Menopause Racial Disparities: What U.S. Data Show in 2026

Menopause racial disparities are real and measurable across at least six separate outcomes — but the size of any gap depends on what was measured and how the study was designed. Among 1,449 SWAN participants who reported frequent hot flashes or night sweats, the study’s African American group had a median symptom duration of 10.1 years versus 6.5 years in its non-Hispanic White group. That is a 3.6-year, or 55.4%, difference— our calculation from the published medians.

Menopause timing is less simple. A 2013 SWAN analysis found a 0.26-year adjusted difference in natural final-menstrual-period medians between its African American and Caucasian groups and no statistically significant adjusted racial or ethnic difference overall. A 2023 analysis of the same cohort, built to correct for enrollment exclusions and unobserved final menstrual periods, predicted natural-menopause medians of 51.4 years for Black participants and 52.0 years for White participants and reported a 1.2-year overall Black–White difference in a framework that included both natural and surgical menopause.

Both of those things are true. Neither is contradictory. Understanding why is the point of this page.

One sentence that captures the whole page:Black participants in SWAN had longer frequent vasomotor symptoms, while Black groups in national VA and NHANES analyses had lower measured documentation, systemic-hormone prescribing or MHT use than the studies’ White comparison groups — but those are separate outcomes from separate populations, not one causal chain.

U.S. research documents racial and ethnic differences across six distinct outcomes — symptom frequency, symptom duration, symptom severity, menopause timing, clinical documentation and prescribing, and population-level hormone therapy use. These findings come from different populations, time periods and study designs. They should not be collapsed into a single “racial disparity percentage” or a single “average menopause age by race.”

A common error is to blur six different things together. Here is what each one actually means:

• Frequency— how often symptoms were reported.
• Duration— how many total years frequent symptoms persisted.
• Severity— how strongly symptoms were rated by the people experiencing them.
• Timing— the age at the final menstrual period, which can be natural or surgical.
• Documentation and prescribing— what appeared in medical records and what clinicians prescribed.
• Population use— the share of a defined population currently using, or reporting a history of using, hormone therapy.

A person can have symptoms that are more frequent but not more severe. A group can have a large gap in prescribing and no gap in natural-menopause timing. Treating these as one phenomenon is why two well-sourced articles can appear to contradict each other.

Two terms worth defining now. Vasomotor symptoms (VMS) are hot flashes and night sweats. Menopausal hormone therapy (MHT) is the scientific term used here; hormone replacement therapy (HRT) is the common alternative term. The included studies operationalized MHT through prescription records or self-reported use of female hormones for menopause, so each estimate keeps its own source definition.

Because different adjustments answer different questions. In the studies below, adjusting for measured health and socioeconomic covariates shrank one natural-menopause timing difference; correcting for who was excluded from the SWAN cohort expanded the estimated timing difference; adding one neighborhood-affluence measure barely changed a symptom-severity estimate; and adding insurance attenuated a Black–White MHT-use estimate. None of those adjustments is interchangeable with another.

This is one of the most useful findings on this page, so it belongs up front rather than buried in a methods note.

We identified four places where researchers published a disparity estimate before and after a consequential analytical change. Lining them up shows what disappears when the papers are read one at a time.

What this means in practice:

• When a covariate-adjusted estimate moves toward zero, the model has statistically accounted for part of the difference using the variables it included. That does not prove those variables caused the original difference.
• When a selection correction makes a difference larger, the study is saying the original sample missed people in a way that distorted the result.
• When one neighborhood proxy changes an estimate by 0.04, that model did not erase the association. It also did not test every socioeconomic or structural condition that could matter.
• When an odds-ratio confidence interval crosses 1 after insurance enters a model, the study no longer detects a statistically significant group difference in that model. It does not prove the groups became equivalent.

The practical rule:the model must travel with the number. “The unadjusted medians differed by 8.5 months” and “the 2013 study found no significant adjusted racial difference” are both accurate descriptions of the same paper.

