Introduction
Background
Cognitive impairment refers to the decline of one or more 1 cognitive functions, including attention, executive function, or memory [1]. This decline may be a natural part of aging [2]. However, with the increase in average lifespan, the prevalence of pathological cognitive decline also rises due to excessive neural damage [3]. This pathological decline is known as dementia, a general term for a variety of neurodegenerative diseases with different causes characterized by “an impaired ability to remember, think, or make decisions interfering with daily activities” [4]. Most types of dementia develop gradually over time [5]. The prestage of dementia is called mild cognitive impairment (MCI). More particularly, MCI is defined as “a syndrome of self-reported cognitive complaint with one or more objective cognitive impairments but preserved independence in functional abilities” [6]. Hence, individuals with MCI, despite experiencing cognitive impairment, still successfully perform activities of daily living (cooking, toileting, walking, visiting friends, etc), and not all individuals with MCI go on to develop dementia [6]. However, they are at a higher risk, with a reported annual conversion rate from MCI to dementia varying from 2% to 31% [7].
Currently, pharmaceutical interventions to cure MCI or dementia are still under research [8]. Nevertheless, early assessment of cognitive decline, along with early intervention methods, such as cognitive training and medication, can effectively slow down the progression of the condition [9-12]. The current gold standard for diagnosing MCI involves a detailed anamnesis, followed by a neuropsychological examination, and is often complemented by brain imaging to detect structural changes, along with blood and cerebrospinal fluid analyses [13,14]. This comprehensive assessment necessitates a team of medical specialists, consuming a significant number of medical resources [15]. Therefore, before a detailed examination is performed, it is advised to administer a quicker screening test and to continue with a full examination only if necessary [16]. The Mini-Mental State Examination (MMSE) [17] is currently the most widely used instrument for rapid screening of Alzheimer disease (AD) and MCI [18], assessing cognitive functions such as attention, registration and recall, orientation, and language. It is fast to administer, taking only 10 minutes, but it has also shown a lack of sensitivity, misclassifying older adults with MCI as healthy [19]. The Montreal Cognitive Assessment (MoCA) [20] is another popular instrument that evaluates similar cognitive functions in up to 20 minutes. Several studies have shown MoCA to be superior to the MMSE in terms of sensitivity for detecting MCI [20-22]. However, the MoCA has shown lower specificity, misclassifying healthy older adults (with natural cognitive decline) as at risk for MCI.
The suboptimal accuracy of these quick screening instruments may be partly due to the limited assessment time inherent in their design. However, another factor negatively impacting the accuracy of neuropsychological tests may be anxiety. Patients who are required to perform abstract tasks at a clinic in the presence of medical experts may experience heightened stress levels and perform suboptimally [23]. This is known as the white coat effect [24] and is known to disturb neuropsychological test performance. Moreover, neuropsychological tests are designed as one-time assessments and can be disturbed by circumstances such as lack of sleep or traumatic events (eg, loss of a spouse). However, the already long waiting lists make frequent and repeated neuropsychological tests infeasible.
Therefore, to complement screening instruments for MCI and alleviate the resources needed for a full examination, researchers are exploring alternative sources of information on cognitive functioning. Such solutions may come in the form of dedicated sensors installed in the home or wearables to detect movements [25,26], embedded software trackers that capture mouse movements [27,28], keyboard loggers installed on computers [29], conversational humanoids [30], or the use of interactive virtual simulations and games. Particularly, the latter has stirred the interest of researchers; games and simulations have long been used in understanding, measuring, and training cognition [31-33]. In addition to providing interactive, immersive environments to engage participants, they provide challenges bounded by a fixed set of rules, which can be highly indicative of targeted cognitive functioning. Research has indicated that games are more engaging than classical neuropsychological tests [32-34] and can be used to assess cognitive skills [35]. Moreover, the entertaining nature of gamified interactive systems (GISs) might reduce stress. They can be played in the comfort of the patient’s home, making the presence of a trained administrator redundant [36]. Finally, GISs also encourage more frequent interaction, providing additional testing data. Thus, they have the potential to complement traditional screening instruments by offering longitudinal assessment, which provides health care professionals with a new perspective for their case findings and diagnosis.
