Age of menopause and dementia risk in 10,832 women from the Swedish Twin Registry

Abstract

INTRODUCTION

An earlier age of menopause (AOM) is hypothesized to increase vulnerability to the neuropathological processes of dementia, which begin in midlife.

METHODS

We tested this hypothesis in a sample of 10,832 women from the Swedish Twin Registry, stratified by menopause etiology.

RESULTS

Survival models showed that a U‐shaped association was present for women whose menopause occurred spontaneously. Sensitivity analyses conducted in hormone naïve, apolipoprotein E ε4+, and AOM restricted subsamples showed largely analogous patterns of results.

DISCUSSION

Supporting conclusions from basic research, our results suggest that estrogens (proxied here by AOM) influence several biological pathways mediating dementia disease processes. In line with trends in hormone research across the past century, our findings challenge the oversimplified “more‐is‐better” perspective on hormone exposure and highlight the need for cross‐disciplinary approaches to better understand the interacting endocrine and biopsychosocial factors that underlie the association between AOM and dementia pathogenesis.

Highlights

• We found a U‐shaped association of timing of spontaneous menopause and dementia risk.

• We also found a negative linear association of age of induced menopause and dementia risk.

• Restriction to a hormone‐naïve sample did not alter the pattern of results.

• Conducted exploration of the impact of common survival model parameter choices on results.

1. BACKGROUND

In contrast to the steady decline of testosterone levels observed in men across the adult lifespan ¹ , women experience a sharp drop in the production of estrogens by the ovaries at the time of menopause. ² , ³ Estrogens play a pivotal role in neuro‐ and cardio‐protective pathways; therefore, an earlier age of menopause (AOM) and thus the prolonged experience of a relative poverty of endogenous estrogens is hypothesized to increase a woman's vulnerability to dementia‐related pathophysiological processes that often begin in midlife. ⁴ , ⁵ , ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³

Despite plausible mechanisms emerging from preclinical research, examinations of the association between dementia risk and AOM (a proxy for the precipitous drop in circulating estrogens) have produced mixed results, and attempts to meaningfully synthesize this literature have been severely hindered by the methodological heterogeneity of the primary studies. For example, a 2016 meta‐analysis of 13 studies including 19,449 women found no difference in dementia risk between the youngest and oldest AOM groups. ¹⁴ Conversely, two subsequent meta‐analyses, including 14 and 11 primary studies, respectively, reported a pattern of increased dementia risk associated with “early” menopause (AOM < 45) and primary ovarian insufficiency (AOM < 40). ¹⁵ , ¹⁶ However, sensitivity analyses revealed that the exclusion of an exceptionally large cohort contributing roughly 99% of the participant‐level data to both analyses, ¹⁷ nullified the effect.

A woman's AOM is defined by the age of her final menstrual period, provided amenorrhea occurred for 12 consecutive months post‐final menstrual period. AOM is inherently a continuous variable without biologically meaningful categories. Therefore, in addition to reducing statistical power and increasing the probability of false positives, the use of arbitrary categories, as was done in many of the previous investigations, limits statistical power and hinders between‐study comparisons. ¹⁸ , ¹⁹

Further, permanent amenorrhea can result from the spontaneous depletion of ovarian follicles (i.e., spontaneous or natural menopause), the removal of the ovaries via bilateral oophorectomy (i.e., induced or surgical menopause), or the removal of the uterus via hysterectomy. Despite producing analogous phenotypes, each of the three etiologies is associated with a distinct hormone profile. ²⁰ , ²¹ Given that the change in estrogen levels is hypothesized to mediate the association between AOM and dementia risk, it is crucial to consider menopause etiology so that the results can be meaningfully interpreted in the context of plausible mechanisms. However, few studies collect sufficient gynecological surgery histories to characterize women's menopause etiology. Similarly, the use of exogenous hormones, either through oral contraceptives (OCs) or menopausal hormone therapy (MHT), has widespread impacts on the endocrine system, but a lifetime history of hormone use is often unavailable.

Given these noted limitations, we aim to contribute to the literature by examining associations between continuous AOM and dementia risk within a large sample of Swedish women with detailed reproductive health histories, including previous gynecological surgeries and lifetime history of exogenous hormone use. This comprehensive data set enabled continuous analyses stratified by menopause etiology and sensitivity analyses conducted in a sample of exogenous hormone‐naïve women to understand the influence of menopause etiology and hormone use on the AOM and dementia connection.

RESEARCH IN CONTEXT

1. Systematic review: The authors began the literature review with the cited meta‐analyses and review papers, the primary studies cited therein, and relevant articles published in the interim.