SWAN produced different Black–White timing figures because the analyses measured different outcomes and corrected different biases. For natural final menstrual periods, the 2013 unadjusted median difference was 0.71 years and the covariate-adjusted difference was 0.26 years, with no significant adjusted racial or ethnic difference. A 2023 selection-bias analysis predicted a 0.60-year natural-FMP difference, a 1.80-year surgical-FMP difference and a 1.20-year overall difference. Those values must not be presented as a single range measuring the same thing.

The table is deliberately not labeled a “range.” The first three rows concern natural FMP, the fourth is an overall summary, and the fifth concerns surgical FMP. Comparing 3.1 months with 21.6 months as though they were low and high estimates of one identical outcome would be wrong.

This is the mechanism that makes the timing literature make sense.

To enter the longitudinal SWAN cohort, a participant had to be 42 to 52 years old, have no prior hysterectomy or bilateral oophorectomy, not be using reproductive hormones, not be pregnant, and have had a menstrual period in the previous three months. People who had already had a hysterectomy or bilateral oophorectomy could not enter under those rules, and natural FMP could no longer be observed after hysterectomy.

That exclusion did not fall evenly.

The Black surgical-menopause exclusion percentage was about 1.8 timesthe White percentage — our calculation from 30.9% and 17.0%.

Reeves and colleagues quantified the resulting distortion. Models that ignored selection overestimated menopause age by an average of 0.72 years overall, including 1.10 years for Black women and 0.50 years for White women. Because the size of the error differed by group, the error also changed the measured disparity.

Two technical terms, in plain language. Left truncation means people who had already experienced the relevant outcome before cohort entry could not be represented in the same way as those who entered outcome-free. Right censoringmeans some participants’ final menstrual periods were not observed during follow-up. Reeves et al. addressed left truncation with inverse-probability weighting and right censoring with multiple imputation.

One source-table discrepancy is worth recording because the numbers have been repeated elsewhere. The Reeves Results text reports Japanese eligibility as 54.4% and Black eligibility as 38.9%; Table 1’s counts reproduce as 55.4% and 38.8%. We use the reproducible Table 1 values. Table 2 also reverses the Chinese and Japanese cohort counts shown in Table 1. None of the Black–White calculations on this page depends on those reversed subgroup counts.

We could produce one. It would be widely shared and it would be misleading.

Group estimates of menopause age depend on whether surgical menopause is included, whether timing is observed or model-predicted, who was eligible for the cohort, how unobserved final periods were handled, the age range recruited, which covariates were adjusted and the historical period studied. Change any one and the number moves. A table that hides all of that is not a reference; it is a rumor with a citation attached.

Among SWAN participants who met the study’s threshold for frequent vasomotor symptoms, the published median total duration was 10.1 years for the African American group, 8.9 years for the Hispanic group, 6.5 years for the non-Hispanic White group, 5.4 years for the Chinese group and 4.8 years for the Japanese group. The African American median was 3.6 years, or 55.4%, longerthan the White-group median — a calculation from the published figures.

The arithmetic, shown so it can be checked:

The African American median was also 2.1 timesthe Japanese median — another calculation from the published point estimates.

What these medians do not tell you. The analysis included participants who reported frequent symptoms, defined as hot flashes or night sweats on at least six days during the previous two weeks. Not everyone experiences frequent symptoms. A group median is not an individual prognosis. The relative difference is not a risk ratio. And no confidence interval for our derived difference can be inferred from the published medians alone, so we present it as arithmetic rather than inventing uncertainty bounds.

SWAN found African American participants had 1.63 times the adjusted odds of frequent vasomotor symptoms compared with White participants. A separate 2024 analysis of a large online care-seeking sample found Black participants had 1.87 times the adjusted odds of severe hot flashes and 1.78 times the adjusted odds of severe night sweats. These are different constructs — how often symptoms occur versus how severe they are — measured in different populations.

The 2006 odds ratio must remain an odds ratio. Because frequent VMS were not rare, the authors warned that the odds ratios overstate relative risks. OR 1.63 does not mean a 63-percentage-point difference and should not be rewritten as one.

