Detection of Amyloid status in patients with uncertain diagnosis and associations with remotely‐collected cognitive composites

Abstract

AbstractBackgroundIn an effort to build efficient and scalable screening tools for early detection of cognitive impairment, the Brain Health Registry (BHR) was created to collect demographic and cognitive information in a remote setting. A select cohort of participants enrolled in the BHR also underwent amyloid PET scans, allowing the utilization of remotely collected cognitive data for amyloid enrichment.MethodSelf‐reported measures were collected through the BHR from 656 older adults co‐enrolled in the Imaging Dementia ‐ Evidence For Amyloid Scanning (IDEAS) study with an in‐clinic diagnosis of unexplained mild cognitive impairment (MCI; n = 522) or dementia of uncertain cause (n = 134) (Table 1). Global Amyloid‐PET centiloid values were obtained via a published PET‐Only processing method. An independent pathology‐based cut‐off of 24.4 was applied to define amyloid status. A comprehensive set of BHR variables were regressed on the continuous centiloid values to quantify the association between online measures with global amyloid burden. Then, logistic regression was employed to determine the ability of BHR variables to detect amyloid positivity. Results from both regression models are provided in Table 2. Accuracy measures are reported in Table 3.ResultThe overall model fit of the BHR data with centiloid values was statistically significant (F = 4.69; p‐val<0.001), while R^2 = 0.09, highlighting the strong association with global amyloid burden and the high variability for this cohort. For both dichotomous and continuous models, family history of AD and subjective memory concern are associated with higher amyloid accumulation; higher Geriatric Depression Scale scores are associated with lower amyloid, while low scores of Memtrax N‐back test are related to higher amounts of amyloid. When applying 5‐fold cross‐validation to the logistic model, the classification accuracy of the BHR variables is 64% (cvAUC = 64%; spec = 67%; sens = 60%).ConclusionDespite high variability in BHR data, remotely collected subjective and cognitive information are associated with amyloid in older adults with unexplained MCI or dementia of an uncertain cause. The ability to detect amyloid status with online data points to the large potential for building an online screener and motivates further online data collection.