2. Interpretation: After summarizing the existing body of knowledge, our analyses apply the analytic and content‐specific recommendations developed by experts in the field. We found an association between a woman's late‐life dementia risk and the age of her menopause transition. Consistent with previous findings, “early” menopause was associated with an increased risk of dementia, but the significant quadratic term indicates a U‐shaped relationship such that women who experienced late menopause are also at increased risk for dementia.

3. Future directions: Large‐scale efforts are needed to better understand the associations between women's lifetime hormonal experience, including the timing, symptoms, and potential treatment of their menopause, and to translate these findings to clinical care considerations that benefit women's later‐life cognitive health.

2. METHODS

Data were obtained from female individuals from the Swedish Twin Registry (STR) who completed the Screening Across the Lifespan Twin (SALT) study interview between 1998 and 2002 (see Figure 1A). Detailed information regarding the STR research infrastructure and SALT protocols can be found elsewhere. ²² , ²³ Briefly, SALT aimed to screen for the most common complex diseases among twins enrolled in the STR born prior to 1959. Through a computer‐assisted structured telephone interview, researchers screened participants for cognitive impairment and collected detailed self‐reported health data, including female‐specific health factors such as AOM, births, history of gynecological surgeries, and OC and MHT use.

FIGURE 1.

Graphical depiction of the Swedish Twin Registry (STR) research infrastructure, survival model states, and hypothetical example individuals (HEIs) considered for inclusion current analyses. A, HEIs aligned with the calendar year. B, HEIs aligned to age. HEIs A–D reflect typical entrants. That is, individuals who were ≥ 65 at the time of the Screening Across the Lifespan Twin (SALT) study interview and therefore enter the “at risk” for dementia set at the time of interview. HEI E reflects a delayed entrant. That is, an individual who was younger than 65 at the time of her SALT interview and therefore was delayed from entering the “at risk” set until age 65. HEIs A and E were administratively censored at the time of data compilation, meaning that no evidence of dementia was observed directly through an STR study and no dementia indicators appeared in any of the National Patient Registers on or prior to December 31, 2016. HEIs B and D showed evidence of dementia during follow up, and were therefore removed from the “at risk” set at the time of estimated dementia onset. HEI C was censored prior to data compilation and was therefore removed from the “at risk” set at the time of death or emigration. HEIs F–H reflect STR participants that completed at least some portion of the SALT interview but were excluded from the survival analyses. HEI F and G were excluded because they experienced an event prior to age 65 (the start of time). HEI H was excluded due to dementia onset prior to the SALT interview.

2.1. Data acquisition and model parameters

2.1.1. Menopause

If women reported 12 consecutive months of amenorrhea that did not result from medical treatment or gynecological surgery, they were classified as having experienced spontaneous menopause (SM). If women reported a history of unilateral or bilateral oophorectomy (removal of one or both of the ovaries) or saplingo‐oophorectomy (removal of the ovaries and fallopian tube[s]) concurrent with or prior to spontaneous amenorrhea, they were classified as having experienced induced menopause (IM), regardless of whether a hysterectomy (removal of the uterus) was performed concurrently. Women who reported a hysterectomy (without oophorectomy) prior to SM or IM were excluded from analyses to ensure that the AOM demarcates a hormonal, rather than phenotypic, change (i.e., the absence of menses). Women who were still in the reproductive or menopause transition stages at the time of the SALT interview (i.e., those still experiencing regular periods and those experiencing irregular periods with < 12 consecutive months of amenorrhea, respectively) were classified as premenopausal and were excluded from the analyses.

Consistent with the stages of reproductive aging workshop + 10 staging, ²⁴ the spontaneous AOM was defined as women's self‐reported age of final menstrual period, provided that at least 12 consecutive months of amenorrhea occurred between the final menstrual period and the time of report. The self‐reported age of oophorectomy was used to define the AOM within the IM sample.

2.1.2. Dementia

The protocols for diagnosing dementia in STR have been described in detail elsewhere. ²⁵ , ²⁶ Briefly, individuals ≥ 65 and older were given a cognitive screening instrument as part of the SALT protocol. Individuals showing evidence of impairment and their co‐twin were referred to an in‐person clinical visit with a physician and psychologist, which resulted in a differential dementia diagnosis through a multidisciplinary clinical consensus conference (MCCC). Where possible, the etiology of the dementia was ascertained, and those presenting with dementia resulting from causes independent of aging (e.g., brain trauma, multiple sclerosis, hydrocephalus, Parkinson's disease, Korsakoff's syndrome, etc.) were excluded.

