Association between menopausal hormone therapy and Alzheimer’s disease neuropathology

Abstract

Background and Objectives:

While evidence suggests that the neurophysiological impact of estrogen decline during menopause may contribute to increased risk of Alzheimer’s Disease (AD) in women, the effect of menopausal hormonal therapy (MHT) on AD risk requires further study. We sought to examine the associations between MHT use and neuropathological, clinical and imaging/fluid biomarker outcomes.

Methods:

In this cohort study, we tested the association between estrogen-only MHT use and dementia outcomes in female participants using two independent, large-scale datasets: National Alzheimer’s Coordinating Center (NACC) and Alzheimer’s Disease Neuroimaging Initiative (ADNI). Participants included females over 50 years with self-reported use of estrogen-only MHT or no self-report of MHT. Clinical, imaging/fluid biomarker, and neuropathological outcomes were examined. Research was performed at academic medical centers.

Results:

Neuropathological data was collected from NACC for 258 MHT users (mean age of death = 81.9, SD = 19.5) and 2701 non-MHT users (mean age of death = 82.2, SD = 11.0). The ADNI cohort included 110 MHT users (mean age = 76.5, SD = 7.5) and 1948 non-MHT users (mean age = 73.2, SD = 8.9). The odds of increased AD pathology on autopsy (primary outcome) were significantly decreased in MHT users relative to non-users (OR 0.65, 95% CI, 0.48–0.88, p=0.005). MHT use was associated with secondary outcomes including significantly decreased amyloid pathological load assessed via plasma (ß=0.44, 95% CI, 0.16–0.73, p=0.0025) and cerebrospinal fluid (ß=0.07 0.002–0.13, p=0.030). MHT use was associated with significantly lower odds of clinical dementia diagnoses (OR 0.61, 95% CI, 0.55–0.67), p<0.0001) and lower odds of symptoms of memory/functional decline (OR 0.67, 95% CI, 0.61–0.74, p<0.0001).

Discussion:

Our findings demonstrate small but significant associations between MHT use during later life and a range of AD-related neuropathological and clinical outcomes in two large cohorts of female participants. Although our results do not address causality and have limited generalizability due to the retrospective nature of the study, they suggest a protective effect of MHT use in the dementia course.

INTRODUCTION

Alzheimer’s Disease (AD) is a leading cause of morbidity and mortality in the aging population worldwide.^1,2 More than 2/3 of AD patients are women, which has led to increased interest in the neurophysiological impact of estrogen decline during menopause.^1,2 Estrogen decline is postulated to impact multiple physiological processes relevant to brain aging, including neuronal synaptic plasticity, neuroinflammation, and blood brain barrier (BBB) integrity.³ While early evidence suggested that menopausal hormone therapy (MHT) may be protective against dementia,⁴ more recent studies have found inconclusive or even harmful effects.^5–8 Factors contributing to contrasting findings on the impact of MHT on brain aging and dementia include incomplete understanding of the optimal timing of its initiation as well as limited data on risk and protective factors that may mediate its impact, such as clinical and genetic risk factors. For example, subgroup analyses from the Women’s Health Initiative suggests that MHT initiation at older age (i.e., 65–79 years of age) may be associated with an increased risk of dementia, while larger samples comprising postmenopausal women over the age of 50 demonstrate no increased risk of dementia⁹ and small, positive associations with memory performance¹⁰. The formulation of MHT may also play a role, with studies examining estrogen-only MHT generally reporting improved dementia-related outcomes in MHT users while studies examining estrogen and progesterone combination MHT reporting increased risk of AD in MHT users or no association.⁸

Recent literature links potential benefits of MHT to underlying neurophysiological mechanisms. A recent UK biobank analysis examined mortality related to different classes of medications and found that estrogens, as a class, reduced mortality and further that 4 of the top 9 drugs that reduced mortality were estrogen preparations.¹¹ A proteomics study examining a small sample of women (N=47) found beneficial effects of estrogen therapy on immune aging and, to a less prominent degree, brain aging profiles.¹² Finally, a prospective treatment study indicated a significant positive effect of estrogen therapy on plasma amyloid beta levels.¹³

Although a definitive diagnosis of AD can only be obtained at autopsy,¹⁴ no previous study to date has measured the association between MHT use and AD-related neuropathological outcomes measured on autopsy data. Moreover, a recent review highlighted the need to adjust for key variables known to impact cognitive trajectories in later life, such as number of Apolipoprotein E (APOE) ε4 alleles and vascular risk factors such as hypertension, in order to more fully understand the potential role of MHT in the AD process¹⁵.

Our primary aim was to examine the association between estrogen-only MHT (estradiol or conjugated estrogens) and neuropathological outcomes in female participants from the National Alzheimer’s Coordinating Center (NACC). Second, we sought to evaluate the association between MHT use and molecular imaging and fluid biomarker measures of AD-related neuropathology measured in-vivo using data from female participants in the Alzheimer’s Disease Neuroimaging Initiative (ADNI). Finally, we examined the association between MHT use and clinical outcomes in both ADNI and NACC datasets, including clinical diagnosis of dementia, Clinical Dementia Rating-Global (CDR-G) scores and Rey Auditory Verbal Learning Test (RAVLT) scores. We hypothesized that women who used MHT during the study period would display less evidence of AD neuropathology on autopsy and, secondarily, less evidence of AD neuropathology assessed by in-vivo biomarkers. We also hypothesized that MHT users would demonstrate improved clinical outcomes relative to MHT non-users.

METHODS

Participants and Study Design

This retrospective longitudinal cohort study utilized participant information from the NACC and the ADNI. The NACC dataset involved data from 37 past and present Alzheimer’s Disease Research Centers (ADRCs) funded by the National Institute on Aging collected between June 21, 2005, and February 22, 2023. Data from the multisite ADNI database¹⁶ including ADNI1, ADNI2 and ADNIGO and ADNI 3 were collected between 2003 and 2018. The ADNI study aims to investigate biomarkers for early detection and treatment of AD. Detailed overall inclusion and exclusion criteria are described in the eMethods and in previous publications.^17,18

Standard Protocol Approvals, Registrations, and Patient Consents

All participants or their legally authorized representative completed written informed consent, and Institutional Review Boards at each participating NACC and ADNI institution reviewed and approved the respective protocols. The Stanford Institutional Review Board also approved the data analysis presented here (protocol 69421).

