Background Over the past several decades, treatment development for Alzheimer’s disease (AD) has been largely focused on modifying amyloid‐beta (Aβ), but no drugs that modify the pathophysiological processes underlying the disease have been FDA approved; it is therefore possible that Aβ may not be the optimal target for treating AD. The NIA‐AA consortium has proposed the use of amyloid, tau, and neurodegenerative (A/T/N) biomarkers in diagnosis and treatment of AD. However, it remains unclear whether each arm of the A/T/N framework has an equally weighted contribution to the progression of AD or rather a stage‐dependent importance to AD development. Methods Here we use random forest, a machine learning algorithm, in participants from the ADNI dataset (Table 1) to predict AD cognitive decline using integrated biomarkers from the A/T/N framework: Aβ‐PET, CSF‐pTau, and FDG‐PET and MRI‐Structural, respectively (Table 2). We chose random forest for its high prediction accuracy and interpretability. We also analyzed the relationship between A/T/N biomarkers and memory composite and executive functioning composite scores. Results We show that the A/T/N biomarkers have stage‐dependent importance to AD development, with Aβ and phosphorylated‐tau (pTau) better predicting early dementia status (i.e. mild cognitive impairment) and neurodegeneration, especially low glucose uptake, better predicting later dementia status (i.e. clinical AD) (Table 3). We show a similar pattern when correlating markers to performance on memory and executive functioning tests. Conclusions Our results provide evidence that AD treatments may need to be stage‐oriented to match the natural disease progression. (Figure 1) Aβ and tau may be appropriate targets early in the disease course, but brain metabolic restoration should be explored as a treatment target later on in the disease process.

No abstract

No abstract

No abstract