Despite their potential, it remains unclear to what extent GISs can be effectively used for the assessment of MCI, particularly as an early indicator of dementia. Researchers are still investigating the effective methods for system integration [37] and the accuracy of GISs in assessing cognitive decline [31,37]. In summary, while the of GISs for assessing MCI holds potential clinical significance, it remains in its early stages of development.
Objectives
Therefore, to provide a more comprehensive understanding of the potential of GISs, we aimed to present a scoping review of their applications in assessing MCI, addressing the following research questions (RQs).
What are the different game types (genres and interaction paradigms) of GISs used to assess MCI?
Which cognitive functions are assessed by GISs?
How are GISs evaluated (eg, longitudinal or cross-sectional), and how reliable are these studies in reporting diagnostic performance?
Methods
Overview
The review followed the latest PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines (checklist 1) [38], focusing on GISs for assessing MCI. The protocol is provided in Multimedia Appendix 1 [9-12,20-22,38,39] and can also be accessed online [40]. The databases queried include the Scopus database (eg, PubMed, MEDLINE, and Embase) as well as more technical libraries, specifically IEEE Xplore and ACM Digital Library. The screening process for queried articles consisted of 3 phases: title screening, abstract screening, and full-text eligibility. Ambiguous articles at each phase were advanced to the next stage for further review. In the full-text eligibility, articles were reassessed to ensure that all 3 inclusion criteria were met, in addition to the 4 exclusion criteria.
Search Queries Used
The query terms used in the search strategy are presented in Textbox 1. For the population, the search focused on patients with either “Mild Cognitive Impairment” or “MCI,” deliberately excluding dementia to maintain an emphasis on the prestage of dementia. Regarding the instrument used for assessment, the search included the terms “Gamif*,” “Game,” “Video game,” or “Videogame.” In addition, the purpose of the instrument was specified as “assessment,” with variations such as “Evaluation,” “Screen*,” and “Diagnos*” included as synonyms in the queries. These terms were searched across metadata, titles, abstracts, keywords, and full text. Because the syntax of each database varied slightly, the queries were adapted to match the specific requirements of each library, with the exact queries provided in the Multimedia Appendix 2.
Summary of search terms.
Disease
Mild Cognitive Impairment
MCI
Instrument
Gamif*
Game
Video game
Videogame
Purpose
Assess*
Evaluat*
Measure*
Screen*
Diagnos*
Eligibility Criteria
We focused on GISs designed to assess MCI and their potential for such assessment in this review. As a result, only articles meeting the predefined inclusion and exclusion criteria were included, as detailed in Textbox 2.
For systems described in multiple articles, such as separate publications for system design and diagnostic performance, all related articles were included and considered collectively as the entirety of a system. For systems that the authors named, we use the name in the remainder of the paper. For those systems that were not named by the authors, we refer to the system number given in this review.
Inclusion and exclusion criteria.
Inclusion criteria
Full-text, peer-reviewed articles published in English
Systems assessing mild cognitive impairment (MCI) as part or all of their function
Explicit inclusion of game or gamified elements in the article
Exclusion criteria
Studies featuring games or gamified elements in nonelectronic forms, such as pen-and-paper games
Game or gamified systems designed only for rehabilitation or training of specific cognitive functions, or targeting dementia instead of MCI
Articles that did not claim implementation
Book chapters that are inaccessible or retracted articles
Data Extraction and Synthesis
The following subsections describe how data were extracted and analyzed for each RQ, in alignment with the Results section by RQs.
RQ1: Identifying Game Types and Interaction Paradigms
We first extracted the information needed to address RQ1. We described the GISs used in articles based on genre (eg, casual games or simulation games [41]) and interaction paradigm [42].