Participant Assessment and Inclusion Criteria

Demographic characteristics including participant sex, age, years of education, race and ethnicity were collected via participant self-report or caregiver report. Female participants aged 50 years or older were included in this analysis if they had available clinical and/or neuropathologic data for at least one major study outcome and had at least one visit of data on MHT use (see Figure 1). A participant was defined as an MHT user if they reported current use of estrogen-only hormone therapy at any study visit. Estrogen-only hormone therapy across both datasets included estradiol, conjugated estrogens and esterified estrogens, or estrogen; subgroups in the NACC dataset were created for conjugated estrogen and estradiol users. Topical preparations were not included for classification of MHT users. MHT non-users were defined as female participants who did not report using MHT as defined above at any NACC or ADNI visit. Female participants who reported estrogen + progestin combination therapy, irrespective of use of estrogen-alone MHT, were excluded as previous findings indicate this therapy may increase AD risk^5,8. Participants in both cohorts were subsequently included in the analysis of each outcome if the relevant data was found in the respective clinical database at a time point coinciding with or after which they reported use of MHT. Both databases collect data on a longitudinal basis, and we selected the last available assessment of each outcome to allow for accumulation of age-related neuropathology and/or cognitive decline. Male participants were excluded, as were participants without data on covariate variables, which are listed below.

Figure 1.

Inclusion criteria for NACC and ADNI participants included in this analysis. MHT use included self-reported current use of estrogen-only therapy at any NACC visit. Menopausal hormonal therapy (MHT) National Alzheimer’s Coordinating Center (NACC); Clinical Dementia Rating-Global (CDR-G).

Neuropathologic Outcomes

A subgroup of individuals had neuropathological data measured via brain autopsy for at least one outcome of interest. Our primary outcome was AD pathology assessed via the NIA-AA AD Neuropathologic Change score (ABC score), which is a composite of three components of AD pathology: amyloid-β (Aβ) deposits (“A” for Amyloid), neurofibrillary degeneration (“B” for Braak stage), and neuritic plaques (“C” for CERAD rating of plaque distribution). The ABC score is rated on a severity scale of 0–3 and is associated with cognitive decline in older adults.¹⁹ We also assessed several exploratory neuropathological variables including Frontotemporal Lobar degeneration, Cerebral Amyloid Angiopathy (CAA), Cerebrovascular Disease, and Hippocampal Atrophy. See eTable 1 for more details and descriptions for all secondary/exploratory outcomes.

Statistical Analysis

To assess differences in demographic characteristics between MHT users and MHT non-users, chi-squared tests and two-sample t-tests were used for categorical (i.e., sex) and continuous (i.e., age at initial NACC visit) variables, respectively. Demographic characteristics between participants with and without neuropathological data were also compared using the tests outlined above.

To examine the associations between MHT use and neuropathological outcomes, ordinal or binomial logistic regressions were performed when appropriate for ordinal (i.e., ABC score) and binary (i.e., presence of lacunes/infarcts) outcome variables, respectively, with each outcome of interest as the dependent variable. For all analyses involving the NACC dataset regressions included the following covariates which are known to be associated with trajectories of cognitive decline in later life: 1) current age (for clinical data at index visit) or age at death (for post-mortem neuropathological data)²⁰, 2) educational level²¹, 3) APOE genotype (number of ε4 alleles)²², 4) race (white vs. non-white race)²³, and 5) self-reported presence of recent or active hypertension²⁴. For analyses involving the ADNI dataset, covariates included 1) current age, 2) educational level, 3) APOE genotype (number of ε4 alleles); race and hypertension were included as additional covariates in sensitivity analysis due to the limited sample size and missing information for race and hypertension. MHT use status (users vs non-users) was the independent variable, with non-MHT users as the reference group.

For ordinal regressions, we used the R polr package; model coefficients and confidence intervals were scaled logarithmically. For binomial categorical variables, binary logistic regression models were fit using generalized linear models with a logit link. Odds ratios correspond to unique contributions of MHT use status to the model after accounting for covariates. Finally, for continuous outcome variables (i.e., RAVLT immediate memory scores), hierarchical linear regressions were performed, with the same independent variable and covariates as above. In hierarchical regressions, control variables were added sequentially and MHT use status was always added last within the R lm function from the stats package. The model significance was evaluated and R² change statistics were computed corresponding to the proportion of unique variance of MHT use after accounting for control variables.

For ordinal regression analyses, the proportional odds assumption was assessed using the Brant test prior to inference²⁵. When violations of the proportional odds assumption were identified, partial proportional odds models were fit, allowing selected covariates to vary across outcome thresholds. For all regression analyses, linearity of continuous covariates was examined using restricted cubic splines, and multicollinearity was evaluated using variance inflation factors. When evidence of nonlinearity was observed, variables were modeled using appropriate transformations or spline terms. In the presence of multicollinearity, model specifications were adjusted to improve stability. Hierarchical linear regression models were additionally evaluated for normality of residuals and homoscedasticity using Q-Q plots and residuals-versus-fitted-plots, respectively²⁶.

For our primary outcome – AD pathology measured via the ABC score – we performed inverse probability weighting to account for differences in deceased individuals with and without autopsy data. First, we computed a logistic regression model predicting autopsy status among all deceased NACC participants based on MHT use and known demographic factors related to autopsy selection: age, APOE genotype, race and education. We then used predicted probabilities from this model to calculate stabilized weights equal to the ratio of the marginal probability of autopsy to the individual predicted probability. Finally, stabilized weights were included along with other covariates in the model predicting ABC score from MHT use.

