JMIR Publications

ABSTRACT

Background:

High-resolution digital cognitive tools administered at frequent intervals enable the study of age-related cognitive change at a previously unattainable level of granularity. Computational modeling further enables the simultaneous examination of multiple aspects of cognitive performance.

Objective:

The objective of this study was to examine whether repeated digital multiday learning assessments can detect normative age-related cognitive decline over short time frames and to quantify age-related differences in process-based cognitive markers derived from computational modeling.

Methods:

Two hundred and nine cognitively unimpaired (CU) older adults (Mage = 73.6 ± 8.2, 65% female) from the Harvard Aging Brain Study and affiliated studies were administered the Boston Remote Assessment for NeuroCognitive Health (BRANCH), which includes the Face–Name task, a paired associative learning task. Daily median response times to correct responses were used as the primary outcome. A new version of the Face-Name task was administered across up to four testing intervals over M=2.8±0.4 years. Using computational modeling, we estimated longitudinal changes in peak speed, a practice-adjusted marker of best expected performance, as an indicator of age-related cognitive decline. We further examined age-related differences in peak speed, stability (within-person variability), learning within and across testing intervals, and between-interval forgetting, with embedded linear regressions for age, sex, education, and device.

Results:

Peak speed slowed by 2%/year, indicating age-related cognitive decline. Higher age (+1 standard deviation, % change relative to each parameter’s mean) was associated with slower peak speed (+9%), steeper annual slowing in peak speed (+85%), greater within-person variability (+18%), reduced learning across testing intervals (-24%), and more forgetting (+13%).

Conclusions:

Our findings indicate that a digital multiday learning paradigm, combined with computational modeling of response times, sensitively detects age-related cognitive decline over short timescales (<3 years), which are difficult to capture in conventional study designs relying on infrequent assessments. In addition, this approach identifies specific process-based cognitive markers associated with aging, including slower speed, reduced stability, and diminished learning across testing intervals.