Chen, X.; Nishimaki, K.; Tomoshige, S.; Chotiyanonta, J. S.; Uchida, Y.; Wasserman, B. A.; Gottesman, R. F.; Guallar, E.; Windham, B. G.; Oishi, K.

Abstract Objectives White matter hyperintensities (WMH) are associated with cognitive decline in older adults. We examined whether anatomically defined spatial WMH patterns are associated with cognitive performance and subsequent decline in a large community-based cohort with approximately six years of follow-up. Methods We analyzed 1,667 participants (mean age 76 years) from the Atherosclerosis Risk in Communities Neurocognitive Study. WMH were segmented from FLAIR images and mapped to 112 white matter regions using deep-learning-based parcellation. Principal component analysis identified dominant WMH distribution patterns; candidate components were selected by Horn's parallel analysis and retained if stable across resampled split samples. Linear mixed-effects models examined associations between WMH patterns and cognitive performance at baseline and follow-up, adjusting for demographic, genomic, and vascular risk factors. Robustness was assessed using analyses accounting for total WMH burden, study center, and selective attrition. Results Of 14 candidate principal components (PCs) identified by parallel analysis, eight met the resampling stability criterion and were retained, together explaining 42% of the variance. Three components were associated with cross-sectional or future cognition. PC3, characterized by deep white matter burden in the corpus callosum and anterior corona radiata with relative preservation of superficial white matter, showed no cross-sectional association at baseline but was associated with accelerated decline across multiple domains. Interpretation A deep spatial pattern of WMH loading on the corpus callosum and anterior corona radiata was associated with subsequent cognitive decline in the absence of a cross-sectional association at baseline, independent of total lesion burden.