Statistical values for secondary outcomes (exploratory neuropathological, molecular imaging, and clinical variables) were corrected for multiple comparisons within each dataset (ADNI and NACC) using the false discovery rate procedure²⁷. We also conducted several exploratory, post-hoc analyses in the NACC dataset. Previous research has indicated that the protective effects of estrogen may be most significant for APOE ε4 carriers who are also at higher risk for AD¹³, but other research has indicated worse outcomes for APOE ε4 carriers who use MHT²⁸. Thus, we performed exploratory subgroup analysis for APOE ε4 allele carriers and non-carriers. In addition, given the demographic skew of our cohort towards self-report of MHT at older age (average age of initiation: 69.9 years in NACC), we performed exploratory subgroup analysis for the primary outcome. Specifically, we examined association between MHT use and ABC neuropathology score among participants who first reported MHT use prior to 60 years of age. Additional subgroup analyses for the ABC score included examining outcomes by type of estrogen only therapy (estradiol and conjugated estrogen). Finally, we examined estrogen-progesterone combined MHT use relative to MHT non-users but only for clinical outcomes due to lack of sufficient sample sizes for neuropathological data among the estrogen-progesterone combined MHT subgroup. Since these analyses were purely exploratory and separate from the main analysis, they were not included in multiple comparison correction. A detailed descriptions of secondary outcomes is included in the eMethods section. Due to limited sample sizes, we did not perform these secondary analyses in the ADNI cohort.

All hypothesis tests were two-sided and overall alpha was 0.05. We used R, version 4.2.1²⁹. When data was missing for a given participant, the participant was removed case-wise from the relevant analysis. The STROBE reporting guidelines³⁰ were used to draft this manuscript, and the STROBE reporting checklist was used during editing.³¹

Data Availability

The data used in this manuscript including de-identified individual participant data and data dictionaries defining each field in the set are available to qualified researchers upon request directly to NACC³² or ADNI¹⁶. The investigators for NACC and ADNI contributed to data used in this study; they did not participate in the analysis or writing of this report.

RESULTS

Demographic information of all eligible participants (NACC: N=19,402, N=2959 with neuropathological data, ADNI: N=2058) is presented in Table 1, with comparisons between MHT and non-MHT users. In NACC, participants were predominantly white (15,138, 78.0%), with a mean [SD] age of study initiation of 71.9 [9.4] years and a mean [SD] education of 14.8 [3.3] years. In ADNI, participants were predominantly white (1053, 76.6%), with a mean [SD] age of study initiation of 71.1 [7.62] years and a mean [SD] education of 15.7 [2.69] years. In addition, comparison of demographic characteristics between deceased participants with and without available autopsy data is presented in eTable 2.

Table 1.

Demographic characteristics of all NACC and ADNI participants with available clinical and/or neuropathological data included in this analysis. Continuous variables are presented as mean ± SD while categorical variables are presented as column-based percentages. NA represents cells that are not available or applicable.

	NACC Dataset (Clinical)			NACC Dataset (Neuropathological)				ADNI Dataset
	MHT non-users (N=17,561)	MHT users (N=1,841)	t-Value (df) or X2 (df), p-value	MHT non-users (N=2,701)	MHT users (N=258)	t-Value (df) or X2 (df), p-value	MHT non-users (N=1948)	MHT users (N=112)	t-Value (df) or X2 (df), p-value
Age at first visit, Mean Years ± SD [range]	72.1 ± 9.5 [50–106]	69.8 ± 8.9 [50–97]	−10.5 (2297.3), <0.0001	76.7 ± 10.6 [50–106]	75.0 ± 10.1 [51–94]	−2.6 (312.8), 0.009	70.7±7.9 [50–90]	72.2±6.8 [56–89]	−2.13 (126) <0.035
Age at first MHT use, Mean Years ± SD [range]	NA	71.0 ± 9.2 [50–100]	NA	NA	75.9 ± 10.4 [51–100]	NA	NA	72.2±6.8 [56–89]	NA
Study follow-up time, Mean years ± SD [range]	3.9 ± 3.9 [0–17.2]	5.6 ± 4.5 [0–17.2]	1545 (2149.5), <0.0001	4.1 ± 3.5 [0–16.1]	5.6 ± 3.7 [0–15.0]	6.2 (303.6), <0.0001	2.8±3.7 [0–29]	4.7±3.9 [0–19]	−4.72 (109) <0.001
Number of visits, Mean years ± SD [range]	4.2 ± 3.4 [1–18]	5.7 ± 3.9 [1–18]	15.4 (2141), <0.0001	4.5 ± 3.0 [1–15]	5.8 ± 3.3 [1–15]	6.3 (301.1), <0.0001	3.3±3.1 [0–21]	4.9±3.2 [0–17]	−5.13 (122), <0.001
Age at Death, Mean Years ± SD [range]	NA	NA	NA	82.2 ± 11.0 [51.2–107.6]	81.9 ± 10.5 [55.3–103.4]	−0.5 (312.7), 0.64	NA	NA	NA
Education, Mean Years ± SD [range]	14.7 ± 3.3 [0–27]	15.7 ± 2.8 [1–29]	14.7 (2428.7), <0.0001	14.7 ± 2.9 [0–24]	15.5 ± 2.4 [9–20]	5.0 (331.7), <0.0001	15.7 ± 2.7 [4–20]	15.2 ± 2.7 [8–20]	1.8 (119), 0.07
Race, N (%)
White	13475 (76.7)	1663 (90.3)	179.7 (1), <0.0001	2497 (92.4)	251 (97.3)	8.3 (1), 0.004	1521(78.1)	87 (79.1)	0.66 (3), >0.10
Black/African American	3076 (17.5)	141(7.7)		154 (5.7)	7 (2.7)		309 (15.5)	17 (15.5)
Asian	466 (2.7)	25 (1.4)		22 (0.8)	0 (0.0)		59 (3.0)	4 (3.6)
Other/Unknown	544 (3.1)	12 (0.7)		28 (1.0)	0 (0.0)		59 (3.0)	2 (1.8)
Ethnicity, N (%)
Non-Hispanic	15922 (90.7)	1749 (945.0)	37.1 (1), <0.0001	2601 (96.3)	253 (98.1)	1.2 (1), 0.28	157 (7.9)	4 (3.6)	2.8 (2), >0.10
Hispanic	1572 (9.0)	91 (4.9)		85 (3.1)	5 (1.9)		1798 (91.0)	105 (95.5)
Other/Unknown	67 (0.4)	4 (0.2)		15 (0.6)	0 (0.0)		20 (1.0)	1 (0.9)
APOE e4 Alleles, N (%)
0	10404 (59.2)	1138 (61.8)	9.2 (2), 0.01	1525 (56.5)	155 (60.1)	3.8 (2), 0.15	752 (54.7)	62 (62)	4.9 (2), 0.08
1	5991 (34.1)	611 (33.2)		968 (35.8)	78 (30.2)		498 (36.2)	35 (35)
2	1166 (6.6)	92 (5.0)		208 (7.7)	25 (9.7)		124 (9)	3 (3)

