Scenario-Based Serious Game for Screening Mild Cognitive Impairment in Older Adults: Cross-Sectional Preliminary...

Many existing digital cognitive assessments rely on isolated or abstract tasks, and primarily use accuracy-based outcome measures, despite recent advances in the field. Few have been culturally adapted to reflect the familiar daily-life experiences of Korean older adults. Integrating culturally meaningful scenarios with task-specific accuracy and reaction time (RT) measures may provide a more ecologically grounded approach to mild cognitive impairment (MCI) screening.

This study aimed to develop and preliminarily evaluate a tablet-based, scenario-based serious game for community-based MCI screening among Korean older adults.

In this cross-sectional diagnostic accuracy study, community-dwelling adults aged 60-84 years were recruited through convenience sampling from a senior welfare center in Daejeon, South Korea, between July 2024 and June 2025. Of 87 participants who underwent eligibility assessment, 67 participants were included in the analysis: 47 cognitively normal participants, and 20 participants with MCI. Clinical classification was determined according to Petersen criteria based on the Korean version of the Consortium to Establish a Registry for Alzheimer's Disease battery, second edition, structured clinical interviews, and clinician judgment blinded to the serious game results. The serious game initially comprised 5 culturally familiar and ecologically grounded daily-life tasks, from which task-specific accuracy and RT features were extracted. A reduced 4-task model informed by exploratory misclassification analysis served as the primary classification model and was evaluated using random forest classification with out-of-bag validation. A full 5-task model and a demographic-adjusted model were also evaluated. Convergent validity and perceived workload were assessed.

The group with MCI was older (77.3 vs 74.5 y; P=.002) and had fewer years of education (9.4 vs 11.2 y; P=.04). The primary 4-task model achieved an overall accuracy of 82.1% (55/67), balanced accuracy of 80.1%, sensitivity of 75% (15/20), specificity of 85.1% (40/47), and an area under the receiver operating characteristic curve of 0.787 (95% CI 0.643-0.932). The demographic-adjusted model showed broadly comparable discrimination (area under the receiver operating characteristic curve 0.794, 95% CI 0.654-0.934). Accuracy and RT measures from the sale items task ranked among the most important predictors. Digital composite scores were significantly correlated with the Korean version of the Consortium to Establish a Registry for Alzheimer's Disease battery, second edition total score (r=.655 and r=.447; both P<.001). Perceived workload did not significantly differ between groups (P=.97).

This study provides preliminary evidence for the diagnostic utility and feasibility of a tablet-based serious game integrating culturally familiar daily-life scenarios with task-specific accuracy and RT measures, thereby capturing both performance accuracy and potential differences in processing efficiency. The automated format may offer a scalable and accessible screening option for community settings without specialized equipment. Independent, multisite, and longitudinal validation is required to establish its generalizability and predictive utility.