Abstract
Objectives:
While premature menopause is associated with adverse cognitive outcomes, less is known about the impact of late menopause and whether associations vary by socioeconomic status. This study examines the association between the timing of natural menopause and cognitive decline, focusing on the modifying role of education.
Study design:
The study included 5082 U.S. women from the Health and Retirement Study who had experienced natural menopause by 2008 and were followed through 2020. Age at menopause was self-reported and categorized as <40 (premature), 40–44 (early), 45–49, 50–55, and > 55 (late). Multivariable-adjusted mixed-effects linear regression models estimated associations with baseline cognition and rate of cognitive decline, including interaction terms to assess effect modification by education.
Main outcome measure:
Score on a 27-point cognitive scale that includes word recalls, serial 7 s, and backward counting.
Results:
On average, participants were aged 69.9 years and had 12.6 years of education. Compared to those with menopause at 50–55, women with menopause before age 50 had significantly lower baseline cognitive scores. Late menopause was associated with faster cognitive decline (−0.01 standard deviations/year; 95% CI −0.017, −0.002), equivalent to 0.25 years of cognitive aging annually. Each additional year of education reduced this decline by 0.004 standard deviations/year (95% CI 0.001, 0.006), equivalent to 0.1 fewer years of cognitive aging annually.
Conclusions:
Earlier menopause was associated with a lower level of cognitive function, and late menopause with faster cognitive decline. Higher education may buffer adverse effects of late menopause. Dementia prevention strategies may need to consider menopause timing, especially for women with lower levels of education.
Keywords: Age at menopause, Cognitive decline, U.S. women, Education, Effect modification
1. Introduction
In the United States, about 1 in 10 adults aged 65 and older are living with Alzheimer’s Disease and related dementias (ADRD) [1]. By 2060, the annual number of new ADRD cases is projected to nearly double, from 0.5 million to about 1 million, with women having a much higher lifetime risk of dementia [2].
Growing evidence shows that earlier age at menopause is associated with poorer cognitive performance, earlier onset of cognitive decline, and mixed findings regarding ADRD risk [3–5]. Relevant U.S. research has relied primarily on regional cohorts [6–8] and small sample sizes [6,8], limiting the generalizability of their findings. Moreover, U.S.-based studies have yet to establish whether age at menopause is associated with long-term cognitive decline [9] and findings on cognitive performance are mixed [6,8]. In addition, both U.S. and global evidence remain limited in their exploration of resilience factors that may buffer the impact of earlier menopause on cognitive function and AD/ADRD risk, particularly social determinants such as education. Prior studies have primarily focused on biological modifiers, such as vascular risks, synaptic integrity, and amyloid-beta level [10–12]. Identifying individuals who are most resilient or vulnerable is essential for designing targeted prevention strategies.
To address these research gaps, we utilized data from the Health and Retirement Study (HRS), a nationally representative longitudinal study of adults aged over 50 [13] to examine the association between age at menopause and cognitive function at baseline and decline over time over a 12-year period (2008–2020). Additionally, we examined whether individual educational attainment modifies this relationship, given that education contributes to cognitive reserve, or one’s ability to maintain cognitive function despite accumulating aging-related brain pathology [14].
We hypothesized that earlier age at menopause would be associated with both worse baseline cognitive function and more rapid cognitive decline, and more years of education would mitigate these associations.
2. Methods
2.1. Data
The HRS is an ongoing nationally representative, longitudinal study of U.S. adults over age 50 [13], conducting biennial interviews to collect health, cognition, employment, retirement, income, wealth, family structure, and more. Launched in 1992, the study has recruited over 37,000 individuals from over 23,000 households using a multistage area probability sampling design with geographic stratification and clustering. Within each sampled household, one age-eligible respondent is randomly selected, and their spouse or partner is also included regardless of age.
We used HRS 2008 wave as our analytic baseline because it was the first interview wave to capture age at natural menopause - the key exposure variable for our analysis. To assess cognitive decline, we followed participants across biennial HRS waves until 2020. We used the RAND HRS Longitudinal File, supplemented with the Harmonized HRS dataset (Version D, 1992–2021), developed by the Gateway to Global Aging Data.
2.2. Study sample
HRS participants eligible for this study were those who self-identified as female, who reported being menopausal, and had complete data for analysis (Fig. 1). To ensure the inclusion of natural menopause, we excluded participants who reported having hysterectomy before menopause. The final analytic sample included 5082 female participants, contributing a total of 25,136 repeated observations of cognitive performance across seven waves from 2008 to 2020, with a median follow-up of 10 years.
Fig. 1.
Study sample flow chart.
2.3. Measurements
2.3.1. Exposure
Beginning in HRS 2008, female participants who either had not had a hysterectomy or had a hysterectomy after menopause were asked to report the age or year of their last menstrual period (LMP), as an indicator of age at natural menopause. Participants (46, 0.9% of the analysis sample) who could not recall their exact age at menopause were offered bracketed age categories: younger than 45, about 45, older than 45 and younger than 50, about 50, older than 50 and younger than 55, about 55, and older than 55. We used exact ages or brackets to classify menopause timing into clinically defined categories: <40 years (premature), 40–44 years (early), 45–49 years, 50–55 years (reference), and > 55 years (late) [15].