Neuropathologic outcomes

Statistical outcomes for all neuropathologic findings are found in Table 2 and Figure 2 (MHT users: N=258, non-MHT users: N=2701). The odds of having increased AD neuropathology on the ABC score were decreased in MHT users relative to non-MHT users (OR [95% CI]: 0.65 [0.48–0.88], p=0.005). This result remained significant after applying inverse probability weighting to account for potential selection bias in participants undergoing autopsy (0.61 [0.45–0.84], p=0.0023).

Table 2.

Neuropathological outcomes for NACC participants (N=2959) and molecular and fluid biomarkers for ADNI participants. A proportional odds model was used for categorical variables (ABC score) thus odds ratios reflect the odds of being in a higher category of the outcome (vs all lower categories combined), for MHT use vs no use.

	MHT non-users	MHT users	OR, 95% CI	p value
Alzheimer’s Disease (ABC score)
Not AD, N (%)	154 (10.2%)	30 (17.9%)	0.65, 0.48 – 0.88	0.005*a
Low ADNC, N (%)	262 (17.4%)	32 (19.0%)
Intermediate ADNC, N (%)	328 (21.7%)	39 (23.2%)
High ADNC, N (%)	765 (50.7%)	67 (39.9%)
Hippocampal Atrophy
None, N (%)	361 (24.6%)	56 (34.8%)	0.75, 0.56 – 1.01	0.061
Mild, N (%)	385 (26.2%)	38 (23.6%)
Moderate, N (%)	395 (26.9%)	33 (20.5%)
Severe, N (%)	328 (22.3%)	34 (21.1%)
Cerebral Amyloid Angiopathy
None, N (%)	1015 (38.3%)	120 (47.6%)	0.77, 0.60 – 0.98	0.035*
Mild, N (%)	811 (30.6%)	66 (26.2%)
Moderate, N (%)	542 (20.5%)	47 (18.7%)
Severe, N (%)	279 (10.5%)	19 (7.5%)
Frontotemporal Lobar Degeneration
TDP-43 Pathology, N (%)	97 (7.3%)	9 (6.1%)	0.75, 0.36 – 1.57	0.45
Tau Pathology, N (%)	249 (16.3%)	29 (17.2%)	1.01, 0.66 – 1.56	0.95
Cerebrovascular Disease
Single or Multiple Hemorrhages, N (%)	38 (2.5%)	4 (2.4%)	0.95, 0.33 – 2.71	0.92
Lacunes/Infarcts, N (%)	535 (20.0%)	32 (12.5%)	0.59, 0.40 – 0.88	0.009**
	MHT users	MHT non-users
	N, Mean± SD	N, Mean± SD	β, CI	R2 Change, P value
Plasma AB42/AB40	46, 0.093±0.03	692, 0.086±0.015	0.443 (0.157, 0.729)	0.011, 0.003**b
Plasma pT217/AB42	46, 0.01±0.01	692, 0.014±0.019	−0.219 (−0.507, 0.069)	0.003, >0.10
CSF ABETA42	59, 1284.5±674.2	677, 1084.5±649.2	0.066 (0.002, 0.131)	0.004, 0.04*
CSF PTAU	59, 24.88±14.43	669, 28.71±16.15	−0.051 (−0.119, 0.017)	0.003, >0.10
CSF PTAU/ABETA40	24, 0.001±0.0003	249, 0.002±0.002	−0.038 (−0.155, 0.079)	0.001, >0.10
Amyloid PET	50, 0.795±0.154	528, 0.835±0.205	−0.038 (−0.113, 0.037)	0.175, >0.10
Tau PET	42,1.272±0.275	605, 1.320±0.362	−0.14 (−0.435, 0.155)	0.001, >0.10
FDG PET	53, 1.177±0.174	623, 1.183±0.186	−0.053 (−0.319, 0.212)	0, >0.10

^*p<0.05 at the nominal level;

^**significant after correction for multiple comparisons across all secondary/exploratory outcomes in NACC dataset, p<0.0045. Note that our primary outcome, ABC score, is not subject to multiple comparisons. Two subtypes of Frontotemporal Lobar degeneration pathology were included: tau and transactive response DNA binding protein of 43 kDa (TDP-43).

^aResult remains significant after inverse probability weighting is added to regression model to account for differences between autopsied and non-autopsied individuals.

^bResult is significant (p<0.05 at the nominal level) when race and hypertension are included as covariates.

Figure 2.

Neuropathological and molecular and fluid biomarker outcomes. A. Adjusted odds of neuropathological outcomes for menopausal hormone therapy (MHT) use. Dots indicate odds ratios and tails indicate 95% confidence intervals. Odds ratios below 1 indicate the likelihood of respective outcome is significantly less for MHT users relative to non-users. All outcomes listed in this table are pathological; thus, a lower likelihood is consistent with improved outcomes in MHT users. FTD = Frontotemporal lobar degeneration; CVD=Cerebrovascular Disease; CAA = Cerebral Amyloid Angiopathy; Transactive response DNA binding protein of 43 kDa (TDP-43). B. Standardized betas associated with MHT use for molecular imaging and fluid biomarkers. Beta values greater than 0 indicate that MHT use was associated with higher scores on a given outcome. Amyloid burden = Florbetapir PET standardized uptake value ratio. Greater score indicates higher amyloid burden in the gray matter relative to a white matter reference region; CSF = cerebrospinal fluid, Higher CSF amyloid beta indicates less pathology. FDG=18F-fluorodeoxyglucose positron emission tomography. Higher FDG indicates higher brain metabolism.

