We adopted a quasi-experimental pretest-posttest design with a subsequent follow-up to test the feasibility of the exergames in a case series of stroke patients with VSN symptoms. As we aimed to assess implementation of the exergames, thus testing if our intervention can be fully implemented as planned and proposed, an uncontrolled pretest-posttest design is appropriate [57]. A broad variety of definitions exist for the concept of case series in literature [58]. For our study, we used the definition of a case series as being a “report on a series of patients with an outcome of interest” [59]. Recruiting a small convenience sample was ideal for the planned limited efficacy testing, as we aimed to gain intermediate rather than final outcomes in this feasibility project, which allowed us to plan a shorter follow-up period [57].

Identification of potential patients for this project was carried out by staff neuropsychologists and occupational therapists in 2 collaborating rehabilitation clinics (Klinik Bethesda Neurorehabilitation, Parkinson-Zentrum, Epileptologie, Tschugg, Bern, and Zürcher RehaZentrum Wald, Faltigberg-Wald, Zurich). They screened all incoming stroke patients with a diagnosis of VSN for eligibility in this study. We aimed for at least 5 participants, as this amount is considered the minimum reasonable number of independent subjects in a group to combine their data [58]. Fewer than 5 patients are usually presented in a descriptive, narrative form of individual case reports. We strived for a maximum of 10 patients as recommended by Abu-Zidan et al [58]. We included patients with a right brain lesion (RBL) due to a first stroke 15 to 180 days after the cerebral event and a diagnosed VSN as measured by the Catherine Bergego Scale (CBS) [60]. Inclusion criteria were being able to sit in a chair or wheelchair with a backrest for 45 minutes, being at least 18 years old, and having a clear view (with or without vision aids) of a computer screen placed at a distance of 60 to 65 centimeters from patient’s face. VSN patients were excluded if their neglect was diagnosed as due to brain injury other than stroke, if severe apraxia was present—measured as less than 5 points on the Apraxia Screen of TULIA (test for upper limb apraxia) (AST) [61]—or if other noncontrolled medical conditions (eg, chronic pain, drug abuse) were present. Patients with a left brain lesion due to a first stroke were excluded because the exergame difficulty levels were designed to progress from the right (easy) to the left (difficult) side of the computer screen. An option to run the games vice-versa (from left to right) was not available.

All patients signed written informed consent before study entry. Ethical approval for the study was received from the local ethics committees (Zurich No. 2014-0543 and Bern No. 389/2014) as well as from the Swiss agency for the authorisation and supervision of therapeutic products (Swissmedic, 2015-MD-0003). The latter approval was required as the software was not yet certified with the European Community marking for medical devices. The study is registered at ClinicalTrials.gov [NCT02353962].

Patients were seated at a table in front of a 21-inch computer monitor at a distance of 60 to 65 cm in order to provide optimal eye tracking (Figure 1). We chose a seated position to allow more patients to participate (eg, wheelchair users) and avoid exhaustion through standing in an upright position. A height-adjustable chin rest (Novavision GmbH) was mounted on the table to avoid compensatory head movements while playing the exergames. Instead of a mouse to control games, a haptic Falcon Novint device (Novint Technologies) was used. This enabled individuals to experience a realistic sense of touch by providing simulated sensory feedback when reaching for and grasping virtual objects [62]. The Falcon Novint device can be handled with one hand only, allowing stroke patients to play the exergames with their nonaffected hand. The device was placed at the side of the computer monitor at a distance allowing ease of reach for the patients. The nonaffected upper extremity was positioned in approximately 45° shoulder abduction, 70° to 90° elbow flexion, and the forearm fully pronated. All participants were expected to independently complete 15 training sessions while being monitored by a supervising therapist. The supervision included observation of the patient during the intervention giving assistance where appropriate (eg, using the menu to start a new game). Observation was necessary for assistance if potential software difficulties occurred and for safety reasons for the patient, the latter being a regulation of the collaborating clinics.