The self-reported age at menopause demonstrated acceptable reliability. Each woman reported her age at LMP for up to four times across HRS waves (2008, 2012, 2016, and 2020). The intraclass correlation coefficient (ICC) was 0.71, indicating that 71% of the total variability in reported age at LMP reflects between-woman differences rather than within-woman variation over time. Consistent with this finding, the median difference in reported age at menopause between the 2008 and 2012 wave was 0 years (interquartile range: 0–2 years), suggesting high temporal stability of reports.
2.3.2. Cognition outcomes
The HRS core survey assesses cognition using selected items from the modified Telephone Interview for Cognitive Status (TICS-M). We constructed a composite cognitive score with a range from 0 to 27, including immediate recall (0–10), delayed recall (0–10), serial 7’s (0–5), and backward counting (0–2). This composite score is the basis of the widely used Langa-Weir classification of cognitive function [16].
2.3.3. Covariates
We accounted for covariates associated with both menopause timing and cognitive outcomes, all measured in the HRS 2008 wave, the study baseline. Covariates included age, and self-reported race at the time of the first interview, categorized in the RAND HRS Longitudinal File as White, Black, or Other, based on the 2000 U.S. Census. The “Other” category combined American Indian or Alaska Native, Asian, Native Hawaiian or Pacific Islander, and other self-described racial groups due to small sample sizes. Additional covariates included marital status (married, separated/divorced, widowed, or never married), total number of biological children, number of household residents, years of education, maximum years of parental education, quintiles of household assets (e.g., the net value of the residence, vehicles, stocks, mortgages), quintiles of individual or couple income (e.g., earnings, pensions) from the previous calendar year, and urbanicity (urban, suburban, or ex-urban residence)., Health-related covariates included ever smoking, body mass index (BMI), calculated from self-reported height and weight as kilograms divided by meters squared and categorized as not overweight or obese (<25.0), overweight (25.0–29.9), and obese (≥30), ever diagnosis of diabetes, and number of depressive symptoms measured by the 8-item Center for Epidemiologic Studies Depression (CES—D) scale.
2.4. Analysis strategy
We used mixed-effects linear regression models with random intercepts. The outcome was the 27-point global cognitive score, standardized to the 2008 HRS baseline means and standard deviation to facilitate comparability across waves. All models adjusted the covariates described above and clustered at the participant level to account for within-person correlation and obtain robust standard errors.
To assess cognitive decline over time, we included an interaction term between age at menopause and years since baseline. To better account for potential confounding effects changing over time, we evaluated interaction terms between time and all covariates, retaining those that were statistically significant interactions at P < 0.05 for model simplicity: years since baseline × marital status, urbanicity, smoking history, and diabetes.
To evaluate effect modification by education, we included a three-way interaction term among years since baseline, age at menopause, and years of education. To improve the interpretation of the triple interaction, years of education were centered at 12 years, the sample median. To visualize this interaction, we estimated predicted cognitive scores at baseline and at 2, 4, 6, 8, and 10 years since baseline, across different age-at-menopause groups and at two levels of educational attainment: 12 years (the sample median) and 13 years. These values were chosen to represent a meaningful and interpretable contrast—comparing individuals with median educational attainment (typically corresponding to a high school education) to those with slightly higher education. Because education was modeled as a continuous linear variable, the interaction effect reflects the expected change in the association for each additional year of education.
We conducted several sensitivity analyses to assess the robustness of our findings. These included: 1) Using individual cognitive test items (e. g., word recall, serial 7’s) as alternative outcomes to evaluate if results are driven by a particular cognitive domain; 2) Excluding participants with a history of cancer to avoid the potential for chemotherapy-induced menopause; 3) restricting to women with consistent (≤1 year difference) reported age at menopause at baseline and four-year follow-up to assess the reliability of self-reported menopause timing; 4) Excluding individuals in the lowest 10th percentile of baseline cognitive scores to reduce recall bias of menopause timing; 5) Excluding individuals who reported age at menopause with potential digit preference, such as heaping at ages ending in 0 or 5, to reduce recall-related measurement error; 6) Excluding participants who reported entering menopause within two year prior to baseline to reduce the risk of misclassifying individuals with temporary amenorrhea as postmenopausal; 7) Stratifying the sample by years since menopause occurrence and baseline (equal or under versus above the median years since menopause: 19 years) to address recall bias and assess how long the association between age at menopause and cognition persists; 8) Excluding baseline health covariates and adjusting only for early-life confounders to avoid over adjustment of potential mediators.
3. Results
At baseline, participants had a mean age of 69.9 years and 12.6 years of education; women with late menopause were older (mean 72.4 years) and more educated. Most participants were White (81.9%), who tended to report late menopause (Table 1). Menopause occurred at ages 50–55 for 55% of women, 45–49 for 22.3%, >55 for 11.0%, 40–44 for 8.4%, and < 40 for 3.1%. About 54% were married or partnered and 30% widowed; the mean number of children was 2.9 and household size was 2. Approximately 30% were working, particularly among women with menopause at ages 45–55. One-third were overweight and 30% obese; 20% had diabetes, with higher prevalence among women with premature or late menopause. Women with earlier menopause reported more depressive symptoms.
