Clinical expression beyond measured Alzheimer disease molecular pathology: a discovery and external evaluation study

Background Cognitive expression varies substantially among people with comparable Alzheimer disease molecular pathology. We evaluated whether a prespecified profile integrating hippocampal structure, education, and plasma glial fibrillary acidic protein (GFAP) captured cognition beyond measured phosphorylated tau 217 (pTau217), amyloid PET, and tau PET burden. Methods We used the Alzheimer’s Disease Neuroimaging Initiative (ADNI; n = 436) for discovery and the Health and Aging Brain Study: Health Disparities (HABS-HD; n = 491) for external evaluation. The profile was the equal-weight mean of intracranial-volume-normalised hippocampal volume, education, and inverse log-transformed plasma GFAP. The primary outcome was cohort-standardised four-item Preclinical Alzheimer Cognitive Composite (PACC4). Linear models adjusted for pTau217, amyloid PET, tau PET, age, and sex. Prespecified analyses examined pathology-balanced residual cognition, selection weighting, longitudinal cognition, and clinical progression. Results Higher profile values were associated with higher PACC4 in ADNI (β = 0.703, 95% CI 0.573–0.834; ΔR²=12.1%; P < 0.001) and HABS-HD (β = 0.818, 95% CI 0.657–0.979; ΔR²=15.5%; P < 0.001), with no evidence of between-cohort heterogeneity (P = 0.274; I²=16.3%). In pathology-balanced samples, residual cognition differed by 0.695 SD in ADNI (53 pairs) and 0.823 SD in HABS-HD (79 pairs). The HABS-HD association was similar after inverse selection-probability weighting (β = 0.837, 95% CI 0.656–1.017; P < 0.001). Associations with progression attenuated after additional adjustment for baseline cognition and clinical status. Conclusions A prespecified structure–reserve–glial profile captured reproducible cognitive variation beyond measured Alzheimer disease molecular pathology across two cohorts. It provides an explanatory clinical-expression framework, not evidence of a causal resilience mechanism or a validated individual-level prognostic tool. Trial registration Not applicable.