DOES HIPPOCAMPAL VOLUME PREDICT POSITIVE AMYLOID STATUS ON FLORBETAPIR-PET IN HEALTHY CONTROLS AND PRODROMAL STAGES OF ALZHEIMER'S DISEASE?

Reduced hippocampal volume (HV) on MRI is associated with Alzheimer's disease (AD). We evaluated HV prediction of amyloid status using florbetapir (FBP)-PET in healthy controls (HC), early and late mild cognitive impairment (MCI) and AD dementia groups. Baseline cross-sectional data for 170 HC, 252 early MCI (EMCI), 136 late MCI (LMCI), and 75 AD subjects from ADNI-2 were used. FBP-PET amyloid status was determined using ≥1.1 threshold for positivity from a composite SUVR (average of 6 regions of interest: anterior cingulate, posterior cingulate, precuneus, frontal medial-orbital, parietal, and temporal) normalized to whole cerebellum. HV was measured using FreeSurfer and corrected by intracranial volume. Elastic net logistic regression with 10-fold cross validation was used to evaluate prediction of amyloid positivity using HV, age and education as covariates in the four patient groups separately. Classification performance in each model was compared to random variables. The ranking of these three variables were used to compare the prediction power of each. FBP-PET was positive in 27% HC, 48% EMCI, 68% LMCI, and 95% AD. The model (HV, age and education) predicted amyloid status better than random variables in EMCI with positive predictive value (PPV)=63% versus 44% and negative predictive value (NPV)=63% versus 50%, respectively. For LMCI, the prediction performance using HV and other covariates compared to random variables was PPV=73% versus 67% and NPV=61% versus none. In EMCI, age was more predictive than HV, while the converse occurred in LMCI based on the ranking of these input variables. In both MCI groups, education was least predictive. HV and other covariates did not provide any predictive value compared with random variables in either HC (PPV=25%, NPV=74%) or AD (PPV=95%, NPV=none) because a majority of HC were negative and a majority of AD were positive. A model including HV was better than random variables in predicting high amyloid burden in the EMCI and LMCI groups, but not in the HC or AD groups. Age predicted pathological amyloid presence better than HV in EMCI. However, the modest PPV and NPV values obtained using the HV model limit its utility.