Sequential Analysis of Partial Least Squares Structural Equation Modeling and Self-Organizing Maps Using a Mediation Approach and Latent Variables for Alzheimer’s Disease Analysis

Abstrak

Alzheimer's disease heterogeneity challenges clinical management.This study introduces an integrative framework combining Partial Least Squares Structural Equation Modeling (PLS-SEM) and Self-Organizing Maps (SOM) for patient stratification using National Alzheimer's Coordinating Center data.PLS-SEM validated four latent constructs, identifying functional disability as the principal predictor of clinical severity, with indirect effects mediated by cognitive and social activities.Comparing raw indicators against validated latent scores, SOM outperformed K-Means and hierarchical clustering, exhibiting enhanced robustness and clustering quality.Utilizing latent scores, SOM delineated three statistically distinct clinical phenotypes: a mild phenotype comprising the vast majority, a severe phenotype, and a small high-risk phenotype.This integrated approach offers a clinically relevant framework for precise patient segmentation, facilitating targeted therapeutic interventions.

Identifier

• DOI: https://doi.org/10.22266/inassexpress.2026.003
• OpenAlex: https://openalex.org/W7131635404