Table 1.
Baseline characteristics, overall and by age at menopause, Health and Retirement Study (HRS), 2008–2020.
|
|
<40, premature |
40–44, early |
45–49 |
50–55 |
>55, late |
Total |
|---|---|---|---|---|---|---|
| (N = 160) | (N = 429) | (N = 1,133) | (N = 2,801) | (N = 559) | (N = 5,082) | |
| Age at baseline, mean (SD) | 70.6 (10.1) | 69.5 (10.2) | 69.3 (10.2) | 69.6 (10.1) | 72.4 (9.1) | 69.9 (10.1) |
| Race | ||||||
| White | 120 (75.0%) | 351 (81.8%) | 908 (80.1%) | 2,333 (83.3%) | 451 (80.7%) | 4,163 (81.9%) |
| Black | 30 (18.8%) | 54 (12.6%) | 149 (13.2%) | 353 (12.6%) | 88 (15.7%) | 674 (13.3%) |
| Other | 10 (6.2%) | 24 (5.6%) | 76 (6.7%) | 115 (4.1%) | 20 (3.6%) | 245 (4.8%) |
| Marital status | ||||||
| Married/partnered | 74 (46.2%) | 230 (53.6%) | 585 (51.6%) | 1,550 (55.3%) | 297 (53.1%) | 2,736 (53.8%) |
| Separated/divorced | 25 (15.6%) | 52 (12.1%) | 167 (14.7%) | 373 (13.3%) | 56 (10.0%) | 673 (13.2%) |
| Widowed | 53 (33.1%) | 131 (30.5%) | 326 (28.8%) | 796 (28.4%) | 195 (34.9%) | 1,501 (29.5%) |
| Never married | 8 (5.0%) | 16 (3.7%) | 55 (4.9%) | 82 (2.9%) | 11 (2.0%) | 172 (3.4%) |
| Total number of biological children ever born, mean (SD) | 2.9 (2.3) | 3.1 (2.2) | 2.9 (2.0) | 2.8 (1.9) | 2.9 (1.9) | 2.9 (2.0) |
| Number of household residents | 2.0 (1.3) | 2.1 (1.2) | 2.1 (1.1) | 2.1 (1.1) | 1.9 (0.9) | 2.0 (1.1) |
| Years of education, mean (SD) | 11.2 (3.5) | 11.9 (3.4) | 12.3 (3.2) | 12.9 (3.0) | 12.8 (2.8) | 12.6 (3.1) |
| Maximum years of parental education, mean (SD) | 9.3 (3.7) | 9.5 (3.9) | 9.9 (3.9) | 10.3 (3.6) | 10.0 (3.6) | 10.1 (3.7) |
| Quintiles of household assets | ||||||
| Q1 | 54 (33.8%) | 133 (31.0%) | 309 (27.3%) | 532 (19.0%) | 108 (19.3%) | 1,136 (22.4%) |
| Q2 | 34 (21.2%) | 99 (23.1%) | 232 (20.5%) | 568 (20.3%) | 105 (18.8%) | 1,038 (20.4%) |
| Q3 | 32 (20.0%) | 72 (16.8%) | 227 (20.0%) | 554 (19.8%) | 123 (22.0%) | 1,008 (19.8%) |
| Q4 | 23 (14.4%) | 67 (15.6%) | 197 (17.4%) | 554 (19.8%) | 112 (20.0%) | 953 (18.8%) |
| Q5 | 17 (10.6%) | 58 (13.5%) | 168 (14.8%) | 593 (21.2%) | 111 (19.9%) | 947 (18.6%) |
| Quintiles of total annual income (respondent and spouse if married) | ||||||
| Q1 | 61 (38.1%) | 124 (28.9%) | 297 (26.2%) | 584 (20.8%) | 113 (20.2%) | 1,179 (23.2%) |
| Q2 | 35 (21.9%) | 94 (21.9%) | 257 (22.7%) | 583 (20.8%) | 136 (24.3%) | 1,105 (21.7%) |
| Q3 | 32 (20.0%) | 91 (21.2%) | 223 (19.7%) | 518 (18.5%) | 110 (19.7%) | 974 (19.2%) |
| Q4 | 14 (8.8%) | 66 (15.4%) | 189 (16.7%) | 541 (19.3%) | 113 (20.2%) | 923 (18.2%) |
| Q5 | 18 (11.2%) | 54 (12.6%) | 167 (14.7%) | 575 (20.5%) | 87 (15.6%) | 901 (17.7%) |
| Work for pay | ||||||
| No | 127 (79.4%) | 313 (73.0%) | 792 (69.9%) | 1,885 (67.3%) | 424 (75.8%) | 3,541 (69.7%) |
| Yes | 33 (20.6%) | 116 (27.0%) | 341 (30.1%) | 916 (32.7%) | 135 (24.2%) | 1,541 (30.3%) |
| Urban residence | ||||||
| Urban | 77 (48.1%) | 198 (46.2%) | 539 (47.6%) | 1,369 (48.9%) | 257 (46.0%) | 2,440 (48.0%) |
| Suburban | 34 (21.2%) | 99 (23.1%) | 262 (23.1%) | 671 (24.0%) | 126 (22.5%) | 1,192 (23.5%) |
| Ex-urban | 49 (30.6%) | 132 (30.8%) | 332 (29.3%) | 761 (27.2%) | 176 (31.5%) | 1,450 (28.5%) |
| Ever smoking | ||||||
| No | 71 (44.4%) | 220 (51.3%) | 548 (48.4%) | 1,522 (54.3%) | 312 (55.8%) | 2,673 (52.6%) |
| Yes | 89 (55.6%) | 209 (48.7%) | 585 (51.6%) | 1,279 (45.7%) | 247 (44.2%) | 2,409 (47.4%) |
| Categorical Body Mass Index | ||||||
| Not overweight or obese | 67 (41.9%) | 163 (38.0%) | 434 (38.3%) | 998 (35.6%) | 221 (39.5%) | 1,883 (37.1%) |
| Overweight | 43 (26.9%) | 137 (31.9%) | 371 (32.7%) | 946 (33.8%) | 170 (30.4%) | 1,667 (32.8%) |
| Obesity | 50 (31.2%) | 129 (30.1%) | 328 (28.9%) | 857 (30.6%) | 168 (30.1%) | 1,532 (30.1%) |
| Ever diagnosis of diabetes | ||||||
| No | 119 (74.4%) | 339 (79.0%) | 896 (79.1%) | 2,249 (80.3%) | 436 (78.0%) | 4,039 (79.5%) |
| Yes | 41 (25.6%) | 90 (21.0%) | 237 (20.9%) | 552 (19.7%) | 123 (22.0%) | 1,043 (20.5%) |
| Depressive symptoms (CESD score) | 2.1 (2.3) | 1.5 (2.1) | 1.6 (2.1) | 1.3 (1.9) | 1.4 (1.9) | 1.5 (2.0) |
Note: SD = Standard Deviation. CESD = Center for Epidemiologic Studies Depression Scale.