In addition to the screening of those over age 65 at the time of the SALT interview, the cognitive status of all participants, regardless of the age of the SALT interview, was monitored through linkages with national patient registers until December 31, 2016. Individuals were given a dementia diagnosis if (1) diagnostic codes associated with dementia appeared in in‐ or out‐patient health records (see Table S1 in supporting information), (2) individuals were prescribed drugs used to treat dementia (see Table S2 in supporting information), or (3) if dementia was listed as a contributing factor to death. ²⁶

When participants were diagnosed with dementia via STR MCCC, patient and informant interviews as well as medical records were used to derive the most accurate estimate of the age of dementia onset as possible. ²⁷

For those participants diagnosed with dementia only through the National Patient Register, the age of onset was estimated using a linear model that leveraged the observed discrepancies between STR MCCC estimated onset and the age of register appearance among the 419 women with both sources of data and an MCCC estimated age of onset > 65. Congruent with a previous investigation, ²⁸ the difference between the MCCC estimated age of dementia onset and the appearance of a dementia‐associated diagnostic code or prescription record in a patient register averaged 4 to 5 years, and an average of 9 years elapsed between the MCCC age of onset and death. See Figure S1 in supporting information.

For cases in which the National Patient Register linear model adjustment predicted dementia onset after age 65 but before the SALT interview, we estimated the age of onset as occurring approximately three quarters of the way through the interval between the two time points (see Figure 2). This decision was informed by the increasing likelihood of dementia onset with advancing age, making later points in the interval statistically more probable than earlier ones. As noted above, participants aged ≥ 65 underwent cognitive screening, with impaired individuals referred for further clinical evaluation. If dementia was evident at the SALT interview, a diagnosis via the MCCC protocol would be available, eliminating the need for estimating the age of onset through the National Patient Registers.

FIGURE 2.

Participant inclusion flow chart for the primary and sensitivity analyses. APOE, apolipoprotein E; dx, diagnosis; MHT, menopausal hormone therapy; OC, oral contraceptive; SALT, Screening Across the Lifespan Twin study; STR, Swedish Twin Registry.

2.1.3. Exclusionary criteria

No missing data imputation was conducted. Therefore, only participants with complete data relating to the age and etiology of menopause, the diagnosis (or lack thereof) of dementia, an estimated age of onset when relevant, and all relevant covariates were included. All analyses were restricted to women with an AOM between 30 and 64 (inclusive). Women who reported a lifetime history of medical conditions known to impact menstrual cycles, such as polycystic ovarian syndrome or cancer, were excluded. Finally, women for whom no follow‐up data were available after age 65, who experienced the onset of dementia prior to age 65 or prior to the SALT interview, were also excluded.

2.2. Summary of statistical analyses

Data processing and analyses were conducted in R version 4.3.2 ²⁹ using the survival and survminer packages. ³⁰ , ³¹ Cox proportional hazards models typically use the duration of time‐on‐study as the model timescale and only include baseline age as a covariate. ³² This practice stems from the original application of survival models to capture the impact of medical interventions on patient survival. In this context, the implementation of the intervention creates a meaningful model start time. However, this approach can lead to inaccurate results in the context of cohort studies such as SALT because the time of the interview is somewhat random and thus does not provide a meaningful start time. Instead, it is recommended to use participant age as the timescale, as depicted in Figure 1B. Therefore, age was used as the timescale for all Cox proportional hazards models, with age 65 demarcating the model start‐of‐time. Women whose SALT interview occurred after age 65 were treated as delayed entrants and were not included in the at‐risk sample until their age at the SALT interview. Dementia diagnosis was used as the model event, and the estimated age of onset was used as the event time. Women who were not diagnosed with dementia during the observational period were censored (i.e., removed from the at‐risk sample) at the age of death, emigration, or age on December 31, 2016. Observations were clustered by twin pair, ties were handled using exact partial likelihood, and the proportional hazards assumption was tested for all models using the method described by Grambsch and Therneau ³³ via the cox.zph function.

For each question, an unadjusted and a covariate‐adjusted model was applied to the relevant data. All covariate‐adjusted models included three ordered factor covariates: the self‐reported number of births (0, 1, 2, or 3+), lifetime education attainment (primary or less, some or complete secondary, or post‐secondary), and smoking history (ever regular smoker or never regular smoker).

2.2.1. Primary analyses

To examine whether IM impacts dementia risk, we applied a model with menopause etiology as a binary variable (IM vs. SM). AOM (centered at age 30) was included in the adjusted model.

