A convenience sample of community-dwelling patients was recruited through members of the Association of Cerebrovascular Disease Patients. Adults in the chronic phase post stroke (>6 mo) were eligible to participate if they had a stable health condition, could follow simple instructions, and were able to walk with supervision or independently, with or without a walking aid, on flat ground or all surfaces (Functional Ambulation Categories [FAC] 3‐5) [24]. Participants were excluded if they could not play games on the balance boards or had additional neurologic or musculoskeletal impairments that would interfere with performing the required tasks or completing the questionnaires. Participants were block randomized into 4 combinations (Figure 1).

The study complied with the Declaration of Helsinki and was approved by the National Medical Ethics Committee of the Republic of Slovenia (0120-303/2021/3). All participants provided written informed consent to participate in the study. All data were anonymized. Participants did not receive any compensation.

The custom system, classified as a class 1 medical device and developed for rehabilitation, was the Equio (Kinestica, d.o.o.). The noncustom system was the Nintendo Wii with the Wii Balance Board and Wii Fit games. These systems were chosen because they allow comparisons in game mechanics, as both use a balance board to interface with the game. The balance boards of both systems have 4 pressure sensors at the bottom, which calculate the COP position of the user based on body-weight transfers during standing, and both are portable. The custom board has a sampling frequency of 200 Hz [25], while the noncustom board has a sampling frequency between 35 and 100 Hz (average 63 Hz) [26]. The custom board is made of aluminum, is thinner, and has a larger surface area (496 mm × 496 mm × 12 mm; 6635 g) [25], while the noncustom board is made of plastic, is lighter, but thicker (510 mm × 310 mm × 55 mm; 4000 g; Figure 2A). The systems were connected to a TV screen 220 cm away from the participant standing on the balance boards.

Comparable self-paced and game-paced games from each system were selected based on the required movements of the center of COP direction (see Figure 2B and C). For example, the tilt table game, present in both systems, required the COP to be moved in all directions. For the noncustom games, the basic difficulty level was used for all participants, as recommended by the developer for initial gameplay. The custom games have 3 difficulty levels. Their difficulty was adjusted for accuracy requirements (eg, size of obstacles or targets), number of targets, and the addition of a dual task. Game-paced games were played at the speed required by the game, while self-paced games were played at the participants’ preferred speed. Details of custom and noncustom games are described in Multimedia Appendix 1.

Participants attended a familiarization session and 2 gameplay sessions, approximately 1 week apart, in the Physiotherapy Laboratory of the university (see Figure 3). Familiarization consisted of playing each of the 12 games for 2 minutes. To set the range of motion needed during the custom games, the COP range of motion in the mediolateral and anteroposterior directions was measured for each participant using the custom system. These values were automatically translated to the custom game scenarios and adjusted based on accuracy at the start of gameplay in all sessions. The difficulty level of each custom game was determined based on the capabilities of participants during the familiarization session.

Gameplay was studied in 2 sessions. To ensure movement quality and safety, during the first minute of each game, a physiotherapist guided movement with either verbal or manual cues and positioned the participant’s feet. Both the order of game types (self-paced and game-paced) and the order of the systems (custom and noncustom) were randomized (see Figure 1). In each gameplay session, participants played 1 block consisting of 3 games (5 min each) on the noncustom balance board and 3 comparable games on the custom balance board. Each gameplay session consisted of 30 minutes of training with a 10-minute rest in the middle. During the rest periods, the assessment of the user’s experience was performed in a sitting position (Figure 3).

Participants’ balance and gait were measured during the familiarization session. Balance was assessed using the Mini Balance Evaluation System Test [27,28]. Walking capacity was rated with the FAC, and comfortable and fast walking speeds were measured with the 10- meter walk test. The latter was conducted on a 14-meter walking path [28], with 1 familiarization trial and 1 test trial performed at a comfortable speed, followed by a fast speed walking trial.

The primary outcome measures were movement frequency (number of weight-shift repetitions to the affected lower limb) and movement amplitude of pelvic displacement to the affected side. Movement performance during gameplay was recorded using a digital video camera (Canon PowerShot SX620 HS; Figure 4, first), recording at 30 frames per second. A marker used to monitor pelvic movement was placed on the sacrum at the height of the center of body mass (S2). The camera was positioned behind the participant to clearly view the participant’s pelvis and torso, the marker, the balance board, the position of the feet, and the TV screen. The TV screen, balance boards, and camera were always positioned in the same place; however, the camera was moved back from this position if the participant’s head was not visible in the video. The first minute of the recordings was excluded from the analysis because participants had to adjust to the game at the beginning (described above).