Compared to women who experienced menopause at ages 50–55 (reference group), those with earlier menopause had lower baseline cognitive scores: −0.130 SD (95% CI: −0.250 to −0.009) for premature menopause (<40 years), −0.081 SD (95% CI: −0.155 to −0.006) for early menopause (40–44 years), and −0.070 SD (95% CI: −0.121 to −0.019) for menopause at 45–49 years (Table 2). There was no difference in baseline cognition for those with late menopause (>55 years). Over the follow-up period, women in the reference group experienced an average annual cognitive decline of −0.022 SD units (95% CI: −0.027 to −0.018), sharing the similar trends of cognitive decline with other women experiencing menopause at younger ages. Those with late menopause showed a faster decline of −0.010 SD units per year (95% CI: −0.017 to −0.002) relative to the reference, analogous to 0.25 more years of cognitive aging per year, based on the cross-sectional association of age with baseline cognition (−0.04 SD per year, 95% CI: −0.041 to −0.036, not shown in Table 2).
Table 2.
Years of education as an effect modifier of the association between age at menopause and cognitive score at baseline and rate of change over time, HRS 2008–2020.
| Coefficients (95% Confidence Intervals) | p values | |
|---|---|---|
| Independent effects | ||
| Cognitive score at baseline: menopause at 50–55 years (reference group) | ||
| Difference compared to those with menopause <40 years (premature) | −0.130 (−0.250, −0.009) | 0.035 |
| Difference compared to those with menopause at 40–44 years (early) | −0.081 (−0.155, −0.006) | 0.034 |
| Difference compared to those with menopause at 45–49 years | −0.070 (−0.121, −0.019) | 0.008 |
| Difference compared to those with menopause >55 years (late) | 0.042 (−0.027, 0.111) | 0.234 |
| Years of education | 0.077 (0.066, 0.087) | <0.001 |
| Time since baseline, SD units/year | −0.022 (−0.027, −0.018) | <0.001 |
| Interaction effects | ||
| Time * years of education | 0.0000 (−0.001, 0.001) | 0.916 |
| Time * age at menopause (reference: 50–55 years) | ||
| Difference for those with menopause <40 years (premature) | 0.0063 (−0.006, 0.019) | 0.335 |
| Difference for those with menopause at 40–44 years (early) | −0.0033 (−0.011, 0.004) | 0.392 |
| Difference for those with menopause at 45–49 years | 0.0036 (−0.001, 0.009) | 0.160 |
| Difference for those with menopause >55 years (late) | −0.010 (−0.017, −0.002) | 0.009 |
| Years of education * age at menopause (reference: 50–55 years) | ||
| Difference for those with menopause <40 years (premature) | 0.024 (−0.010, 0.059) | 0.161 |
| Difference for those with menopause at 40–44 years (early) | 0.006 (−0.016, 0.028) | 0.605 |
| Difference for those with menopause at 45–49 years | 0.003 (−0.013, 0.019) | 0.739 |
| Difference for those with menopause >55 years (late) | −0.001 (−0.024, 0.023) | 0.942 |
| Time * years of education * age at menopause (reference: menopause at 50–55 years with median years of education at 12 years) | ||
| Difference for those with menopause <40 years (premature) and one additional year of education | −0.001 (−0.005, 0.002) | 0.427 |
| Difference for those with menopause at 40–44 years (early) and one additional year of education | −0.0003 (−0.002, 0.002) | 0.796 |
| Difference for those with menopause at 45–49 years and one additional year of education | 0.0002 (−0.001, 0.002) | 0.814 |
| Difference for those with menopause >55 years (late) and one additional year of education | 0.004 (0.001, 0.006) | 0.003 |
Note: N = 5,082. Cognition outcome was standardized according to its’ baseline distribution. A multivariable-adjusted mixed-effects linear regression model with random intercepts was used. Covariates were measured at study baseline and included age, race (white, black, or other), marital status (married, separated/divorced, widowed, or never married), total number of biological children, total number of household residents, maximum years of parental education, urbanicity (urban, suburban, or ex-urban residence), quintiles of household assets, quintiles of total annual income (respondent and spouse if married), working for pay, ever smoking, body mass index (BMI) categorized as not overweight or obese (<25.0), overweight (25.0–29.9), and obese (≥30), ever diagnosis of diabetes, and depressive symptoms measured as by the 8-item Center for Epidemiologic Studies Depression (CES-D) scale, as well as interaction terms between years since baseline and each of the following covariates: marital status, urbanicity, smoking, and diabetes.
