There is growing interest in interventions that aim to mitigate the physical and cognitive decline associated with aging in older adults [1]. Applying gamified digital health technologies with real-time user behavior and gameplay interactions such as computerized cognitive training games or exergame technology is becoming increasingly popular to address the challenges of physical and cognitive decline in the aging population [2-7]. Within this field, considerable attention has been directed toward exergame-based physical and motor-cognitive training, which is considered more promising than conventional physical, cognitive, or motor-cognitive training to improve cognitive and physical performance in diverse user populations [8-10]. Exergames are defined as “technology-driven physical activities, such as video game play, that require participants to be physically active or exercise in order to play the game” [11]. Exergame-based training typically focuses on incorporating cognitive tasks into motor tasks than can be designed to specifically train a variety of physical (eg, cardiorespiratory, motor, balance, strength, or multicomponent exercises) and cognitive (eg, learning and memory, complex attention, executive function, and visuospatial skills) domains [5]. The positive impact of exergaming on physical and cognitive functions has been observed in healthy older adults [5,8,12-14], as well as in older adults with cognitive impairments such as mild neurocognitive disorder (mNCD) or more severe forms such as dementia [5,15-17].

mNCD describes a critical stage of cognitive decline that goes beyond normal aging without affecting the capacity for independence in everyday activities [18]. Early interventions may effectively influence the course of cognitive decline, making those with mNCD an optimal target population for secondary prevention [19-22]. Several pharmacological and nonpharmacological methods have been proposed to slow down the progression of symptoms accompanying mNCD and improve overall quality of life [20,23]. According to current research, the most effective type of exercise to achieve this goal is to implement physical training, ideally with integrated cognitive tasks (simultaneous motor-cognitive training) [21,24-30]. This type of training has been proposed to operate through various disease-modifying mechanisms, including multisystem effects that reduce mNCD-related neuropathological damage [31,32] and promote synaptogenesis, neurogenesis, and angiogenesis [33,34]. It also helps maintain or increase cognitive reserve [26,35,36], empowering individuals to retain independence in everyday functioning despite neurodegenerative changes [33,34].

Physical and motor-cognitive training may represent the predominant nonpharmacological intervention that effectively mitigates cognitive decline in mNCD [27,30,37,38] and is recommended for secondary prevention of mNCD by a collaborative international guideline [39] as well as a global consensus on optimal exercise recommendations for enhancing healthy longevity in older adults [22]. In this context, exergames provide a promising solution to deliver this type of training [30]. Recent systematic reviews support the use of exergame-based training for improving both physical and cognitive performance in individuals with mNCD. However, studies in this field exhibit high heterogeneity, which weakens the strength of the evidence [15,16,40-42]. Furthermore, recent research suggests that, while exergame-based training may indeed improve cognitive performance, its superiority over traditional training remains uncertain, and further research should investigate the conditions in which exergames are most effective [12].

Several exergame systems have been investigated for older adults and have been shown to effectively promote engagement and motivation [43-46], which largely explains the superior adherence to exergame- or technology-enhanced training compared to adherence to conventional approaches of physical and motor-cognitive activities, exercise, or training [46,47]. Furthermore, exergames offer high value due to the technical adaptability of various elements. However, while many exergames are technically well designed for attractiveness and entertainment, they often lack systematic user adaptability and specificity [48]. Therefore, future exergames should be better tailored to specific target groups as well as individual goals and abilities [5,8,48,49].

To enhance the user adaptability of exergames, it has been recommended to implement real-time adaptive exergame mechanisms that support the dual-flow concept proposed by Sinclair et al [50]. According to this concept, an optimal exergame experience requires a balance between the challenge of the game and the player’s skills, as well as between the intensity of the game and the player’s fitness. A common method for implementing adaptive exergame mechanics is to use the data that are directly collected during exergaming [49]. These data are commonly referred to as in-game metrics or performance measures and capture the real-time interaction between the user and the serious game. Game metrics are quantitative measures that can range from simple measures such as point scores or reaction time to more complex measures that combine various parameters [51]. Game metrics have the potential to improve personalization and individualized progression of training according to users’ performance, provide (real-time) feedback, and monitor gameplay [51-53]. However, there are currently no standards or guidelines for what data should be collected and for what specific purpose [52,54].