For the 8 games that required movement of the COP in either the mediolateral direction or all directions, the number of weight shifts to the affected lower limb (repetitions/min) and the amplitude of pelvic displacement to the affected side in the frontal plane were extracted from video recordings using the Kinovea software (version 0.9.1) [29]. For each game, the last 4 minutes of gameplay were analyzed. Any voluntary movement of the pelvis to the affected side in the mediolateral direction, from the starting point to the farthest point and back to the starting position, was considered a correct repetition. The amplitude of pelvic displacement was determined by measuring the maximum displacement of the marker from the start to the end position (Figure 5). The analysis was performed for games in which the markers were not concealed in ≥50% of the recording (eg, due to compensatory rotation of the pelvis or while the physiotherapist provided additional guidance or safety).

The secondary outcome measures assessed user experience: the Borg Rating of Perceived Exertion Scale (RPE) [30], the Modified Physical Activity Enjoyment Scale (mPACES) [31], and the Flow State Scale for Occupational Tasks (FSSOT) [32]. These measures were collected twice in each study block, immediately after playing the games. The Borg Scale (ranging from 6 to 20) was used to estimate the cardiovascular intensity of physical activity through RPE. The mPACES is a 5-item scale, with scores expressed as percentages, that rates enjoyment during gameplay and has been used previously in VR studies [23,31]. The FSSOT is a 14-item questionnaire assessing flow state, with scores expressed as a total or in 3 domains: sense of control, emotional experience, and absorption [32]. At the end of each gameplay session, participants viewed a figure displaying the 6 games and ranked them from most liked (6 points) to least liked (1 point).

Data were assessed for normality using the Shapiro-Wilk test. Paired t tests with CIs were used to determine differences between custom and noncustom games for the number of repetitions and amplitude of pelvic displacement within the game pairs. Missing data were excluded from the analysis. Paired t tests or Wilcoxon signed rank tests were conducted to test for differences in mPACES, FSSOT, and RPE scores between games of custom and noncustom systems within self-paced and game-paced games. Cohen d was calculated to estimate the ES (0.2 small; 0.5 medium; >0.8 large). Cohen d and CIs for nonparametric data were estimated using the Bootstrap method. To correct for multiple comparisons, the α was adjusted to .025 within pairs. To identify differences in likability within self-paced and game-paced games, the total score of each game was calculated for all participants and then ranked within each block from most liked (highest total score) to least liked (lowest total score). IBM SPSS Statistics version 29.0 was used for the analysis.

Twenty-six participants completed the study. Two individuals were excluded from the original 28 because they were unable to play the games on the balance board. There were no adverse events (eg, falls) or side effects (eg, motion sickness). Some movement performance data were missing due to concealed pelvic marker.

Participants’ ages ranged from 27 to 70 years. The time since experiencing a stroke ranged from 0.7 to 37.8 years. Most participants were able to walk independently on both level and nonlevel surfaces, stairs, and inclines (FAC 5; Table 1). Participants’ balance ranged from moderate (8‐13 points; n=4) to mild balance deficits (15‐23 points; n=14) and very mild balance deficits to normal balance (24‐27 points; n=8) [33].

Participants demonstrated better movement performance, including higher movement frequency (more weight-shift repetitions) and greater pelvic displacement to the affected side during the game-paced noncustom games. Movement performance between custom and noncustom games was more similar during self-paced games. User experience, specifically flow state and preference, favored the game-paced custom games.

As hypothesized, movement frequency based on the number of weight shifts to the affected lower limb during custom games was on average 21% to 61% lower than during noncustom games, with large ESs (ES 0.8‐2.5). These findings are consistent with our previous study [23], where the total number of step repetitions was significantly lower for the custom game compared to the noncustom Kinect game (average difference 33%). The custom system allowed calibration for movement displacement within the user’s limits of support as well as selection of movement speed (game difficulty level). In contrast, noncustom games lack this customization, requiring players to move faster and farther to meet the game mechanics. This could, however, result in poor movement control, as shown in our previous work [23].

Contrary to our hypothesis, the amplitude of pelvic displacement toward the affected side was on average 13% to 21% greater for the noncustom games than for the custom games, with medium-to-large ESs (ES 0.6‐0.9), except for one case (self-paced games: Penguin Slide vs Lights), where there was no difference. This may be due to a conservative range of motion settings for custom games, which could only be adjusted in 5 centimeter increments. The movement boundary set by customization likely limited potential displacements. However, without a 3D motion analysis, no definitive statement can be made about movement quality or lack of control in relation to movement amplitude.

As hypothesized, user experience, specifically flow state, was better for the custom game-paced games than for the noncustom equivalent games, and there was no difference between custom and noncustom self-paced games. These findings support the claim that the lack of user control is especially evident in game-paced, rather than self-paced, games [23]. An important finding is that the flow state (FSSOT) data deepen the understanding of the user experience during custom game-paced games, in which subscores for sense of control and emotional experience were significantly higher than for noncustom games. This suggests that, regardless of the serious game system used, the ability of the user to control game speed is an important feature.

The findings of flow-state–favoring, custom game–paced games in this study were supported by a medium ES (ES 0.5) in enjoyment. The construct of enjoyment has been reported to be related to engagement for training and motor learning in virtual environments and likely promotes motivation [12]. VR-game balance training, specifically, was reported to be more enjoyable than conventional weight-shift training [34]. Therefore, it is important to consider the type of game pacing as a factor for the usability of serious games for patients with neurological conditions. Game-paced games use speed to determine the level of difficulty, which, when not adjusted, can lead to frustration and affect game enjoyment and motivation [19].