To address our primary hypotheses regarding the association between continuous AOM and dementia, we fit two model sets, each to the SM and IM samples. The first model imposed a linear association between AOM (centered at age 30) and dementia risk, and the second imposed a quadratic association.

2.2.2. Sensitivity analyses

Planned sensitivity analyses applied the same series of models to three defined subsamples of women with SM: apolipoprotein E (APOE) ε4+, APOE ε4–, and hormone naïve (i.e., never users of OC or MHT). Post hoc sensitivity analyses were conducted in subsamples of the SM and IM groups restricted to those with an AOM between 40 and 60 to examine whether the pattern of results is influenced by the AOM range. Finally, based on previous research, we anticipate either a linear association or a quadratic association between AOM and dementia risk, and therefore chose to include AOM as a continuous and quadratic term in the primary models. However, it is possible that the association between AOM and dementia risk is non‐linear, but in a form not well captured by a quadratic term. Therefore, AOM was modeled using a restricted cubic spline with three, four, five, and six knots to test for non‐linearity.

2.2.3. Supplemental analyses

For comparison to existing literature that used a categorical definition of age at menopause, a second series of two model sets was fit to the SM sample. The first modeled primary ovarian insufficiency diagnosis (defined as AOM < 40) as a binary variable. The second modeled early menopause status (defined as AOM < 45) as a binary variable.

To provide real‐world exploration of the impact of key model choices in the context of age‐dependent disease outcomes, we modified the primary models (linear and quadratic associations between AOM and dementia) by (1) the use of time‐on‐study as timescale (rather than age, which we believe to be the appropriate timescale and therefore applied to all previous models), and/or (2) using “complete‐cases” analyses (rather than delayed‐entry) which necessitate the restriction of the sample to either: (a) those women who were under study before becoming “at risk” of the event, defined as age 65 here. That is, those women who were postmenopausal but not yet 65 at the time of their SALT interview. Or, (b) a sample restricted to those with equivalent opportunity to experience the factor of interest (post‐menopausal status) prior to observation. That is, women who were ≥ 65 at the time of their SALT interview.

3. RESULTS

A total of 10,832 women met the criteria for inclusion. As shown in Figure 3, the most common reasons for exclusion were being premenopausal at the time of the SALT interview or having an unknown AOM, accounting for 37% and 18% of excluded cases, respectively. Consistent with historical trends among Swedish women, ³⁴ 94% of women experienced SM (see Table 1). Of the 10,173 women who experienced SM, 58% were considered exogenous hormone naïve (i.e., they reported never using exogenous hormones). Compared to hormone users, the hormone‐naïve participants were born earlier (average birth year 1933 vs. 1939; see Table S3 in supporting information), which is consistent with the introduction of OC in Sweden in the mid 1960s ³⁵ and the increase in popularity of MHT during that same time. ³⁶

FIGURE 3.

Distribution of age of menopause (AOM) by menopause etiology and dementia diagnosis. The dashed gray lines reflect the cut points for premature and early menopause, while the dotted gray line reflects the sample (i.e., induced or spontaneous menopause) mean age of menopause. The dashed black line reflects the start of study time. To be included in these analyses, women had to have a menopause age between 30 and 64.9 and an estimated age of dementia onset or censoring after 65 (i.e., the start of time).

TABLE 1.

Participant demographics for the spontaneous and induced menopause samples.

