What does menopausal hormone therapy improve? Timing and longevity evidence

Our assessment: timing matters, but the evidence establishes neither universal life extension nor an absence of every possible benefit in every age group. ELITE supports an arterial-wall benefit in a particular early-menopause population; other early-treatment trials are not uniformly positive. Overall long-term WHI mortality is close to null, while uncertainty remains in younger women. We checked primary reports, age-stratified figures, registration and aggregate calculations, separating statistical estimands from clinical extrapolation.

01Treatment was randomized; starting now versus a decade later was not

ELITE randomized treatment separately among women less than six years and at least ten years past menopause. This can test whether treatment effects differ between these populations. It did not randomize comparable women to start immediately or ten years later. The strata already differ in age, vascular state and other characteristics, so heterogeneity does not establish an individual's optimal starting year. ELITE

WHI age at randomization is also different from ELITE's years since menopause. Some WHI participants had used hormones before enrolment, making randomization age different from first-ever use. Ages 50–59 cannot simply replace “less than six years since menopause.”

Regimens differ too. ELITE used oral 17β-estradiol, with cyclic vaginal micronized progesterone for women with a uterus. WHI studied oral conjugated equine estrogens, or CEE, plus medroxyprogesterone acetate, or MPA, and CEE alone in women with prior hysterectomy. ELITE's progestogen was not MPA. These are not interchangeable interventions.

02Arterial measurements: ELITE supports heterogeneity, not a universal early benefit

ELITE randomized 643 women; 596 had carotid intima-media thickness, or CIMT, follow-up for the primary analysis. Median intervention lasted approximately five years.

The formal treatment-by-menopause-stratum-by-follow-up-time interaction has p=0.007. Assuming independent, approximately normal stratum estimates, the reported intervals give a difference in treatment slopes of −0.0046 mm/year, approximate 95% CI −0.00799 to −0.00121, and p≈0.0078. This targets the same between-stratum difference in treatment slopes, not another estimand. It is an aggregate approximation, not a refit of the participant-level mixed model or recovery of its full covariance structure.

The late-stratum interval allows small benefit or harm; nonsignificance does not establish exactly zero effect. CIMT is a subclinical vascular measurement, so improving it does not directly demonstrate fewer myocardial infarctions. ELITE's CT measures of coronary calcium, stenosis and plaque showed no clear treatment advantage in either stratum. Those findings belong alongside the positive CIMT result. Results and Table 2

KEEPS provides important contrasting evidence. It randomized 727 healthy women six to 36 months after their last menstrual period to lower-dose oral CEE, transdermal estradiol, or placebo; active regimens included cyclic oral micronized progesterone. Its formal abstract reports similar mean CIMT increases of approximately 0.007 mm/year across groups over four years, without clear differences in coronary-calcium progression despite some metabolic improvements. We did not obtain the full report and use these results at abstract level. Differences in dose, population and follow-up make this an imperfect replication of ELITE, but it limits the claim that starting early guarantees arterial benefit from any regimen. KEEPS

The timing hypothesis proposes favourable effects on early vascular disease and endothelium but potentially different effects on advanced plaques. Some trial patterns are compatible with that mechanism. They do not establish one causal chain connecting CIMT, lipids and clinical events. Coagulation, breast tissue, bone and symptoms can also respond through different pathways, rather than one measurable “ageing speed.”

03Cognition: no support for the proposed timing benefit, rather than proof of exact zero

ELITE-Cog's modified intention-to-treat analysis included 567 women, with mean treatment duration of 57 months. The standardized verbal-memory difference was −0.06 (95% CI −0.22 to 0.09; p=0.33), with timing-interaction p=0.88. Executive function and global cognition also did not support benefit. Cognitive report

This does not support the hypothesized cognitive improvement from that regimen in these healthy women. It does not exclude every small effect or every formulation, and did not directly test dementia prevention decades later. The paper's “Class I” designation describes its evidence classification, not certainty of no biological effect. Cognition is another outcome here, not automatically a formal negative-control outcome known to be unaffected by treatment.

WHI dementia findings concern an ancillary population aged at least 65: HR approximately 2.01 (1.19–3.42) for CEE+MPA and 1.47 (0.85–2.52) for CEE alone. They must be retained without claiming dementia harm has been established at every age. WHI synthesis

04WHI: more absolute events need not mean a different relative treatment effect

The two WHI trials randomized 16,608 women with a uterus and 10,739 with prior hysterectomy, with median interventions of 5.6 and 7.2 years. Each regimen had its own placebo comparison. CEE versus CEE+MPA was not a head-to-head randomized comparison.

The following consistently uses the intervention phase of the 2013 report. HR confidence intervals are not simultaneous intervals adjusted across every outcome.