There was a small but statistically significant difference in perceived exertion, with lower RPE scores for the custom games, indicating a medium ES (ES 0.6). Participants perceived the game-paced noncustom games as somewhat more exerting (a mean difference of 1.19 points) than the custom games. This confirms previous findings on perceived exertion [23] and is consistent with the greater movement frequency and pelvic displacement observed in this study. However, perceived effort was quite similar across the games, and the difference was not clinically meaningful. The range of RPE scores for the game-paced noncustom games was wide, from 7 (very, very light) to 17 (very hard), and similar for the custom games, from 7 to 15 (hard) [30]. Because VR balance games are not expected to influence cardiovascular capacity [35], we propose that other factors, such as the difficulty of the task for each individual, influenced the level of exertion experienced by participants. None of the participants needed or requested an extension of the 10-minute rest period during the study protocol.

Clinical reasoning for selecting patients after a stroke for whom serious games are meaningful is challenging, as the patient population is heterogeneous [5]. While we did not analyze subgroups in this study due to sample size limitations, we note that participants’ abilities influenced gameplay. Participants in our study with moderate balance deficits (Mini-BESTest [Mini Balance Evaluation Systems Test] 8‐13 points; n=4) frequently experienced difficulties while playing games on both balance boards, particularly in placing their feet correctly due to muscle shortening, foot deformities, and increased muscle tone. For participants with very mild balance deficits to normal balance [33] (Mini-BESTest 24‐27 points; n=8), the custom system games were often too easy or monotonous. Therefore, we speculated that for highly functioning patients after a stroke, noncustom games are likely more suitable. However, this is strictly an observation that requires further investigation with a larger sample size.

Taken together, the findings suggest that both custom and noncustom serious games have a role in balance training for persons after a stroke. Noncustom games promote greater movement amplitude and frequency toward the stroke-affected side. Custom games may promote greater adherence based on positive flow, enjoyment, and likability findings. There has been a tendency to dismiss noncustom games, particularly those that are game-paced. However, this study suggests they have a role in stimulating exercise intensity with movement toward the affected side. Game designers may also need to guard against constraining movement displacement.

While the findings of this study generalize to patients in the early-to-late chronic phase after a stroke with moderate-to-very mild balance deficits or normal balance (Mini-BESTest 8–27/28 points), the sample size did not allow for the interpretation of the effect of balance deficit severity on gameplay movement performance and user experience. Future studies may consider the relationship between balance ability and performance on serious games with a strong balance requirement. Movement performance should also be analyzed for games that encourage COP movement in the anteroposterior direction.

The findings of this study generalize to games that use the COP as the marker of movement. These game mechanics are shared by many other games that use some form of platform to track movement. However, it is unclear whether the findings would be similar in games that use total body tracking. Differences between the 2 VR systems used, such as sampling frequency, board dimensions, surface area, and material properties, could influence COP sensitivity and thereby affect gameplay mechanics independently of game customization. These factors were not isolated in this study, and their potential contribution cannot be fully excluded. It was not possible to perfectly match all the custom and noncustom games.

The analysis of video recordings was challenging, even though Kinovea software is a valid and reliable tool for measurements within an angle range of 90˚ to 45˚ [36], with accuracy for measuring linear positions and velocities from 30 Hz video within 9% [37]. The use of 3D motion analysis systems would increase the validity of the results. Another limitation of this study is the reduced number of participants available for movement analysis due to the concealed pelvic marker. We encountered several challenges due to compensatory pelvis rotations, which caused the marker to fall outside the appropriate plane or resulted in movements too small for measurement. Additionally, the marker was sometimes covered by the physiotherapist while providing physical guidance or ensuring safety. Consequently, for some recordings—particularly those of more demanding, faster games—we did not perform the analysis due to missing data.

The findings of this study indicate that both customization and type of game pacing influence movement frequency, movement amplitude, enjoyment, flow state, and perceived exertion in VR balance games after a stroke. To our knowledge, this is the first study to parse the effects of game customization and type of game pacing on balance performance and the experience of patients after a stroke. Movement frequency and movement amplitude were significantly greater for all noncustom game–paced games and for half of the noncustom self-paced games. This is important to consider because weight-shifting toward the affected side is a common problem for patients after a stroke. Users’ flow state (including the sense of control and emotional experience subdomains) and likability (user preference) were significantly greater for custom game–paced games. There was no significant difference between custom and noncustom games for enjoyment. Therefore, it is important to consider both game customization (primarily for user experience, with custom games preferred) and type of game pacing (primarily for movement performance) when selecting serious games as interventions to improve balance for patients after a stroke. These findings contribute to the body of knowledge that guides clinicians’ clinical reasoning in selecting serious games for rehabilitation.