By investigating game metrics and understanding how well they can capture true cognitive and motor-cognitive performance, we can provide a robust basis for in-game adjustments and monitoring of performance progression [54,55]. While many studies emphasize the importance of individually tailored training and commonly use in-game metrics for this purpose, only a limited number have analyzed the psychometric properties of in-game metrics. Previous studies that have investigated game metrics from exergames for older adults with and without physical or cognitive impairments have found moderate to strong correlations between some game metrics and clinical assessments [54,56,57]. This suggests that certain game metrics may be useful for measuring physical and cognitive performance. Furthermore, one study demonstrated that game metrics can distinguish between healthy older adults and those with mNCD [56]. In addition, another study explored a difference in exergame performance progression between older adults with and without cognitive impairment [57]. However, further studies are needed with domain-specific evaluation, larger participant samples, and inclusion of participants with a clinical diagnosis of mNCD, as stated by Guimarães et al [57].

The primary objective was to identify valid game metrics as indicators for in-game domain-specific cognitive performance during exergaming in older adults with mNCD. We hypothesized statistically significant correlations between specific game metrics and reference assessments within their corresponding neurocognitive subdomains. As a secondary objective, we aimed to explore game metric performance change over time during a 12-week exergame-based training intervention for older adults with mNCD.

This study is a secondary analysis of data from 2 randomized controlled trials (RCTs) that aimed to evaluate the feasibility (feasibility study [FS]; NCT04996654) [58] and effectiveness (effectiveness study [ES]; NCT0538707) [59] of an individually tailored home-based exergame-based training concept specifically for the secondary prevention of mNCD developed within the Brain-IT project [60]. The training concept is rooted in years of iterative and user-centered co-design, development, testing, and evaluation with continuous patient and public involvement [61]. It serves as a guideline for the implementation of the training by providing algorithmic decision trees for the structure, content, and individualized tailoring of the training and can be implemented using different commercially available hardware [59].

The project was structured in 3 phases. In phase 1, we combined a comprehensive literature synthesis [61] with qualitative research, including primary end users (individuals with mNCD), secondary end users (multidisciplinary health and care professionals), other exergaming researchers, and experts from the exergaming industry [62], to elaborate a set of design requirements for the Brain-IT training concept. In phase 2, possible concepts were co-designed and developed based on the requirements defined in phase 1. The first prototype of the Brain-IT training concept [59,61] then entered the iterative cycle of feasibility, usability, safety, and acceptance testing and integrating of findings for further development until an “acceptable” solution was achieved (refer to the FS [58], the data from which are analyzed in this study). This process resulted in a novel intervention type specifically targeting various relevant mechanisms of action to alleviate the pathological state of mNCD by combining, for the first time, exergame-based motor-cognitive training with biofeedback-guided resonance breathing as an adjunct neuromodulatory intervention [59]. In phase 3, we confirmed the effectiveness (refer to the ES [59], the data from which are analyzed in this study) of the Brain-IT training in improving global cognitive performance as well as immediate and delayed verbal recall. We observed that the training not only could effectively slow down cognitive decline in comparison to usual care but also resulted in 55% of participants showing clinically relevant improvements in cognitive performance [59].

We compiled this manuscript according to the latest version of the CONSORT (Consolidated Standards of Reporting Trials) Statement for Randomized Trials of Nonpharmacologic Treatments (Tables S1 in Multimedia Appendix 1). As this is a secondary analysis of 2 RCTs, not all original trial design elements may be fully applicable or reported in this paper as they were not relevant for the specific analysis conducted (for full reproducibility, please also refer to the published FS [58], the study protocol for the ES [63], the ES itself [59], and the training concept—Supplementary File 2 of the ES [59]).

All original study procedures were carried out in accordance with the Declaration of Helsinki. The original study protocols were approved by the ETH Zurich Ethics Commission (EK 2021-N-79) for the FS and the Ethics Committees of Zurich and Eastern Switzerland (EK-2022-00386) for the ES.

All study participants were fully informed about the study procedures in person (at the interested person’s home or at one of the study centers depending on their preference) through verbal explanations and an information sheet. After sufficient time for consideration (ie, at least 24 hours after handing out the study information sheet [which was approved by the corresponding ethics committee] but, on average, approximately 1 week), suitable patients willing to participate in the study provided written informed consent in a second in-person meeting with one of the trained investigators of the study team at the home of the interested person or at one of the study centers. The original informed consent forms allowed for the secondary analyses conducted in this study as all study participants provided consent for (1) original datasets being made available in a publicly accessible repository in deidentified form and (2) the transfer of encrypted data for research purposes. Only deidentified data were analyzed for this study (for more detail, see the Data Availability section).

