Detection of mild cognitive impairment using digital assessments of cognition

Abstract

AbstractBackgroundConventional cognitive assessment typically provides a single measure of performance at a single point in time in a laboratory setting. Smartphone‐based ecological momentary assessments (EMA) facilitate short‐term intensive assessment of cognitive performance in people’s natural environments. Though heterogeneous variance in digital cognitive assessments has been observed between older adults with mild cognitive impairment (MCI) and cognitively unimpaired (CU), the independent contribution of intra‐individual variability in performance to detection of MCI is poorly characterized in diverse populations of older adults. We examine the ability of digital cognitive assessments of processing speed in discriminating MCI from CU in older adults from a racially/ethnically diverse community sample.MethodEinstein Aging Study (EAS), smartphone‐based digital assessments of cognition, included the response time‐based symbol match test of processing speed. This was assessed 6 times per day over two weeks and was repeated annually. MCI was defined based on in‐house neuropsychological evaluations using Jak‐Bondi criteria. Statistical methods for modeling MCI using digital cognitive assessments include pattern‐mixture discriminant analysis based on heterogeneous linear mixed effects model for digital cognitive measures given MCI, joint modeling using shared random effects with random location scale for digital cognitive measures, and quadratic logistic regression models. All methods take into account the unbalanced adherence rate among individuals and control for covariates including age, sex, years of education, and race/ethnicity. Stratified analyses by race/ethnicity among non‐Hispanic (NH) Whites and Blacks were performed.ResultAmong 311 systematically recruited community‐dwelling non‐demented older adults from the EAS (Mean age = 77.5 years, SD = 4.8; 67.5% Female; 45.3% non‐Hispanic (NH) White, 40.8% NH‐Black), 97 (31.2%) individuals had MCI. Adherence to the EMA protocol was slightly better in individuals who were CU compared to those with MCI. Results from the quadratic logistic model (the easiest to apply) showed evidence of contribution from the quadratic terms that reflect heterogeneous intra‐individual variability in all samples and particularly among NH‐Blacks (p = 0.008).ConclusionWe observe evidence that intra‐individual variability in processing speed improves the ability to discriminate cognitive impairment, particularly among NH‐Black older adults. Leveraging digital technology and performance variability across frequent assessments in the smartphone‐based EMA offers novel opportunities to identify cognitive impairment.