rehabilitation exercises using the VRRS Tablet home telerehabilitation system (Khymeia Srl, Noventa Padovana, Italy). The motor exercises were performed using inertial sensors for the acquisition and processing of the movement performed by the patient. These data were shown to the patient with visual and auditory feedback in a serious game environment. The exercises covered the rehabilitation of balance and lower limbs (e.g., maintaining balance on one leg, marching in place, standing on tiptoe, squatting, etc.). The therapists involved in the study defined the protocol of exercises in telerehabilitation mode customized according to the characteristics and needs of the patient.

Control Group (CG)
The CG underwent 30-40 sessions (5 days/week, 6-8 weeks; 45 min each session) of at-home treatments without the use of any technological devices. The CG rehabilitation was an active comparator treatment and consisted of a written home-based self-administered booklet with conventional motor and cognitive activities tailored for each patient. The motor exercises included the rehabilitation of balance and lower limbs (e.g., maintaining balance on one leg, marching in place, standing on tiptoe, squatting, etc.). The intensity and duration of the CG were the same as the IG.

Outcome Measures
Clinical assessment was performed at baseline (T1) and at the end of the treatment (T2) both in people with PD and MS. All investigators and outcome assessors were blinded to the type of treatment. The following outcome measures were administered by assessors blinded to the intervention groups: • The mini-Balance Evaluation Systems Test (mini-BESTest) is a shortened version of the Balance Evaluation Systems Test. It is composed of a 14-item scale that evaluates balance with a total score of 28. Items are grouped into the following four subcomponents: anticipatory postural control (max score = 6), reactive postural control (max score = 6), somatosensory orientation (max score = 6), and dynamic walking (max score = 10). A summary of the subcomponents and the items of the mini-BESTest is depicted in Table 1. The mini-BESTest has been shown to have good psychometric properties in both PD and MS [45,46]. • The Timed Up-and-Go (TUG) test which involves rising from a seated position, walking to a pre-determined location, turning, and returning to a seated position, is a common test used to assess functional mobility, dynamic balance, and walking ability. The score is the time required to perform the following tasks: standing up from a chair; walking 3 m: turning around, walking back to the chair and sitting down. The validity and reliability of the TUG in people with PD and MS have been published [47,48]. TUG performance has been associated with mobility status and fall risk [49,50]. The TUG test used was the subtest included in the mini-BESTest.

•
The Timed Up-and-Go-test Dual-task (TUG-D) is a dual-task measure of functional mobility that evaluates balance with a simultaneous cognitive task. The TUG-D score is the time required to perform the TUG when the following cognitive task is added: while walking, the participant counts backward in threes from a randomly chosen start number between 60 and 100 to avoid a learning effect. The TUG-D performance on the TUG-D represents a significant predictor of future falls in people with PD and MS [51,52]. The TUG-D test used was the subtest included in the mini-BESTest.

•
The Montreal Cognitive Assessment (MoCA) is a rapid screening instrument for mild cognitive dysfunction. It assesses different cognitive domains: attention and concentration, executive functions, memory, language, visuo-constructional skills, conceptual thinking, calculations, and orientation. The total MoCA score is 30 points. The MoCA has been recognized as a valid and sensitive instrument to identify cognitive impairment in people with PD and MS [53,54]. Table 1. Summary of the subcomponents and the items of the mini-Balance Evaluation Systems Test (mini-BESTest).

Mini-BESTest (Max 28 Points)
Anticipatory postural control (max 6 points) -Sit to stand -Rise to toes -Stand on one leg (right and left) * The primary outcome of the study was the mini-BESTest total score.

Statistical Analysis
The database analyzed in this study was composed of people with MS from our previous study [38] to which people with PD have been added. A sample size of 60 patients (30 per arm) achieves 95% power to detect a difference of 2.0 (standard deviation = 2.0; Cohen's d = 1) in the primary outcome [58] in a design with two repeated measurements according to the literature on the psychometric performance of the Mini-BESTest in patients with balance disorders [59], assuming an alpha error of 0.05, and 5% dropout rate of patients. A priori sample size was calculated using G*Power 3.1 software. Statistical analyses were carried out by using jamovi (Version 2.3) software (https://www.jamovi.org; accessed on 7 February 2023). Summary statistics are expressed as frequencies, percentages, means and Standard Deviations (SD), median and Interquartile Range (IQR). Comparisons between the two groups (IG and CG) of baseline clinical features in the full sample and people with PD and MS separately using parametric (independent samples t-tests, analysis of covariance-ANCOVA) or corresponding non-parametric (independent samples Mann-Whitney U test) tests, as appropriate. Variable distribution was inspected through histograms and skewness and kurtosis statistics were calculated to assess normality. When variables violated the assumption of normal distribution, the natural logarithmic [ln] transformation was applied. A modified intention-to-treat method was implemented [60]. Missing data for the outcome measures comprised in the principal dataset were handled according to a single imputation procedure replacing missing values with the median value in each group at a specific time point in case of missing data less than 5% [61]. Outcome measures were analyzed using the jamovi module General analyses for linear model [62]  appropriate. Variable distribution was inspected through histograms and skewness and kurtosis statistics were calculated to assess normality. When variables violated the assumption of normal distribution, the natural logarithmic [ln] transformation was applied. A modified intention-to-treat method was implemented [60]. Missing data for the outcome measures comprised in the principal dataset were handled according to a single imputation procedure replacing missing values with the median value in each group at a specific time point in case of missing data less than 5% [61]. Outcome measures were analyzed using the jamovi module General analyses for linear model [62]  , and the magnitude of effects was interpreted as follows: small (pη 2 = 0.01), medium (pη 2 = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-value < 0.05 for all analyses.