RQ2: Mapping Cognitive Functions to the MoCA Domains
Second, we extracted the information needed to address RQ2. We drew on the descriptions provided in each paper and mapped the mentioned cognitive functions to those assessed by the MoCA [20], as outlined in Textbox 3.
We chose MoCA over the MMSE because it provides a more sensitive and comprehensive assessment of MCI [43], particularly in domains relevant to many GISs [44]. MoCA includes dedicated tasks for executive functions (eg, trail making, verbal fluency, and abstraction) and more detailed visuospatial assessments (eg, cube copying and clock drawing) than MMSE. These domains are frequently targeted by GISs, making MoCA a more suitable framework for mapping cognitive coverage.
Mapping of cognitive functions to the Montreal Cognitive Assessment domains.
Test
Short-term memory: assesses the ability to recall information after a short delay.
Visuospatial abilities: evaluates skills such as drawing and understanding spatial relationships.
Executive functions: tests planning, problem-solving, and organizing abilities.
Attention, concentration, and working memory: measures the ability to focus, maintain attention, and manipulate information.
Language: assesses naming, fluency, and comprehension.
Orientation to time and place: evaluates awareness of current date, location, and situation.
RQ3: Evaluation Design, Quality Assessment, and Diagnostic Performance Reporting
Finally, we extracted the information to address RQ3. Here, only systems that included evaluations for assessing MCI were coded. Each system may have multiple studies encompassing different research designs that have been published. If multiple studies of 1 system were included in the review, we chose articles focusing on the report of diagnosis. Usability studies, or studies focused on collecting other data (eg, physiological signals), were not considered to answer RQ3.
We first investigated how GISs are evaluated. Therefore, we coded the different research study designs used to assess accuracy, for example, cross-sectional studies or longitudinal studies. Second, we investigated how reliable these studies are. The articles were evaluated across 3 dimensions, each scored on a scale from 1 to 3, with higher scores indicating higher quality, based on a quality assessment method adapted from the studies by Bezabih et al [45], Johnson et al [46], and Boyle et al [47].
The detailed dimensions and scoring methods are provided in Textbox 4.
Hence, the score for each article can range from 4 to 12. We consider scores from 3 to 5 as “weak evidence,” 6 to 7 as “moderate evidence,” and scores 8 to 9 as “strong evidence.”
Finally, we investigated the diagnostic performances that were reported. The actual outcomes of the studies were coded, including sensitivity, specificity, and area under the curve (AUC).
Dimensions and scoring methods.
How appropriate is the reference standard? (1 to 3 out of 3 points)
Reference standard is based on fast screening instruments (eg, Mini-Mental State Examination or Montreal Cognitive Assessment)
Reference standard is based on a comprehensive neuropsychological test
Reference standard is based on full diagnosis by clinicians
How relevant are the comparative groups? (1 to 3 out of 3 points)
No controls, or it is unspecified who or how many participants have mild cognitive impairment (MCI)
MCI versus healthy controls, MCI versus older adults with subjective cognitive decline, or MCI versus older adults with dementia MCI versus healthy controls, with additional subdistinctions, for example, additional persons with subjective cognitive decline or persons with Alzheimer disease, or intraperson measurement (in case of long-term measurement in persons with MCI)
How generalizable are the findings of this study to the target population, considering the method of analysis and sample size of persons with MCI? (1 to 3 out of 3 points; for studies other than cross-sectional studies, the score was given case by case)
Cross-sectional study with <10 participants with MCI
Cross-sectional study with 10 to 30 participants with MCI
Cross-sectional study with >30 MCI participants with MCI
Discussion
Principal Findings
We identified 49 GISs for assessing MCI in this scoping review, mostly casual games and simulations, that support engagement through simple game mechanics or familiarity with daily tasks, such as shopping. We further identified that GISs mainly targeted cognitive functions, such as memory, attention, and executive function, but often lacked depth in language and orientation assessment. Finally, we found that only 14 systems reported diagnostic performance, with other studies remaining in early design stages or offering limited correlation studies. Moreover, although several GISs reported strong diagnostic performances, many of these studies (9/31, 29%) were limited by small samples, methodological flaws, and overfitting risks. In the following sections, we revisit our 3 RQs in more depth.
