Longitudinal risk estimation of progression from normal cognition to amnestic mild cognitive impairment: The independent contribution of ATN biomarkers.

Background: It is still not clear about the independent contribution of ATN biomarkers (amyloid beta [A], pathologic tau [T], and neurodegeneration [N]) on top of neuropsychological tests for risk estimation of amnestic mild cognitive impairment (aMCI) in subjects with normal cognition (NC). Method: We included 306 NC subjects who underwent neuropsychological tests and were available for ATN biomarkers with cerebrospinal fluid (CSF) examinations, MRI and FDG‐PET at baseline. The involved ATN biomarkers included CSF‐Aβ, CSF‐Ptau, CSF‐Tau, FDG‐PET and an MRI‐based Alzheimer's disease resemblance atrophy index (AD‐RAI) calculated by AccuBrain. The study cohort was divided into training cohort (n = 204, 19.6% developed aMCI within 4.73±2.83 years) and validation cohort (n = 102, 20.6% developed aMCI within 4.67±2.72 years). All the demographic features and neuropsychological tests (Table 1) were entered into the cox regression model with backward elimination to predict time to incident aMCI (the subscores of FAQ and ADAS‐cog were also entered as candidate predictors), and the resulting model based on the training cohort was treated as the basic model. Individual ATN biomarkers were added on top of the basic model for cox regressions, and the enriched models would be tested in the validation cohort using time‐dependent AUC (t‐AUC) if they presented higher Chi‐squares than the basic model in training cohort. Result: The features that survived backward elimination for the basic model were shown in Table 2. Only the enriched model with Ptau or AD‐RAI presented higher Chi‐square than the basic model in the training cohort (p<0.05, Table 3). In the validation cohort, the model Basic+(AD‐RAI) presented significantly higher t‐AUC (0.770∼0.859) than the basic model through year 2∼7 and Basic+Ptau through year 1∼4 (p<0.05, Figure 1). The model with Ptau only presented significantly higher t‐AUC than the basic model for year 7 (p = 0.02). Also, adding Ptau on top of Basic+(AD‐RAI) presented no significant increase of t‐AUC for any time point (Figure 1). These findings indicated that Basic+(AD‐RAI) may be used as the optimal model for predicting risk of aMCI (nomogram shown in Figure 2). Conclusion: AD‐RAI consistently presented independent contribution to predicting aMCI among NC subjects on top of demographic characteristics and neuropsychological tests. [ABSTRACT FROM AUTHOR]