Results
A sample of 150 participants met the inclusion criteria and were included in the appropriate. Variable distribution was inspected through histograms and skewness an kurtosis statistics were calculated to assess normality. When variables violated th assumption of normal distribution, the natural logarithmic [ln] transformation w applied. A modified intention-to-treat method was implemented [60]. Missing data fo the outcome measures comprised in the principal dataset were handled according to single imputation procedure replacing missing values with the median value in eac group at a specific time point in case of missing data less than 5% [61]. Outcome measur were analyzed using the jamovi module General analyses for linear model [62] (retrieve from https://gamlj.github.io; accessed on on 7 February 2023). Generalized Linear Mixe Models (GLMMs) were performed to evaluate score differences across the two time poin (T1 and T2). Different test scores were used as dependent variables (one for each mode and the effects (Time, Group, Pathology) were considered independent variables. Tim Group, Pathology, and their interaction (Time✻Group, Time✻Pathology, Group✻P thology, Time✻Group✻Pathology) were included in each model as fixed effects. To a count for subject specific variability, each subject was used as a random factor in all th models. GLMMs for the cognitive outcome (MoCa Test), also included age and educatio as covariates. The final models were the following: Motor Outcome measure ~1 + Time Group + Pathology + Time:Group + Time:Pathology + Group:Pathology + Time:Group:P thology +( 1|Subject), Cognitive Outcome measure ~1 + Time + Group + Pathology + Ag + Education + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Patholog + (1|Subject). Effects sizes (partial eta-squared pη 2 ) for the posthoc tests performed to i terpret significant fixed effects were calculated by R Studio software (Version 1.4.1106 and the magnitude of effects was interpreted as follows: small (pη 2 = 0.01), medium (p = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-value 0.05 for all analyses.

Results
A sample of 150 participants met the inclusion criteria and were included in th appropriate. Variable distribution was inspected through histogram kurtosis statistics were calculated to assess normality. When va assumption of normal distribution, the natural logarithmic [ln] applied. A modified intention-to-treat method was implemented [6 the outcome measures comprised in the principal dataset were han single imputation procedure replacing missing values with the m group at a specific time point in case of missing data less than 5% [61] were analyzed using the jamovi module General analyses for linear m from https://gamlj.github.io; accessed on on 7 February 2023). Gener Models (GLMMs) were performed to evaluate score differences acros (T1 and T2). Different test scores were used as dependent variables ( and the effects (Time, Group, Pathology) were considered independ Group, Pathology, and their interaction (Time✻Group, Time✻Pat thology, Time✻Group✻Pathology) were included in each model as count for subject specific variability, each subject was used as a ran models. GLMMs for the cognitive outcome (MoCa Test), also include as covariates. The final models were the following: Motor Outcome Group + Pathology + Time:Group + Time:Pathology + Group:Patholo thology +( 1|Subject), Cognitive Outcome measure ~1 + Time + Grou + Education + Time:Group + Time:Pathology + Group:Pathology + Tim + (1|Subject). Effects sizes (partial eta-squared pη 2 ) for the posthoc te terpret significant fixed effects were calculated by R Studio softwar and the magnitude of effects was interpreted as follows: small (pη 2 = = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance 0.05 for all analyses.

Results
A sample of 150 participants met the inclusion criteria and w priate. Variable distribution was inspected through histograms and skewness and is statistics were calculated to assess normality. When variables violated the ption of normal distribution, the natural logarithmic [ln] transformation was d. A modified intention-to-treat method was implemented [60]. Missing data for tcome measures comprised in the principal dataset were handled according to a imputation procedure replacing missing values with the median value in each at a specific time point in case of missing data less than 5% [61]. Outcome measures nalyzed using the jamovi module General analyses for linear model [62] (retrieved ttps://gamlj.github.io; accessed on on 7 February 2023). Generalized Linear Mixed s (GLMMs) were performed to evaluate score differences across the two time points d T2). Different test scores were used as dependent variables (one for each model), e effects (Time, Group, Pathology) were considered independent variables. Time, , Pathology, and their interaction (Time✻Group, Time✻Pathology, Group✻Pay, Time✻Group✻Pathology) were included in each model as fixed effects. To acfor subject specific variability, each subject was used as a random factor in all the s. GLMMs for the cognitive outcome (MoCa Test), also included age and education ariates. The final models were the following: Motor Outcome measure ~1 + Time + + Pathology + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pay +( 1|Subject), Cognitive Outcome measure ~1 + Time + Group + Pathology + Age ation + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology ubject). Effects sizes (partial eta-squared pη 2 ) for the posthoc tests performed to int significant fixed effects were calculated by R Studio software (Version 1.4.1106), e magnitude of effects was interpreted as follows: small (pη 2 = 0.01), medium (pη 2 , and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-value < r all analyses.
ults Group EVIEW 6 of 16 appropriate. Variable distribution was inspected through histograms and skewness and kurtosis statistics were calculated to assess normality. When variables violated the assumption of normal distribution, the natural logarithmic [ln] transformation was applied. A modified intention-to-treat method was implemented [60]. Missing data for the outcome measures comprised in the principal dataset were handled according to a single imputation procedure replacing missing values with the median value in each group at a specific time point in case of missing data less than 5% [61]. Outcome measures were analyzed using the jamovi module General analyses for linear model [62] (retrieved from https://gamlj.github.io; accessed on on 7 February 2023). Generalized Linear Mixed Models (GLMMs) were performed to evaluate score differences across the two time points (T1 and T2). Different test scores were used as dependent variables (one for each model), and the effects (Time, Group, Pathology) were considered independent variables. Time, Group, Pathology, and their interaction (Time✻Group, Time✻Pathology, Group✻Pathology, Time✻Group✻Pathology) were included in each model as fixed effects. To account for subject specific variability, each subject was used as a random factor in all the models. GLMMs for the cognitive outcome (MoCa Test), also included age and education as covariates. The final models were the following: Motor Outcome measure ~1 + Time + Group + Pathology + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology +( 1|Subject), Cognitive Outcome measure ~1 + Time + Group + Pathology + Age + Education + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology + (1|Subject). Effects sizes (partial eta-squared pη 2 ) for the posthoc tests performed to interpret significant fixed effects were calculated by R Studio software (Version 1.4.1106), and the magnitude of effects was interpreted as follows: small (pη 2 = 0.01), medium (pη 2 = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-value < 0.05 for all analyses.