RQ1: What Are the Different Game Types (Genres and Interaction Paradigms) of GISs Used to Assess MCI?
We found that most GISs (30/49, 61%) consist of casual games with clear rules and straightforward gameplay, making them accessible and easy to play without requiring extensive learning or specialized technology. This simplicity may encourage older adults to engage with these systems, addressing the challenge that older adults generally have a lower game participation rate than other age groups [133]. We also noted the popularity of simulations (17/49, 35%). These offer a unique opportunity to represent and assess daily activities in a more controlled environment, providing valuable supportive data with higher ecological validity for evaluating MCI. In particular, shopping experiences are commonly integrated into many simulation game systems (7/49, 14%). The shopping experience is indeed a significant daily activity that requires short-term memory to memorize the shopping list, executive functions to recognize items, and spatial orientation to find them. Only a few systems (4/49, 8%) use full-body movement interaction paradigms to assess MCI. These include exergames and Kinect-based quiz games. Finally, only a few systems (3/49, 6%) were developed with dedicated hardware for the assessment of MCI, representing an extension of traditional GISs design.
In addition, we identified shared characteristics among the games examined in this study. All the GISs included in this research were noncompetitive and offered a single-player mode. Each gaming session was relatively brief, lasting around 15 minutes, and designed to be easy to understand. These observations align with the preference of older adults for puzzles and intellectually stimulating games [134]. Nevertheless, we like to emphasize the ongoing debate regarding whether merely simulation experiences can genuinely be considered games [135]. When daily activities are simulated without a well-integrated game design, it may lead to a lackluster gaming experience [136,137]. Furthermore, the gaming experience of casual games may not provide meaningful play for all older audiences. As articulated in the gerontoludic manifesto [138], game and mental health researchers may want to put more effort into understanding what differentiates older players rather than considering them as united in their age-related impairments.
Although we categorized gamified systems into casual, simulation, full-body movement, or dedicated interactive games for descriptive purposes, it is important to note that these genre distinctions are blurry in practice. Many games combine features from multiple genres, and genre labels do not necessarily reflect the specific cognitive demands or mechanisms embedded within gameplay. In addition, most studies did not describe game mechanics in sufficient detail to allow for systematic mapping to targeted cognitive functions. This limits the extent to which we can draw mechanistic conclusions about genre-specific diagnostic performance.
RQ2: Which Cognitive Functions Are Assessed by GISs?
The scoping review revealed that a diverse range of cognitive functions were assessed by GISs, including short-term memory (28/49, 57% GISs), visuospatial skills (20/49, 41% GISs), executive functions (23/49, 47% GISs), and attention and working memory (25/49, 51% GISs). However, a significant gap exists in evaluating language abilities (10/49, 20% GISs) and orientation to time and space (7/49, 14% GISs).
Moreover, the GISs reviewed in this study vary in how comprehensively they address the 6 cognitive domains evaluated by the MoCA (ie, short-term memory, visuospatial abilities, executive functions, attention, language, and orientation). For example, some simulation games incorporate daily tasks such as shopping, cooking, or wayfinding that naturally engage multiple domains, particularly short-term memory, executive functions, and visuospatial abilities. In contrast, casual games more often target specific domains, such as attention or working memory, through task-focused mechanics such as matching, sequencing, or quick-response challenges. Full-body movement games and dedicated interactive systems were fewer in number, and their alignment with MoCA domains varied by design, with some addressing attention or memory through physical interaction.
It is important to note that many GISs incorporate features from multiple game types, and genre labels do not consistently reflect the underlying cognitive functions. In addition, we could not draw definitive conclusions about the relative diagnostic effectiveness of different game types due to heterogeneity in the reporting of GISs designs. A better understanding of how specific game elements engage specific cognitive functions can guide future design and evaluation efforts. Hence, a more granular analysis of game mechanics in relation to targeted cognitive functions—beyond genre—would strengthen future investigations.
