This clinical study was registered on ClinicalTrials.gov (identifier number: NCT02077049). The design and treatments used in this study have been described in more detail previously .
A single-center randomized controlled clinical trial was performed comparing self-regulated conventional exercises with self-regulated exergames at Walenstadtberg Rehabilitation Clinic in Switzerland. Many of its clientele are persons over 65 years of age with musculoskeletal impairment (due to ortho-traumatology, internal medicine, oncology, or pulmonology) who are referred for inpatient rehabilitation from acute hospitals or by general practitioners. Patients were allocated by central randomization to the exergame group or to the conventional exercise group in a ratio of 1:1. Randomization was stratified according to balance and computer skills . A research assistant collected patient ratings during the intervention phase and performed clinical assessments pre- and post-intervention. Figure 1 shows the study design and patient flow.
Participants and recruitment
All patients over 65 years of age referred for inpatient rehabilitation from June 2014 to December 2015 were evaluated for inclusion in the study. Following medical screening by the doctor, patients were checked for inclusion and exclusion criteria by the study researcher. Inclusion criteria were: ability to walk independently over 20 m (with or without walking aids) and written informed consent. Exclusion criteria were: disorders limiting the use of computer games (e.g. neurological disorders, visual impairment, deafness, vertigo, cognitive impairment), or specific medical contra-indications, such as open wounds or severe pain, preventing prescription of self-regulated training. For the purpose of this study, cognitive impairment was defined as a Mini-Mental State Examination (MMSE) score <26 .
Randomization and blinding
An independent investigator who was not involved in the trial generated a randomization schedule with four strata according to computer and balance skills using blocks of two. A research assistant blinded to the randomization schedule checked inclusion criteria in eligible patients, asked for written informed consent, and informed the therapy secretariat about the included patient and stratum according to computer and balance skills. The therapy secretariat performed randomization according to the schedule and planned patient treatments according to group assignment.
Participants and physiotherapists involved in the study were inevitably aware of treatment allocation. Due to the small size of the rehabilitation center, it was impossible to blind the study researcher to the treatment allocation over the whole trial period. However, the therapists, the study researcher and the patients were not able to influence the pre- and post-intervention measurements. The ActiGraph® mobility tracker recorded and transmitted all data directly to the computer in an encrypted form. An external researcher, blinded to participants’ group allocation, analyzed the ActiGraph® data.
Measurements of training volume, recorded in the logbook, and of enjoyment and motivation were patient-reported. Although patient-reported outcomes may be biased, systematic errors are likely to be comparable in both groups. Therefore, we considered between-group comparisons based on these self-reported outcomes to be valid.
All study participants were allocated two time-slots (2 × 30 min/d) from Monday to Friday, dedicated to self-regulated training, conventional or exergames, for ten working days. This was communicated to patients via a printed weekly therapy program containing the various medical appointments and therapy sessions for each day. Before the intervention started, all patients underwent two instruction sessions with a trained physiotherapist on how to perform the self-regulated training (conventional or exergames). Patients were instructed to repeat self-regulated exercises, conventional or exergames, during the allocated time-slots as many times as possible. In addition, all patients were encouraged to walk and climb stairs instead of using the elevator. This protocol ensured that all participants received the same attention at the beginning of the study and were equally motivated to perform self-regulated training (Fig. 1).
Exercises in both groups (conventional or exergames) aimed to improve balance, strength and mobility, based on the same physiological assumptions about, and physical requirements of, elderly people. Affordance levels of the exergames were therefore comparable to the conventional exercises.
In order to ensure safety during self-regulated training, different levels of exercise difficulty were developed for both the conventional and the exergames programs. The physiotherapist selected appropriate exercises to tailor the exercise program to the patient’s balance skills, as assessed with the Berg Balance Scale (BBS). A BBS score <45 indicates a risk of falling, and thus those patients performed self-exercise in a sitting position only. Patients scoring between 45 and 55 points performed the exercises in a static standing position, whereas patients reaching the maximum score of 56 performed exercises in a dynamic standing position. Furthermore, a therapist assistant was always present during the scheduled exercise sessions to offer assistance if needed and to record adverse events.