The intervention program consisted of a series of 9 exergames performed while seated. The games were designed to simulate real-world tasks, such as cooking from a recipe, going for a walk with a dog, or doing a puzzle (for detailed game information on 4 games, see Pirovano et al [51]). The Falcon Novint represented, for example, a dog leash by simulating a pull from the dog to the left or right side of a virtual walking path. The increase in difficulty of all games during the training course was accomplished according to Gentile’s taxonomy of motor skills [56]. Using this systematic classification to design the exergames allowed us to design a theory-based rehabilitation program that followed the principles of exercise training (see Hoffman [53] and Ammann et al [63] for detailed descriptions of these principles). Playing time per game was adjusted from 1 to 10 minutes per game depending on the patient’s ability to concentrate playing the VR game while maintaining a seated position. After initial training and instructions were given by the research team, all subsequent game adjustments during the intervention were performed by clinic staff (occupational therapists and neuropsychologists) in accordance with the patient’s wishes.

The VR-based VSN training intervention took place in the 2 collaborating rehabilitation clinics serving as an additional therapy option to the standard program, which comprised daily occupational, physical, and neuropsychological therapy. Each patient was asked to attend 5 30- to 45-minute sessions per week for 3 weeks. The supervising therapist individually adjusted the intensity of playing the exergames by changing the difficulty level or game duration in the game menu and by deciding if short breaks between each game would be necessary or not. In accordance with the training principle of individuality [53], which states that people respond differently to the same training stimulus, the patient selected 3 to 4 REWIRE VSN exergames from the game menu to be played in each session. The choice was based on personal interest of the patient, which was assumed to enhance motivation while playing. During the 3-week intervention time, patients were allowed to change games if they wanted to test another one or felt bored with the previously played game. After a break of 4 weeks, a follow-up measure was performed aiming to test the training principle of reversibility, which states that the ability to maintain performance is reduced when the training stimulus (the exergames) is removed.

Data analysis was carried out using SPSS for Windows version 23.0 (IBM Corp). A Shapiro-Wilk test was administered and quantile-quantile plots were drawn to test normality of the data. The results confirmed our assumption of nonnormally distributed data due to the small sample size (P ≤.05 for most parameters). We therefore used nonparametric tests for data analysis. Accordingly, a Wilcoxon signed rank test was adopted to compare post- with preintervention results and follow-up with postintervention results. The Friedman test was used to test for differences in the NET and ZüMAX between all 3 measurement time points [71]. We analyzed the data for each individual and for the whole group. The fact that the ZüMAX and NET comprise subtests and the NET additionally provides a conversion table to transform raw scores into standard scores allowed us to use the Wilcoxon signed rank test not only on a group level but also on an individual level. For calculations per patient, we used the achieved standard scores of each subtest as variables, resulting in 17 variables (corresponding to the 17 NET subtests) for the NET and 5 variables (corresponding to the 5 ZüMAX domains) for the ZüMAX. For the analysis on a group level, we compared the achieved total scores per measurement point of the 7 patients.

The ETNT data were provided by the software described in Rabuffetti and colleagues [34]. A subset of the relevant indexes—namely those that were related to visual perception—was used for data analysis as only these items were suitable for the adapted test version. As the ETNT software provided 1 value per index and patient, we used the Wilcoxon signed rank test to analyze post- to preintervention changes within the sample. Additionally, we graphically displayed the individual changes post-pre intervention by drawing the performed search path and fixation points and creating heat maps to visualize group changes post-pre intervention.

In order to perform an a priori power analysis to determine the minimum sample size for a future randomized controlled trial, we calculated the effect sizes for the secondary outcome measures. We applied the Cohen formula for nonparametric tests [72] (see Figure 4). Accordingly, small, medium, and large effect sizes were labeled as r=0.1, 0.3, and 0.5, respectively [73]. The level of significance was set at P ≤.05.