Results
Pathology) were included in each model as fixed effects. To account for subject specific variability, each subject was used as a random factor in all the models. GLMMs for the cognitive outcome (MoCa Test), also included age and education as covariates. The final models were the following: Motor Outcome measure~1 + Time + Group + Pathology + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology + (1|Subject), Cognitive Outcome measure~1 + Time + Group + Pathology + Age + Education + Time:Group + Time:Pathology + Group:Path-ology + Time:Group:Pa-thology + (1|Subject). Effects sizes (partial eta-squared pη 2 ) for the posthoc tests performed to interpret significant fixed effects were calculated by R Studio software (Version 1.4.1106), and the magnitude of effects was interpreted as follows: small (pη 2 = 0.01), medium (pη 2 = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-value < 0.05 for all analyses.

Results
A sample of 150 participants met the inclusion criteria and were included in the study: 80 people with PD and 70 with SM. Of the sample, 75 participants were allocated to the IG (PD = 40; MS = 35) and 75 to the CG (PD = 40; MS = 35). A sample of 132 participants (IG = 65; CG = 67) completed the baseline (T0) and post-treatment (T1) evaluations, and 18 subjects dropped out (IG = 10; CG = 8). None of the drop-outs occurred during the study for treatment-related reasons, and no participant experienced any adverse event during treatment. Baseline demographics and clinical data in the full sample, people with PD, and people with MS are detailed in Table 2. No differences were registered between IG and CG at T0. The baseline demographics and clinical data in the IG and CG are detailed in the Supplementary materilas (Table S1). Results of the GLMMs were performed on each outcome measure to verify whether the two treatments had different effects on the common outcomes in the full sample of patients with chronic neurological diseases (Table 3).    0.001, pη 2 = 0.09). No significant differences were found in the people with PD Figure 2 depicts the results on TUG-D[ln].  0.001, pη 2 = 0.09). No significant differences were found in the people with PD Figure 2 depicts the results on TUG-D[ln].    All clinical outcomes significantly improved between T0 and T1 in both groups except for Mini-BESTest Reactive postural control and Somatosensory orientation subcomponents. When considering the primary outcome (i.e., the Mini-BESTest total score), the IG showed a greater improvement than the CG (effect of interaction Time J. Clin. Med. 2023, 12, x FOR PEER REVIEW appropriate. Variable distribution was inspected through histograms and skewne kurtosis statistics were calculated to assess normality. When variables violat assumption of normal distribution, the natural logarithmic [ln] transformatio applied. A modified intention-to-treat method was implemented [60]. Missing d the outcome measures comprised in the principal dataset were handled accordin single imputation procedure replacing missing values with the median value i group at a specific time point in case of missing data less than 5% [61]. Outcome me were analyzed using the jamovi module General analyses for linear model [62] (ret from https://gamlj.github.io; accessed on on 7 February 2023). Generalized Linear Models (GLMMs) were performed to evaluate score differences across the two time (T1 and T2). Different test scores were used as dependent variables (one for each m and the effects (Time, Group, Pathology) were considered independent variables Group, Pathology, and their interaction (Time✻Group, Time✻Pathology, Grou thology, Time✻Group✻Pathology) were included in each model as fixed effects. count for subject specific variability, each subject was used as a random factor in models. GLMMs for the cognitive outcome (MoCa Test), also included age and edu as covariates. The final models were the following: Motor Outcome measure ~1 + Group + Pathology + Time:Group + Time:Pathology + Group:Pathology + Time:Gro thology +( 1|Subject), Cognitive Outcome measure ~1 + Time + Group + Pathology + Education + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pat + (1|Subject). Effects sizes (partial eta-squared pη 2 ) for the posthoc tests performed terpret significant fixed effects were calculated by R Studio software (Version 1.4 and the magnitude of effects was interpreted as follows: small (pη 2 = 0.01), mediu = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-v 0.05 for all analyses.  Table 2. No differences were registered be IG and CG at T0. The baseline demographics and clinical data in the IG and CG tailed in the Supplementary materilas (Table S1). Results of the GLMMs were perf on each outcome measure to verify whether the two treatments had different eff the common outcomes in the full sample of patients with chronic neurological d ( Table 3).