Each additional year of education was associated with a 0.077 SD (95% CI, 0.066 to 0.087) higher baseline cognitive score (Table 2). Education was not associated with differences in the rate of cognitive decline overall, nor did it substantially modify the relationship between age at menopause and baseline cognition. However, during follow-up, among women with late menopause (>55 years), each additional year of education was associated with a 0.004 standard deviation (SD) per year slower rate of cognitive decline (95% CI: 0.001 to 0.006), equivalent to 0.1 fewer years of cognitive aging per year, compared to the cross-sectional association of age with baseline cognition (−0.04 SD per year).
Fig. 2 illustrates baseline cognitive levels and trajectories derived from the models in Table 2. Women with late menopause showed faster cognitive decline over follow-up. After six years, those with late menopause and higher education (solid black line) declined below similarly educated women with menopause at ages 50–55 (solid gray line), with the gap widening thereafter. Among those with late menopause, individuals with median education (dashed black line) experienced even steeper declines, crossing below the 50–55 group after four years. By the end of follow-up, this group had lower cognitive scores than women with earlier menopause and higher education.
Fig. 2.
Initial cognitive scores and rate of cognitive decline over time, by age at menopause and education
Notes: Results are based on the model estimates shown in Table 2. N = 5082. Cognition outcome was standardized according to its’ baseline distribution. A multivariable-adjusted mixed-effects linear regression model with random intercepts was used. Covariates were measured at study baseline and included age, race (white, black, or other), marital status (married, separated/divorced, widowed, or never married), total number of biological children, total number of household residents, maximum years of parental education, urbanicity (urban, suburban, or ex-urban residence), quintiles of household assets, quintiles of total annual income (respondent and spouse if married), working for pay, ever smoking, body mass index (BMI) categorized as not overweight or obese (<25.0), overweight (25.0–29.9), and obese (≥30), ever diagnosis of diabetes, and depressive symptoms measured as by the 8-item Center for Epidemiologic Studies Depression (CES—D) scale, as well as interaction terms between years since baseline and each of the following covariates: marital status, urbanicity, smoking, and diabetes.
Sensitivity analyses supported the robustness of the main findings (Supplemental Tables 1–10). Earlier menopause showed a stronger association with poorer performance on serial 7’s (attention domain) scores than on recall memory (memory domain). Notably, the modifying effect of education on the relationship between late menopause and cognition appeared stronger when recall memory was the cognitive outcome (Supplemental Tables 1 and 2). Results were consistent after addressing potential measurement error in self-reported menopause age, including excluding inconsistent reports, chemotherapy-related menopause, low baseline cognition, digit preference, temporary amenorrhea, and longer time since menopause (Supplemental Tables 3–9), as well as potential overadjustment (Supplemental Table 10). Furthermore, findings also persisted among women more than 19 years post-menopause (Supplemental Table 9).
4. Discussion
In this large, nationally representative US study with up to 12 years of follow-up, we observed a strong association between earlier age at natural menopause and worse later-life cognitive function. American women who experienced menopause after age 55 experienced more rapid cognitive decline than those who experienced menopause at typical ages, a new finding that had not yet been fully studied in the U.S. [9] Our findings highlight the protective role of education in mitigating cognitive decline among this population, addressing a key research gap in the menopause-dementia literature, which previously lacked examinations of resilience factors. This study highlights the importance of incorporating reproductive aging and social determinants in dementia risk prediction and prevention strategies for women.
This study extends prior U.S. research limited to regional samples or shorter follow-up [6–8]. Our finding that younger age at natural menopause is associated with poorer baseline cognitive function aligns with previous studies from the U.S. [8,9] and meta-analyses in high-income countries [3]. A key mechanism is earlier estrogen loss: estradiol declines sharply during menopause, and its neuroprotective and neurotrophic effects supports brain health by promoting cell growth, connectivity, and communication [17].Early estrogen decline may therefore accelerate brain aging. Earlier menopause is also associated with sleep disturbances, mental health problems, frailty, chronic pain, and metabolic syndrome, which may further contribute to neurodegeneration [5,9].