Kochersberger and colleagues ran their symptom-severity model with and without a ZIP-code neighborhood-affluence score. For severe hot flashes among Black participants compared with White participants:

• Without the neighborhood-affluence score: OR 1.91(97.5% CI, 1.75–2.09)
• With the neighborhood-affluence score: OR 1.87(97.5% CI, 1.72–2.04)

The point estimate moved by 0.04. That model did not erase the association. It also does not show that socioeconomic conditions are unimportant: the added variable was one ZIP-code affluence proxy, not a complete measure of income, wealth, discrimination, healthcare access or accumulated structural exposure.

Sample and conflict disclosure. The Evernow intake contained 68,864 registrants; 67,867 were included in the analysis after exclusions. Participants were English-speaking U.S. adults who sought care through one commercial telehealth platform, reported at least one symptom and were reached through digital marketing or word of mouth. Several authors held paid or leadership roles with Evernow, including its chief medical officer, founder and chief executive officer, chief science advisor and a paid medical advisor. The paper discloses those relationships. The sample supports symptom-severity associations among these care seekers; it is not a nationally representative prevalence survey.

Why the Indigenous/First Nations row uses 1.74.Table 3 reports OR 1.77 before the neighborhood-affluence variable was added, and Table 4 reports OR 1.74 after it was added. The abstract contains an inconsistent 1.39 value paired with Table 3’s 1.19–2.75 interval. The detailed tables resolve the models, so we use the Table 4 estimate because the other severity rows in this table use the same model.

The national VA and NHANES studies examined here each report racial or ethnic differences in documentation, prescribing or MHT use. In VA records, Black women veterans had lower adjusted odds of meeting the study’s documented-symptom definition and of receiving systemic hormone therapy. In NHANES 2017–March 2020, current MHT use was 0.5% among non-Hispanic Black women, 2.6% among Hispanic women and 5.8% among non-Hispanic White women. Lower measured use is a fact; whether it represents undertreatment in an individual case is not something these data can establish.

Blanken and colleagues analyzed Veterans Health Administration records for 200,901 women veterans ages 45 to 64, with a mean age of 54.3. The study defined documented menopause symptoms through relevant ICD-9 codes at two or more encounters. In the full cohort, 5.2% met that definition, 5.1% had a systemic hormone-therapy prescription and 5.0% had a vaginal-estrogen prescription during fiscal years 2014–2015.

That last row runs in the opposite direction from the systemic-therapy rows. Hispanic/Latinx women veterans had higher adjusted odds of vaginal-estrogen prescribing despite no detected difference in the study’s documented-symptom outcome. Any claim that minority groups uniformly received less of every menopause treatment would be wrong.

The 5.2% documentation figure also needs its definition attached. It means 94.8% did not meet the requirement of menopause-related codes at two or more VA encounters during the study period. It does not mean 94.8% had no symptoms, never mentioned symptoms or received no care outside VA records.

Yang and Toriola analyzed 13,048 postmenopausal women across ten NHANES cycles. Overall current MHT use fell from 26.9% in 1999–2000 to 4.7% in 2017–March 2020.

Here is the original comparison the separate percentages do not state on their own:

The absolute White–Black gap narrowed while the point-estimate ratio widened. The White-minus-Black difference fell from 19.5 percentage points to 5.3 percentage points. The White-to-Black point-estimate ratio rose from about 2.6 to 11.6.

That 11.6 is arithmetic from 5.8 divided by 0.5. It is not a confidence interval and it should not be treated as a precisely estimated population ratio. The source published separate subgroup confidence intervals, not an interval for this derived ratio.

Chesnokova and colleagues pooled NHANES 2013–March 2020 data for 1,666 respondents ages 45 to 64 with Medicaid or private insurance who met the paper’s study-defined eligibility rules. The weighted analytic population was about 22.3 million.

The study’s primary outcome was not current MHT use. It was self-reported history of ever using female hormones such as estrogen or progesterone for menopause, excluding birth-control and infertility use.

Medicaid-insured participants reported prior MHT use at 10.7% versus 20.7% for privately insured participants. After full adjustment, Medicaid coverage was associated with half the oddsof prior use: OR 0.50 (95% CI, 0.28–0.87).

