Computational Phenotyping of Cognitive Decline With Retest Learning.

Objectives: Cognitive change is a complex phenomenon encompassing both retest-related performance gains and potential cognitive decline. Disentangling these dynamics is necessary for effective tracking of subtle cognitive change and risk factors for Alzheimer's Disease and Related Dementias (ADRD).

Method: We applied a computational cognitive model of learning and forgetting to data from Einstein Aging Study (EAS; n = 316). EAS participants completed multiple bursts of ultra-brief, high-frequency cognitive assessments on smartphones. Analyzing response time data from a measure of visual short-term working memory, the Color Shapes task, and from a measure of processing speed, the Symbol Search task, we extracted several key cognitive markers: short-term intraindividual variability in performance, within-burst retest learning and asymptotic (peak) performance, across-burst change in asymptote and forgetting of retest gains.

Results: Asymptotic performance was related to both mild cognitive impairment (MCI) and age, and there was evidence of asymptotic slowing over time. Long-term forgetting, learning rate, and within-person variability uniquely signified MCI, irrespective of age.

Discussion: Computational cognitive markers hold promise as sensitive and specific indicators of preclinical cognitive change, aiding risk identification and targeted interventions.