	Spontaneous menopause				Induced menopause
	Total	Censored	Dementia	p a	Total	Censored	Dementia	p b	p c
N	10,173	8884	1289		659	591	68
Years observed	14.4 (4.7)	15.5 (3.7)	6.8 (3.7)	<0.001	14.5 (4.6)	15.4 (3.8)	7.0 (3.5)	<0.001	0.747
Birth year
Age	1935.7 (9.7)	1937.0 (9.4)	1927.4 (7.5)	<0.001	1936.1 (9.5)	1937.1 (9.3)	1927.5 (5.9)	<0.001	0.401
Menopause	50.5 (4.0)	50.5 (3.9)	50.4 (4.5)	0.155	47.2 (6.3)	47.3 (6.3)	46.1 (6.2)	0.138	<0.001
SALT interview	63.9 (9.0)	62.7 (8.7)	71.6 (7.4)	<0.001	63.6 (8.9)	62.6 (8.7)	71.6 (5.8)	<0.001	0.410
Event d	78.3 (7.5)	78.3 (7.8)	78.5 (5.5)	0.347	78.1 (7.5)	78.0 (7.8)	78.6 (5.0)	0.548	0.430
Births				<0.001				0.770	0.007
0	1,287 (12.7)	1,099 (12.4)	188 (14.6)		103 (15.6)	90 (15.2)	13 (19.1)
1	1,624 (16.0)	1,383 (15.6)	241 (18.7)		124 (18.8)	111 (18.8)	13 (19.1)
2	4,063 (39.9)	3,645 (41.0)	418 (32.4)		256 (38.8)	233 (39.4)	23 (33.8)
3+	3,199 (31.4)	2,757 (31.0)	442 (34.3)		176 (26.7)	157 (26.6)	19 (27.9)
Education				<0.001				0.300	0.124
Primary or less	4,059 (39.9)	3,323 (37.4)	736 (57.1)		281 (42.6)	248 (42.0)	33 (48.5)
Secondary e	4,119 (40.5)	3,683 (41.5)	436 (33.8)		269 (40.8)	241 (40.8)	28 (41.2)
Post‐secondary	1,995 (19.6)	1,878 (21.1)	117 (9.1)		109 (16.5)	102 (17.3)	7 (10.3)
Ever smoker f	3,818 (37.5)	3,460 (38.9)	358 (27.8)	<0.001	278 (42.2)	259 (43.8)	19 (27.9)	0.017	0.019
APOE ε4	948 (29.1)	789 (27.3)	159 (44.3)	<0.001	59 (26.0)	50 (24.3)	9 (42.9)	0.112	0.350
MHT use	4,268 (42.0)	3,868 (43.6)	400 (31.1)	<0.001	374 (56.8)	350 (59.2)	24 (35.3)	<0.001	<0.001
OC use	567 (6.2)	545 (6.8)	22 (1.8)	<0.001	38 (6.7)	37 (7.4)	1 (1.6)	0.144	0.671

^a Censored versus dementia for the spontaneous menopause sample.

^b Censored versus dementia for the induced menopause sample.

^c Spontaneous menopause versus induced menopause samples.

^d Dementia or censoring.

^e Includes some secondary education.

^f Self‐reported ever regularly smoking.

Abbreviations: APOE, apolipoprotein E; MHT, menopausal hormone therapy; OC, oral contraceptive.

As anticipated, given that evidence of dementia at the time of the SALT interview was an exclusion criterion, only 40 of the 1357 dementia cases had an onset date estimated by MCCC; the remaining estimated onset ages were based on the National Patient Register indicators. Of these estimates, 138 predicted a dementia onset age prior to the age of the SALT interview were adjusted to the three‐quarter mark of the interval between the SALT interview and register indicator appearance as described in section 2.1.2. The age of dementia onset ranged from 66.02 to 95.64 years, with 508 unique event times (see Figure 4).

FIGURE 4.

Relative risk of dementia by age of menopause (AOM) as predicted by the unadjusted quadratic model. The risk of dementia is shown relative to the predicted risk for a woman with the sample mean age of menopause (illustrated by the single point). The model indicates that a woman who experienced menopause at age 31 has > 2.5 times the risk of developing dementia compared to a woman with an average age of menopause. Similarly, a woman experiencing menopause at age 64 has an almost 2‐fold greater risk of developing dementia compared to a woman with an average age of menopause across follow‐up.

3.1. Primary analyses

There was some evidence for the violation of the proportional hazards assumption, suggesting that the influence of ever regularly smoking on dementia risk is not consistent over time. Therefore, the coefficients related to smoking will not be interpreted. However, this violation does not impact the interpretation of the findings relevant to menopause age, and therefore, stratified analyses were not conducted. No other variables or global tests violated the proportional hazards assumption (ps > 0.05).

3.1.1. Spontaneous menopause

As shown in Table 2, among women with SM, the linear model was a poor fit. Applying a quadratic model improved the fit, and this was unchanged in the adjusted models. Figure 4 shows the relative risk ratios derived from the quadratic model. The model indicates that a woman who experienced menopause at age 31 has > 2.5 times the risk of developing dementia compared to a woman with an average age of menopause (≈ age 51). Similarly, a woman experiencing menopause at age 64 has almost a 2‐fold greater risk of developing dementia compared to a woman with an average age of menopause across follow‐up.

TABLE 2.

Cox model results of the associations between menopause terms and dementia risk.

