Baseline neuroimaging profiles and predictive modelling for early Alzheimer disease prognosis

Alzheimer's disease (AD) presents significant clinical heterogeneity in cognitive decline. Neuroimaging profiles and peripheral biomarkers, such as plasma oxidative stress biomarkers could provide information on AD progression. This study evaluates the contribution of baseline neuroimaging features and plasma lipid peroxidation biomarkers to modelling cognitive decline trajectories in early AD. Data were collected from patients with early AD (n = 195), and they were neuropsychologically evaluated to assess their progression. Baseline data included demographics, neuropsychological evaluations, neuroimaging measures and plasma lipid peroxidation levels. Follow-up neuropsychological evaluations were conducted biennially. Individual cognitive decline trajectories were estimated using MMSE and CDR Sum of Boxes (CDR-SB) scores. Multivariate modelling and hierarchical clustering analyses were applied to identify prognostic signatures and patient subgroups. Baseline neuroimaging variables showed significant associations with longitudinal cognitive trajectories, as reflected in MMSE and CDR Sum of Boxes slopes, whereas plasma lipid peroxidation biomarkers did not. Multivariate models based on neuroimaging features demonstrated substantial predictive contribution (conditional R² 0.825 for MMSE and 0.808 for CDR Sum of Boxes). Hierarchical clustering identified four distinct groups, primarily driven by left Koedam scores, enlarged perivascular spaces, and Fazekas scores. These clusters were significantly associated with differential rates of cognitive decline, particularly for MMSE. In conclusion, structural neuroimaging measures at baseline are robust indicators of cognitive decline in early AD, supporting its utility for individualized prognosis in early AD.