No compensation was granted to participants, but detailed feedback on individual performance as well as the study outcomes in general was provided at the end of the studies.

Recruitment for the 2 parallel-group and single-blinded (the outcome evaluator of pre- and postintervention measurements was blinded to group allocation) RCTs took place between July 2021 and October 2023 in collaboration with health care facilities in the larger area of Zurich, Switzerland. The eligibility criteria for study participants have been published previously [58,59,63] and are listed in Textbox 1 [64-68].

After successful recruitment, participants underwent a premeasurement (see the Reference Assessments subsection in the Outcomes section) at the study center located at ETH Hönggerberg and were randomly assigned in a 1:2 (FS) or 1:1 (ES) ratio to either the control or intervention group. This secondary analysis investigated only the data from participants allocated to the intervention groups who engaged with the exergame-based training.

There were no important changes to the trial design and study setting after commencement that were relevant to the analyses presented in this paper. More details can be found in the published FS [58] and ES [59].

Following the premeasurement and group allocation, a study investigator installed the exergame system in the participants’ homes, where the training sessions took place. A standardized familiarization session was conducted to introduce the exergame system (ie, 4 games from the first scheduled training session [at level 1] were introduced and played for 2 minutes each). The exergame system Senso Flex (Dividat AG; hardware: prototype version 2; software: version 22.4.0-360-gf9df00d5b) was used [69]. The participants were instructed to follow the Brain-IT training concept [59] for the 12-week intervention, which consisted of completing an instructed minimum of 5 training sessions per week. Each training session had a duration of 24 minutes and included 6 motor-cognitive games and 3 resonance breathing exercises. Approximately one-third of the sessions were supervised by a study investigator, with more supervision provided at the beginning of the intervention and less toward the end [59]. The intervention was personalized and individually adapted based on several predefined parameters and included games for the neurocognitive domains of learning and memory, executive function, complex attention, and visuospatial skills [59,70]. The relevant parameters for this study were (1) the focus on participants’ main cognitive deficits through an individual constellation of games and (2) the individual adaptation of the standardized difficulty levels (levels 1-10), with each participant starting at level 1 [59].

The parameters for determining the 10 game levels were defined through agreement between an experienced neuropsychologist and the research team and were extensively tested [71]. According to the Brain-IT training concept, level 1 should refer to an introductory level (ie, even most impaired patients with mNCD should be able to play the game at the first trial), and in level 10 (ie, “healthy functioning” level), the game demands are expected to be challenging but doable for an average healthy older adult. The remaining levels are defined to increase neurocognitive demands consecutively and regularly from level to level. In the definition of these 10 game levels, we integrated the methodology by Huber et al [72], who established an extended taxonomy supporting motor-cognitive learning. This methodology provides a framework for structuring training to specifically integrate models of skill acquisition [73,74] that has also been described to apply to cognitive skill learning and relearning [75].

For more details on the specific exergames, as well as all algorithmic decisions for the personalized design and individualized progression of the training, including a description of all games and game level settings, we kindly refer interested readers to the published Brain-IT training concept [59]. To ensure replicability, the Brain-IT training concept was planned and reported according to the Consensus on Exercise Reporting Template [76]. The training concept provides specific instructions on how to adapt the training when implemented with alternative hardware and software solutions (Supplementary File 2 of the ES [59]).

A summary of the participant flow is provided in Figure 1. Data were available for a total of 31 participants who were allocated to the intervention group in the 2 RCTs analyzed. The time between the initial premeasurement with the group allocation and the start of the intervention ranged from 1 to 2 weeks.

The demographic and clinical characteristics of all participants are listed in Table 2.

Table 3 presents the results of the correlation analyses between game metrics and neurocognitive reference assessments. In the neurocognitive domain of learning and memory, 3 metrics of the game Shopping Tour (ie, mean reaction time, precision score, and collected items) and 1 metric of the game Simon (ie, point rate) met the alternative hypothesis criteria by exhibiting statistically significant moderate to strong correlations with the corresponding reference assessments (see Table 3 for all P values). In the neurocognitive domains of executive function and visuospatial skills, the metric point rate of the games Targets and Gears met the alternative hypothesis criteria by exhibiting statistically significant moderate correlations with the corresponding reference assessments. None of the metrics of the games targeting the neurocognitive domain of complex attention or of the remaining 2 games, Nomis and Tetris, met the alternative hypothesis criteria. Descriptive statistics for all outcome variables are provided in Tables S2 and S3 in Multimedia Appendix 1.

