The role of age in the relationship between brain structure and cognition: moderator or confound?

Models trained on younger subjects successfully predicted cognition in older subjects, but models trained on older subjects failed to generalize to younger individuals.

• This asymmetry in generalisability was found using structural brain imaging data from the UK Biobank.
• The finding held when age was treated as a moderator rather than solely as a confound.
• The asymmetry suggests that brain-structure–cognition relationships in younger subjects capture more generalizable patterns than those in older subjects.
• The authors interpret this as reflecting that age-related (e.g. disease-related) variations in older subjects introduce specificity not present in younger subjects.

Age was examined both as a confound (via deconfounding) and as a moderator (via age-stratified modeling) in brain-cognition predictive models.

• Deconfounding removes the effects of age from brain structure and cognition measures before modeling.
• Treating age as a moderator involved estimating brain-cognition associations separately across age groups.
• The moderator approach captures age-stratified changes in how brain structure and cognitive performance are statistically connected.
• The study used structural brain imaging data from the UK Biobank as its primary dataset.

There is a trade-off between model specificity and generalisability depending on whether age-specific or pooled models are used.

• Age-specific models may offer greater specificity for a target population but at the cost of generalisability.
• Pooled models may generalize more broadly but may miss age-specific brain-cognition associations.
• The authors conclude that 'the optimal approach—whether age-specific or pooled—depends on the research or clinical goal for the target population.'
• This trade-off is relevant for addressing age-related cognitive decline in clinical contexts.

Age is strongly associated with both brain structure and cognition, making it a key variable that predictive models risk simply capturing as an age effect.

• Standard practice applies deconfounding to remove age effects from brain-cognition models.
• Without mitigation, predictive models of cognition from brain structure risk reflecting age associations rather than direct brain-cognition relationships.
• The paper motivates treating age as a moderator specifically because variations in brain structure linked to cognitive performance in older subjects (e.g. related to disease) may differ from those in younger subjects.
• Understanding these relationships is described as 'essential for addressing age-related cognitive decline.'

Brain-structure–cognition associations in older subjects may reflect disease-related variations that differ from those underlying cognition in younger subjects.

• The authors propose that for the moderator view to hold, variations in brain structure linked to cognitive performance differences in older subjects (e.g. related to disease) would differ from those in younger subjects.
• This conceptual distinction motivates age-stratified modeling as a meaningful analytical choice beyond statistical deconfounding.
• The failure of older-subject models to generalize to younger individuals is consistent with older-subject models capturing disease- or aging-specific structural patterns.

The optimal analytical approach—age-specific or pooled modeling—depends on the research or clinical goal for the target population.

• Age-specific models are more appropriate when the clinical goal targets a specific age group, such as older adults with cognitive decline.
• Pooled models may be preferred when generalisability across the lifespan is the primary goal.
• The findings suggest neither approach is universally superior, and the choice should be driven by the intended application.