The study population of interest will be individuals who have had a stroke and stayed in a hospital linked to one of the following sites: 1) Sherbrooke site (Centre Hospitalier Universitaire de Sherbrooke), 2) Montreal sites (Jewish Rehabilitation Hospital, Institut de Réadaptation Gingras-Lindsay-de-Montreal and Centre de Santé et de Services Sociaux Champlain-Charles-Le Moyne), and 3) Toronto site (Sunnybrook Health Sciences Center). Special care will be taken to ensure that participant eligibility will target individuals who have had a stroke and returned home without requiring intensive rehabilitation. We will base our selection on the Rankin classification  established by the treating neurologist or trained therapist during the hospitalization period. This classification system allows us to assign individuals who have had a stroke to one of four groups according to their level of functional impairment. Patients with a Rankin score of 2 generally present with mild impairments, maintain their autonomy and can perform all previous activities. Patients with a Rankin score of 3 present with moderate impairments and generally require moderate assistance with their activities of daily living. These patients are usually sent home with or without referral for physiotherapy services at home.
Therefore, a sample of participants aged 45 years and older will be recruited based on the following inclusion criteria: 1) have had a stroke with a Rankin score of 2 or 3; 2) not be referred to an IFRU and return home following discharge from hospital; 3) understand instructions to allow participation in evaluations and interventions; 4) have a balance problem as evidenced by a score between 46 and 54 on the Berg Balance Scale ; 5) have a caregiver who would be available during the telerehabilitation sessions to ensure safety during exercises; and 6) live in an area served by high speed Internet. Excluded will be individuals who present with: 1) a previous stroke episode in the last 12 months (other than the present one); 2) severe body hemineglect; 3) significant hemianopsia visual problems accompanied by hemineglect; 4) uncontrolled medical problems; and 5) moderate to severe aphasia.
Participants will be recruited during the hospitalization period by the team of treating nurses, physiotherapists and neurologists at each of the three sites (Sherbrooke, Montreal and Toronto). When the team decides that a patient will not require a referral to an IFRU and will return home (inclusion criteria 1 and 2), they will verify inclusion criteria 3 and 4 and exclusion criteria 1 to 5 from the medical chart. If patients are eligible based on these criteria, the hospital team will ask if they agree to be contacted by a research associate. If they agree, a research associate will visit them in hospital to inform them of the project and ask them to participate. The research associate will verify inclusion criteria 5 and 6. If a patient meets all the criteria and is interested in participating in the study, they will be asked to sign the consent form previously approved by the ethics committee at each recruitment site.
Once included, participants will be randomly assigned to either the telerehabilitation group or the home visits group by the coordinator of each recruitment site. This randomization will be performed using block randomization of sizes 2 and 4 done by a computer. Block size will not be known by the evaluators. This randomization will be performed following stratification based on the Rankin score (2 or 3), a variable that is likely to influence recovery prognosis. A system of numbered, sealed envelopes will be put in place.
All evaluations will be completed in the research centers/hospitals of the three sites and will last approximately three hours. The first evaluation will be conducted at baseline, prior to starting intervention (T1) to collect base measures on the participants’ condition. A second evaluation will be conducted just after the two-month intervention (T2) and the third evaluation will be four months (T3) following discharge from the hospital. All evaluations will be carried out by evaluators trained in the standard procedures for all measures, who are independent and blinded to the intervention. Specific directions will be provided to the evaluators and the participants to ensure that intervention allocation remains anonymous.
To measure balance and functional mobility, the Community Balance and Mobility Scale (CB&M) [43–45] will be used. This tool was developed for ambulatory individuals following a traumatic brain injury . It has been validated in individuals following a stroke [44, 46]. It assesses 19 tasks. The items are rated on a six-point scale (from 0 = unable to perform, to 5 = able to perform independently), for a maximum total score of 95 points. The CB&M has demonstrated good intra-rater, inter-rater and test-retest reliability (ICC = 0.98 for all three cases) and high internal consistency (Cronbach’s alpha = 0.96) . It has good convergent validity with the Chedoke-McMaster Stroke Assessment  for leg and foot (r = 0.61 and 0.63, respectively) and for lower extremity strength (r = 0.67) . Furthermore, the CB&M has been shown to be sensitive to change from baseline at follow-up in community-dwelling, ambulatory individuals post-stroke (SRM (standardized response mean) = 0.83) . In addition, the Berg Balance Scale [42, 48] will be used to evaluate balance in patients. The Four-Squares Test will also be used. This test consists of moving in different preset directions: forward, backward and to the sides. The time required to complete the sequence is recorded. An execution time greater than 15 seconds is predictive of falling. This test has high inter-rater (ICC = 0.99), intra-rater (ICC = 0.98) and test-retest (ICC = 0.98) reliability in community-dwelling older individuals aged 65 years and over . Also, for predicting falls, this test has good specificity ranging from 88% to 100% and a positive predictive value of 86% .