We also found that MHT use was associated with lower odds of CAA neuropathology (0.77 [0.60 – 0.98], p=0.035) and lacunes/infarcts (0.59 [0.40 – 0.88], p=0.009). MHT use was not associated with hippocampal atrophy (0.75 [0.56–1.01], p=0.06), frontotemporal lobar degeneration with TDP-43 pathology or tau pathology (p’s>0.10) or presence of multiple hemorrhages (p>0.10). Given that our primary aim was to assess the association between MHT use and AD-related neuropathology (the ABC score) independent of vascular burden, we re-examined associations between MHT use and ABC score while adding CAA severity and presence of lacunes/infarcts as covariates. When covarying for degree of CAA neuropathology, the association between MHT use and lower odds of AD neuropathology remained significant (0.67 [0.50–0.92], p=0.013). When covarying for presence of lacunes/infarcts, the association between MHT use and lower odds of AD neuropathology remained significant (0.66 [0.49–0.90], p=0.008).

Molecular imaging and fluid biomarker outcomes

Statistical outcomes for all molecular imaging and fluid biomarker variables are found in Table 2 and Figure 2. Amyloid brain pathological load was decreased among MHT users relative to non-MHT users when examining 1) plasma Aβ42/40 (ß=0.44, 95% CI, 0.16–0.73, p=0.0025), 2) cerebrospinal fluid (CSF) Aβ1–42, (ß=0.07 0.002–0.13, p=0.030). Lower plasma Aβ42/40 ratio and CSF Aβ1–42 are indicative of greater AD brain pathology because Aβ1–42 reduction indicates that the least soluble Aβ1–42 is being converted into insoluble plaques in the brain.³³ We did not find evidence for an association between MHT use and other molecular or fluid biomarkers as described in Table 2.

Clinical outcomes

Statistical outcomes for all NACC clinical variables are found in Table 3 and Figure 3 (MHT users: N=1,841, non-MHT users: N=17,561). We found that the odds of a clinician-defined dementia diagnosis at the last available clinical visit was significantly lower for MHT users within the NACC dataset (OR [95% CI]: 0.61 [0.55 – 0.67], p<0.0001). The result was not significant within the smaller ADNI dataset (0.66 [0.437–1.02], p=0.06). MHT use was associated with lower odds of clinical decline as measured via the CDR-G within the NACC dataset (0.67 [0.61– 0.74], p<0.0001). These results were not significant in the ADNI dataset (p>0.10).

Table 3.

Clinical outcomes including diagnosis, dementia-related cognitive decline and verbal memory performance in NACC participants (N=19402) and ADNI participants (N=2351) at their last clinical visit. CDR = Clinical Dementia Rating, higher score = greater evidence of dementia; RAVLT = Rey Auditory Verbal Learning Test, higher score on RAVLT total indicates higher immediate memory performance. Higher score on RAVLT learning indicates greater learning across the 5 immediate memory trials. Higher score on RAVLT forgetting indicates more items forgotten between immediate memory trial 5 and delayed memory.

	NACC Dataset				ADNI Dataset
Outcome	N MHT Users	N MHT non-users	OR, 95% CI	p-value	N MHT Users	N MHT non-users	OR, 95% CI	p-value
Clinical Diagnosis	1841	17,561	0.61, 0.55 – 0.67	<0.0001**	91	1253	0.667 (0.437–1.018)	0.060
CDR-Global	1841	17,561	0.67, 0.61 – 0.74	<0.0001**	98	1324	0.814 (0.55, 1.205)	>0.10
	NACC Dataset				ADNI Dataset
Outcome	MHT Users (N=77)	MHT non-users (N=408)	β, CI	R2 Change, p-value	MHT Group (N, mean ± SD)	No MHT Group (N, mean ± SD)	β, CI	R2 Change, p-value
RAVLT Immediate, mean ± SD	53.0 ± 13.2	49.2 ± 14,1	0.14, −0.089 – 0.37	0.003, 0.23	85, 39.81±15.74	985, 35.78±16.44	0.251 (0.051, 0.451)	0.005, 0.0412*
RAVLT Learning, mean ± SD	5.8 ± 2.7	5.9 ± 2.8	−0.04, −0.38 – 0.30	0, 0.82	85,4.86±2.93	985, 4.31±2.88	0.194, (−0.014, 0.402)	0.003, 0.067
RAVLT Forgetting, mean ± SD	2.2 ± 2.1	2.3 ± 2.7	0.05, −0.28 – 0.38	0, 0.77	85, 3.88±2.71	985, 4.10±2.70	−0.086, (−0.305, 0.133)	0.001, >0.10

^*p<0.05 at the nominal level;

^**p<0.05 after correction for multiple comparisons across all exploratory outcomes in NACC or ADNI dataset.

^a.Results are significant (p<0.05 at the nominal level) when race and hypertension are included as covariates.

Figure 3.

Impact of MHT use on clinical outcomes. A. Adjusted odds of clinical outcomes for the MHT group. Odds ratios below 1 indicate likelihood of given outcome is significantly less for the MHT group. B. Standardized beta values evaluating the association between MHT use and memory performance. Beta value greater than 0 indicates that MHT use was significantly associated with higher scores on a given outcome. CDR = Clinical Dementia Rating, higher score = greater evidence of dementia; RAVLT = Rey Auditory Verbal Learning Test. Higher score on RAVLT total indicates higher immediate memory performance. Higher score on RAVLT learning indicates greater learning across the 5 immediate memory trials. Higher score on RAVLT forgetting indicates more items forgotten between immediate memory trial 5 and delayed memory.