From the 18 VSN patients consecutively screened for eligibility in both clinics from March 2015 to March 2016, 7 patients (39%) were eligible and consented to participate in this study, therefore taking part in the VR exergaming program including baseline, postintervention, and 3-month follow-up measures. Reasons preventing patients from participating were suffering from a right-sided VSN due to a left brain lesion, having a severe apraxia (fewer than 5 points on the TULIA (AST) screening instrument, and being in a poor health condition confining them to bed. Patient characteristics are presented in Table 1.

An overview of individual (P1-P7) and group (interquartile range [IQR], mean) results in the training protocol is shown in Multimedia Appendix 1. There were no adverse events and drop-outs during the intervention. A median attendance of 14 (IQR 12-15) training sessions (maximum 15 sessions) was achieved, which corresponds to a median adherence of 93% (IQR 80%-100%). Reasons for nonparticipation were of organizational or medical nature (eg, overlap with other therapy sessions or due to fatigue) rather than because of motivational factors. All patients played 2 to 4 games and repeated at least 1 game per training session. The supervising therapists adapted and individually progressed the patient training protocols on a weekly basis during the exergames intervention in accordance with patient progress. If, for example, the patient got bored with the current difficulty level of the played game or the therapist observed that the game was played without effort, the therapist modified the difficulty level within each game. However, if the patient had reached the most difficult level, the therapist replaced easy games with more complex ones (ie, games including more distractors or moving objects). An analysis of the progress as measured by the achieved game scores was therefore not feasible, as progression in difficulty resulted in a temporary decrease in game scores. Instead, the progress in the exergames training of the 7 individual patients was documented weekly according to the Gentile’s taxonomy of motor skills. These results are shown in Multimedia Appendix 1.

This study evaluated the feasibility of an exergames intervention aimed to affect VSN symptoms in patients early poststroke in terms of implementation (adherence, attrition, and safety) and limited efficacy testing by documenting changes in VSN symptoms. The exergames intervention was tolerated well by all participants and was mainly performed without major difficulties, showing that its implementation in the clinical setting was feasible. With 0 out of 7 (0%) attrition, no adverse events, and a median adherence rate of 14 out of 15 sessions (93%), the compliance of the patients to the exergames was excellent. Such a result was possibly due to the clinic staff’s commitment, as the VR intervention was smoothly integrated into the daily therapy schedule of the clinic. However, as we aimed to test the fit of our intervention in a real-world setting, we prioritized clinic constraints over optimal conditions and settings. As a consequence, this priority reduced potential omissions of training sessions as described in purely home-based VR interventions [46]. There, the level of use of the VR system was variable and fell far short of recommendations, despite the weekly or biweekly visits of a researcher to the patients’ homes to check progress and retrieve data.

Other studies testing novel VR systems for upper limb stroke rehabilitation have also shown high levels of adherence to the training intervention [74-76]. However, these patients were in the chronic stage of recovery and did not suffer from USN. The single participant with USN of the Duckneglect study (Mainetti et al [41]), who also was in the chronic stage of stroke recovery, showed an excellent adherence to the exergames, in keeping with our case series.

Regarding the exergames training, the median duration per session was 30 (IQR 23-30) minutes, which fell short of the planned 30 to 45 minutes of training time. For our study, we decided to set a timeframe rather than an exact exposure time, because little is known about the optimum duration and patterning of training exposure to virtual environments [77]. Possible reasons for the rather short training sessions in our sample were twofold: either patients (eg, P2 and P3) were quite fit and finished the planned exergames session early or, conversely, were too tired to play longer (eg, P1). In our sample, 5 out of 7 (71%) patients indicated being fit after most trainings (median 13 out of 15 [87%] possible sessions), excluding fatigue as being a main reason for the rather short training time. P7, however, being the only participant with a highly distinctive VSN as measured with the NET, needed on average 34 minutes to accomplish the games. These results highlight the importance of adjustability of the difficulty level of the exergames to the functional ability of the stroke patient [76], thus avoiding a decline in enjoyment level while playing [77]. Given the uncertainty about optimal exposure together with our small and highly varied sample (Table 1, CBS scores), future trials should focus on investigating optimal thresholds of exposure time corresponding to the degree of VSN severity.