However, such theory does not explain why women with menopause after age 55 experienced faster cognitive decline. While consistent with evidence from China [18], Sweden [19], and the Netherlands [20], this finding contrasts with research reporting either reduced dementia risk [3] or no difference by menopause timing [8,9]. These inconsistencies may reflect differences in sample size, follow-up duration, menopause age categorizations, and the confounding adjustment. Late menopause coincides with the onset of aging-related health issues. Estrogen may benefit cognition only when brain cells are healthy and could be harmful once damage has occurred [21]. Additionally, late menopause is associated with increased risks of diabetes, a known contributor to cognitive decline [22], and may coincide with the early accumulation of Alzheimer’s biomarkers. Furthermore, more severe menopausal symptoms at older ages [23] may further compound cognitive challenges.
This study further found that higher education mitigates the accelerated cognitive decline associated with late menopause. Education, a key modifiable dementia risk factor [14], may confer protection through greater cognitive reserve, potentially delaying menopause-related cognitive decline. Furthermore, higher education is associated with better access to menopause-related healthcare and treatment, including menopause hormone therapy (MHT) [24] and reduced frequency and severity of menopausal symptoms [25], which may further reduce cognitive vulnerability associated with the late menopausal transition.
Inconsistent with our hypothesis, premature or early menopause was not associated with faster cognitive decline. Cognitive changes may occur more rapidly in the early postmenopausal period, as suggested by neuroimaging studies [26]. Our analysis may have missed this critical window, as HRS assessments begin at age ≥ 51. Women with earlier menopause were therefore observed later in the postmenopausal course, whereas women with late menopause were assessed closer to the early postmenopausal window, potentially capturing more rapid decline.
We also did not observe the effect modification by education among individuals with premature or early menopause, likely due in part to limited sample size within these subgroups, which reduced power to detect interaction effects. Additionally, individuals with earlier menopause often face a greater burden of cardiovascular morbidity [27] and depression [28], which are known risk factors for dementia [14]. These adverse health conditions may outweigh the protective effects of education on cognitive health in this population.
This study has several limitations. Age at natural menopause was self-reported and subject to recall error. We mitigate this by modeling menopause age categorically, and conducting extensive sensitivity analyses, all of which showed consistent results (Supplemental Tables 3–9). Such limitations are common in large cohort studies of reproductive aging [4,6,10], though our study applied more extensive strategies to address measurement concerns than prior work. Nonetheless, these limitations underscore the need for improved measurement of reproductive aging in future research. The HRS also lacks detailed data on hormonal contraception (e.g., intrauterine devices), endometrial ablation, ovarian suppression, and other hormone-modifying therapies, which may cause temporary amenorrhea and misclassification of menopause age. To address this concern, we conducted a sensitivity analysis excluding women who reported menopause within two years prior to baseline (Supplemental Table 7), thereby reducing misclassification from temporary amenorrhea; results were consistent. In addition, hormonal intrauterine device use was uncommon among study participants during their 40s [29], suggesting limited impact of unmeasured hormonal contraception. Any remaining misclassification is most likely non-differential and would bias estimates toward the null, implying our results may underestimate the true association between earlier menopause and cognition. Finally, the use of MHT may misclassify menopause age and confound or modify associations, but detailed perimenopausal MHT data are unavailable in HRS. Given the sharp decline in MHT use after 2002 and prevalence below 10% among adults aged 51+ over the past two decades [30], unmeasured MHT is unlikely to substantially affect our findings.
In conclusion, both early and late menopause were associated with adverse cognitive function in later life among U.S. women. Importantly, higher educational attainment appears to mitigate the accelerated cognitive decline observed among women with late menopause. Future research should further elucidate the mechanisms through which education and other modifiable factors confer cognitive resilience across the menopausal transition. Such work will be critical for informing targeted and equitable dementia prevention strategies for aging women.
Supplementary Material
Funding
This work is funded by the National Institute on Aging at the U.S. National Institutes of Health, under grant number R01 AG070953 (Kobayashi, Gross), and U01 AG009740 (Langa). The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication.
Ethical approval
This study used data from the US Health and Retirement Study (HRS), which is funded by the National Institute on Aging (U01 AG009740) and was approved by the Institutional Review Board at the University of Michigan (Ann Arbor, MI, USA; HUM00061128). All recruited individuals provided written consent to participate in the HRS. This analysis was deemed exempt from human subject regulation by the Health Sciences and Behavioral Sciences Institutional Review Board at the University of Michigan (HUM00178420).
Footnotes
Declaration of competing interest
The authors declare that they have no competing interest.
Data availability
There are no linked research data sets for this paper. Data will be made available on request.
