Predicting amyloid status using remote online self‐report and cognitive assessment: The Brain Health Registry

Abstract

AbstractBackgroundThe ability to identify older adults with elevated brain amyloid beta (Ab) load can facilitate Alzheimer’s disease (AD) prodromal and preclinical trial recruitment and has possible clinical benefits. Development and validation of non‐invasive, cost‐effective methods to accurately predict brain Ab status are crucially needed. The overall goal of this study was to assess the predictive value of online, remotely‐collected information in determining Ab‐PET status. We compared the predictive performance of online measures to the performance of similar in‐clinic measures.MethodThe Brain Health Registry (BHR) is an online registry for recruitment and longitudinal assessment in cognitive aging research. In BHR participants who had an available Ab status from in‐clinic PET scans, logistic regression was used to determine the optimal combination of online variables to predict Ab‐PET (Ab+ and Ab‐) status in all participants. BHR predictors included: demographics (age, gender, education), self‐reported health information (memory concern, family history of AD, Geriatric Depression Scale (GDS) score, Everyday Cognition Scale score), and a cognitive assessment (Cogstate Brief Battery (CBB)). Area under the receiver operator curve (AUC) evaluated discrimination accuracy. The model with the highest AUC was cross‐validated using 10‐fold cross‐validation.Result354 BHR participants had available Ab data and sufficient additional BHR data (see Table 1 for participant characteristics). 48.9% of the participants were Ab+. Model accuracy ranged from AUC 0.55‐0.65 (see Table 2). The most accurate predictive model (AUC=0.65) included demographics, self‐reported health information, and CBB. In this model, GDS score, memory concern, and gender were all significantly associated with Ab status. The cross‐validated estimated AUC was 0.58, which is noticeably lower than the uncorrected estimate. In a larger cohort (n=635) of participants who did not take Cogstate, other BHR variables predicted Ab+ (AUC=0.79). The discrimination ability of our online data was comparable to studies that used similar information collected in‐clinic, with in‐clinic AUCs ranging from 0.59‐0.76.ConclusionOur results suggest that a novel, online approach can aid in the prediction of Ab status. Since online variables are low‐cost and non‐invasive, there is an advantage to using this approach to facilitate AD trial screening than prediction models that use in‐clinic measures.