Scientific Reports

When an older individual retakes a cognitive test, they usually improve the second time, and the magnitude of that improvement can predict future cognitive trajectory. In 178 healthy older adults retested overnight, the one-night improvement predicted who would decline over the next 2.5 years.

Nearly everyone improved on the overnight retest (99%), but the amount of improvement varied widely and tracked performance over the following 2.5 years.

Smaller overnight gains were linked to a six-fold increase in odds of landing in the worst trajectory group.

The groups looked similar at session one. Overnight improvement, not the starting score, separated long-term outcomes.

Standard cognitive scores correct for age, education, and sex, but often ignore premorbid ability, race/ethnicity, and other important predictors. The C-model integrates these variables into scoring.

The model explained nearly twice as much score variance as standard scoring, improving signal over noise.

It reduced racial disparity in MCI classification by more than three-fold and improved detection in high-vocabulary participants.

Benefits replicated across independently recruited samples, supporting robustness rather than overfitting.

In 1,916 adults ages 18 to 89, domain scores in older adults increasingly reflected general ability rather than domain-specific function.

This dedifferentiation effect was strongest for language and memory and accelerated after age 70.

A bifactor model separated general and domain-specific variance and identified many more selective weaknesses than domain-averaged scores.

Most selective weaknesses occurred in people with intact general ability, showing how focal deficits can be masked by standard methods.

Traditional batteries often capture fewer than 15 scores. This analysis used 107 measures from 22 tests in 1,916 adults, including speech and language biomarkers.

Confirmatory factor analysis supported five domains: general ability, processing speed, memory, language content, and speech fluency.

Demographic correction improved model fit and domain separation.

The structure replicated in a subgroup tested with expanded measures, indicating strong model stability.

A scalable computer-vision pipeline scored 9,117 drawings from 2,011 adults with agreement comparable to expert raters.

Unlike black-box systems, scoring preserved element-level localization and interpretation aligned to manual protocols.

The memory trial declined with age faster than copy performance and showed reliable longitudinal sensitivity.

Touchscreen stroke data added kinetic markers such as pauses and stroke smoothness that are not visible in static scoring.