The Better the Study, the Smaller the Effect: What 20 Years of Hormone-and-Alzheimer's Research Actually Shows

The August 2026 paper is a retrospective cohort study drawing on two long-running research databases: the National Alzheimer's Coordinating Center and the Alzheimer's Disease Neuroimaging Initiative. Its authors compared women aged 50 and over who self-reported using estrogen-only hormone therapy against women who reported no use, and looked at neuropathology collected at autopsy alongside clinical diagnoses and fluid biomarkers.

The primary result was a genuine one. Among the women with autopsy data, hormone therapy users had significantly lower odds of showing increased Alzheimer's pathology, an odds ratio of 0.65 with a confidence interval running from 0.48 to 0.88. That is the 35% figure. The same analysis found lower odds of a clinical dementia diagnosis and lower odds of reported memory or functional decline, with amyloid measured in plasma and cerebrospinal fluid pointing the same way.

The researchers were notably careful about what this means. They described their findings as small but significant associations, and wrote plainly that their results "do not address causality and have limited generalizability." Senior author Hadi Hosseini told Science News that the findings "should not be taken as proof that hormone therapy prevents Alzheimer's disease". The distance between that sentence and the headlines it generated is most of what this article is about.

One Scope Limit Worth Holding Onto

The study examined estrogen-only therapy, which in practice is prescribed almost entirely to women who have had a hysterectomy. The researchers did not have enough data to analyze combined estrogen-plus-progestogen therapy, which is what roughly 80% of hormone therapy users take because they still have a uterus. So the finding describes a clinically distinct group, and it is silent on the formulation most women encounter. If you have followed how medical thinking on hormone therapy has shifted over the past two decades, you will recognize this as the kind of detail that tends to disappear between a paper and its coverage.

Both of the figures that traveled furthest this month are checkable in a few minutes, and both come back slightly different from how they were reported.

Where 21,462 Came From

Much of the coverage described the study as analyzing 21,462 participants. That number appears in the institution's own news release about the research, and it accurately describes the combined size of the two databases the researchers drew from. It does not describe the number of women in the analysis.

The paper reports its groups directly. The autopsy comparison rests on 258 hormone therapy users against 2,701 non-users. The biomarker cohort adds 110 users and 1,948 non-users. Add them together and the analytic sample is roughly 5,017 women, with the headline neuropathology result resting on 258. That is not a small study, and the finding is not invalidated by its size. But 258 and 21,462 produce very different impressions of how settled something is, and only one of them is what was measured.

The ages are worth the same attention. In the autopsy sample, the mean age of death was 81.9 years among hormone therapy users and 82.2 among non-users. This is a study of women who lived into their eighties, which is a meaningful distance from the perimenopausal reader most likely to have the headline forwarded to her, and from what the research shows about memory during the menopause transition itself.

Where 58 Percent Came From

The second figure has a different problem: it is not new, and it is not about Alzheimer's specifically. It comes from a 2021 analysis of a ten-year Humana insurance claims dataset covering 379,352 women, which reported a relative risk of 0.42 for neurodegenerative disease among hormone therapy users.

That 0.42 is the 58% reduction. But the outcome it describes is a pooled category combining Alzheimer's disease, Parkinson's disease, dementia, multiple sclerosis and ALS. Those are five conditions with different mechanisms, different ages of onset, and no shared biology that a single hormone would be expected to move all at once. A 58% reduction across all of them, in billing records, is an extraordinary claim about human physiology. It is a much more ordinary claim about who fills prescriptions and stays continuously insured.

Here is every major investigation of hormone therapy and dementia from the past two decades, arranged not by date but by how much control each design has over confounding. Randomized trials sit at the top because assigning treatment by chance is the only method that reliably separates a drug's effect from the characteristics of the people who take it. Below them sit systematic reviews, then large observational studies, then retrospective cohorts assembled from research volunteers.

The Rung That Proves the Point on Its Own

If the ladder above looks like an argument built by choosing convenient studies, there is a cleaner way to see it. One 2023 meta-analysis pooled both kinds of evidence separately, using a single set of inclusion rules. Its six randomized trial reports, covering roughly 42,000 participants, produced a pooled relative risk of 1.38 — increased dementia risk. Its 45 observational studies, covering 768,866 cases, produced a pooled relative risk of 0.78 — reduced Alzheimer's risk.

Same authors, same protocol, same question, two evidence types, and the estimates land on opposite sides of no effect. The gradient is not something assembled here from selectively chosen papers. It is visible inside one publication, by a team whose broader work is sympathetic to hormone therapy's potential.

Four randomized trials have tested hormone therapy against placebo and then measured cognition. Read in order, they answer the objections readers most often raise.

The first and largest was the Women's Health Initiative Memory Study. It randomized 4,532 women aged 65 and older to combined therapy or placebo, and reported a hazard ratio of 2.05 for probable dementia — the finding that reshaped a generation of prescribing. It also deserves the same scrutiny applied to everything else here: that doubling rests on 61 dementia diagnoses in total, 40 in the hormone group and 21 in the placebo group. The direction is clear and the precision is not.

The obvious objection is that these women were 65 or older, and the theory in circulation holds that hormone therapy protects the brain when started near menopause. Three trials have tested exactly that. A 2013 follow-up assessed 1,326 women who had been randomized at ages 50 to 55 and found no overall sustained benefit or risk to cognitive function.

The ELITE trial went further and randomized women into an early group within six years of menopause and a late group more than a decade past it. Over five years of testing, it found no meaningful difference from placebo in either group. Lead investigator Victor Henderson of Stanford said the results suggest "healthy women at all stages after menopause should not take estrogen to improve memory".