		Induced menopause		Spontaneous menopause		Induced (vs. spontaneous)	POI (vs. AOM > 40)	Early (vs. AOM > 45)
		Linear	Quadratic	Linear	Quadratic
N events		68	68	1289	1289	1357	1357	1357
		Unadjusted
BIC		735.3	739.2	20910.7	20909.2	22189.4	22187.9	22184.3
Menopause type	HR					0.781 ^	1.101	1.189 ^
	95% CI					0.528: 1.035	0.733: 1.468	1.013: 1.365
Menopause age	HR	0.966 ^	1.001	0.992	0.899 **	0.990
	95% CI	0.929: 1.004	0.844: 1.157	0.978: 1.007	0.830: 0.968	0.976: 1.003
Menopause age2	HR		0.999		1.003 **
	95% CI		0.993: 1.004		1.001: 1.004
		Adjusted
BIC		756.5	760.4	20877.3	20874.7	22161.1	22159.5	22157.0
Menopause type	HR					0.761 *	1.080	1.151
	95% CI					0.505: 1.016	0.709: 1.450	0.975: 1.328
Menopause age	HR	0.960 ^	0.996	0.996	0.895 **	0.993
	95% CI	0.919: 1.002	0.826: 1.167	0.981: 1.011	0.826: 0.965	0.979: 1.007
Menopause age2	HR		0.999		1.003 **
	95% CI		0.993: 1.005		1.001: 1.005
Ever smoker	HR	0.707	0.706	0.983	0.986	0.968	0.966	0.964
	95% CI	0.162: 1.253	0.161: 1.251	0.856: 1.110	0.858: 1.113	0.844: 1.092	0.843: 1.090	0.841: 1.088
Education (L)	HR	0.981	1.004	0.595 **	0.592 **	0.610 **	0.610 **	0.612 **
	95% CI	0.369: 1.593	0.388: 1.621	0.452: 0.738	0.449: 0.736	0.471: 0.749	0.471: 0.749	0.472: 0.751
Education (Q)	HR	0.742	0.754	0.883 *	0.880 *	0.876 *	0.876 *	0.876 *
	95% CI	0.269: 1.215	0.273: 1.234	0.769: 0.996	0.767: 0.993	0.766: 0.986	0.766: 0.986	0.766: 0.986
Births (L)	HR	1.098	1.084	0.886 ^	0.886 ^	0.893 ^	0.894 ^	0.898 ^
	95% CI	0.540: 1.656	0.531: 1.638	0.759: 1.013	0.759: 1.013	0.769: 1.017	0.771: 1.018	0.775: 1.022
Births (Q)	HR	1.049	1.073	1.077	1.076	1.079	1.080	1.078
	95% CI	0.549: 1.549	0.549: 1.597	0.955: 1.198	0.955: 1.198	0.960: 1.197	0.962: 1.199	0.960: 1.196
Births (C)	HR	0.829	0.820	1.143 *	1.141 *	1.126 *	1.123 *	1.122 *
	95% CI	0.347: 1.310	0.338: 1.302	1.026: 1.260	1.024: 1.258	1.012: 1.240	1.009: 1.237	1.008: 1.236

Notes: The linear (L), quadratic (Q), and cubic (C) terms reflect the pattern of the relative change in hazard across the ordered categories but are not interpreted as typical hazard ratios.

Abbreviations: AOM, age at menopause; BIC, Bayesian information criterion; CI, confidence interval; HR, hazard ratio; POI, primary ovarian insufficiency.

^{^} P < 0.10.

^* P < 0.05.

^** P < 0.01.

3.2. Induced menopause

IM was not independently associated with dementia risk when covarying for AOM. Among women with IM, there was a negative linear association between the AOM and dementia risk. Specifically, each additional year of premenopausal status was associated with a 5.8% lower dementia risk. There was no evidence of a quadratic association between AOM and dementia risk among women with IM (see Table 2).

3.3. Sensitivity analyses

When stratified by APOE ε4 carrier status, the quadratic association between AOM and dementia risk only remained marginally significant for ε4 carriers (see Table 3). Restricting analyses to the hormone‐naïve subsample did not meaningfully change the pattern of results, nor did restricting analyses to the women who experienced SM between ages 40 and 60 (see Table S4 in supporting information). Finally, the results of all four restricted cubic spline models indicate that the association between AOM and dementia risk is non‐linear among women with SM (all p < 0.058). Of the four models, the model with three knots provided the best fit and illustrated a gentle U‐shaped association between AOM and dementia risk, supporting the main finding (see Figure S2 in supporting information). However, the quadratic model was a slightly better fit (three‐knot spline model Bayesian information criterion [BIC]: 20960.3, quadratic model BIC: 20909.2).

TABLE 3.

Stratified analysis results for APOE ε4+ and ε4– samples.

