Variability in working memory performance on mobile devices is sensitive to mild cognitive impairment: Results from the Einstein Aging Study

Abstract Background Most prior research on distinguishing cognitive impairment from healthy controls focuses on measuring mean‐level of cognitive performance on one or relatively few assessments, ignoring potentially valuable information about variability. Within‐person variability in cognitive performance is emerging as a promising indicator of mild cognitive impairment (MCI) and dementia. The limited research on this topic, however, typically examines speeded responses and variability transpiring across moments during a single test administration. The current study extends prior work by using ecological momentary assessments (EMA) to examine cognitive status differences in variability across hours and days in three dimensions of working memory (WM): object‐location binding, object‐feature binding, and spatial WM. Method In a sample of 243 systematically recruited, community‐dwelling adults age 70+ from the Einstein Aging Study (Mean age=77.46 years, SD=5.13), we identified 68 individuals meeting Jak‐Bondi criteria for MCI and 175 cognitively normal controls (Table 1). All participants performed object‐location binding, object‐feature binding, and spatial WM tasks on a smartphone up to four times daily for 14 days, totaling up to 56 assessments per person. We employed heterogeneous variance models using log‐linear prediction of residual variance to simultaneously assess MCI differences in mean performance and within‐person variability. Models adjusted for linear trends across assessments and study days, as well as age, education, gender, and ethnicity (Table 2; Figure 1). Results For mean performance, individuals with MCI exhibited worse performance than healthy controls in measures of object‐location binding (Est.=0.14, SE=0.02, p<.001), object‐feature binding (Est.=‐0.27, SE=0.04, p<.001), and spatial WM (Est.=0.90, SE=0.20, p<.001). For variability across assessments, individuals with MCI exhibited greater within‐person variability than controls in object‐location binding (Est.=0.89, SE=0.03, p<.001) and object‐feature binding (Est.=0.37, SE=0.03, p<.001) performance. Variability in spatial WM performance did not vary as a function of cognitive status (Est.=0.02, SE=0.03, p=.63). Conclusion Results suggest both mean and variability in performance are sensitive to MCI and exhibit dissociative patterns by WM dimension. Variability in object‐location and object‐feature binding performance may provide specific detection of MCI. The 14‐day smartphone‐based EMA offers novel opportunity to leverage performance variability across frequent assessments for identifying early clinical manifestations of cognitive impairment.