Comparing operational definitions of screening-based cognitive impairment in the Tohoku Medical Megabank cohort: Focus on digital engagement

Background Dementia prevention is a global health priority, yet most large-scale population studies come from Western settings. Japan, a rapidly aging society, provides a unique non-Western epidemiological context for characterizing cognitive decline. Objective To assess whether the Tohoku Medical Megabank (TMM) cohorts replicate established dementia-related associations, and to identify correlates that are statistically significant across three prespecified operational classifications. Methods We analyzed data from >81,000 participants in the TMM cohorts. Participants were classified using (i) self-reported physician-diagnosed dementia and screening-defined impairment using (ii) Mini-Mental State Examination (MMSE) ≤ 27 and (iii) Japanese version of the Montreal Cognitive Assessment (MoCA-J) ≤ 25. Associations with 14 established dementia-related correlates and >3,800 exploratory variables were assessed using false-discovery-rate control, semantic clustering, and sequential multivariable logistic regression. Results Participants with self-reported dementia had lower MMSE and MoCA-J scores. Established correlates replicated in expected directions for at least one classification, including lower education, hypertension, depressive symptoms, and hearing difficulty. We identified 106 significant correlates across all three classifications; for self-reported dementia, digital device non-use was associated with higher odds (fully adjusted OR up to 2.97, 95% CI: 1.42–6.18), while smartphone use showed inverse associations (OR = 0.47, 95% CI: 0.26–0.86); directions were consistent for screening-defined impairment. Conclusions In this study, leveraging existing cognitive classifications available in TMM cohorts, we found that established dementia-related associations largely replicated and identified correlates across three operational classifications. Digital disengagement emerged as a correlate that may reflect early functional change; given the cross-sectional design, causal inference requires further examination. Longitudinal validation with adjudicated diagnoses is needed.