Results
All clinical outcomes significantly improved between T0 and T1 in both grou cept for Mini-BESTest Reactive postural control and Somatosensory orientation su ponents. When considering the primary outcome (i.e., the Mini-BESTest total scor IG showed a greater improvement than the CG (effect of interaction Time✻Group: p = 0.020; posthoc comparison: IG T0 vs. T1 p-value < 0.001, pη 2 = 0.13) with an improv of about 24% concerning the maximum score achievable. A significant effect of facto was observed for the Mini-BESTest Anticipatory postural control subcomponent (p < 0.001, pη 2 = 0.10) showing an improvement in both groups. When considering the appropriate. Variable distribution was inspected through histograms and skewness and kurtosis statistics were calculated to assess normality. When variables violated the assumption of normal distribution, the natural logarithmic [ln] transformation was applied. A modified intention-to-treat method was implemented [60]. Missing data for the outcome measures comprised in the principal dataset were handled according to a single imputation procedure replacing missing values with the median value in each group at a specific time point in case of missing data less than 5% [61]. Outcome measures were analyzed using the jamovi module General analyses for linear model [62] (retrieved from https://gamlj.github.io; accessed on on 7 February 2023). Generalized Linear Mixed Models (GLMMs) were performed to evaluate score differences across the two time points (T1 and T2). Different test scores were used as dependent variables (one for each model), and the effects (Time, Group, Pathology) were considered independent variables. Time, Group, Pathology, and their interaction (Time✻Group, Time✻Pathology, Group✻Pathology, Time✻Group✻Pathology) were included in each model as fixed effects. To account for subject specific variability, each subject was used as a random factor in all the models. GLMMs for the cognitive outcome (MoCa Test), also included age and education as covariates. The final models were the following: Motor Outcome measure ~1 + Time + Group + Pathology + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology +( 1|Subject), Cognitive Outcome measure ~1 + Time + Group + Pathology + Age + Education + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology + (1|Subject). Effects sizes (partial eta-squared pη 2 ) for the posthoc tests performed to interpret significant fixed effects were calculated by R Studio software (Version 1.4.1106), and the magnitude of effects was interpreted as follows: small (pη 2 = 0.01), medium (pη 2 = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-value < 0.05 for all analyses. None of the drop-outs occurred during the study for treatment-related reasons, and no participant experienced any adverse event during treatment. Baseline demographics and clinical data in the full sample, people with PD, and people with MS are detailed in Table 2. No differences were registered between IG and CG at T0. The baseline demographics and clinical data in the IG and CG are detailed in the Supplementary materilas (Table S1). Results of the GLMMs were performed on each outcome measure to verify whether the two treatments had different effects on the common outcomes in the full sample of patients with chronic neurological diseases ( Table 3).