Three things in that table matter:

The Black–White estimate stopped being statistically significant after insurance entered the model.It moved from OR 0.68 to 0.72, and the confidence interval then included 1. A formal adjusted mediation analysis reported an indirect effect through Medicaid coverage of −0.009 (95% CI, −0.017 to −0.002; P = .01), which the authors interpreted as insurance statistically accounting for part of the difference. Cross-sectional mediation does not establish a causal mechanism.

The lowest Model 1 point estimate belonged to the non-Hispanic Asian group.Its OR was 0.36. In the fully adjusted model the estimate was 0.74 with a wide confidence interval that included 1, so the study does not establish a persistent adjusted Asian–White difference.

The Hispanic point estimate crossed from below 1 to above 1 after full adjustment. The confidence interval spanned both directions. That is a warning against treating any one point estimate as a fixed group property.

The eligibility language also needs a hard limitation attached. Medical-history variables used to identify conditions that could preclude MHT were missing for 85.1%of respondents; participants were excluded when they affirmatively reported an exclusion condition, while those missing that information remained included. Separately, 239 of 1,905 otherwise eligible respondents — 12.5%— were missing the primary outcome and excluded. The authors’ sensitivity analysis found the Medicaid association materially unchanged, but selection bias from missingness could not be ruled out.

The evidence points to overlapping pathways — everyday discrimination, insurance coverage, socioeconomic conditions, clinical communication and documentation, reproductive surgery patterns, and the design of the research itself. No study in this evidence base establishes race as a self-contained biological mechanism.

Reeves and colleagues analyzed 2,377 SWAN participants followed for roughly two decades. Higher everyday discrimination was associated with reporting any VMS after adjustment for known risk factors: OR 1.30(95% CI, 1.09–1.54). The adjusted association with frequent VMS was no longer statistically significant.

Chronic discrimination was also associated with two any-VMS trajectories:

• Continuously high: OR 1.69(95% CI, 1.03–2.77)
• High before FMP, declining after FMP: OR 1.70(95% CI, 1.01–2.88)

Source: Reeves et al., Menopause, 2024.

Adjustment for discrimination attenuated but did not eliminate the Black–White symptom difference. This is observational evidence consistent with a pathway. It is not proof that discrimination alone caused the measured disparity.

The 2026 NHANES analysis shows why insurance cannot be treated as a minor administrative variable. In the weighted study sample, non-Hispanic Black participants made up 25.5% of Medicaid enrollees and 9.2% of privately insured participants. Medicaid enrollees reported food insecurity at 65.1% versus 12.7%, while 7.4% versus 43.2% held a college degree or higher.

Those differences do not make Medicaid a biological explanation. They show that insurance type sits inside a larger pattern of materially different circumstances. The statistical model can measure attenuation after adding insurance; it cannot fully separate coverage, income, care access, clinician participation, continuity, discrimination and patient preference.

The VA study measured coded documentation and prescriptions, not everything that happened in an encounter. Menopause symptoms had to appear through qualifying codes at two or more visits. Diagnoses and prescriptions outside the VA were unavailable. A racial difference in that outcome can arise through differences in symptoms, reporting, conversation, coding, prescribing, care outside the system or some combination.

The SWAN eligibility table is the clearest example of a disparity being hidden before the analysis starts. When 30.9% of screened Black women and 17.0% of screened White women were ineligible because of surgical menopause, the cohort did not lose the same kind of information from each group. The later selection analysis changed the result because it modeled the people the original longitudinal cohort could not observe in the same way.

No included study establishes a genetic explanation for the observed group differences. The 2026 NHANES authors state the underlying problem directly: race and ethnicity are socially constructed variables with no intrinsic biological basis, while regression categories cannot untangle discrimination, cultural context and historical inequities by themselves.

That does not mean biology never matters in menopause. It means a race label is not a biological mechanism, and these studies do not justify writing as though it is.