In particular, GISs could extend the duration of their assessment. The engaging nature of GISs can potentially encourage older adults to use the system for extended periods, creating opportunities to collect data over a longer time, in a manner that might not be feasible through traditional assessment tools. Thus, GISs could play a valuable role in evaluating long-term cognitive functions by tracking subtle changes over time, providing extra information that current screening instruments, such as MMSE and MoCA, are not designed to capture. Despite this potential, no existing system was identified in this review that specifically targets long-term MCI evaluation. Hence, future studies should prioritize the development and validation of GISs that leverage game mechanics to promote repeated use and sustained engagement, while also exploring how these systems can complement traditional diagnostic methods.
RQ3: How Are GISs Evaluated (eg, Longitudinal or Cross-Sectional), and How Reliable Are These Studies in Reporting Diagnostic Performance?
Out of a total of 49 GISs, 29 (59%) remain in the system design stage or present a preliminary user study. Hence, these GISs lack evidence of accuracy. Moreover, many studies (6/49, 12%) conducted on GIS were limited to correlating system metrics with traditional screening instruments, reducing their scientific rigor. In total, only 29% (14/49) of the GISs reported clinical validation to classify patients with MCI or intraperson comparisons.
The quality scores presented in Table 2 highlight notable variations in the methodological rigor of studies evaluating GISs for MCI assessment. While some studies used well-defined reference standards, relatively larger sample sizes, and appropriate validation strategies, others lacked clarity in participant selection or relied on less robust evaluation methods. This variability should be considered when interpreting reported diagnostic performance; higher-quality studies are more likely to yield reliable and generalizable findings, yet the results may be more nuanced. For researchers and clinicians, these scores offer a practical reference point for assessing the strength of evidence and identifying areas where future studies can improve in terms of design transparency and reporting consistency.
While clinical validation studies indicate that GISs may be a useful tool for assessing MCI, in particular, the risk of model overfitting and methodological bias remains a critical concern. For example, the virtual action planning supermarket model achieved an extraordinary F1-score of 1.0 across 4 machine learning algorithms: decision trees, random forests, adaptive boosting, and extreme gradient boosting. Such a result must be interpreted with extreme caution. The study included only 12 participants and used data enhancement techniques, yet identified 45 out of 46 features as significant—without any dimensionality reduction. Such a high feature-to-sample ratio introduces a substantial risk of overfitting, undermining the robustness of the findings.
More broadly, studies reporting near-perfect performance (eg, 1.0 accuracy or F1-score) are likely reflecting limitations in sample size, study design, or evaluation procedures rather than actual diagnostic capability. To prevent misinterpretation, especially by third-party stakeholders, it is essential to frame such results as preliminary and methodologically constrained results. These findings should not be taken as evidence of clinical readiness.
A robust study design is crucial for the clinical validation of GISs. First, the use of age-matched control groups is essential to minimize confounding variables. A key limitation in interpreting the performance of GISs-based cognitive assessments is the potential confounding effect of previous gaming experience or digital proficiency. Younger participants may outperform older adults not solely due to better cognitive functioning but also because of greater familiarity with digital devices and interaction patterns common in games. Only a few of the reviewed studies (4/31, 13%) assessed or controlled for previous gaming experience, making it difficult to disentangle cognitive performance from gameplay proficiency, as shown in the results for RQ3. This issue is further supported by recent findings from the study by Murata et al [139], which demonstrate that peripheral metrics such as swipe speed—independent of game content—can predict cognitive impairment. Future research should consider including baseline assessments of digital literacy or gaming familiarity and explore interface designs that minimize performance biases related to device handling rather than cognitive ability. In addition, previous research has highlighted sex-related differences in the risk, progression, and presentation of MCI [140,141]; most of the included studies (30/31, 97%) did not report results disaggregated by sex or analyze potential sex-based differences in diagnostic performance. As such, our review was unable to assess how sex may influence the effectiveness or generalizability of GISs for MCI assessment. The lack of such analyses represents a critical gap in the literature. Future research should explicitly report and analyze sex-disaggregated data to ensure that these systems are equitably designed and validated across diverse populations. Nevertheless, it is encouraging to see the inclusion of diverse groups, such as participants with AD, as MCI is regarded as a transitional phase between healthy aging and AD.