Results
All clinical outcomes significantly improved between T0 and T1 in both groups except for Mini-BESTest Reactive postural control and Somatosensory orientation subcom- appropriate. Variable distribution was inspected through histograms and skewness and kurtosis statistics were calculated to assess normality. When variables violated the assumption of normal distribution, the natural logarithmic [ln] transformation was applied. A modified intention-to-treat method was implemented [60]. Missing data for the outcome measures comprised in the principal dataset were handled according to a single imputation procedure replacing missing values with the median value in each group at a specific time point in case of missing data less than 5% [61]. Outcome measures were analyzed using the jamovi module General analyses for linear model [62] (retrieved from https://gamlj.github.io; accessed on on 7 February 2023). Generalized Linear Mixed Models (GLMMs) were performed to evaluate score differences across the two time points (T1 and T2). Different test scores were used as dependent variables (one for each model), and the effects (Time, Group, Pathology) were considered independent variables. Time, Group, Pathology, and their interaction (Time✻Group, Time✻Pathology, Group✻Pathology, Time✻Group✻Pathology) were included in each model as fixed effects. To account for subject specific variability, each subject was used as a random factor in all the models. GLMMs for the cognitive outcome (MoCa Test), also included age and education as covariates. The final models were the following: Motor Outcome measure ~1 + Time + Group + Pathology + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology +( 1|Subject), Cognitive Outcome measure ~1 + Time + Group + Pathology + Age + Education + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology + (1|Subject). Effects sizes (partial eta-squared pη 2 ) for the posthoc tests performed to interpret significant fixed effects were calculated by R Studio software (Version 1.4.1106), and the magnitude of effects was interpreted as follows: small (pη 2 = 0.01), medium (pη 2 = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-value < 0.05 for all analyses. None of the drop-outs occurred during the study for treatment-related reasons, and no participant experienced any adverse event during treatment. Baseline demographics and clinical data in the full sample, people with PD, and people with MS are detailed in Table 2. No differences were registered between IG and CG at T0. The baseline demographics and clinical data in the IG and CG are detailed in the Supplementary materilas (Table S1). Results of the GLMMs were performed on each outcome measure to verify whether the two treatments had different effects on appropriate. Variable distribution was inspected through histograms and skewness and kurtosis statistics were calculated to assess normality. When variables violated the assumption of normal distribution, the natural logarithmic [ln] transformation was applied. A modified intention-to-treat method was implemented [60]. Missing data for the outcome measures comprised in the principal dataset were handled according to a single imputation procedure replacing missing values with the median value in each group at a specific time point in case of missing data less than 5% [61]. Outcome measures were analyzed using the jamovi module General analyses for linear model [62] (retrieved from https://gamlj.github.io; accessed on on 7 February 2023). Generalized Linear Mixed Models (GLMMs) were performed to evaluate score differences across the two time points (T1 and T2). Different test scores were used as dependent variables (one for each model), and the effects (Time, Group, Pathology) were considered independent variables. Time, Group, Pathology, and their interaction (Time✻Group, Time✻Pathology, Group✻Pathology, Time✻Group✻Pathology) were included in each model as fixed effects. To account for subject specific variability, each subject was used as a random factor in all the models. GLMMs for the cognitive outcome (MoCa Test), also included age and education as covariates. The final models were the following: Motor Outcome measure ~1 + Time + Group + Pathology + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology +( 1|Subject), Cognitive Outcome measure ~1 + Time + Group + Pathology + Age + Education + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology + (1|Subject). Effects sizes (partial eta-squared pη 2 ) for the posthoc tests performed to interpret significant fixed effects were calculated by R Studio software (Version 1.4.1106), and the magnitude of effects was interpreted as follows: small (pη 2 = 0.01), medium (pη 2 = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-value < 0.05 for all analyses. None of the drop-outs occurred during the study for treatment-related reasons, and no participant experienced any adverse event during treatment. Baseline demographics and clinical data in the full sample, people with appropriate. Variable distribution was inspected through histograms and skewness and kurtosis statistics were calculated to assess normality. When variables violated the assumption of normal distribution, the natural logarithmic [ln] transformation was applied. A modified intention-to-treat method was implemented [60]. Missing data for the outcome measures comprised in the principal dataset were handled according to a single imputation procedure replacing missing values with the median value in each group at a specific time point in case of missing data less than 5% [61]. Outcome measures were analyzed using the jamovi module General analyses for linear model [62] (retrieved from https://gamlj.github.io; accessed on on 7 February 2023). Generalized Linear Mixed Models (GLMMs) were performed to evaluate score differences across the two time points (T1 and T2). Different test scores were used as dependent variables (one for each model), and the effects (Time, Group, Pathology) were considered independent variables. Time, Group, Pathology, and their interaction (Time✻Group, Time✻Pathology, Group✻Pathology, Time✻Group✻Pathology) were included in each model as fixed effects. To account for subject specific variability, each subject was used as a random factor in all the models. GLMMs for the cognitive outcome (MoCa Test), also included age and education as covariates. The final models were the following: Motor Outcome measure ~1 + Time + Group + Pathology + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology +( 1|Subject), Cognitive Outcome measure ~1 + Time + Group + Pathology + Age + Education + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology + (1|Subject). Effects sizes (partial eta-squared pη 2 ) for the posthoc tests performed to interpret significant fixed effects were calculated by R Studio software (Version 1.4.1106), and the magnitude of effects was interpreted as follows: small (pη 2 = 0.01), medium (pη 2 = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-value < 0.05 for all analyses. appropriate. Variable distribution was inspected through histograms and skewness and kurtosis statistics were calculated to assess normality. When variables violated the assumption of normal distribution, the natural logarithmic [ln] transformation was applied. A modified intention-to-treat method was implemented [60]. Missing data for the outcome measures comprised in the principal dataset were handled according to a single imputation procedure replacing missing values with the median value in each group at a specific time point in case of missing data less than 5% [61]. Outcome measures were analyzed using the jamovi module General analyses for linear model [62] (retrieved from https://gamlj.github.io; accessed on on 7 February 2023). Generalized Linear Mixed Models (GLMMs) were performed to evaluate score differences across the two time points (T1 and T2). Different test scores were used as dependent variables (one for each model), and the effects (Time, Group, Pathology) were considered independent variables. Time, Group, Pathology, and their interaction (Time✻Group, Time✻Pathology, Group✻Pathology, Time✻Group✻Pathology) were included in each model as fixed effects. To account for subject specific variability, each subject was used as a random factor in all the models. GLMMs for the cognitive outcome (MoCa Test), also included age and education as covariates. The final models were the following: Motor Outcome measure ~1 + Time + Group + Pathology + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology +( 1|Subject), Cognitive Outcome measure ~1 + Time + Group + Pathology + Age + Education + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology + (1|Subject). Effects sizes (partial eta-squared pη 2 ) for the posthoc tests performed to interpret significant fixed effects were calculated by R Studio software (Version 1.4.1106), and the magnitude of effects was interpreted as follows: small (pη 2 = 0.01), medium (pη 2 = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-value < 0.05 for all analyses. None of the drop-outs occurred during the study for treatment-related reasons, and no participant experienced any adverse event Group J. Clin. Med. 2023, 12, x FOR PEER REVIEW 6 appropriate. Variable distribution was inspected through histograms and skewness kurtosis statistics were calculated to assess normality. When variables violated assumption of normal distribution, the natural logarithmic [ln] transformation applied. A modified intention-to-treat method was implemented [60]. Missing data the outcome measures comprised in the principal dataset were handled according single imputation procedure replacing missing values with the median value in group at a specific time point in case of missing data less than 5% [61]. Outcome meas were analyzed using the jamovi module General analyses for linear model [62] (retrie from https://gamlj.github.io; accessed on on 7 February 2023). Generalized Linear M Models (GLMMs) were performed to evaluate score differences across the two time po (T1 and T2). Different test scores were used as dependent variables (one for each mo and the effects (Time, Group, Pathology) were considered independent variables. T Group, Pathology, and their interaction (Time✻Group, Time✻Pathology, Group✻ thology, Time✻Group✻Pathology) were included in each model as fixed effects. T count for subject specific variability, each subject was used as a random factor in al models. GLMMs for the cognitive outcome (MoCa Test), also included age and educa as covariates. The final models were the following: Motor Outcome measure ~1 + Tim Group + Pathology + Time:Group + Time:Pathology + Group:Pathology + Time:Group thology +( 1|Subject), Cognitive Outcome measure ~1 + Time + Group + Pathology + + Education + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Patho + (1|Subject). Effects sizes (partial eta-squared pη 2 ) for the posthoc tests performed t terpret significant fixed effects were calculated by R Studio software (Version 1.4.1 and the magnitude of effects was interpreted as follows: small (pη 2 = 0.01), medium = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-val 0.05 for all analyses.

