There is a comfortable assumption about dementia: that it is a disease of the seventies and eighties, and that reaching ninety with your mind intact means you have effectively escaped it.
A study following more than 800 people past their ninetieth birthdays finds that is not true — and that the risk factors operating in the tenth decade of life are not the ones the field expected.
The study
LifeAfter90 is a prospective cohort study run by researchers at UC Davis Health, published in The Lancet Healthy Longevity online on 4 August 2026 (DOI 10.1016/j.lanhl.2026.100882).
- More than 800 participants, all Kaiser Permanente members, aged 90 or older, median age 92 at enrolment.
- All were dementia-free at the start.
- Cognitive assessments every six months, drawing on health records reaching back decades.
- Running since 2018. Lead author Hilary L. Colbeth; senior author Rachel Whitmer, PhD, professor of public health sciences and neurology.
- Funded by the National Institute on Aging (R01AG056519, P30AG072972).
The design matters. Studies of very old populations are rare and difficult, because the cohort is small, attrition is high, and enrolling people already past ninety means only survivors are studied — a selection effect the researchers have to work around rather than eliminate. The historical records reaching back decades are what make it possible to connect midlife exposures to tenth-decade outcomes.
What they found
Sex. Women faced approximately double the dementia risk of men even after age 90. This is not simply because women live longer — the comparison is within a cohort already past ninety.
Race and ethnicity. Black participants had about 75 per cent greater risk than Asian participants. The disparities documented in midlife populations did not narrow with extreme age.
Genetics, unexpectedly. APOE is the strongest known genetic risk factor for Alzheimer’s disease. In this cohort:
- APOE4, normally the high-risk variant, was not a significant risk factor across the cohort as a whole.
- But it did raise risk among men specifically, and nearly doubled risk among Black participants.
- APOE2, the protective variant, cut risk by roughly 60 per cent across all groups.
Why the APOE4 result is the interesting one
The finding that APOE4 loses its overall predictive power after ninety, while remaining potent within particular subgroups, is genuinely counterintuitive and has a plausible explanation.
It is likely, at least in part, a survival effect. If APOE4 raises dementia risk substantially in the seventies and eighties, then people who carry it and reach ninety dementia-free are, by definition, unusual — they carry something else protective, genetic or environmental, that has offset it. The variant’s average effect in a ninety-plus cohort is therefore muted, not because the biology changed but because the population was filtered.
That it remains powerful among Black participants and among men points to something the averaging obscures. It is also consistent with a longer-standing observation in the literature: APOE4’s effect on Alzheimer’s risk varies meaningfully across ancestral populations, and has generally been weaker in African-ancestry groups in earlier-life studies. Why the pattern appears to invert at extreme age is not explained by this study.
A necessary caution about categories
Race and ethnicity in epidemiology are social categories, not biological ones. When a study reports differing risk by racial group, it is capturing some mixture of genetic ancestry, lifetime socioeconomic conditions, healthcare access and quality, education, chronic stress, environmental exposure, and measurement artefacts in how dementia is diagnosed across populations.
The finding that disparities persist into the tenth decade is real and important. Attributing it to any single mechanism is not supported by this design, and the authors do not do so.
Colbeth’s framing is the accurate one: racial and ethnic disparities in dementia risk “continue into the tenth decade of life.”
What the researchers say they still cannot explain
The team is explicit that a central puzzle remains open: why some participants who carried classic midlife risk factors — hypertension, high cholesterol in their sixties and seventies — nonetheless remained cognitively intact into their nineties.
Whatever is protecting those people is not understood. It is arguably the most valuable question the cohort could answer, because a protective mechanism is a target for intervention in a way that an unmodifiable risk factor is not.
Whitmer’s stated motivation is the practical one: “We need to understand who is most affected by dementia after 90.”
Limitations, stated plainly
Three constraints belong alongside the results.
The cohort is insured. Participants are Kaiser Permanente members — people with continuous health coverage and documented care histories. That is what makes the decades of records available, and it also means the findings may not transfer to uninsured or low-access populations, who are plausibly at higher risk and are absent from the sample.
Only survivors are observed. Everyone in the study reached ninety without dementia. They are not a random sample of people born ninety years ago; they are the ones who made it. This limits what can be inferred about the general population.
The precise statistics here come from institutional summaries. The subgroup figures reported above — the doubled risk for women, the 75 per cent difference, the 60 per cent APOE2 reduction — are drawn from UC Davis Health’s description of the paper. Exact hazard ratios and confidence intervals from the peer-reviewed text could not be independently confirmed for this article. Readers who need the precise effect sizes should consult the paper directly.
Why this matters now
The population aged 90 and over is among the fastest-growing age groups in most high-income countries. It is also among the least studied, because clinical trials and cohort studies routinely impose upper age limits and because recruitment is hard.