The fact that there were no adverse events during the training period in our case series was encouraging. Besides being lucky that no recurrent stroke or other medical emergency happened during the intervention, the design of the games might have contributed as well to the safety of our participants. For example, implementing both stationary and in-motion conditions of the virtual scenario together with the option of choosing between intertrial variability and no intertrial variability while gaming allowed the patient to choose the optimal virtual environment to be challenged on the one hand but not be overwhelmed on the other hand (see Multimedia Appendix 1). Allocating these options in difficulty level might have contributed to reducing the risk of cybersickness while playing despite the stationary condition of the patient [64]. When designing the setup of our study, we intentionally planned a seated position to play the games. This allowed patients to fully concentrate on the exploration of the neglected space without having to invest energy standing in an upright position. Furthermore, playing the games in a seated position contributed to the prevention of falls. Prahm et al [78], for example, also designed a game-based intervention in a seated position reporting no adverse events. However, their participants were able-bodied adults. Wiloth et al [79] reported no adverse events in their game-based assessment to measure motor-cognitive function in people with dementia while they were standing on a movable platform. Although falls are highly prevalent in people with cognitive impairment such as dementia [80], people poststroke additionally suffer from motor impairment. Despite their hemiparesis, our participants were all able to perform the games and handle the Falcon Novint with the nonaffected hand. The clinic staff reported that sometimes the more concentrated or—toward the end of the training session—the more tired the participants became while playing the exergames, the more they tilted to the left side with their upper body. As this is a well-known phenomenon in VSN patients poststroke [81], we think our approach of offering gameplay in a seated position guaranteed patient safety. The rather tilted position, however, did not prevent the participants from continuing to play the exergames.

Our limited efficacy testing showed a group trend of improvements in cognitive and spatial exploration skills. However, these changes cannot be exclusively attributed to the exergames intervention. One reason is the ongoing VSN treatment in the rehabilitation clinic that might also explain some of the improvements. A further possible reason is spontaneous recovery of VSN symptoms not only during the acute phase after stroke but also during the following few weeks. Paolucci et al [82], for example, reported a decrease of VSN symptoms to 20% from 45% after 1 month poststroke, which may also have occurred in our sample. Additionally, the heterogeneity in our sample regarding neglect severity—4 out of 7 (57%) were only mildly affected with CBS scores between 5 to 7 points (Table 1), while 2 out of 7 (29%) were severely affected (16-17 scores on the CBS)—might have influenced the rate of improvements, too. However, our sample did not show a ceiling effect—being present if 15% or more participants achieve the highest possible score [83]—as no participant achieved the highest score in any of the outcome measurements. Furthermore, most patients continued improving their scores in the NET and ZüMAX assessments postintervention, achieving the highest scores after a break of 4 weeks (follow-up). Only P1 in both tests and P2 (NET) and P7 (ZüMAX) showed a decline in scores from post to follow-up, which would correspond to the training principle of reversibility [53,63] that states that once a training stimulus is removed, performance levels will eventually return to or below baseline. Comparing our efficacy testing results with literature was difficult, as studies with a similar setup, time point of measurement, and target group are scarce. There is evidence that training with VR methods improve spatial attention and show transfer of improved spatial attention in activities of daily living in chronic neglect [30,32]. Kim et al [84] showed additional benefit for treating cognitive impairment in stroke patients without VSN in the subacute phase of recovery when adding VR training to classical cognitive rehabilitation. The evidence supports our findings that using VR systems to treat cognition in stroke patients is promising and feasible; however, further research is warranted and necessary to test its use in patients with VSN symptoms early poststroke. Future studies with a focus on treatment effects using controlled research designs should be used to assess causal relationships between the game-based interventions and important patient outcomes.

