Objective: Royall and colleagues identified the latent dementia phenotype, δ, which represents the concomitant cognitive and functional changes of dementia. The phenotype has been cross-validated using clinical and neuropathological outcomes in the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. The present study longitudinally validates δ and examines its association with biomarkers in a heterogeneous sample representing mild cognitive impairment (MCI), Alzheimer's disease (AD), and normal cognition. Reduced cerebral metabolic rates of glucose occur early in AD and correlate with disease progression. We hypothesized that δ would fit the neuropsychological test data well across time and that fludeoxyglucose (FDG) uptake via positron emission tomography (PET) imaging would predict the rate of change in dementia severity. Method: Three models were fit to the ADNI dataset, using four time points (baseline and 6, 12, and 18 months): individual growth curve models for δ (n = 1724) and FDG-PET (n = 1296) as well as a parallel process growth curve model examining the relationship between latent intercepts and slopes for FDG-PET uptake values. Results: The growth models fit well for both δ (CFI = 0.996, TLI = 0.995, RMSEA = 0.067) and FDG-PET (CFI = 0.999, TLI = 0.999, RMSEA = 0.020) individually. The combination of the two growth curve models into a parallel process growth model also fit the data well (CFI = 0.989, TLI = 0.987, RMSEA = 0.057). Conclusion: Longitudinal validation of δ provides further evidence for its use as a clinical marker for dementia. Glucose uptake by the brain predicts the rate of change in dementia severity, suggesting that changes in an individual's δ score is highly representative of the underlying functional pathological changes.