Both regimens also reduced hip fractures and increased gallbladder disease and urinary incontinence. Benefits and harms need joint consideration. A lack of clear age interaction does not mean every age subgroup separately demonstrates the same statistically significant harm: younger groups have fewer events and often wider intervals. Nor does it guarantee that early use is risk-free.

The global index records the first monitored event among coronary disease, breast cancer, stroke, pulmonary embolism, selected cancers, hip fracture or other death. It does not weight symptom relief, event severity and personal preferences into an individual net-benefit score. Definition and original figures

At ages 50–59, 60–69 and 70–79, CEE+MPA's absolute excesses are +12, +22 and +38 events per 10,000 person-years, but corresponding HRs are approximately 1.12, 1.13 and 1.12, with age-trend p>0.99. An absolute gradient does not establish relative effect modification by age. CEE alone gives −19, −1 and +51, with age-trend p=0.02, a different finding. Age and time-since-menopause analyses are also distinct.

A simple counterexample shows why. Hold treatment HR at 1.2 across three hypothetical groups with baseline hazards of 5, 10 and 20 per 1,000 person-years. Their absolute excess rates become 10, 20 and 40 per 10,000 person-years. These are stated assumptions, not a fit to WHI or predictions for individual women. Absolute differences alone cannot establish relative interaction.

05Mortality: the overall result is near null, with uncertainty in younger women

WHI's 18-year cumulative follow-up recorded 7,489 deaths. The pooled all-cause HR was 0.99 (0.94–1.03), with 1.02 (0.96–1.08) for CEE+MPA and 0.94 (0.88–1.01) for CEE. No clear overall mortality difference was detected. This does not establish an exactly zero effect or supply an estimate of added years of life. 18-year report

Age-trend p values are 0.01 and 0.06. The younger group's cumulative interval still allows meaningful benefit. Crossing the 0.05 significance threshold does not prove that benefit disappeared. The windows overlap and cannot be treated as independent estimates.

Likewise, 1,088/7,489 deaths during intervention means approximately 14.5% of deaths occurred in that phase; it does not identify what proportion of treatment benefit occurred or disappeared when. “Intervention phase” refers to follow-up under original randomized assignment, not verified uninterrupted medication use. Attributing an effect change to stopping requires appropriate period comparisons and information on adherence and crossover, rather than the distribution of total deaths.

The reported reduction in postintervention dementia-related mortality is another endpoint, not the inverse of incident dementia in the older ancillary study. The authors specifically caution about cause-of-death underreporting, competing mortality and multiple comparisons. We do not treat it as established neuroprotection.

Positive counterevidence also matters. DOPS randomized 1,006 recently menopausal women. At ten years, death/heart-failure admission/myocardial-infarction composite events numbered 16 versus 33, HR 0.48 (0.26–0.87); deaths alone numbered 15 versus 26, HR 0.57 (0.30–1.08). This supports further investigation of potential early-treatment clinical benefit, while a positive composite does not establish mortality benefit separately. The study was open-label with few events. The accessible registration was posted after trial initiation and cessation of the randomized intervention and lists fracture/bone density as primary outcomes; it cannot verify prospective registration of the cardiovascular composite. The report states that cardiovascular safety outcomes were prespecified, but we did not obtain the original protocol to resolve this further. DOPS report Registration

Sparse safety events also cannot prove absence of harm: deep-vein thrombosis occurred in two versus one participant, with an HR interval of 0.18–22.16. DOPS had no randomized late-initiation group, so it cannot establish timing interaction by itself.

06Longer breast follow-up reinforces the need to distinguish regimens

The 2020 WHI breast report extends follow-up beyond twenty years. In its own trial population, CEE alone had breast-cancer incidence HR 0.78 (0.65–0.93) and breast-cancer mortality HR 0.60 (0.37–0.97). CEE+MPA had corresponding HRs of 1.28 (1.13–1.45) and 1.35 (0.94–1.95). A nonsignificant mortality interval for the combined regimen does not establish no mortality risk. Long-term report

Different uterus-status populations prevent this contrast from isolating MPA's causal contribution or transferring the estrogen-only result directly to women with a uterus. More than 98% ascertainment concerns deaths; extended breast-cancer incidence follow-up still required renewed consent, addressed through weighting and other sensitivity analyses. The 2013, 2017 and 2020 papers follow the same WHI project, not three independent randomized replications.

07Denmark 2026: three mortality ratios, rather than one confounding calculation

The Danish cohort includes 876,805 women, of whom 104,086 redeemed a hormone prescription during follow-up. Table 2 contains three quantities:

• Pooled mortality rates of approximately 54.9 versus 35.5 per 10,000 person-years; original deaths/person-time give a rate ratio of 1.548.
• The Cox model labelled “crude,” which already uses age as the underlying time axis: HR 1.04 (1.01–1.06).
• The further multivariable-adjusted Cox model: HR 0.96 (0.93–0.98).