Second, comprehensive cognitive assessments should be administered to all participants, with those diagnosed with MCI specifically evaluated by health care professionals to ensure diagnostic precision. Solely relying on screening instruments for labeling MCI is inadequate and may overlook the complexity of cognitive decline.
Furthermore, most existing studies involve small, single-center cohorts, which limits generalizability. In addition, all the studies in the review used retrospective case-control designs, which can overestimate diagnostic performance by comparing well-defined cases and controls in artificial conditions. This introduces a high risk of bias and limits applicability in real-world diagnostic contexts. Future research would benefit from adopting larger, blinded, and multicenter studies—ideally through prospective diagnostic cohort designs—to strengthen the evidence base and enhance the clinical relevance of GISs.
Longitudinal studies, in particular, could facilitate a deeper understanding of long-term memory and its association with other cognitive functions, broadening the scope of cognitive domains assessed through GISs. By leveraging the engaging nature of gamified systems, researchers can explore how sustained interactions might reveal patterns or markers indicative of MCI progression or even early-stage cognitive decline. Future research could focus on designing and implementing longitudinal studies that incorporate diverse user populations, validate their findings against established clinical benchmarks, and investigate the potential of GISs to serve as an early warning system for cognitive decline. However, longitudinal studies present several challenges, such as participant retention, data standardization, and the ethical implications of prolonged data collection. Addressing these will be essential to maximize the utility and reliability of GISs in clinical practice.
Limitations
This review has several limitations. First, we did not expand our search beyond the initially included articles, which may have resulted in the exclusion of relevant studies related to specific GISs. For instance, in some cases, we only captured validation studies, while related user studies could not be identified from our database search. Second, our understanding of games within GISs was primarily derived from text descriptions and screenshots in the articles, without any direct interaction with the systems themselves. Consequently, the findings from RQ1 and RQ2 rely on the authors’ interpretations rather than firsthand experiences. This reliance on secondary information may restrict the accuracy and completeness of our descriptions of GISs. The community has also noted the challenges in articulating artifacts in experimental game research, and efforts are currently underway to refine standards [142].
Conclusions
GISs hold considerable potential to enhance the diagnostic process for MCI by capturing high-frequency, rich, and ecologically valid data, offering new insights beyond traditional screening instruments and test batteries. This rich data collection could provide health care professionals with a complementary perspective for diagnosis, enabling a more dynamic understanding of cognitive changes over time. To this end, this paper presents a scoping review of GISs used for assessing MCI. In total, 49 GISs were identified from 81 published articles and categorized into 4 types: casual games, simulation games, full-body movement games, and dedicated interactive games. We also examined the cognitive functions targeted and the accompanying diagnostic performance validation studies. Findings indicate a broad spectrum of GISs available for evaluating MCI with a wide range of cognitive functions. In addition, we identified several GISs reporting diagnostic performances comparable to screening instruments such as MMSE and MoCA—including a few studies (6/31, 19%) with near-perfect results. However, these findings should be interpreted with caution, as they often stem from studies with small sample sizes and limited methodological rigor. The analysis also identified limitations in the GISs validation; only 2 systems included in this review conducted intraperson comparison studies, and no rigorous longitudinal study was identified. This gap represents a missed opportunity to investigate the full potential of GISs in assessing MCI progression. Such studies could provide critical insights into how gameplay proficiency, learning effects, and adaptation to game mechanics influence the evaluation of cognitive functions, thereby enhancing the interpretability and accuracy of these GISs. Overall, this review underscores the significant potential of GISs in assessing MCI, while highlighting the need for continued research to yield more conclusive results across various facets of these interactive systems.