The median AI values in our sample ranged from 0.08 preintervention to 0.0 at follow-up. Comparing those results to bigger RBL stroke samples with USN—mean (SD) lower AI –0.28 (0.5) for n=34 and mean (SD) higher AI –0.47 (0.5) for n=22; Vossel et al [69] and mean (SD) AI –0.16 (0.38) for n=55; Vossel et al [6]—our 7 patients showed quite a high level of self-awareness for their visuospatial deficits, including P3 and P7 who scored below zero (indicating anosognosia) during all 3 measurements. Of the 7 patients, 3 continued increasing their level of self-awareness postintervention to zero (P1, P2, P4), suggesting positive effects of time poststroke on anosognosia rather than our training intervention. However, time poststroke might not be a viable indicator for those continued improvements in self-awareness of neglect, as Vossel and colleagues [6] found no differences in their AI values across their 3 patient subgroups differing in time since stroke onset. Therefore, continued rehabilitation might be a plausible reason for those further improvements. However, looking at the other 4 patients, AI values showed a tendency toward worsening from preintervention to follow-up, with P3 and P7 overestimating and P5 and P6 underestimating their NET performance, although they also had continued rehabilitation. There is evidence that anosognosia for spatial deficits is not predominant, with different tasks evoking different degrees of awareness about the neglect symptoms [85]. As the AI is calculated on the basis of 4 paper-and-pencil and 2 behavioral tasks of the NET, it might be that this mix of tasks also influenced the miscellaneous AI results. For example, Ronchi et al [85] found that anosognosia level improved after performance of complex visuomotor (eg, cancellation and drawing) and reading tests. By contrast, the self-rating in line bisection tasks was not related to actual task performance [85]. Furthermore, as the AI test is designed to be performed after task execution, it might be that the patients were able to correct their erroneous self-rating to some extent at least in the complex visuomotor tasks. Last but not least, the repetition of the NET test over a relatively short time span might have influenced the AI results, too, as the patients knew that a self-rating would follow after certain tasks.

The ETNT results are to be considered preliminary and should be interpreted with caution. Calibration difficulties with the Eye Tribe Tracker system (eg, most patients only reached 3 out of 5 points in calibration quality scores) may have influenced the reliability of the setup and the accuracy of the results. The calibration consisted of eye-tracking a circle that moved around the whole display. The difficulties experienced by patients in following the rapidly moving circle and the requirement to look at calibration points at the very left of the computer screen were the main reasons for the rather poor calibration results. Such difficulties may produce visible effects (in Figure 5, for example, P4 evidences a dense cloud of gaze points on the bottom right corner during the postintervention assessment, where P4 unsuccessfully tried to tag the 3 targets he could point at with his hand). Baheux et al [86] also reported calibration problems with their 3-D haptic VR system coupled with an eye-tracking device. They assumed that the VSN patient spectacle wear or eye color might have been reasons for calibration difficulties. However, these calibration difficulties notwithstanding, our ETNT results showed trends toward slight improvements in both total located and missed targets on the left side of the screen. The heat maps display the increased ability to detect targets in the upper left portion of the computer screen postintervention but remained substantially unmodified otherwise. However, the preintervention performance was already fairly good in our sample. Furthermore, the heat maps show that the ETNT can identify neglected areas. This is in line with the Rehabilitation Gaming System by Maier et al [31] using the Kinect motion capture system being equally able to measure symptoms of neglect. In contrast to our test, the stroke patients in the chronic stage of recovery explored the neglected side with the paretic arm.

