Identifying cognitive subgroups in older adults from community data with hierarchical cluster analysis

This study applied Ward's hierarchical clustering to detailed neuropsychological data from a community-recruited sample of 180 older adults to identify naturally occurring cognitive subgroups and characterize multidomain cognitive heterogeneity beyond conventional mild cognitive impairment (MCI) criteria. Twenty-two demographically adjusted z-scores spanning memory, language, executive function, and attention/processing speed were submitted to Ward's method, and the resulting four-cluster solution was cross-validated using an independent k-means analysis. The identified subgroups included an average balanced profile cluster (ABP), an average with relative non-memory weakness cluster (A-nonMEM↓), an average with memory-specific weakness cluster (A-MEM↓), and a global multidomain weakness cluster (G-MD↓). All four subgroup patterns were reproduced in the k-means solution, and the dissociation between the A-nonMEM↓ and A-MEM↓ paralleled the non-amnestic versus amnestic distinction reported in prior MCI and cluster-analytic studies. Cluster membership explained substantially more variance in domain-level cognitive performance than the Montreal cognitive assessment (MoCA), indicating that the identified subgroup structure captured multidomain cognitive heterogeneity beyond global screening alone. These findings demonstrate that unsupervised, data-driven clustering can identify reproducible multidomain cognitive phenotypes within a community-recruited aging sample without reliance on predefined diagnostic categories. Hierarchical clustering may therefore provide a complementary framework for characterizing subtle patterns of cognitive heterogeneity across the broader spectrum of cognitive aging.