This analysis included the 6 game metrics that met the criteria for the primary objective. Table 4 presents the results for the number of times that the participants played each game. Figure 2 shows the performance progression curves over gameplay time in percentages. For all game metrics, there was an initial increase in average performance progression, or a decrease in the case of the metric “mean reaction time.” This was followed by a plateau (see Simon, Targets, and the mean reaction time metric of Shopping Tour) or even a slight decline (see Gears and the precision score and collected items metrics of Shopping Tour) in performance. Interindividual differences in performance progression were observed, particularly for the metric “point rate.” While some individuals exhibited a flatter curve, others had a more pronounced initial increase or a sigmoidal curve pattern. Figure 3 shows the performance progression within difficulty levels. In addition to the findings shown in Figure 2, we observed that, in the game Shopping Tour, the metrics “precision score” and “collected items” increased within most levels, whereas the metric of mean reaction time initially declined and then leveled off. For the metric of point rate in the games Gears and Simon, we observed considerable drops in performance between certain levels of difficulty.

One of the main strengths of this study was that we conducted the correlation analyses between game metrics and clinical assessments only within specific preassigned neurocognitive domains or subdomains. This approach was enabled by the already available content validation of the games to their primary trained neurocognitive domain or subdomain within the Brain-IT training concept [61]. Therefore, the primary objective was hypothesis driven, with predefined criteria for interpretation of the results. In addition, we used data from clinically validated and well-established neuropsychological assessments across all neurocognitive domains.

There are also some limitations that need to be discussed. First and most importantly, the sample size was small, and we did not conduct an a priori sample size calculation as we analyzed existing datasets from studies conducted as part of the Brain-IT project. To ensure appropriate interpretation of our results, we conducted a post hoc power analysis using G*Power (version 3.1) [90,91], which revealed that most of our analyses were underpowered. This might have resulted in potential false-negative or false-positive results, affects the robustness and generalizability of our findings, and warrants confirmatory studies with a priori sample size calculations to ensure more reliable and robust conclusions. Considering our criteria to confirm the alternative hypotheses as assumptions for a power analysis, a sample size of 37 or 50 would be required to show (1) a statistically significant correlation (P≤.05; uncorrected; 1-sided) with (2) a correlation coefficient of ρ≥0.4 with sufficient power of β>.80 or β>.90, respectively, which may be used as a reference for planning future studies on this topic.

Second, this study was a secondary analysis, and the original studies as well as the intervention were not originally designed a priori to investigate the objectives of this study, which restricted our methodological choices in this secondary analysis. Although the game sessions were standardized in terms of game design and intervention delivery, there were interindividual differences in gameplay due to personalization and individual progression of the intervention.