The pooled rate ratio and age-timescale HR use different comparisons and time weighting. The shift from 1.55 to 0.96 is not simply the same estimand before and after one adjustment, and cannot quantify the fraction “explained by confounding.” Population composition and analysis choices matter; adjustment still does not equal randomization. Methods and Table 2

Cumulative treatment duration is updated over time, with time before first prescription classified as unexposed. After cessation, the primary analysis retains prior/current use status. This avoids simply assigning future treatment time as already exposed, while leaving potential unmeasured confounding, treatment-affected covariates and stopping selection. The authors' use of “intention-to-treat” for that exposure definition does not supply randomized-assignment protection.

The sibling subset of 57,773 women gives adjusted HR 1.06 (0.94–1.19); censoring at cessation gives 0.83 (0.76–0.89). Populations, follow-up and selection differ, so the most favourable estimate cannot simply be preferred. Median age at the end of follow-up was only 59.4 years, limiting inference about older-age mortality and lifetime effects.

In the bilateral-oophorectomy-at-45–54 subgroup, 703 counts deaths (386+208+109), not all subgroup participants. Median ages of 60.9 versus 56.6 among exposed and unexposed decedents condition on having died and cannot be interpreted as 4.3 additional years of life. This surgical population is also not interchangeable with natural menopause generally.

08Where the conclusions should stop

These studies support evaluating regimens, populations, symptoms, fractures, vascular outcomes, cognition and mortality separately. Some earlier populations may have more favourable findings, but “earlier” alone does not guarantee efficacy for every regimen. Increasing absolute risk with age does not always mean relative treatment effects interact with age.

Universal life extension has not been established. Conversely, nonsignificant overall or subgroup results do not exclude every possible benefit. Transdermal estradiol and micronized progesterone are not wholly absent from randomized evidence—KEEPS includes them. The important gaps concern sufficiently large, long-duration comparisons of relevant clinical events and mortality, which routes of administration and intermediate markers cannot fill by themselves.

Scope and limitations: aggregate reanalysis and explicit hypothetical risk scenarios; no participant data, mixed/Cox model refits or real life-years calculation. Some subgroup, multiple-endpoint and extended-follow-up results face selection and precision limitations. KEEPS was read at formal-abstract level; the original DOPS protocol was unavailable. This is not an exhaustive assessment of every special population or regimen and provides no personal medication advice.

Download checked inputs, calculations, independent validation and six figures.

09Sources

• <a id="source-s1"></a>s1: Hodis HN, et al. Vascular Effects of Early versus Late Postmenopausal Treatment with Estradiol. NEJM 2016. Full text. Primary model, Table 2, null CT results, regimen and limits; late-stratum interval/p value checked.
• <a id="source-s2"></a>s2: Henderson VW, et al. Cognitive effects of estradiol after menopause. Neurology 2016. Full text. Methods, results, regimen and power limitations read.
• <a id="source-s3"></a>s3: Manson JE, et al. Menopausal hormone therapy and health outcomes during the intervention and extended poststopping phases. JAMA 2013. Full text. Phases, global-index definition, original Figures 2/5a and age trends checked.
• <a id="source-s4"></a>s4: Manson JE, et al. Menopausal Hormone Therapy and Long-term All-Cause and Cause-Specific Mortality. JAMA 2017. Full text. Overall/age-specific estimates, cumulative/period definitions, cause-of-death and multiplicity limits.
• <a id="source-s5"></a>s5: Mikkelsen AP, et al. Menopausal hormone therapy and long term mortality. BMJ 2026. Full text. Methods, Table 2, time updates, sibling/cessation sensitivities, the 703 deaths and follow-up age.
• <a id="source-s6"></a>s6: Harman SM, et al. Arterial imaging outcomes and cardiovascular risk factors in recently menopausal women: a randomized trial. Ann Intern Med 2014. Formal abstract. KEEPS design/regimens and null imaging outcomes; full report not obtained and no participant reanalysis.
• <a id="source-s7"></a>s7: Schierbeck LL, et al. Effect of hormone replacement therapy on cardiovascular events in recently postmenopausal women: randomised trial. BMJ 2012. Official report. Read available official abstract/methods/results/discussion; direct HTML fetch restricted, readable web extraction retained. Original protocol unavailable.
• <a id="source-s8"></a>s8: Chlebowski RT, et al. Association of Menopausal Hormone Therapy With Breast Cancer Incidence and Mortality During Long-term Follow-up of the WHI Randomized Clinical Trials. JAMA 2020. Full text. Incidence/mortality, reconsent and sensitivities; same WHI populations.
• <a id="source-s9"></a>s9: ClinicalTrials.gov, DOPS NCT00252408. Current API record: study start in 1990, first posting in 2005, primary fracture and bone-density outcomes. Current registration does not reconstruct a prospective historical protocol.