Conventional self-regulated exercises
Conventional self-regulated exercises using instruction leaflets are routinely prescribed at Walenstadtberg Rehabilitation Clinic to all suitable inpatients, with the objective of improving balance, strength and mobility. Exercises were performed in sitting, standing or walking, depending on the patient’s balance skills. All patients performed individual exercises, adapted to their balance abilities, within the gymnastics room of the clinic, according to a printed instruction sheet. Most of the time, several patients attended conventional self-regulated exercise sessions at the same time.
The GameUp project  developed seven mini-games for balance, leg strength and flexibility with a user-centered approach, making sure that elderly users could not only perform the exercises, but could also read, see and enjoy the graphics and sounds of the game. Mini-games were combined into three different exercise levels, performed in sitting, standing or walking, thereby allowing affordance adjustments according to the balance abilities of the individual player. The total length of the exercise program was 12 min. Players exercised in a special room and they were asked to repeat the program during the allocated 30 min exercise time. Termination of the program was possible at any time. In case of technical problems, a therapy assistant was available to help.
The primary outcome of this randomized controlled trial was adherence, defined as the duration of daily self-regulated training in minutes. Patients recorded the duration of self-regulated exercise during each session in minutes and the number of sessions per day. We considered using the log data from the Kinect-based game to measure daily training volume in the exergames group. However, these measurements could not be used in the conventional exercise group. Therefore, we decided to use a standardized logbook with self-reported measures in both groups. The feasibility of the logbook was tested in a pilot study comparing conventional balance training with Nintendo Wii balance games in patients with stroke .
Secondary outcomes were motivation and enjoyment for each training day, and objective balance skills of the participants. After each training patients rated motivation and enjoyment on a five-level Likert scale, ranging from 1 = very low to 5 = very high, in the logbook. Objective dynamic balance (local dynamic stability) skills were assessed at pre- and post-intervention with a tri-axial ActiGraph accelerometer, size 4.6 × 3.3 × 1.5 cm, weight 19 g (ActiGraph LLC, Pensacola, FL 32502, USA ). The accelerometer measures trunk acceleration in medio-lateral, vertical and antero-posterior directions. The sampling rate was 100 Hz. The accelerometer was attached to the participants’ lower back at the level of the third lumbar vertebra and they were instructed to walk as fast as safely possible along a 50-m corridor.
The sample size calculation was based on a previous feasibility study  evaluating motivation and time spent in self-regulated exercise using exergames. Assuming a medium effect size (d = 0.5), a statistical power of 0.80, and a type I error risk of 0.05, a sample size of 64 subjects per group would be needed for between-group comparison with two measurement points. Although repeated measures with 10 time-points reduce the required sample size, we did not perform a sample size calculation for repeated measures because these are very unreliable due to the difficulty of estimating covariance between repeated measures a priori.
Descriptive statistics were recorded at baseline for the two groups. Missing outcome data in patients who disliked treatment were substituted by lowest values. Missing data unrelated to treatment were substituted using expectation maximization including residual error. The assumption of normality was tested for the outcomes by visual inspection of histograms with a normality curve. Mixed model analysis was used for repeated-measures to account for the dependency of repeated measures within patients. Models included fixed effects for the interaction between group and time, time and group and random intercepts. A homogeneous autoregressive (order 1) or diagonal covariance structure was used, depending on the model fit evaluated, with -2log likelihood. As the aim of the study was to compare outcomes between groups during the 10-day treatment phase, hypotheses testing focused on the interaction between group and time. Cohen’s d effect sizes were computed using bootstrapped means and SDs of the first and last measurement and were considered small if >0.2, moderate if >0.5 and large if >0.8 . Analyses were performed with a critical value of 0.05 using SPSS Version 23.
The three-dimensional acceleration data collected with the accelerometer ActiGraph® was used to calculate the largest Lyapunov exponent (λS, “divergence exponent”). The λS is a non-linear gait stability index, which has been advocated as an early indicator of the risk of falls . Lower values of λS indicate better stability. The peak in the acceleration signal corresponds to the heel strike in the gait cycle. After graphical inspection of the acceleration signal, 60 consecutive steps were selected for analysis. Data analysis was performed with the package tseriesChaos in R .