References
- [1].Manly JJ, Jones RN, Langa KM, Ryan LH, Levine DA, McCammon R, Heeringa SG, Weir D, Estimating the prevalence of dementia and mild cognitive impairment in the US: the 2016 health and retirement study harmonized cognitive assessment protocol project, JAMA Neurol. 79 (2022) 1242–1249, 10.1001/jamaneurol.2022.3543. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [2].Fang M, Hu J, Weiss J, Knopman DS, Albert M, Windham BG, Walker KA, Sharrett AR, Gottesman RF, Lutsey PL, Mosley T, Selvin E, Coresh J, Lifetime risk and projected burden of dementia, Nat. Med 31 (2025) 772–776, 10.1038/s41591-024-03340-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [3].Georgakis MK, Kalogirou EI, Diamantaras A-A, Daskalopoulou SS, Munro CA, Lyketsos CG, Skalkidou A, Th E. Petridou, Age at menopause and duration of reproductive period in association with dementia and cognitive function: a systematic review and meta-analysis, Psychoneuroendocrinology 73 (2016) 224–243, 10.1016/j.psyneuen.2016.08.003. [DOI] [PubMed] [Google Scholar]
- [4].Hao W, Fu C, Dong C, Zhou C, Sun H, Xie Z, Zhu D, Age at menopause and all-cause and cause-specific dementia: a prospective analysis of the UK biobank cohort, Hum. Reprod 38 (2023) 1746–1754, 10.1093/humrep/dead130. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [5].Liao H, Cheng J, Pan D, Deng Z, Liu Y, Jiang J, Cai J, He B, Lei M, Li H, Li Y, Xu Y, Tang Y, Association of earlier age at menopause with risk of incident dementia, brain structural indices and the potential mediators: a prospective community-based cohort study, eClinicalMedicine 60 (2023), 10.1016/j.eclinm.2023.102033. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [6].Bove R, Secor E, Chibnik LB, Barnes LL, Schneider JA, Bennett DA, De Jager PL, Age at surgical menopause influences cognitive decline and Alzheimer pathology in older women, Neurology 82 (2014) 222–229, 10.1212/WNL.0000000000000033. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [7].Gilsanz P, Lee C, Corrada MM, Kawas CH, Quesenberry CP, Whitmer RA, Reproductive period and risk of dementia in a diverse cohort of health care members, Neurology 92 (2019) e2005–e2014, 10.1212/WNL.0000000000007326. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [8].Stickel AM, Tarraf W, Kuwayama S, Wu B, Sundermann EE, Gallo LC, Lamar M, Daviglus M, Zeng D, Thyagarajan B, Isasi CR, Lipton RB, Cordero C, Perreira KM, Gonzalez HM, Banks SJ, Connections between reproductive health and cognitive aging among women enrolled in the HCHS/SOL and SOL-INCA, Alzheimers Dement. 20 (2024) 1944–1957, 10.1002/alz.13575. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [9].Shieu MM, Braley J. Tiffany, Jill Becker, Dunietz GL, The interplay among natural menopause, insomnia, and cognitive health: a population-based study, Nat. Sci. Sleep 15 (2023) 39–48, 10.2147/NSS.S398019. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [10].Wood Alexander M, Wu C-Y, Coughlan GT, Puri T, Buckley RF, Palta P, Swardfager W, Masellis M, Galea LAM, Einstein G, Black SE, Rabin JS, Associations between age at menopause, vascular risk, and 3-year cognitive change in the Canadian longitudinal study on aging, Neurology 102 (2024) e209298, 10.1212/WNL.0000000000209298. [DOI] [PubMed] [Google Scholar]
- [11].Coughlan GT, Betthauser TJ, Boyle R, Koscik RL, Klinger HM, Chibnik LB, Jonaitis EM, Yau W-YW, Wenzel A, Christian BT, Gleason CE, Saelzler UG, Properzi MJ, Schultz AP, Hanseeuw BJ, Manson JE, Rentz DM, Johnson KA, Sperling R, Johnson SC, Buckley RF, Association of age at menopause and hormone therapy use with tau and β-amyloid positron emission tomography, JAMA Neurol. 80 (2023) 462–473, 10.1001/jamaneurol.2023.0455. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [12].Wood Alexander M, Honer WG, Saloner R, Galea LAM, Bennett DA, Rabin JS, Casaletto KB, The interplay between age at menopause and synaptic integrity on Alzheimer’s disease risk in women, Sci. Adv 11 (2025) eadt0757, 10.1126/sciadv.adt0757. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [13].Sonnega A, Faul JD, Ofstedal MB, Langa KM, Phillips JW, Weir DR, Cohort profile: the health and retirement study (HRS), Int. J. Epidemiol 43 (2014) 576–585, 10.1093/ije/dyu067. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [14].Livingston G, Huntley J, Liu KY, Costafreda SG, Selbæk G, Alladi S, Ames D, Banerjee S, Burns A, Brayne C, Fox NC, Ferri CP, Gitlin LN, Howard R, Kales HC, Kivimäki M, Larson EB, Nakasujja N, Rockwood K, Samus Q, Shirai K, Singh-Manoux A, Schneider LS, Walsh S, Yao Y, Sommerlad A, Mukadam N, Dementia prevention, intervention, and care: 2024 report of the lancet standing commission, Lancet 404 (2024) 572–628, 10.1016/S0140-6736(24)01296-0. [DOI] [PubMed] [Google Scholar]