		APOE ε4+		APOE ε4–
		Linear	Quadratic	Linear	Quadratic
N events		159	159	200	200
		Unadjusted
BIC		1829.3	1831.2	2609.4	2613.2
Menopause age	HR	1.011	0.842^	0.976	0.894
	95% CI	0.968: 1.054	0.634: 1.050	0.941: 1.011	0.751: 1.037
Menopause age2	HR		1.005 ^		1.002
	95% CI		0.999: 1.010		0.999: 1.006
		Adjusted
BIC		1850.9	1852.6	2620.5	2623.8
Menopause age	HR	1.009	0.834 ^	0.983	0.881 ^
	95% CI	0.965: 1.053	0.623: 1.044	0.948: 1.019	0.735: 1.026
Menopause age2	HR		1.005 ^		1.003
	95% CI		0.999: 1.010		0.999: 1.007
Ever smoker	HR	1.251	1.288	0.814	0.826
	95% CI	0.908: 1.594	0.944: 1.632	0.466: 1.162	0.477: 1.175
Education (L)	HR	0.753 ^	0.761	0.577 **	0.569 **
	95% CI	0.418: 1.087	0.423: 1.098	0.228: 0.925	0.217: 0.921
Education (Q)	HR	0.981	0.976	0.778 ^	0.772 ^
	95% CI	0.692: 1.271	0.686: 1.267	0.504: 1.053	0.495: 1.048
Births (L)	HR	1.166	1.200	0.707 *	0.700 *
	95% CI	0.781: 1.552	0.814: 1.586	0.406: 1.008	0.398: 1.002
Births (Q)	HR	1.204	1.197	1.282	1.294
	95% CI	0.825: 1.582	0.817: 1.577	0.972: 1.591	0.982: 1.606
Births (C)	HR	1.157	1.144	0.842	0.850
	95% CI	0.783: 1.532	0.770: 1.518	0.533: 1.150	0.540: 1.159

Notes: The linear (L), quadratic (Q), and cubic (C) terms reflect the pattern of the relative change in hazard across the ordered categories but are not interpreted as typical hazard ratios.

Abbreviations: APOE, apolipoprotein E; BIC, Bayesian information criterion; CI, confidence interval; HR, hazard ratio.

^{^} P< 0.10.

^* P< 0.05.

^** P < 0.01.

3.4. Supplemental analyses

3.4.1. Primary ovarian insufficiency and early menopause

Early menopause (AOM < 45) was associated with an ≈ 20% increase in dementia risk, while primary ovarian insufficiency (AOM < 40) was not significantly associated with dementia risk (see Table 2).

3.4.2. Influence of modeling parameters

As shown in Table S5 in supporting information, when time‐on‐study rather than age was used as the model time, there was evidence that the proportional hazards assumption was violated, driven by interview age, for both the linear and quadratic models in both the SM and IM samples. However, the Cox proportional hazards model results were numerically similar to those observed in the primary analyses (see Table S6 in supporting information).

Further, as anticipated, when analyses were conducted using complete cases rather than delayed entry, the sample size, and therefore the statistical power, dropped (see Table S7 in supporting information). Notably, just 4% of individuals who completed the SALT interview prior to age 66 received a dementia diagnosis during follow‐up, compared to almost 25% of participants interviewed on or after age 65. This discrepancy is likely due to the increased risk of dementia with increasing age. Specifically, those with a younger interview age would also be followed at a younger age interval, wherein they would be less at risk for developing dementia than those interviewed at older ages. However, the quadratic association was only statistically significant in the sample interviewed prior to age 66 using time‐on‐study as model time (see Table S8 in supporting information).

4. DISCUSSION

Our analysis found that dementia risk increases with deviation from the typical AOM, regardless of direction, among women with SM. Experiencing menopause 10 years earlier or later than the typical AOM was associated with a nearly 30% increase in dementia risk. The pattern of results was unchanged when restricted to hormone‐naïve women, but only remained marginally significant among APOE ε4 carriers, consistent with the evidence for interactive effects of estrogen with ε4 allele carriership. ³⁷ , ³⁸ Replicating prior findings, there was no difference in dementia risk between women with IM and those with SM when covarying for AOM. ³⁹ However, consistent with findings that early IM is associated with a greater risk of dementia, our data support a linear association between AOM and dementia among women with IM, such that each additional reproductive year was linked to a 5% decrease in dementia risk.

