Insights into the Heterogeneity of Cognitive Aging: A Comparative Analysis of Two Data-Driven Clustering Algorithms.

Abstract

Objectives: Cognitive aging entails diverse patterns of cognitive profiles, brain imaging, and biomarkers. Yet, few studies have explored the performance of multiple clustering algorithms on a single dataset. Here, we employ data-driven methods to analyze neuropsychological performance in older individuals with normal cognition (NC) and mild cognitive impairment (MCI).

Methods: A total of 311 older adults without dementia completed a comprehensive assessment, consisting of 17 cognitive tests and a memory complaint questionnaire. We utilized two clustering algorithms: nonnegative matrix factorization (NMF) and model-based clustering (MBC). Cluster characteristics were examined in demographic, clinical, and brain morphometric data.

Results: Both NMF and MBC uncovered two- and three-cluster solutions, with satisfactory data fit. The two-cluster profiles encompassed a cognitively intact (CI) group and a cognitively suboptimal (CS) group, distinguished by cognitive performance. The three-cluster solutions included CI-memory proficient, CI-nonmemory proficient, and CS groups. Remarkably, patterns of cognitive heterogeneity and their association with demographic and neuroimaging variables were highly comparable across NMF and MBC. Phenotypic homogeneity improved after identifying participants with consistent and mismatched memberships from the two algorithms.

Discussion: The results indicate that two distinct data-driven algorithms, with different heuristics, generated comparable patterns regarding cognitive heterogeneity within NC and MCI. These findings may inform future subtyping studies in cognitive aging, where replication of stratifications found across different methods is strongly recommended.