The practical consequence is that a growing population is being cared for on the basis of evidence generated in younger people. Findings like the APOE4 result illustrate the danger: a risk factor that is well established at 70 may behave differently at 92, and clinical guidance built on the former may misdirect at the latter.
What comes next
The research team says it is now examining how APOE genotype affects mortality risk in people with and without dementia after age 90 — a follow-on analysis of the same cohort, and one that speaks to a question families actually ask.
What APOE is and what it does
The gene is central to Alzheimer’s research and is worth understanding beyond the label.
APOE codes for apolipoprotein E, a protein that transports lipids and cholesterol in the bloodstream and, importantly, in the brain. It exists in three common variants: ε2, ε3 and ε4. Everyone carries two copies, one from each parent.
ε3 is the most common and serves as the reference. ε4 is associated with substantially increased Alzheimer’s risk — one copy raises it, two copies raise it considerably more. ε2 appears protective.
The mechanism is not fully settled, but the leading account is that the ε4 variant is less effective at clearing amyloid-beta from the brain and may promote its aggregation, while also affecting lipid handling, inflammation and blood-brain barrier integrity.
Crucially, APOE4 is a risk factor, not a diagnosis. Many carriers never develop Alzheimer’s. Many people with Alzheimer’s carry no ε4 allele. It shifts probability; it does not determine outcome — which is why genetic testing for it is not routinely recommended outside research, and why the finding that it loses predictive power after ninety is scientifically interesting rather than clinically actionable.
Survivor bias, explained properly
The central interpretive issue in this study deserves more than a sentence, because it applies to almost all research on the very old.
Everyone in the cohort reached ninety without dementia. That is not a neutral entry criterion; it is a severe filter.
Suppose a risk factor substantially raises dementia risk in the seventies and eighties. People who carry it and develop dementia at 78 are, by construction, never enrolled. The carriers who do enrol are those who somehow avoided that outcome — and something must explain how. Whatever that something is, protective genetics, favourable environment, or chance, it is over-represented among carriers in the sample.
The measured effect of the risk factor within the cohort is therefore attenuated, not because its biology changed at ninety but because the sample no longer represents carriers generally.
This is the most plausible reading of the APOE4 result, and it is why the finding should not be summarised as “APOE4 stops mattering after 90.” A more accurate statement is that among people who reached ninety dementia-free, APOE4 no longer distinguished who would go on to develop it — which is a narrower and more useful claim.
That it did still distinguish outcomes among Black participants and among men is the genuinely unexplained part, and the part worth following.
Why the very old are studied so little
The evidence gap this study addresses is structural, and it is larger than most people assume.
Clinical trials routinely impose upper age limits, often at 75 or 80. The stated reasons are defensible on their own terms: multiple coexisting conditions complicate interpretation, polypharmacy creates interaction risks, higher mortality produces attrition, and consent can be complicated by cognitive change.
The cumulative effect is that the fastest-growing age group in most high-income countries is treated on evidence generated in younger people, with the assumption that findings extrapolate.
Results like this one show that assumption failing. A genetic risk factor with well-established effects at 70 behaves differently at 92. Drug metabolism, comorbidity interactions and competing mortality risks all differ too. Extrapolation is a convenience, not a finding.
What is actually modifiable
Since none of the factors in this study can be changed — sex, ancestry, genotype — it is worth stating what the wider evidence says can be, without overclaiming.
Large reviews have converged on a set of modifiable risk factors associated with dementia risk across the life course: hearing loss, hypertension, smoking, obesity, physical inactivity, diabetes, excessive alcohol, depression, social isolation, air pollution, traumatic brain injury and limited education.
Two honest caveats. These associations come largely from observational data, and observational association is not proof of causation — early undiagnosed dementia can itself cause social withdrawal and inactivity, reversing the presumed direction. And no combination eliminates risk; the estimates concern population-level reduction, not individual guarantees.
What can be said is that several of these are worth addressing regardless of dementia — blood pressure control and hearing correction both have clear independent benefits — which makes them reasonable to act on without needing the dementia question settled.
This article reports research findings and is not medical advice. Genetic testing for APOE has significant implications and is best discussed with a clinician or genetic counsellor.
Sources
- Colbeth, H.L. et al., “Incidence of dementia after age 90 years and association with APOE genotype, race, and sex in the USA: the LifeAfter90 prospective cohort study,” The Lancet Healthy Longevity, 4 August 2026 — thelancet.com
- UC Davis Health, “Dementia after age 90,” August 2026 — health.ucdavis.edu
- ScienceDaily, “Reaching 90 without dementia doesn’t mean you’re in the clear,” 29 August 2026 — sciencedaily.com
- Psychiatry Advisor, “Study Identifies Dementia Risk Factors in Those Older Than 90 Years,” 2026 — psychiatryadvisor.com
