Abstract
Background/Objectives: Population aging requires scalable approaches for early identification of cognitive decline, particularly mild cognitive impairment (MCI). Although the full 11-task BrainOK smartphone assessment showed excellent discrimination (AUC = 0.941), its administration time constrains large-scale use. To develop and validate a brief Short-Form BrainOK (SF-BrainOK) that preserves diagnostic performance while substantially reducing testing time. Methods: We enrolled 168 community-dwelling older adults (≥60 years). MCI was defined using the Montreal Cognitive Assessment (MoCA; score ≤ 23) as the reference standard. Candidate tasks were selected from the original BrainOK via LASSO-based preselection. To maximize data utilization given the limited sample size, model performance was evaluated using Leave-One-Out Cross-Validation (LOOCV). The cut-off value was determined by maximizing Youden's J. Results: The final two-task model combined executive function task (Rule-based Drumming I) and memory task (Password Memory I). On the independent test set, discrimination was robust (AUC = 0.783), with sensitivity = 0.75 (95% CI: 0.63-0.85, specificity = 0.71 (95% CI: 0.62-0.80, and accuracy = 0.765 (95% CI: 0.65-0.79) at the optimal cutoff. Conclusions: SF-BrainOK provides a brief, two-task digital screen that markedly reduces administration time while maintaining effective diagnostic performance. By targeting executive function and memory-domains repeatedly shown to be sensitive to early MCI-related change-SF-BrainOK supports scalable, opportunistic screening and the timely identification of at-risk individuals in resource-constrained settings.