Specifically, these restrictions limited the following methodological aspects for the primary research question. According to the Brain-IT training concept, the games were individually assigned considering the participants’ baseline neuropsychological performance, with all participants starting at level 1 of the game demands in the first training session. Thereafter, either the games progressed or more difficult games were introduced as soon as (1) a plateau in performance was reached on predetermined game metrics, (2) a predetermined target score was reached on a predetermined game metric for progression to the next level, (3) the participant requested an increase in task demands, or (4) the staff supervising the participants deemed the increase in task demands feasible [59]. Given these rules for individualized progression of exergame demands, the path through these games and game levels was different for each participant, with large differences in the time to progress to the next levels and the maximum level reached after the 12-week intervention (as shown in Figure 3). For example, one participant might reach level 5 after only 2 weeks of training, whereas another might not reach this level until after 12 weeks. As the first research objective was tied to baseline cognitive performance as measured using standardized neuropsychological assessments, only the first level provided the same standardized conditions for all study participants that allowed the cross-sectional analyses to address this research objective. This methodological limitation may have resulted in participants with higher baseline functioning being underchallenged. However, we did not observe any floor or ceiling effects in any of the game metrics except for number of misses in the game Targets and point score in the game Tetris (Figures S1-S8 in Multimedia Appendix 1), whereas there was a good amount of variability in all game metrics (Table S3 in Multimedia Appendix 1). Therefore, it can be assumed that this limitation did not have a significant impact on the observed results. Nevertheless, analysis of higher levels of difficulty would undoubtedly provide additional insights. As an example, the findings of Litz et al [94] observed a tendency toward more consistent relationships between in-game metrics and standardized cognitive assessments at higher difficulty levels or when averaging scores over all game levels [94]. However, given the individual paths through the game levels and the corresponding time differences to reach a given level in this study, it would be impossible to disentangle learning effects (for participants who had already played the game several times) and performance improvements over time during the intervention from “true” cognitive performance as measured at baseline. Furthermore, not all participants played the same games, resulting in smaller sample sizes in some games. Games that were considered more difficult, such as Nomis and Tetris, had small sample sizes as they were only played by participants with a higher baseline cognitive performance in the neurocognitive domains of executive functioning and visuospatial skills, which informed the allocation of the games in the first session. This could have biased the analysis. Therefore, future studies should be designed to specifically address the research question of this study rather than relying on existing datasets that might restrict the methodological choices in the analyses of the data. This entails a cross-sectional design with a priori sample size calculations rather than the use of data from longitudinal studies and applying different game levels to investigate whether the game levels induce the appropriate changes in perceived task demands (eg, as has been investigated in the work by Manser and de Bruin [71]) and whether and how performance at higher difficulty levels might differentially relate to cognitive performance assessed using standardized neuropsychological assessments.

The personalization and individualized progression of the training also presented a challenge for the analysis of our secondary research question (ie, performance progression) as players varied in the number of times they played the games and difficulty levels were individually adapted. To account for these interindividual differences, we found a method to present performance progression across gameplay by normalizing time to game completion as a percentage. Using this method, we were able to account for variability in adherence and difficulty levels at a group level. However, this method may result in loss of some information, and individual curves were compressed or stretched. These methodological constraints also prevented us from conducting more in-depth analyses of the observed high interindividual variability in performance progression (see Figures 2 and 3 with color coding of individual participant progression curves), such as exploratory analyses to identify potential factors contributing to the variability (eg, age, years of education, and baseline cognitive status) as these factors (particularly baseline cognitive status) influenced individualized tailoring (personalized allocation as well as individualized progression of games and game levels). Therefore, future studies should be specifically designed to account for these aspects and allow for more in-depth analysis to disentangle the factors that influence the high variability in performance progression over time and derive evidence-based recommendations on the implications of accounting for this variability in the methodological aspects of individually tailoring exergame-based training.

Third, although each exergame focused on 1 specific neurocognitive domain or subdomain, they often indirectly trained other domains, which may have confounded the domain-specific analysis. Fourth, the absence of data from specific motor assessments precluded their inclusion in our analyses. Finally, this study only included older adults with mNCD, and the results cannot be generalized to other populations.

The personalization of training is becoming increasingly popular in the current development of technology-supported interventions, including exergames. Within this field, research should aim to identify valid parameters that can be used for in-game adaptation. The potential of game metrics should be further investigated through studies with structured interventions while also considering the impact of game task design as well as personal and external factors. Given the proposal by Netz [106] that (adherence to) optimal exercise intensity and motor-cognitive task complexity are driving mechanisms for global and task-specific neuroplasticity, respectively, and as discussed in previous research, personalized training adaptations should go beyond game metrics and consider additional aspects of players related to physiological states (e.g., attention, stress) or personal characteristics (e.g., learning style, intelligence) to provide an optimal exercise intensity and motor-cognitive task complexity [49]. Different biofeedback systems, such as biocybernetic adaptation loops, which continuously adjust game difficulty or game content in response to real-time physiological data from the user [107-109], are currently being investigated in this area and should be further explored. Furthermore, previous reviews have identified several features of fully immersive virtual reality (VR) that can support skill learning [110], whereas fully immersive VR or augmented reality has a fundamental advantage in delivering ecologically valid game scenarios via recreating a certain level of naturalistic sensory-motor interaction between the user and graphical user interface [111]. Such features include but are not limited to multisensory interactivity with haptic feedback in a high-fidelity VR with 3D representations and the presence of avatars [110]. Given that specially designed and customized VR interventions have been more effective in neurorehabilitation compared to the use of commercially available VR systems [112], these features should be more thoroughly investigated regarding their capacity to improve tailoring of the exercises to provide optimal exercise intensity and motor-cognitive task complexity.