The physical abilities capacities measured will be: 1) walking speed, 2) walking endurance, 3) lower extremity strength, and 6) lower extremity movement capacity. Each of these abilities plays an important role in achieving optimal functional mobility. Walking speed will be evaluated using the Timed Up and Go test (TUG) . The test consists of standing up from a chair, walking a distance of three meters, turning around and returning to the chair to sit down again. The test is timed to measure walking speed in seconds over a distance of six meters. The TUG demonstrates good convergent construct validity with moderate to strong correlations with walking using the Tinetti test (r = 0.53) and walking speed (r = 0.66) in an elders population aged 65 years and over . It has also been shown that walking speed is related to use of gait aids and fall frequency . It is a measure sensitive to change . Walking endurance will be estimated by the distance covered in meters during the Two-Minute Walk Test [53, 54]. The inter-rater and intra-rater reliability of this test is high (ICC = 0.99) . General lower extremity strength will be measured with the Sit to Stand test (STS) . The time required to complete five successive ‘sit-to-stand’ repetitions will be recorded [55, 56]. The test-retest reliability of the STS is high (ICC = 0.89) . There is also a strong correlation of the STS with walking speed over five meters (r = -0.66) and lower extremity strength (r = 0.47) . Lower extremity movement capacity and postural control will be evaluated using the Chedoke-McMaster Stroke Assessment . The impairment inventory determines the presence and severity of physical deficits, and allows patients to be classified according to Brunnstrom’s seven stages of motor recovery. This tool has good intra- and inter-rater reliability (ICC = 0.98 and 0.97, respectively) and good construct validity (r > 0.60 with different aspects of the Fugl Meyer test) .
Two main aspects of psychological capacities measured will be 1) fear of falling, and 2) self-efficacy. The simplified version of the Activities-specific Balance Confidence (ABC) scale  questionnaire (ABC-S) will measure fear of falling. This questionnaire includes 15 questions rated on a four-point Likert scale and evaluates an individual’s balance confidence during a series of daily activities [60, 61]. The ABC-S has strong internal consistency (Cronbach’s alpha = 0.86) good convergent validity with significant associations with balance and occurrence of falls . Self-efficacy when faced with difficult life situations will be evaluated using the Generalized Self-Efficacy Scale (GSES) . This questionnaire is a ten-item scale that uses a four-point Likert scale (from 1 = not true at all, to 4 = completely true) to assess whether an individual believes that their actions are responsible for their successful outcomes. This scale has been shown to have excellent internal consistency (Cronbach’s alpha = 0.86) .
Quality of life will be measured using the Reintegration to Normal Living Index (RNLI) . This tool consists of 11 items ranked on a three-point Likert scale. The total score ranges from 0 to 22, with higher scores depicting lower quality of life. The psychometric properties range from very good to excellent with good internal consistency (Cronbach’s alpha = 0.90) and good test-retest reliability (r = 0.83) . The construct validity of this tool has been examined in a post-stroke population and shows excellent correlations with the Frenchay Activities Index (r = 0.69) and the Short Form 36 Health Survey (r = 0.74) .
Participant satisfaction with the intervention received will be evaluated using the Health Care Satisfaction Questionnaire, a questionnaire developed and validated in French . The satisfaction construct is determined by three distinct factors which include satisfaction with: 1) the therapist relationship, 2) the services provided, and 3) the organization of services. This tool includes 26 questions, scored on a four-point Likert scale. The total score is calculated based on average satisfaction for the three factors, with higher scores indicating higher levels of satisfaction. Test-retest reliability is documented by an ICC of 0.72 and internal consistency by a Cronbach’s alpha of 0.93 for the entire scale .