The individual search strategy (Multimedia Appendices 3 and 4) of most patients was comparable with those described by Müri et al [65] and Rabuffetti et al [34], namely to start in the (extreme) right sector and scan leftwards vertically. This contrasts with the search strategy demonstrated by the control subjects in these studies, which started in the upper left sector and scanned top-down horizontally (like reading). Only P3 showed a nonneglect-specific search strategy (Figure 5). The 4 missed targets in the left lower corner postintervention, which P3 was able to detect preintervention, were due to calibration difficulties, as P3 was able to point at those 4 targets with his hand. P1 and P6 were the only patients able to shift their search starting points from the right to the left side from pre- to postintervention, indicating improvements in exploring the neglected side [34]. Not surprisingly, the increased total test duration pre- to postintervention in our sample went along with slightly more detected targets. When the participant increases the number of detected targets, it can be the case that previously neglected targets are now detected albeit with a fairly large latency. We interpret this as positive since more visual space is actively explored but latency and concomitantly the total test duration may therefore also increase. A future study including a larger sample and a control condition is, however, warranted to substantiate or refute these findings.

We performed an a priori power analysis to determine the minimum sample size for such a future trial [87]. Specifically, we assessed the requirements for a randomized controlled study with an experimental group (receiving exergame-based therapy and usual stroke rehabilitation) and a control group (receiving usual stroke rehabilitation only). Assuming an effect size of r=0.9 (based on our observed value for NET total scores, Multimedia Appendix 5), acceptable type I and II error probabilities (0.05 and 0.20, respectively) may be obtained with a minimum sample of 34 subjects per group for a 2-group pre-post-test design. To account for attrition, initial sample size should be increased to 45 subjects per group [88]. Given the fact that only 18 potential participants were available within 12 months of recruitment, we recommend collaborating with more than 2 clinics for such a trial.

The length of the training phase was rather short (ie, 3 weeks). We deliberately did not choose a longer training period, as we primarily wanted to test the exergames’ feasibility and not their effect on VSN symptoms. On the other hand, the rather short training phase allowed us to keep the drop-out risk relatively low (eg, due to discharge home during the training phase). In a next step, it would be important to test the exergame system’s feasibility in patient homes to evaluate adherence, safety, and attrition to using the system in this setting, as the provision of novel home-based rehabilitation options was the main goal of REWIRE. In this setting, a longer training phase could be tested. Furthermore, a progression as measured by the game scores should be implemented together with an immediate graphical feedback after each training session to enhance motivation for playing the exergames. For this implementation, ideas from the rehabilitation method of shaping [54], where frequent feedback and encouragement during training are central, could be adopted. In order to maximize confidence that changes in outcomes can be attributed causally to the exergames intervention, a control group in a pilot randomized controlled trial design would be needed. The neglect exergames should further be designed to switch levels of difficulty (ie, progressing from the right to the left side of the screen or vice-versa). Designing this option would allow recruiting stroke patients with a left-sided brain lesion and VSN symptoms, too. By excluding them in our project we were aware that we would probably miss some patients having ipsilesional neglect [89], which would have made a participation in the exergames intervention feasible. However, as left-sided neglect is quite rare compared to right-sided neglect [5], the risk of missing such an ipsilesional neglect patient was relatively low.

The ETNT could further be developed regarding the following:

Additionally, future work could correlate ETNT measures to scores in standardized clinical scales, such as the NET scores, in order to validate the derived ETNT measures of recovery after VSN.

This study showed that patients adhered well to the REWIRE neglect exergames intervention with no drop-outs, no adverse events, and an adherence rate of 14 out of 15 sessions (93%). We therefore judged this intervention to be safe and feasible for VSN patients early poststroke and appropriate for further testing. Cognitive and spatial exploration skills, as evaluated using ETNT, NET (spatial exploration), and ZüMAX (cognition) assessments, improved in most patients from pre- to postintervention. The results of the amount of exergames use is promising for future applications and warrants further investigations, for example, in the home setting of patients as a motivating training tool to complement usual care and support augmenting training frequency and intensity in RBL stroke patients with VSN.