We found a significant positive association between MHT use and verbal memory scores (RAVLT) within the ADNI dataset. MHT use was associated with increased scores on immediate memory (ß=0.336; p=0.002) and learning (ß= 0.233, p=0.033) but not forgetting (p>0.10). In the NACC dataset, MHT use was not associated with immediate memory performance (ß=0.14; p=0.23), learning scores (ß=−0.04; p=0.82), or forgetting scores (ß=0.05; p=0.77). To understand the effect of MHT use on cognition in individuals who did not have dementia, we performed post-hoc analysis in the ADNI dataset including only individuals who did not receive a diagnosis of dementia. In this analysis, MHT use was associated with increased verbal memory scores in the subgroup without dementia (immediate memory: ß=0.361; p=0.004, learning: ß=0.249; p=055).

Exploratory analysis within number of APOE ε4 alleles, type of MHT, and younger age subgroups

MHT users who were APOE ε4 non-carriers (N=11,542) had decreased odds of AD neuropathology compared to non-MHT users, but this result did not hold for APOE ε4 carriers (N=7,860, eTable 3). For clinical outcomes, we found that among both APOE ε4 carrier (N=7860) and non-carrier subgroups (N=11542), MHT use was associated with improved CDR-G scores and less severe clinical diagnoses (eTable 4).

The odds of having increased AD neuropathology on the ABC score were decreased in conjugated estrogen users relative to non-MHT users (0.55 [0.34–0.88], p=0.013), while the odds of AD neuropathology did not differ significantly between estradiol users and non-MHT users (0.72 [0.50–1.05], p=0.086). Given the small sample size of the subgroup of patients who reported both conjugated estrogen and estradiol users (n=25), associations with neuropathological outcomes were not reported. We also performed a sensitivity analysis comparing clinical outcomes between estrogen-progestin users (N=158) and MHT non-users (N=17,561), eTable 5.

In the subgroup of female participants reporting MHT initiation before age 60 (n = 61), MHT use was associated with a lower odds of AD neuropathology (OR = 0.30; 95% CI: 0.08–1.10), representing a suggestive association; however, this did not reach statistical significance (p = 0.059).

DISCUSSION

We found evidence for small but significant associations between estrogen-only MHT use and multiple neuropathological and clinical measures in two large and distinct research cohorts of female older adults who on average reported MHT use after the age of 70. First, MHT use was associated with lower risk of AD neuropathology measured post-mortem. Second, MHT use was associated with lower amyloid pathological burden measured in vivo. Finally, MHT use was associated with improved clinical outcomes including lower odds of clinical diagnosis of dementia, lower odds of clinical and functional decline, (CDR-G scales) and increased memory performance (RAVLT). While the observational nature of our analysis prevents ascertainment of the causal relationship between MHT use and clinical and neuropathological dementia outcomes, we highlight several strengths of our analysis that help to add relevant insight into the use of this controversial therapy in older women.

To our knowledge, the present study represents the first investigation of the associations between MHT use and neuropathological outcomes measured postmortem. After adjusting for relevant covariates including age, education, race, presence of hypertension, and number of APOE ε4 alleles, we found evidence for associations between MHT use and AD neuropathology as measured by the AD Neuropathologic Change (ABC) score, which is accepted as the diagnostic standard for postmortem confirmation of AD pathology^34,35. More research is needed to understand the mechanism by which MHT may act to disrupt the buildup of AD pathologic hallmarks in humans, but animal models suggest that the estrogenic component of MHT may act to shift Aβ towards the non-amyloidogenic pathway, preventing aggregation, promoting dephosphorylation of tau proteins, and limiting formation of hyperphosphorylated intracellular deposits.³⁶

Interestingly, we also found associations between MHT use and previously unexplored outcomes such as pathology associated with cerebrovascular disease (lacunes/infarcts) as well as the severity of CAA pathology, which supports the notion that MHT use may be associated with decreased amyloid accumulation. These novel results warrant follow-up and attention to the timing of initiation of MHT use, particularly given conflicting past research on the impact of MHT on risk of cerebrovascular disease and mortality.^8,37–39 While we cannot confirm the precise timing of hormone therapy initiation, we note that current MHT use was reported approximately seven years prior to neuropathological assessment in the present study, allowing for ample time to observe the potential effects of therapy.

We also found associations between MHT use and amyloid measured via plasma and CSF, without observing associations with accumulation of amyloid on PET or accumulation of Tau (CSF or PET). This finding clarifies our neuropathological results, indicating that MHT may be associated with lower amyloid accumulation during later life. The lack of associations between MHT use and amyloid burden measured via PET may be due to the greater sensitivity of both plasma⁴⁰ and CSF⁴¹ measurements relative to PET for early amyloid accumulation. Thus, this pattern of results may indicate that MHT use is associated lower levels of early amyloid accumulation in this cohort. The association between MHT use and memory scores in the present study (Figure 3) supports the hypothesis that MHT use is associated with higher cognitive reserve which is thought to delay detection and/or progression of AD.⁴² Future longitudinal prospective cohort studies will be important for confirming this hypothesis and assessing the directionality of the observed associations.

In exploratory post-hoc analysis stratified by APOE genotype we revealed patterns that may contribute to understanding the existing literature. First, MHT use was associated with decreased post-mortem AD pathology (ABC score) in women without the high risk APOE ε4 genotype but not in APOE ε4 carriers. These findings partially support previous work which indicated that MHT may be associated with improved dementia outcomes among APOE ε4-negative women only^43,44 although outcomes utilized in those studies were non-specific to AD (memory and cognitive decline) and more recent literature has contradicted these findings.^3,45 On the other hand, in analyses of clinical outcomes, MHT use was associated with lower odds of clinical and functional decline and lower odds of clinical dementia diagnosis in both APOE ε4 carriers and non-carriers. Thus, our results indicate that MHT may be associated with improved cognitive outcomes in later life across a wider population, and that other individual differences, such as timing of MHT initiation, may be equally important for fully understanding the impact of MHT on cognitive aging.

