A multivariate plasma pathology-cognition axis reveals biologically anchored residual cognitive heterogeneity across the Alzheimer's disease clinical spectrum

Background Alzheimer's disease (AD) pathology does not fully determine clinical expression, as individuals with similar biomarker-defined burden may show substantial differences in cognition. Characterizing cognitive variation beyond pathological burden may improve understanding of clinical heterogeneity in AD. Objective To develop a multivariate plasma pathology-cognition framework that derives a pathology-anchored residual phenotype capturing cognitive variation beyond circulating pathological burden and to evaluate its clinical and longitudinal relevance. Methods We analyzed 868 participants from the Alzheimer's Disease Neuroimaging Initiative with plasma p-tau217, glial fibrillary acidic protein, and neurofilament light chain measurements, cognitive assessments, and clinical characterization. Canonical correlation analysis derived a multivariate plasma pathology index and cognitive composite based on memory and language performance. The residual phenotype was defined as the deviation between observed cognition and cognition expected from plasma pathology burden. Clinical correlates and longitudinal trajectories were evaluated using regression and mixed-effects models. Biological validity was assessed by examining associations with amyloid-β and tau positron emission tomography measures. Results The first canonical correlation between plasma markers and cognition was significant (r = 0.610, permutation p < 0.001). The plasma pathology index correlated with amyloid-β and tau positron emission tomography measures. The resulting pathology-anchored residual phenotype was associated with clinical diagnosis, education, APOE ε4 status, and sex. Higher residual scores reflected better-than-expected cognition relative to plasma pathology burden and were associated with slower subsequent decline in memory and language. Conclusions A plasma pathology-anchored residual phenotype captures clinically meaningful cognitive heterogeneity beyond circulating biomarkers and provides a scalable framework for investigating differential clinical manifestations of AD pathology.