Abstract
PURPOSE: The study aims to characterize the spatiotemporal distribution of amyloid deposition, differentiate between its subtypes, and explore their predictive value for patient cognitive outcomes. METHODS: Amyloid PET data from a prospective consortium study, the Alzheimer's Disease Neuroimaging Initiative, were used. A spatial independent component analysis revealed regions of amyloid deposition covariation, which served as regions of interest. A subtype and stage inference analysis was then performed to infer spatiotemporal patterns from cross-sectional data. Multinomial logistic regression models evaluated the impacts of demographics and risk factors on subtype assignment and determined the prognostic value of the subtypes for cognitive decline. Longitudinal data were used for validation. RESULTS: The study included 1,049 participants (466 cognitively normal, 447 mild cognitive impairment, and 136 Alzheimer's disease; 72 ± 8 years; 543 women), with follow-up data available for 643 (915 ± 431 days from baseline). Three distinct spatiotemporal patterns were identified, primarily affecting the parietal, frontotemporal, and occipital lobes, respectively, in the early stages. The amyloid deposition rates differed between the subtypes, even after age, diagnosis, apolipoprotein E ε4 carriership (APOE), baseline amyloid burden, and tracer types were controlled for (occipital vs. parietal: β = 32.6, P < .001; occipital vs. frontotemporal: β = 17.0, P = .017; parietal vs. frontotemporal: β = 15.6, P = .026). The rates of change in cognitive scores, adjusted for age, diagnosis, APOE, baseline amyloid burden, baseline cognitive score, and tracer types also differed between the subtypes (occipital vs. Stage 0: β = 0.162, P = .021; occipital vs. parietal: β = 0.134, P = .013). CONCLUSION: Amyloid PET subtyping may serve as a valuable independent prognostic biomarker.