Additional subgroup analyses revealed that the association between MHT use and neuropathological correlates of dementia varied by type of therapy; conjugated estrogen use only was associated with lower odds of AD (ABC) pathology. Few studies to date have examined the impact of type of MHT on cognition, although one study suggests that postmenopausal women at increased risk for AD receiving estradiol therapy had higher verbal memory performance than those receiving conjugated estrogens⁴⁶. While the results are not conclusive and are limited by small sample size within each estrogen therapy subgroup, we do present important preliminary information that conjugated estrogens may be more strongly associated with lower odds for AD neuropathology.

Limitations

While the use of two independent large-scale studies of aging and AD (ADNI and NACC) represents a strength of our study, we also note several limitations. First, while we did covary for known risk factors for cognitive decline during later life such as age, APOE genotype, education, race and hypertension, it is possible that associations between MHT use and dementia outcomes may be confounded by improved pre-existing health status of the MHT group. Importantly, we did demonstrate that the association between MHT use and post-mortem AD pathology was maintained after controlling for potential selection bias in participants undergoing autopsy and presence of CAA and lacunes/infarcts, suggesting some specificity for the association between MHT and AD pathology. Second, while we were able to mine both datasets to classify participants who had used MHT during the study period, information regarding the duration of MHT use or history of MHT use prior to study initiation was not available. While the coding of the longitudinal data available ensured that MHT use occurred prior to the measurement of each outcome of interest, the results of our analysis cannot be generalized to represent associations between lifetime estrogen exposure and dementia outcomes. Topical estrogen preparations were excluded from the definition of menopausal hormone therapy (MHT) in the NACC dataset; therefore, information on their use was not available. However, systemic absorption of low-dose topical estrogen is minimal⁴⁷ and this limitation is unlikely to have meaningfully biased our results. In the ADNI dataset, topical estrogen preparations were explicitly excluded from both the MHT and non-MHT groups. We also note the relatively small sample size for outcomes in the ADNI dataset and consider these results exploratory in nature.

Although we adjusted for a range of measured covariates, we recognize that unmeasured or imperfectly measured factors may still influence the results. Differences in outcomes between MHT users and non-users could reflect differences such as baseline health status, access to and engagement with healthcare, and survivorship-related factors, rather than a direct causal effect of MHT itself. Estrogen-only MHT is not recommended for individuals with an intact uterus due to the risk of endometrial hyperplasia and cancer⁴⁸ thus standard clinical practice indicates combined estrogen plus progesterone therapy. Our findings are most directly applicable to the minority subset of women who have undergone hysterectomy and may be prescribed unopposed estrogen. Therefore, the generalizability of our results to the broader population of MHT users is substantially limited. Furthermore, our study sample, which included estrogen-only MHT users with a mean age of 70.4 years in NACC, contrasts with current MHT practice which typically involves combined estrogen-progestogen therapy initiated in the perimenopausal period (often in the late 40s to early 50s), with discontinuation generally recommended before age 60.⁴⁹ The potential neuroprotective effects observed in our study with estrogen-only MHT may not extend to combined estrogen-progestogen regimens, which are the predominant formulation prescribed in current clinical practice. Unfortunately, NACC and ADNI did not systematically collect data on history of hysterectomy, and thus we could not report this data accurately for our study samples. An exploratory subgroup analysis among women who reported initiation of MHT before age 60 showed lower odds of AD neuropathology, suggesting that our results may generalize to younger women using MHT. However, it is important to note this difference did not reach statistical significance (p = 0.059), potentially due to the reduced sample size (N=60 with MHT under 60 years), and more research will be important in order to reach a definitive conclusion.

Despite these limitations, our findings present evidence of an association between exposure to estrogen-only MHT during later life and a range of AD-related neuropathological and clinical outcomes in two cohorts of female older adults. Our study is the first to our knowledge to evaluate the relationship between MHT use and a more definitive measure of AD pathology measured postmortem via autopsy, relative to previous work investigating only in-vivo and clinical measures. The addition of molecular imaging and fluid biomarkers help to suggest plausible mechanisms through which MHT may be associated with neuropathology. Prospective trials investigating the impact of MHT on neuroimaging and/or neuropathological markers of AD are needed to better assess the causality of the observed associations. We hope that our work combined with these future investigations will help guide clinicians in their consideration of this controversial therapy for menopausal and post-menopausal women, particularly in those with history of a hysterectomy in which estrogen-only MHT can be safely prescribed.

Appendix 1. Alzheimer’s Disease Neuroimaging Initiative (ADNI) Study Group Members