Black–White comparisons appear across every major domain in this evidence matrix. Evidence for Hispanic/Latina populations spans timing, duration, severity, VA prescribing and national MHT use. Chinese and Japanese SWAN groups have longitudinal timing and symptom-duration estimates, but little group-specific U.S. treatment evidence in the core set. Indigenous/First Nations and multiracial estimates are limited to one selected-platform severity study here.

This gap map is part of the asset, not an afterthought. A clean Black–White story is easier to write than the evidence justifies. The thinness for Indigenous, disaggregated Asian and multiracial populations is itself a finding.

These studies establish measured group differences in specific U.S. cohorts, surveys, a health system and a selected telehealth sample. They cannot tell an individual when menopause will occur, how long symptoms will last, whether hormone therapy is right for that person or which single mechanism produced a group difference.

Group estimates are not individual predictions. A 10.1-year group median does not mean every Black woman experiences 10.1 years of frequent symptoms. The median divides the observed distribution; it is not a countdown.

Odds ratios are not percentage-point differences.OR 0.74 does not mean “26 percentage points less likely.” It means 26% lower odds in that model. Converting it into a probability gap without the underlying baseline risk changes the claim.

Natural and surgical menopause are different outcomes. A 0.26-year adjusted natural-FMP difference and a 1.80-year surgical-FMP difference are not the low and high ends of one interchangeable estimate.

Crude rates and adjusted estimates answer different questions. A point-estimate prevalence ratio and an adjusted odds ratio can come from the same dataset without measuring the same quantity.

Study labels are not interchangeable.Our tables preserve source terminology — African American, Black, non-Hispanic Black, Hispanic, Hispanic/Latinx, Caucasian, White, non-Hispanic White and Indigenous/First Nations. Research teams constructed these categories differently, and the labels do not make the populations perfectly comparable.

Publication year is not data year. The 2026 insurance analysis uses NHANES data collected from 2013 through March 2020. It is current scholarship. It is not a measurement of 2026 treatment use.

Lower use does not automatically prove undertreatment. It can indicate inequity. By itself it cannot establish whether treatment was clinically indicated, whether a contraindication existed, whether a patient wanted treatment, whether a nonhormonal treatment was chosen or whether symptoms were discussed but never coded.

Commercial-platform evidence needs its conflict disclosure. The 2024 severity study analyzed people who sought care through Evernow, and several authors held company roles. The result can still be useful, but it cannot be passed off as a neutral national prevalence sample.

The latest national current-use estimates on this page end in March 2020. They cannot show whether the disparities narrowed or widened after that date.

This is educational content, not medical advice.Decisions about menopause symptoms, diagnosis or hormone therapy belong with a qualified clinician who knows the person’s history.

We identified original quantitative studies with a defined menopause-related outcome, a comparison group and a traceable primary source; extracted one estimate per row while preserving each study’s terminology and confidence intervals; recomputed every derived figure from the published values; and labeled every calculation we performed. We did not pool the results because the outcomes and designs are not comparable enough to justify one summary effect.

• Original quantitative U.S. research
• A defined menopause-related outcome or directly relevant structural correlate
• A traceable primary-source publication
• Enough detail on population, comparator, estimate and method to quote the finding accurately

• Consumer summaries where the primary study was available
• Unsourced statistics
• Narrative reviews used as quantitative evidence rows
• Estimates with no identifiable population, comparator or outcome
• General postmenopausal health differences without a direct menopause-experience or care measure
• Causal or biological claims about race that the study design could not support

• One estimate per row
• Stable record ID
• Source terminology preserved
• Confidence level retained, including the 97.5% intervals in the severity study
• Observed and model-predicted medians kept separate
• Natural and surgical FMP kept separate
• Odds ratios never rewritten as prevalence gaps
• Every derived figure labeled as The HRT Index calculation
• Every row carries a limitation
• No pooled meta-analytic effect

For a difference in years:

For a relative difference:

For a relative decline:

We do not calculate a confidence interval for a derived difference or ratio unless the source supplies enough information to do that correctly.

