Polygenic dementia risk splits into two biologic routes across 26 years
A new dementia preprint separates inherited risk into two very different pathways: one tied to lower cognitive reserve before disease emerges, and another tied to faster decline with aging. Using 26 years of follow-up in an ancestrally diverse US cohort of 11,518 adults, investigators applied joint longitudinal-hazard models to ask a question that routine genetic analyses usually blur together: does a polygenic signal raise dementia risk because it leaves people starting from a lower cognitive baseline, or because it accelerates cognitive loss over time?
The answer was not uniform across genetic risk factors. APOE ε4 and polygenic indices for Alzheimer’s disease and related dementias, type 2 diabetes, stroke, LDL cholesterol, and hearing loss were linked to dementia hazard through the rate of cognitive change. In contrast, polygenic indices for educational attainment, cognitive function, physical activity, BMI, smoking, loneliness, schizophrenia, and externalizing were associated with dementia hazard through premorbid differences in cognitive function.
A model built to separate baseline vulnerability from later-life decline
The analytic framework matters because the same observed dementia hazard can arise from different biological routes. A person who enters late life with lower premorbid cognitive function may cross a clinical threshold sooner even if their slope of decline is ordinary. Another person may begin with average cognition but lose ground more quickly, reaching functional impairment through an active degenerative process. Joint longitudinal-hazard modeling is designed to distinguish those mechanisms rather than collapsing them into a single risk estimate.
In practical terms, this approach reframes genetic association work in dementia. A polygenic association with baseline cognition points toward developmental and life-course influences that likely act long before clinical symptoms. A polygenic association with longitudinal decline points toward processes that continue to operate during aging and could, at least in principle, remain modifiable later in life. The distinction is biologically plausible: reserve and resilience are not the same as active neurodegeneration, and the two can coexist in the same individual.
APOE and vascular-metabolic signals tracked with faster decline
The clearest group of findings clustered around APOE ε4 and several cardiometabolic or sensory traits. Their polygenic indices operated through rate of change rather than through premorbid cognition, aligning them with aging-related decline. This pattern fits the broader clinical picture of dementia as a syndrome in which vascular injury, lipid dysregulation, metabolic stress, and sensory loss can compound neurodegenerative vulnerability over time.
Stroke and LDL cholesterol are especially revealing in this context. Both plausibly map onto cerebrovascular injury and impaired brain maintenance, pathways that can erode cognition progressively rather than simply lower starting ability. Type 2 diabetes adds another layer, given its links to microvascular disease, insulin signaling abnormalities, and chronic inflammatory stress. Hearing loss may reflect both a direct sensory contribution to cognitive load and a marker of broader brain-aging processes. In each case, the polygenic association with change over time suggests a mechanism embedded in the aging trajectory itself.
One sentence captures the clinical weight of the result: some inherited risks appear to speed the descent.
Educational and cognitive polygenic scores tracked with lower starting point
A different set of polygenic indices behaved as markers of premorbid cognitive function. Educational attainment and cognitive function stood out here, along with physical activity, BMI, smoking, loneliness, schizophrenia, and externalizing. These signals imply that inherited liability is shaping the cognitive position from which aging begins, rather than the slope of later decline.
That distinction carries a strong life-course implication. If a genetic profile is associated with a lower cognitive baseline decades before dementia diagnosis, then the relevant intervention window may open early, long before old age. Education, physical activity, smoking exposure, social isolation, and psychiatric or behavioral traits may all influence cognitive reserve through cumulative effects on brain development, socioeconomic opportunity, cardiometabolic health, and social-cognitive engagement. The preprint does not claim that these factors are deterministic, but it does suggest that some polygenic signals are acting upstream of dementia itself, in the architecture of lifelong brain health.
This is also where the notion of reserve becomes clinically useful. Two individuals can show the same level of age-related decline and still differ in whether they cross the diagnostic threshold for dementia, because one began farther from the edge. Genetic effects operating through premorbid cognition help explain why dementia risk is not only a story of neurodegeneration; it is also a story of where a person starts on the cognitive continuum.
Ancestry differences caution against a single genetic story
Some results varied by continental ancestry, a reminder that polygenic architecture does not always behave identically across populations. Multi-ancestry polygenic indices are an important step forward because they attempt to retain utility across broader genetic backgrounds, but the portability of risk scores remains imperfect. Differences in linkage disequilibrium structure, allele frequencies, environmental exposures, and cohort composition can all alter how genetic signals are detected and interpreted.
This matters for dementia genetics in a direct way. A score that predicts premorbid cognition in one ancestry group may not have the same pattern in another if the underlying genetic architecture or measurement context differs. The preprint’s ancestry-stratified variation does not weaken the central conclusion; it strengthens the case for cautious interpretation and for building models that respect heterogeneity rather than assuming one uniform pathway from genotype to dementia.
Why the modeling approach changes the clinical conversation
Traditional genetic association studies often ask whether a variant or score is associated with dementia. This preprint asked a more useful question: how does the association operate? That shift moves the field from broad risk description toward mechanism. It also offers a clearer map for intervention timing. Genetic effects that act through baseline cognitive function are logically tied to earlier-life prevention, educational opportunity, and long-range brain health. Genetic effects that act through decline are more compatible with later-life strategies aimed at vascular risk, metabolic control, sensory preservation, and neuroprotective treatment.
The preprint’s conclusion is careful and clinically grounded. Joint longitudinal-hazard models made it possible to distinguish degenerative pathways that might be therapeutically targeted in later life from premorbid differences that would require action earlier in the life course. That is not a treatment claim. It is a mechanistic map, and in dementia research, mechanistic maps are often what determine whether future trials target the right window.
A final caution remains unavoidable: this is a preprint, not a peer-reviewed clinical directive. The study used valuable long-term cohort data and a sophisticated statistical framework, but it still leaves open how well these pathway distinctions will replicate across settings, how strongly they will translate into individual prediction, and which environmental modifiers may blunt or amplify each genetic route to Alzheimer’s disease and related dementias.