Although primary ovarian insufficiency (POI; AOM < 40) was not associated with an increased risk of dementia, early menopause (AOM < 45) increased dementia risk. However, in line with estimates from similar populations, ⁴⁰ just 2% of the sample experienced POI, and of these women, just 31 also developed dementia during follow‐up, so the robustness of this finding is questionable.

These findings are consistent with reports from preclinical research showing that estrogens interact with numerous biological pathways that mediate dementia pathogenesis. However, we hesitate to narrow our focus or attribute the observed findings to exclusively estrogenic or even hormonal mechanisms, for several reasons. First, our main finding showed a U‐shaped relationship between AOM and dementia risk, indicating that women with the shortest durations of estrogen depletion for their chronological age were also at increased risk of developing dementia. Second, while the menopause transition is characterized by extreme changes in circulating estrogen levels, the decline occurs over a period of several years, and the exact temporal dynamics vary between women. ²⁴ , ⁴¹ , ⁴² Therefore, the age of menopause, which is demarcated phenotypically by the final menstrual period, provides only a limited amount of information with respect to estrogen levels. Third, the menopause transition coincides with a constellation of biopsychosocial changes that may confound or mediate the observed associations, such as changes in cardiometabolic health, anxiety around cognitive changes or menopause symptoms, and major shifts in caregiving roles. Finally, AOM itself may be a marker of health in addition to demarcating a major hormonal shift. ⁴³ In this context, extreme deviance from the mean may reflect poorer health overall, which would increase dementia risk. Therefore, to attribute population‐level variation in dementia risk to any single factor, such as estrogen, risks overlooking this complex interplay of influences.

4.1. Strengths, limitations, and future directions

Among the strengths of the current investigation are the STR's large size, use of health registry linkages, and the development and early adoption of detailed demographic and women's health history questionnaires. This investment enabled the statistical consideration of several key factors shown to impact both AOM and dementia risk, which were not examined by previous studies.

However, because participants only completed one SALT interview, the influence of time‐varying covariates could not be considered. This is particularly disappointing in the case of body mass index (BMI). After menopause, endogenous estrogen is primarily aromatized in adipose tissue, ⁴⁴ resulting in a positive association between post‐menopausal adiposity and endogenous estrogen. ⁴⁵ , ⁴⁶ Therefore, within the estrogen hypothesis framework, higher adiposity should indirectly confer a neuroprotective effect, and there is some evidence of this effect. ⁴⁷ However, neither adiposity nor the association between BMI and fat mass is stable across time. ⁴⁴ Further, established associations between BMI and various health outcomes (including dementia risk) vary as a function of age. ⁴⁸ Given that the age of the SALT interview (and therefore the age of known BMI) ranged from 50 to 97, no cohesive interpretation of the construct reflected by the available measure of BMI could be determined.

Although the limitations of retrospective self‐report of medical histories are not unique to this analysis or the STR, they are important to consider. For example, women are less reliable when reporting their AOM the further they get from the time of menopause, and regress responses toward the mean AOM. ⁴⁹ , ⁵⁰ , ⁵¹ Therefore, beyond replication in more demographically and reproductively diverse samples, subsequent clinical cohort studies would benefit from longitudinal, prospective data collection that follows women throughout the menopause transition, capturing detailed health histories (both self‐reported and via electronic health records) and tracking hormonal changes, including estrogen levels, before, during, and after the transition.

Complementary preclinical work is also needed to further probe the biological pathways mediating the relationships between sex hormones and various pathogeneses, as well as examine the interactive effects of multiple hormone changes, and consider the moderating influences of other modifiable factors.

5. CONCLUSION

Within a sample of > 10,000 postmenopausal women enrolled in the STR, we found that both early and late AOM conferred increased dementia risk compared to more typical AOM. These results are consistent with preclinical work demonstrating the involvement of estrogenic pathways in the pathogenesis of dementia and highlight the importance of considering sex‐specific factors in moderating disease risk.

AUTHOR CONTRIBUTIONS

All authors have seen and approved the manuscript.

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts to disclose. Author disclosures are available in the supporting information.

CONSENT STATEMENT

Consent was not necessary because this project only included secondary data analysis.

Supporting information

Supporting Information

Supporting Information

Saelzler UG, Sundermann EE, Foret JT, et al. Age of menopause and dementia risk in 10,832 women from the Swedish Twin Registry. Alzheimer's Dement. 2025;21:e70541. 10.1002/alz.70541

DATA AVAILABILITY STATEMENT

Data can be requested from the Karolinska Institutet; https://ki.se/en/research/research‐infrastructure‐and‐environments/core‐facilities‐for‐research/the‐swedish‐twin‐registry.