The variability in performance progression across participants presents further aspects that need to be investigated. Research should try to identify factors that can help understand and identify how individuals differ in their cognitive and motor-cognitive skill acquisition and performance progression. Different factors that may contribute to a steeper progression curve should be investigated. This should include subgroup analyses, which we suggest should include assessing (1) the baseline physical and cognitive state of the individuals [57,103,105], (2) age [104,105], (3) adherence, and (4) self-efficacy [95]. Moreover, game metrics obtained in the “game module” only provide information on physical and cognitive performance in a specific task (game). However, solely analyzing game metrics does not allow for any conclusions on the efficacy or effectiveness of the training as the habituation and learning effects cannot be disentangled from “true” physical and motor-cognitive performance improvements. Such “true” performance adaptations would either be assessed using standardized physical, motor, or neuropsychological assessments or obtained in a separate evaluation module that integrates scientifically validated gamified assessments to regularly verify whether the exercise or training stimulus is sufficient to induce the desired skill-related changes or near- and far-transfer effects. Therefore, future longitudinal studies should also seek to investigate to which extent performance improvements measured using game metrics within the “game” module translate to physical, motor, or cognitive performance enhancements measured using validated assessments and elucidate factors for promoting such transference effects.

It has been proposed that (adherence to) optimal intensity and complexity of exercises are the driving mechanisms for global neuroplasticity (ie, exercise intensity) and task-specific neuroplasticity (ie, cognitive and motor-cognitive task complexity) [106]. The ability of technology-enhanced training to individually progress exercises according to performance metrics in real time offers a potential advantage over conventional exercises and is considered a key advantage of serious video games (such as exergames or cognitive training games) [113]. There is initial empirical evidence supporting this assumption by showing that training with adaptive difficulty algorithms may be superior to generic games to improve both cognitive and noncognitive outcomes in individuals with mNCD in the context of computerized cognitive training [114]. Most studies on exergame-based training [5,115] and technology-supported physical activity interventions in general [46] seem to also have recognized this potential benefit and individually tailored or progressed the exercises by adjusting the difficulty or complexity of the task or games according to the users’ ability or performance [115]. However, only 42% [5] or 14% [46] of studies have reported or provided sufficient details on exercise intensity, whereas details on cognitive demands were neither assessed nor reported in any of the analyzed studies in the review by Manser et al [5]. Moreover, while a recent review observed that tailored technology-enhanced physical activity interventions (including exergame-based training) resulted in significantly higher adherence than control interventions [46], a recent systematic review observed that the effects on cognitive performance were moderated not by individualization of training (ie, tailored vs generic [one size fits all]) for exergame-based training in middle-aged to older adults but by other variables (eg, “body position” variable, favoring stepping movements in a standing position, or “exercise intensity” variable, favoring moderate exercise intensity) [5]. Therefore, it remains to be investigated in more detail whether individualization of training translates to moderating the efficacy or effectiveness of technology-enhanced training and how such individualization may be optimally designed to maximize efficacy or effectiveness in exergame-based training [5], as well as the broader context of technology-enhanced physical and motor-cognitive training. Ultimately, these investigations should seek to improve the personalization of technology-enhanced interventions by identifying and adopting strategies that facilitate individual learning success and, thus, promote effectiveness in improving cognitive (and physical) performance.

This secondary analysis of 2 RCTs provided valuable insights into exergame metrics. This study demonstrated that a selection of game metrics can serve as valid indicators of in-game domain-specific cognitive performance during exergaming. Moreover, metrics that reflect the precision of responses overall performed better than metrics reflecting the speed of responses, suggesting that metrics for precision may be better indicators of cognitive performance and performance progression during exergaming. Therefore, it is recommended to carefully select and prevalidate game metrics for purposes beyond game performance measurement and monitoring, such as informing personalized adjustments of the difficulty and complexity of exergames. Incorporating valid game metrics could increase their value in exergame designs, which, in turn, could increase exergame engagement and effectiveness. Due to the complex nature of exergames, in-game performance may depend on additional factors such as game design, external conditions, and personal characteristics of the player. Furthermore, the analysis of performance progression revealed high interindividual differences, underlining the importance of personalized training and suggesting the need for further research to explore characteristics explaining these differences and identify and adopt strategies that facilitate individual learning success and, thus, promote effectiveness in improving health- or disease-related outcomes.