Name	Location	Role	Contribution
Michael Weiner, MD	University of California, San Francisco	Principal Investigator	Study leadership, infrastructure oversight, study design
Paul Aisen, MD	University of Southern California	Clinical Core PI	Clinical coordination and oversight
Ronald Petersen, MD, PhD	Mayo Clinic, Rochester	Clinical Core PI	Clinical leadership and participant oversight
Clifford R. Jack, Jr., MD	Mayo Clinic, Rochester	MRI Core PI	MRI protocol development and imaging oversight
William Jagust, MD	University of California, Berkeley	PET Core PI	PET imaging leadership and biomarker oversight
Susan Landau, PhD	University of California, Berkeley	PET Core PI	PET data processing and imaging oversight
Monica Rivera-Mindt, PhD	Fordham University; Mt. Sinai Medical Center	Engagement Core PI	Community engagement and study coordination
Ozioma Okonkwo, PhD	University of Wisconsin	Engagement Core PI	Community engagement and recruitment oversight
Leslie M. Shaw, PhD	University of Pennsylvania	Biomarkers Core PI	Biomarker assay development and oversight
Edward B. Lee, MD, PhD	University of Pennsylvania	Biomarkers Core PI	Biomarker and neuropathology oversight
Arthur W. Toga, PhD	University of Southern California	Informatics Core PI	Data infrastructure and informatics oversight
Laurel Beckett, PhD	University of California, Davis	Biostatistics Core PI	Statistical analysis oversight
Danielle Harvey, PhD	University of California, Davis	Biostatistics Core PI	Statistical methodology and analysis
Robert C. Green, MD, MPH	Harvard University	Data and Publications Committee Chair	Publication oversight and governance
Andrew J. Saykin, PsyD	Indiana University	Genetics Core PI	Genetics study oversight
Kwangsik Nho, PhD	Indiana University	Genetics Core PI	Genetics data analysis
Richard J. Perrin, MD, PhD	Washington University St. Louis	Neuropathology Core PI	Neuropathology oversight
Duygu Tosun-Turgut, PhD	University of California, San Francisco	MRI Core Investigator	Neuroimaging analysis and methodology
Pallavi Sachdev, PhD	Eisai	PPSB Chair	Scientific advisory oversight
Tom Montine, MD, PhD	University of Washington	Resource Allocation Review Committee Chair	Resource review and oversight
Rachel Nosheny, PhD	University of California, San Francisco	Administrative Core	Administrative coordination and project management
Diana Truran Sacrey	Northern California Institute for Research and Education	Administrative Core	Administrative support
Juliet Fockler	University of California, San Francisco	Administrative Core	Administrative support
Melanie J. Miller, PhD	Northern California Institute for Research and Education	Administrative Core	Project coordination
Catherine (Cat) Conti	Northern California Institute for Research and Education	Administrative Core	Administrative support
Winnie Kwang, MA	University of California, San Francisco	Administrative Core	Data and project coordination
Chengshi Jin, PhD	University of California, San Francisco	Administrative Core	Data management
Adam Diaz, MS	Northern California Institute for Research and Education	Biostatistics Core	Statistical support and data analysis
Miriam Ashford, PhD	Northern California Institute for Research and Education	Administrative Core	Administrative coordination
Derek Flenniken	Northern California Institute for Research and Education	Administrative Core	Administrative support
Adrienne Kormos	Northern California Institute for Research and Education	Administrative Core	Administrative support
Michael Rafii, MD, PhD	University of Southern California	Clinical Core Investigator	Clinical study oversight
Rema Raman, PhD	University of Southern California	Clinical Core Investigator	Statistical and clinical coordination
Gustavo Jimenez, MBS	University of Southern California	Clinical Core	Clinical coordination
Michael Donohue, PhD	University of Southern California	Clinical/Biostatistics Core	Statistical analysis and study coordination
Jennifer Salazar, MBS	University of Southern California	Clinical Core	Clinical coordination
Andrea Fidell, MPH	University of Southern California	Clinical Core	Clinical project management
Virginia Boatwright, BS	University of Southern California	Clinical Core	Clinical coordination
Justin Robison, MS	University of Southern California	Clinical Core	Data management
Caileigh Zimmerman, MS	University of Southern California	Clinical Core	Clinical coordination
Yuliana Cabrera, BS	University of Southern California	Clinical Core	Clinical support
Sarah Walter, MSc	University of Southern California	Clinical Core	Clinical coordination
Taylor Clanton, MPH	University of Southern California	Clinical Core	Clinical operations
Elizabeth Shaffer, BS	University of Southern California	Clinical Core	Clinical support
Caitlin Webb, BA	University of Southern California	Clinical Core	Clinical coordination
Lindsey Hergesheimer, BS	University of Southern California	Clinical Core	Clinical coordination
Stephanie Smith, BS	University of Southern California	Clinical Core	Clinical support
Sheila Ogwang, MPH	University of Southern California	Clinical Core	Clinical coordination
Olusegun Adegoke, MSc	University of Southern California	Clinical Core	Data coordination
Payam Mahboubi, MPH	University of Southern California	Clinical Core	Data management
Jeremy Pizzola, BA	University of Southern California	Clinical Core	Clinical operations
Cecily Jenkins, PhD	University of Southern California	Clinical Core	Clinical coordination
Joel Felmlee, PhD	Mayo Clinic, Rochester	MRI Core Investigator	MRI methods and acquisition
Nick C. Fox, MD	University College London	MRI Core Investigator	MRI methodology
Paul Thompson, PhD	UCLA School of Medicine	MRI Core Investigator	Neuroimaging analysis
Charles DeCarli, MD	University of California, Davis	MRI Core Investigator	MRI acquisition and analysis
Kejal Kantarci, MD	Mayo Clinic, Rochester	MRI Core Investigator	MRI biomarker analysis
Prashanthi Vemuri, PhD	Mayo Clinic, Rochester	MRI Core Investigator	Neuroimaging analysis
Paul A. Yushkevich, PhD	University of Pennsylvania	MRI Core Investigator	Imaging methodology
Sandhitsu Das, PhD	University of Pennsylvania	MRI Core Investigator	Imaging analytics
Robert A. Koeppe, PhD	University of Michigan	PET Core Investigator	PET acquisition and analysis
Gil Rabinovici, MD	University of California San Francisco	PET Core Investigator	PET biomarker interpretation
Victor Villemagne, MD	University of Pittsburgh	PET Core Investigator	PET biomarker analysis
Brian LoPresti, MSNE	University of Pittsburgh	PET Core Investigator	PET imaging support
John Morris, MD	Washington University St. Louis	Neuropathology Core Investigator	Neuropathological assessment
Erin Franklin, MS	Washington University St. Louis	Neuropathology Core	Neuropathology support
Nigel J. Cairns, PhD, MRCPath	Washington University St. Louis	Neuropathology Core	Neuropathological analysis
Virginia M.Y. Lee, PhD, MBA	UPenn School of Medicine	Biomarkers Core Investigator	Biomarker development
Magdalena Korecka, PhD	UPenn School of Medicine	Biomarkers Core Investigator	Biomarker assays
Magdalena Brylska, MS	UPenn School of Medicine	Biomarkers Core	Laboratory analysis
Yang Wan, MS	UPenn School of Medicine	Biomarkers Core	Biomarker analysis
Tatiana M. Foroud, PhD	Indiana University School of Medicine	NCRAD Director	Genetics repository oversight
Shannon L. Risacher, PhD	Indiana University School of Medicine	Genetics Core Investigator	Genetics and imaging analysis
Liana G. Apostolova, MD	Indiana University School of Medicine	Genetics Core Investigator	Neurogenetics
Rima Kaddurah-Daouk, PhD	Duke University/AD Metabolomics Consortium	Genetics/Metabolomics Investigator	Metabolomics analysis