Applying Similarity Network Fusion to 515 ADNI participants, researchers integrated structural MRI metrics (cortical thickness, surface area, volume) with four cognitive batteries (ADAS-Cog, MMSE, MoCA, CDR) to produce four data-driven clusters spanning a continuous gradient of neural and cognitive severity. Group 1 (87% clinically diagnosed with dementia) showed the steepest impairment; Group 4 (96% cognitively unimpaired) showed minimal decline. Critically, Groups 2 and 3 captured transitional biology invisible to standard three-tier clinical labeling, including an early 'at-risk' subgroup with measurable cortical and cognitive changes preceding formal MCI diagnosis.
The finding matters because conventional CU/MCI/AD categories were designed for clinical convenience, not biological precision — they mask enormous within-group heterogeneity that undermines both trial enrollment and therapeutic targeting. This multimodal clustering approach aligns with a growing movement toward biologically defined Alzheimer's staging, including the NIA-AA 2024 biological framework. Identifying a discrete at-risk cluster could sharpen recruitment for secondary prevention trials and personalize monitoring intervals. Limitations are significant: the ADNI cohort skews toward educated, white participants, limiting generalizability; cross-sectional clustering cannot confirm whether individuals actually progress through these groups sequentially; and causal mechanisms remain unresolved. As a preprint posted to medRxiv and not yet peer-reviewed, these cluster definitions and their clinical thresholds require independent validation before any diagnostic application. Still, the methodology is a meaningful incremental advance toward precision stratification in cognitive aging research.