Results
A sample of 150 participants met the inclusion criteria and were included in study: 80  None of the drop-outs occurred during study for treatment-related reasons, and no participant experienced any adverse e Pathology p-value = 0.014). Specifically, the posthoc analysis revealed that the people with MS who underwent the telerehabilitation intervention improved the velocity in performing the TUG-D[ln] after treatment, whereas no amelioration was observed in the people with MS who underwent the CG intervention (posthoc comparison: people with MS IG T0 vs. T1, p-value < 0.001, pη 2 = 0.09). No significant differences were found in the people with PD Figure 2    appropriate. Variable distribution was inspected through histograms and skewness and kurtosis statistics were calculated to assess normality. When variables violated the assumption of normal distribution, the natural logarithmic [ln] transformation was applied. A modified intention-to-treat method was implemented [60]. Missing data for the outcome measures comprised in the principal dataset were handled according to a single imputation procedure replacing missing values with the median value in each group at a specific time point in case of missing data less than 5% [61]. Outcome measures were analyzed using the jamovi module General analyses for linear model [62] (retrieved from https://gamlj.github.io; accessed on on 7 February 2023). Generalized Linear Mixed Models (GLMMs) were performed to evaluate score differences across the two time points (T1 and T2). Different test scores were used as dependent variables (one for each model), and the effects (Time, Group, Pathology) were considered independent variables. Time, Group, Pathology, and their interaction (Time✻Group, Time✻Pathology, Group✻Pathology, Time✻Group✻Pathology) were included in each model as fixed effects. To account for subject specific variability, each subject was used as a random factor in all the models. GLMMs for the cognitive outcome (MoCa Test), also included age and education as covariates. The final models were the following: Motor Outcome measure ~1 + Time + Group + Pathology + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology +( 1|Subject), Cognitive Outcome measure ~1 + Time + Group + Pathology + Age + Education + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology + (1|Subject). Effects sizes (partial eta-squared pη 2 ) for the posthoc tests performed to interpret significant fixed effects were calculated by R Studio software (Version 1.4.1106), and the magnitude of effects was interpreted as follows: small (pη 2 = 0.01), medium (pη 2 = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-value < 0.05 for all analyses.  Table 2. No differences were registered between IG and CG at T0. The baseline demographics and clinical data in the IG and CG are detailed in the Supplementary materilas (Table S1). Results of the GLMMs were performed on each outcome measure to verify whether the two treatments had different effects on the common outcomes in the full sample of patients with chronic neurological diseases ( Table 3).

Results
All clinical outcomes significantly improved between T0 and T1 in both groups except for Mini-BESTest Reactive postural control and Somatosensory orientation subcomponents. When considering the primary outcome (i.e., the Mini-BESTest total score), the IG showed a greater improvement than the CG (effect of interaction Time✻Group: p-value = 0.020; posthoc comparison: IG T0 vs. T1 p-value < 0.001, pη 2 = 0.13) with an improvement of about 24% concerning the maximum score achievable. A significant effect of factor Time was observed for the Mini-BESTest Anticipatory postural control subcomponent (p-value < 0.001, pη 2 = 0. 10) showing an improvement in both groups. When considering the Mini-BESTest Dynamic walking subcomponent, the IG showed a significantly higher improvement (effect of interaction Time✻Group p-value = 0.011; posthoc comparison: IG T0 vs. T1, p-value < 0.001, pη 2 = 0.13) than the CG (see Figure 1)