A modified version of the ‘Cost analysis of telemedicine’ table from the University of Minnesota  will help tabulate the costs associated with the teletreatments. These costs will be calculated to establish a cost differential (that is, cost-effectiveness) between the two different types of intervention. The method for collecting this data has already been tested in a previous study . The variables related to the total cost include those associated with the actual intervention (for example, direct and indirect time for the intervention, professionals’ travel time to the participants’ homes, telerehabilitation equipment, and Internet service including installation and set-up).
The typical sociodemographic (for example, age, sex, marital status, living environment, education, primary occupation) and clinical characteristics (for example, diagnosis, gait aid, medications, self-rated health, co-morbidities) will be collected using an in-house questionnaire, with the exception of comorbidities for which the Functional Comorbidity Index will be used .
The level of impairment post-stroke will be measured by the National Institute of Health Stroke Score (NIHSS) [69, 70]. This questionnaire, administered by a healthcare professional, includes 15 items to evaluate patients in the acute post-stroke phase. It demonstrates good internal consistency (Cronbach’s alpha > 0.5) , good inter-rater (ICC = 0.69) and test-retest (ICC = 0.66 to 0.77) reliability in a post-stroke population  and good convergent validity (r = 0.74 with lesion volume seven days post-stroke) . The level of cognitive impairment will be measured using the Montreal Cognitive Assessment (MoCA) . This test evaluates attention, concentration, executive functions, memory, language, visioconstructive capacities, abstraction capacities, calculation and orientation. The maximum score of 30 corresponds to no cognitive problems. The internal consistency of this test is good (Cronbach’s alpha = 0.83) and test-retest reliability is excellent (ICC = 0.92) . The correlation between the MoCA and Mini-Mental State Examination scores has been shown to be excellent (r = 0.87) . Noteworthy is the fact that this test will be completed during the participant recruitment phase.
We expect to recruit 240 participants, that is, 120 per group. Allowing for a drop-out rate of 10%, this sample size will provide a power of 80% to demonstrate the non-inferiority of Tai Chi provided by telerehabilitation compared to Tai Chi provided in person at home, with a two-group one-sided test at the 2.5% level (nQuery Advisor 7.0; Statistical Solutions, Boston, MA). For this calculation, the non-inferiority limit was set at four points on the CB&M scale and the standard deviation of the gain was estimated to be 10.4, based on findings from Knorr [44
], since SRM = MeanGAIN
First, we will describe the characteristics of each group pre-intervention using averages and standard deviations (continuous variables) or percentages (categorical variables). The groups will then be compared using a t-test or chi-squared test. If the groups differ on certain characteristics pre-intervention despite randomization, subsequent analyses will account for these differences.
Primary analysis will aim to test the non-inferiority of Tai Chi provided via telerehabilitation compared to Tai Chi provided in person at home as evaluated by score gains on the CB&M questionnaire from T1 to T2. Delivery by telerehabilitation will be regarded as non-inferior to in person home delivery if the upper limit of the unilateral confidence interval around the difference in gains between the two groups is less than 4. This non-inferiority limit was set below the CB&M’s minimal clinically significant difference of five points. It is recommended to choose a non-inferiority limit less than the difference judged to be clinically significant [74, 75].
The same analytical strategy will be used to evaluate and compare the effectiveness of the interventions on secondary variables and to compare maintenance over time (that is, the confidence interval around the difference between the two groups from T1 to T3) across all variables. A more conservative level of significance will be used to account for the numerous tests.
All analyses will first be performed according to the treatment received (per protocol = PP) and according to the assigned group (intention to treat = ITT), requiring that non-inferiority be shown in both cases since ITT analyses increase the chance of declaring non-inferiority if the two delivery modes produce a similar pattern of withdrawal, while PP analyses bias towards the null in presence of a different pattern of withdrawal . For the ITT analysis, the effects of withdrawal or low levels of compliance will be explored with sensitivity analyses, where the missing data will be replaced with extreme values (that is, no change following treatment - the most favorable change noted in the study). Any non-robustness of results revealed by the comparison of strategies will be noted and the caveats will be mentioned in the discussion of results. All analyses will be conducted at the end of the study since the proposed treatments are of short duration and involve minimal risk to the participants.
The economic analysis will be a type of cost-minimization study . Assuming that the effectiveness will be the same in the dependent variable (CB&M), the cost will be determined for the two groups and the cost differential will be established.