- [15].Eckhardt S, Wellons M, Defining menopause: What is early, what is late? in: Santoro NF, Cooper AR (Eds.), Primary Ovarian Insufficiency: A Clinical Guide to Early Menopause Springer International Publishing, Cham, 2016, pp. 1–17, 10.1007/978-3-319-22491-6_1. [DOI] [Google Scholar]
- [16].Langa KM, Weir DR, Kabeto M, Sonnega A, Langa-Weir Classification of Cognitive Function (1995–2020). https://hrsdata.isr.umich.edu/data-products/langa-weir-classification-cognitive-function-1995-2020, 2023. [Google Scholar]
- [17].Lee BH, Eid RS, Hodges TE, Barth C, Galea LAM, Leveraging research into sex differences and steroid hormones to improve brain health, Nat. Rev. Endocrinol 21 (2025) 214–229, 10.1038/s41574-024-01061-0. [DOI] [PubMed] [Google Scholar]
- [18].Guo M, Wu Y, Gross AL, Karvonen-Gutierrez C, Kobayashi LC, Age at menopause and cognitive function and decline among middle-aged and older women in the China health and retirement longitudinal study, 2011–2018, Alzheimers Dement. 21 (2025) e14580, 10.1002/alz.14580. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [19].Najar J, Östling S, Waern M, Zettergren A, Kern S, Wetterberg H, Hällström T, Skoog I, Reproductive period and dementia: a 44-year longitudinal population study of Swedish women, Alzheimers Dement. 16 (2020) 1153–1163, 10.1002/alz.12118. [DOI] [PubMed] [Google Scholar]
- [20].Geerlings MI, Ruitenberg A, Witteman JCM, van Swieten JC, Hofman A, van Duijn CM, Breteler MMB, Launer LJ, Reproductive period and risk of dementia in postmenopausal women, JAMA 285 (2001) 1475–1481, 10.1001/jama.285.11.1475. [DOI] [PubMed] [Google Scholar]
- [21].Brinton RD, Investigative models for determining hormone therapy-induced outcomes in brain: evidence in support of a healthy cell Bias of estrogen action, Ann. N. Y. Acad. Sci 1052 (2005) 57–74, 10.1196/annals.1347.005. [DOI] [PubMed] [Google Scholar]
- [22].Yazdkhasti M, Jafarabady K, Shafiee A, Omran SP, Mahmoodi Z, Esmaeilzadeh S, Babaheidari TB, Kabir K, Peisepar M, Bakhtiyari M, The association between age of menopause and type 2 diabetes: a systematic review and meta-analysis, Nutr. Metab. (Lond.) 21 (2024) 87, 10.1186/s12986-024-00858-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [23].Gjelsvik B, Rosvold EO, Straand J, Dalen I, Hunskaar S, Symptom prevalence during menopause and factors associated with symptoms and menopausal age. Results from the Norwegian Hordaland women’s cohort study, Maturitas 70 (2011) 383–390, 10.1016/j.maturitas.2011.09.011. [DOI] [PubMed] [Google Scholar]
- [24].Domm JA, Parker EE, Reed GW, German DC, Eisenberg E, Factors affecting access to menopause information, Menopause 7 (2000) 62. [DOI] [PubMed] [Google Scholar]
- [25].Li L, Wu J, Pu D, Zhao Y, Wan C, Sun L, Shen C, Sun W, Yuan Z, Shen Q, He X, Jiang J, Luo N, He Y, Qian Q, Cai P, Zhang M, Factors associated with the age of natural menopause and menopausal symptoms in Chinese women, Maturitas 73 (2012) 354–360, 10.1016/j.maturitas.2012.09.008. [DOI] [PubMed] [Google Scholar]
- [26].Mosconi L, Berti V, Dyke J, Schelbaum E, Jett S, Loughlin L, Jang G, Rahman A, Hristov H, Pahlajani S, Andrews R, Matthews D, Etingin O, Ganzer C, de Leon M, Isaacson R, Brinton RD, Menopause impacts human brain structure, connectivity, energy metabolism, and amyloid-beta deposition, Sci. Rep 11 (2021) 10867, 10.1038/s41598-021-90084-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [27].Muka T, Oliver-Williams C, Kunutsor S, Laven JSE, Fauser BCJM, Chowdhury R, Kavousi M, Franco OH, Association of age at onset of menopause and time since onset of menopause with cardiovascular outcomes, intermediate vascular traits, and all-cause mortality: a systematic review and meta-analysis, JAMA Cardiol. 1 (2016) 767–776, 10.1001/jamacardio.2016.2415. [DOI] [PubMed] [Google Scholar]
- [28].Georgakis MK, Thomopoulos TP, Diamantaras A-A, Kalogirou EI, Skalkidou A, Daskalopoulou SS, Petridou ET, Association of age at menopause and duration of reproductive period with depression after menopause: a systematic review and Meta-analysis, JAMA Psychiatry 73 (2016) 139–149, 10.1001/jamapsychiatry.2015.2653. [DOI] [PubMed] [Google Scholar]
- [29].Hubacher D, Kavanaugh M, Historical record-setting trends in IUD use in the United States, Contraception 98 (2018) 467–470, 10.1016/j.contraception.2018.05.016. [DOI] [PubMed] [Google Scholar]
- [30].Yang L, Toriola AT, Menopausal hormone therapy use among postmenopausal women, JAMA Health Forum 5 (2024) e243128, 10.1001/jamahealthforum.2024.3128. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
There are no linked research data sets for this paper. Data will be made available on request.