Discussion
Considering the importance of a regular sustained program for balance rehabilitation and fall prevention in chronic neurological diseases [19][20][21], this study moves from previously published findings on the impact of telerehabilitation on quality of life in an MS [40] expanding the research scope and the target sample. Hypothesizing to achieve positive effects of telerehabilitation in balance capacity, we tested the effect of such intervention on postural balance in a larger sample of neurologic patients, including not only patients with MS but also with PD.
An RCT cohort of 132 subjects with chronic neurological diseases was considered, representing we analyzed 87% of the fulfilled dataset (and the remaining 13% of the dataset was dismissed due to dropout). Our results showed that the non-immersive virtual reality telerehabilitation system implemented in this study was feasible and easily accepted in people with PD or MS, in line with previously published studies [34][35][36].
This study showed that the non-immersive virtual reality telerehabilitation system was effective and allowed to optimize the timing and intensity of the rehabilitation intervention. These findings are particularly relevant in the case of economic, geographic, and social-distancing barriers than may hinder people with chronic neurological diseases from appropriate. Variable distribution was inspected through histograms and skewness and kurtosis statistics were calculated to assess normality. When variables violated the assumption of normal distribution, the natural logarithmic [ln] transformation was applied. A modified intention-to-treat method was implemented [60]. Missing data for the outcome measures comprised in the principal dataset were handled according to a single imputation procedure replacing missing values with the median value in each group at a specific time point in case of missing data less than 5% [61]. Outcome measures were analyzed using the jamovi module General analyses for linear model [62] (retrieved from https://gamlj.github.io; accessed on on 7 February 2023). Generalized Linear Mixed Models (GLMMs) were performed to evaluate score differences across the two time points (T1 and T2). Different test scores were used as dependent variables (one for each model), and the effects (Time, Group, Pathology) were considered independent variables. Time, Group, Pathology, and their interaction (Time✻Group, Time✻Pathology, Group✻Pathology, Time✻Group✻Pathology) were included in each model as fixed effects. To account for subject specific variability, each subject was used as a random factor in all the models. GLMMs for the cognitive outcome (MoCa Test), also included age and education as covariates. The final models were the following: Motor Outcome measure ~1 + Time + Group + Pathology + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology +( 1|Subject), Cognitive Outcome measure ~1 + Time + Group + Pathology + Age + Education + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology + (1|Subject). Effects sizes (partial eta-squared pη 2 ) for the posthoc tests performed to interpret significant fixed effects were calculated by R Studio software (Version 1.4.1106), and the magnitude of effects was interpreted as follows: small (pη 2 = 0.01), medium (pη 2 = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-value < 0.05 for all analyses. None of the drop-outs occurred during the study for treatment-related reasons, and no participant experienced any adverse event during treatment. Baseline demographics and clinical data in the full sample, people with PD, and people with MS are detailed in Table 2 appropriate. Variable distribution was inspected through histograms and skewness and kurtosis statistics were calculated to assess normality. When variables violated the assumption of normal distribution, the natural logarithmic [ln] transformation was applied. A modified intention-to-treat method was implemented [60]. Missing data for the outcome measures comprised in the principal dataset were handled according to a single imputation procedure replacing missing values with the median value in each group at a specific time point in case of missing data less than 5% [61]. Outcome measures were analyzed using the jamovi module General analyses for linear model [62] (retrieved from https://gamlj.github.io; accessed on on 7 February 2023). Generalized Linear Mixed Models (GLMMs) were performed to evaluate score differences across the two time points (T1 and T2). Different test scores were used as dependent variables (one for each model), and the effects (Time, Group, Pathology) were considered independent variables. Time, Group, Pathology, and their interaction (Time✻Group, Time✻Pathology, Group✻Pathology, Time✻Group✻Pathology) were included in each model as fixed effects. To account for subject specific variability, each subject was used as a random factor in all the models. GLMMs for the cognitive outcome (MoCa Test), also included age and education as covariates. The final models were the following: Motor Outcome measure ~1 + Time + Group + Pathology + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology +( 1|Subject), Cognitive Outcome measure ~1 + Time + Group + Pathology + Age + Education + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pathology + (1|Subject). Effects sizes (partial eta-squared pη 2 ) for the posthoc tests performed to interpret significant fixed effects were calculated by R Studio software (Version 1.4.1106), and the magnitude of effects was interpreted as follows: small (pη 2 = 0.01), medium (pη 2 = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-value < 0.05 for all analyses. None of the drop-outs occurred during the study for treatment-related reasons, and no participant experienced any adverse event during treatment. Baseline demographics and clinical data in the full sample, people with PD, and people with MS are detailed in Table 2 appropriate. Variable distribution was inspected through histograms and skewness and kurtosis statistics were calculated to assess normality. When variables violated th assumption of normal distribution, the natural logarithmic [ln] transformation wa applied. A modified intention-to-treat method was implemented [60]. Missing data fo the outcome measures comprised in the principal dataset were handled according to single imputation procedure replacing missing values with the median value in eac group at a specific time point in case of missing data less than 5% [61]. Outcome measure were analyzed using the jamovi module General analyses for linear model [62] (retrieved from https://gamlj.github.io; accessed on on 7 February 2023). Generalized Linear Mixed Models (GLMMs) were performed to evaluate score differences across the two time point (T1 and T2). Different test scores were used as dependent variables (one for each model and the effects (Time, Group, Pathology) were considered independent variables. Time Group, Pathology, and their interaction (Time✻Group, Time✻Pathology, Group✻Pa thology, Time✻Group✻Pathology) were included in each model as fixed effects. To ac count for subject specific variability, each subject was used as a random factor in all th models. GLMMs for the cognitive outcome (MoCa Test), also included age and educatio as covariates. The final models were the following: Motor Outcome measure ~1 + Time Group + Pathology + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Pa thology +( 1|Subject), Cognitive Outcome measure ~1 + Time + Group + Pathology + Ag + Education + Time:Group + Time:Pathology + Group:Pathology + Time:Group:Patholog + (1|Subject). Effects sizes (partial eta-squared pη 2 ) for the posthoc tests performed to in terpret significant fixed effects were calculated by R Studio software (Version 1.4.1106 and the magnitude of effects was interpreted as follows: small (pη 2 = 0.01), medium (pη = 0.06), and large (pη 2 = 0.14) effects [63]. The statistical significance was set at p-value 0.05 for all analyses. None of the drop-outs occurred during th study for treatment-related reasons, and no participant experienced any adverse even during treatment. Baseline demographics and clinical data in the full sample, people wit PD, and people with MS are detailed in Table 2. No differences were registered betwee IG and CG at T0. The baseline demographics and clinical data in the IG and CG are de