Finally, the KEEPS Continuation study returned to 275 women roughly a decade after they were randomized in early menopause and found the effects on cognitive change to be, in the authors' words, "all statistically nonsignificant." Its own limitations are real and worth stating: only 41% of the original participants returned, those who returned had scored higher at baseline, and the group was predominantly white, well-educated and at low cardiovascular risk.

In December 2025, a systematic review commissioned by the World Health Organization screened 5,914 records and pooled the ten studies that met its quality bar, covering 1,016,055 participants. Its finding was flat: no significant association between hormone therapy and mild cognitive impairment or dementia, with subgroup analyses by timing, duration and formulation also showing nothing.

The review's authors were careful about their own certainty, rating the evidence from moderate to very low, and they were explicit that important benefits and harms remain possible. That caution is part of why the result is credible rather than a reason to discount it. Their conclusion was that hormone therapy prescription "should be based on other perceived benefits and risks and not for dementia prevention" — which is also the position taken by current UK clinical guidance and by the 2024 Lancet Commission on dementia.

That framing is worth sitting with. It does not say the evidence is discouraging; it says dementia prevention is not among the things this particular decision should turn on, and that the reasons to consider or decline hormone therapy are the same ones that applied last year. Readers looking for where the evidence on brain health is stronger will find it in the everyday habits the research genuinely supports.

This review was published eight months before the news cycle described at the top of this article. It is larger, more rigorous, and more recent than most of the evidence being cited around it, and it went almost entirely uncovered. Newness and rigor are different properties, and only one of them reliably drives a headline.

The gap between the two kinds of evidence is not a mystery to the researchers who produce it. It has a name and a well-documented mechanism.

The Healthy User Problem

Women who take hormone therapy, and especially those who stay on it for years, differ systematically from women who do not. As a group they tend to be more educated, more affluent, better insured, more engaged with preventive care, and more consistent about taking prescriptions as directed. Every one of those characteristics independently predicts better cognitive outcomes in later life, entirely apart from any hormone.

The authors of that 2023 meta-analysis named this directly in their own limitations, citing "confounding by indication" and "healthy user-bias" and noting that hormone therapy users tend to be healthier and more socioeconomically advantaged. They concluded that the literature as a whole provides "an overall low-level evidence." The WHO-commissioned review made the same point about why observational findings keep failing to replicate: it happens, in its phrasing, "largely due to healthy user bias." Randomization removes the difference between the groups. It also removes the effect.

The Prescription Problem

There is a second, subtler version of the same issue, and one large study illustrates it well. A 2023 study using Danish national registries compared 5,589 dementia cases against 55,890 matched controls and found that women who had received estrogen-progestogen therapy had a higher rate of dementia, with a hazard ratio of 1.24 that climbed to 1.74 among the longest-duration users. The association persisted among women treated at 55 or younger.

That result matters for two reasons. First, it is a very large, carefully conducted observational study in the same tier as the claims-database analysis, pointing the opposite way — which means observational studies do not merely disagree with trials, they disagree with each other. Second, its authors offered the cleanest available explanation of why. Further research is needed, they wrote, to determine whether the findings reflect a real effect or "an underlying predisposition in women in need of these treatments." Women who are prescribed hormone therapy may differ from women who are not in ways that affect dementia risk on their own, before the medication is considered at all.

There is a third clue hiding in the claims-database study itself. The theory that hormone therapy protects the brain depends on early initiation; benefit is supposed to shrink or reverse when treatment starts late. That analysis found its greatest apparent risk reduction in women aged 65 and older — the opposite of what the theory predicts. When a result runs against the mechanism proposed to explain it, that is usually a signal about the data rather than about biology.

The newest study's neuropathology came from a network of research centers whose participants agreed, often years in advance, to enroll in aging studies and donate their brains. That is an enormous scientific gift, and it also produces a group of people who are not much like the general population.

The scale of that effect is measurable. A 2024 analysis of 7,178 autopsied participants across 28 research centers found that how a center recruits its volunteers systematically predicts what its autopsy cohort ends up showing, with differences in age, sex, genetics and co-occurring conditions falling out of recruitment strategy alone. Its authors concluded that "it is essential to consider the characteristics of autopsy cohorts."

Two details make this more than a generic caveat. The analysis used the same database that supplied the 2026 study's neuropathology, and one of its authors directs that database. And the editorial Neurology published alongside the new study cites this selection-bias work directly, under a title that asks when, which, and for whom — the journal's own commentary treated the finding as an open question rather than a resolved one.

It is worth being precise about the limits of this argument too. Randomized trials of hormone therapy run for a handful of years, while Alzheimer's disease develops across decades, so a null result over five or ten years has not ruled out something slower. The absence of evidence for a protective window is not evidence that no such window exists. And the trials have their own generalizability problems — a 2025 analysis in PLOS Medicine noted that participants in this literature have been overwhelmingly white, well-educated and healthy, and that European and UK guidance currently disagree with each other about brain-health claims.

What the pattern does support is a way of reading. When a striking effect appears in a study that observes who happened to take something, and shrinks or vanishes in studies that assign it at random, the most economical explanation is usually that the effect belonged to the people rather than the pill. That is the shape of this literature, and it is the same shape that has shown up in other nutrient-and-dementia headlines that arrived with similar confidence.

Estrogen's role in the aging brain remains a serious scientific question, and better-designed trials may yet find something these have not. For now, the honest description of 20 years of research is that the strongest studies keep finding the smallest effects — and that noticing which kind of study you are reading is a skill worth more than any single result it reports.