• Read the full Reeves et al. 2023 and Chesnokova et al. 2026 articles
• Re-summed every SWAN Table 1 eligibility row against the stated totals
• Checked the Gold et al. 2013 figure values against the primary paper
• Recomputed every month conversion, duration difference, relative decline and point-estimate ratio on this page
• Checked the Kochersberger tables against the abstract and used the model-consistent Table 4 value for the Indigenous/First Nations painful-sex row
• Corrected the Evernow analytic sample from 68,864 registrants to 67,867 analyzed participants
• Checked the Harlow et al. 2022 correction notice; it changes that review’s physical-performance paragraph and does not alter a statistic used here
• Checked every DOI and publication identity listed below

The timing figures are not one sevenfold range. A 3.1-month covariate-adjusted natural-FMP difference and a 21.6-month selection-corrected surgical-FMP difference measure different outcomes. This release keeps both figures and separates them.

The Indigenous/First Nations severity estimate is publishable once the model is named.The primary paper’s detailed tables show OR 1.77 before and 1.74 after the neighborhood-affluence variable was added. This release uses OR 1.74 because the neighboring rows use the same Table 4 model and records the abstract inconsistency.

The 11.6 MHT-use ratio has no invented uncertainty range. It is a point-estimate calculation from 5.8% and 0.5%. The source does not publish a confidence interval for that ratio, so neither do we.

These are the follow-up questions the evidence can and cannot resolve. Each answer names the population and method rather than generalizing past what the studies measured.

Symptom frequency

Gold EB, Colvin A, Avis N, et al. Longitudinal analysis of the association between vasomotor symptoms and race/ethnicity across the menopausal transition: Study of Women’s Health Across the Nation. American Journal of Public Health. 2006;96(7):1226–1235. doi:10.2105/AJPH.2005.066936

Symptom duration

Avis NE, Crawford SL, Greendale G, et al. Duration of menopausal vasomotor symptoms over the menopause transition. JAMA Internal Medicine. 2015;175(4):531–539. doi:10.1001/jamainternmed.2014.8063

Symptom severity

Kochersberger A, Coakley A, Millheiser L, et al. The association of race, ethnicity, and socioeconomic status on the severity of menopause symptoms: a study of 68,864 women. Menopause. 2024;31(6):476–483. doi:10.1097/GME.0000000000002349

Menopause timing

Gold EB, Crawford SL, Avis NE, et al. Factors related to age at natural menopause: longitudinal analyses from SWAN. American Journal of Epidemiology. 2013;178(1):70–83. doi:10.1093/aje/kws421

Reeves AN, Elliott MR, Karvonen-Gutierrez CA, Harlow SD. Systematic exclusion at study commencement masks earlier menopause for Black women in the Study of Women’s Health Across the Nation (SWAN). International Journal of Epidemiology. 2023;52(5):1612–1623. doi:10.1093/ije/dyad085

Cross-domain SWAN review

Harlow SD, Burnett-Bowie SM, Greendale GA, et al. Disparities in reproductive aging and midlife health between Black and White women: the Study of Women’s Health Across the Nation (SWAN). Women’s Midlife Health. 2022;8(1):3. doi:10.1186/s40695-022-00073-y — Correction published October 21, 2022: doi:10.1186/s40695-022-00082-x

Discrimination and symptom trajectories

Reeves AN, et al. Does everyday discrimination account for the increased risk of vasomotor symptoms in Black women? The Study of Women’s Health Across the Nation (SWAN). Menopause. 2024;31(6):484–493. doi:10.1097/GME.0000000000002357

Documentation and prescribing

Blanken A, Gibson CJ, Li Y, et al. Racial/ethnic disparities in the diagnosis and management of menopause symptoms among midlife women veterans. Menopause. 2022;29(7):877–882. doi:10.1097/GME.0000000000001978

National hormone-therapy trends

Yang L, Toriola AT. Menopausal hormone therapy use among postmenopausal women. JAMA Health Forum. 2024;5(9):e243128. doi:10.1001/jamahealthforum.2024.3128

Insurance and hormone-therapy use

Chesnokova A, Mumford SL, Schachter A, et al. Insurance type and menopausal hormone therapy use among US women. JAMA Network Open. 2026;9(7):e2623740. doi:10.1001/jamanetworkopen.2026.23740