Discussion
Considering the importance of a regular sustained program for balance rehabilitation and fall prevention in chronic neurological diseases [19][20][21], this study moves from previously published findings on the impact of telerehabilitation on quality of life in an MS [40] expanding the research scope and the target sample. Hypothesizing to achieve positive effects of telerehabilitation in balance capacity, we tested the effect of such intervention on postural balance in a larger sample of neurologic patients, including not only patients with MS but also with PD.
An RCT cohort of 132 subjects with chronic neurological diseases was considered, representing we analyzed 87% of the fulfilled dataset (and the remaining 13% of the dataset was dismissed due to dropout). Our results showed that the non-immersive virtual reality telerehabilitation system implemented in this study was feasible and easily accepted in people with PD or MS, in line with previously published studies [34][35][36].
This study showed that the non-immersive virtual reality telerehabilitation system was effective and allowed to optimize the timing and intensity of the rehabilitation intervention. These findings are particularly relevant in the case of economic, geographic, and social-distancing barriers than may hinder people with chronic neurological diseases from achieving outpatient rehabilitation services [28]. In this context, the non-immersive virtual reality telerehabilitation system is a promising approach to ensure continuity of care [29].
The postural balance assessed with the Mini-BESTest total score registered a higher improvement in the IG compared to the CG: subjects who conducted non-immersive virtual reality telerehabilitation system amended postural control more than their peers who conducted at-home conventional rehabilitation without the use of any technological devices. The outcomes obtained agree with the literature on the effects of telerehabilitation in improving postural control in individuals with chronic neurological diseases [38], PD [39,64], and MS [24,34,40,42,65,66]. On the other hand, in people with PD, the superiority of telerehabilitation in increasing postural stability has not been confirmed in some studies [39,62]. Specifically, Gandolfi et al. [39] and Seidler et al. [62] demonstrated that telerehabilitation improved balance and gait functions similar to conventional treatments.
The impact of the non-immersive virtual reality telerehabilitation system on postural control had been further explored by analyzing the dynamic balance and gait by means of the Mini-BESTest subcomponents, TUG, and TUG-D. The results evidenced that both the TI and the CG improved the Mini-BESTest "anticipatory postural control" subcomponent and TUG at the end of the treatment, without registering any between-group differences. Compared with the CG, the TI performed significantly better at the Mini-BESTest "dynamic walking" subcomponent and TUG-D. Thus, subjects who conducted non-immersive virtual reality telerehabilitation registered a higher performance in the execution of the following motor tasks: change in gait speed; walk with head turns; walk with pivot turns; step over obstacles and timed up and go with a dual task. These results are in agreement with the literature that showed an improvement in the ability to control balance and change gait speed and direction in subjects who underwent telerehabilitation [38,62,67].
Similar outcomes were also obtained in studies on the application of virtual reality technology in rehabilitation, which positively improved movement velocity, balance, and gait [68,69]. The results on the ability to execute a dual task (i.e., the TUG-D) are representative of the potential for non-immersive virtual reality telerehabilitation systems to improve both motor and cognitive functions [52]. To this extent, Intzandt et al. [70] analyzed how different types of rehabilitation can influence motor function (gait) and cognition in people with PD, finding that goal-based training can mitigate both motor and non-motor symptoms (such as fatigue, depression, apathy, and cognitive impairment) which can also influence motor performance.
The importance of the obtained outcome in favor of the non-immersive virtual reality telerehabilitation is also enhanced by the fact that the difficulty in executing a motor task paired with a simultaneous cognitive task has been associated with falls in people with chronic neurological diseases [52,71,72]. Furthermore, dual-task activities involving both motor and cognitive resources constitute a significant portion of most activities of daily living, and thus improving these activities has a major impact on participation and quality of life [73][74][75].
The analysis of the effects of telerehabilitation on cognitive functions revealed that both treatments improved the MoCA total score, and no differences were found between groups. Indeed, this outcome was influenced by the criteria for subject recruitment which included only participants without severe cognitive impairment (MoCA total score ≥ 18). Although the primary goal of the telerehabilitation treatments was the rehabilitation of balance and lower limb functions, the obtained outcomes on cognitive performance are encouraging and consistent with the literature [76,77], calling for future studies on cognitive telerehabilitation in chronic neurological diseases.
The analysis of differences between PD and MS revealed that the positive improvement of telerehabilitation in postural balance (Mini-BESTest total score) and ability to change gait speed and direction (Mini-BESTest "dynamic walking" subcomponent) was registered in both pathologies. Thus, the hypothesis of the study, i.e., telerehabilitation is effective in improving balance in both people with PD and SM, is reinforced and justified. On the other hand, the ability to execute a dual-task (i.e., the TUG-D) registered a significant difference between pathologies: subjects with MS who conducted the non-immersive virtual reality telerehabilitation improved the TUG-D more than their peers with PD. There is supportive evidence in the literature for the use of dual-task interventions to improve both single and dual-task gait speed [78]. The findings from the meta-analysis of Cinnera et al. support dual-task training as a beneficial therapy for improving dynamic balance and functional mobility in people with MS [79]. On the other hand, the impact of dual-task training in PD on dual-task gait speed is controversial [80,81]. The difference between performances in the two disease groups could be influenced by different brain substrates of neuroplasticity in the two diseases. It is well known that both PD and MS have the intrinsic property to structurally and functionally reorganize the damaged brain networks in response to rehabilitation [82,83]. The recovery in mild to moderate MS is achieved and sustained by the repair of damage through remyelination, with a resolution of inflammation and functional reorganization [84]. Whereas in PD the exercise, practice and movement strategy training act on supplementary circuits, thus supporting the sensorimotor integration and reinforcing the coupling of premotor areas, which are typically affected early in PD [85].
Our study had some limitations that are worth mentioning: it included patients with mild disability (H&Y score between 2.5 and 3 in people with PD; EDSS score ≤ 6.5 in people with MS) and the non-immersive virtual reality telerehabilitation system implemented in this study did not focus on upper limb motor therapy and specific cognitive rehabilitation. Therefore, the future research agenda should analyze the efficacy of telerehabilitation in more severely affected patients and increase the panel typology of treatments delivered remotely for complete physical and cognitive telerehabilitation in patients with chronic neurological diseases. Furthermore, the observed dropout rate (slightly more than 10%) was higher than the estimated a priori dropout rate (about 5%), approximately expected in these two neurological populations after virtual reality clinical trials as reported from recent metanalyses [86,87]. In these metanalyses, the typical reasons for dropout were difficulties in reaching the research center, refusal to participate, personal or familial issues, loss of data due to administrative problems, exacerbation of symptoms or other medical complications [86]. However, it is worth noting that in the period in which our study was carried out intervened a worldwide unpredictable adverse event due to the pandemic COVID-19. It has now been acknowledged that the COVID-19 pandemic has hindered the progress and completion of clinical trials [88].

Conclusions
The recent COVID-19 pandemic has underpinned the importance of ensuring a continuum of rehabilitation interventions, in particular for a frail population such as people with PD and MS. While therapeutic intervention in a clinical setting represents the firstline treatment, telerehabilitation intervention could act as the missing parts of the puzzle leading to an optimal continuity of care. Our results demonstrated that the non-immersive virtual reality telerehabilitation system is well tolerated and has positive effects on static and dynamic balance and gait in both people with PD and MS.
Supplementary Materials: The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/jcm12093178/s1, Table S1: Baseline demographics and clinical data in the IG and CG.