Associations of Epigenetic Age Estimators With Cognitive Function Trajectories in the Women's Health Initiative Memory Study

Abstract

Background and Objectives

Epigenetic age estimators indicating faster/slower biological aging vs chronological age independently associate with several age-related outcomes; however, longitudinal associations with cognitive function are understudied. We examined associations of epigenetic age estimators with cognitive function measured annually.

Methods

This longitudinal study consisted of older women enrolled in the Women's Health Initiative Memory Study with DNA methylation (DNAm) collected at baseline (1995–1998) from 3 ancillary studies and were followed up to 13 years. Global cognitive function was measured annually by Modified Mini-Mental State Examination (3MS; baseline–2007) and by modified Telephone Interview for Cognitive Status (TICS-m, 2008–2021). We calculated 5 epigenetic age estimators: extrinsic AgeAccel, intrinsic AgeAccel, AgeAccelPheno, AgeAccelGrim2, Dunedin Pace of Aging Calculated From the Epigenome (DunedinPACE), and AgeAccelGrim2 components (DNA-based plasma protein surrogates). We estimated longitudinal epigenetic age estimator-cognitive function associations using linear mixed-effects models containing age, education, race or ethnicity, and subsequently alcohol, smoking, body mass index, and comorbidities. We examined effect modification by APOE ε4 carriage.

Results

A total of 795 participants were enrolled. The mean baseline age was 70.8 ± 4 years (10.7% Black, 3.9% Hispanic or Latina, 85.4% White), A 1-SD (0.12) increment in DunedinPACE associated with faster annual declines in TICS-m scores in minimally adjusted (β = −0.118, 95% CI −0.202 to −0.034; p = 0.0006) and fully adjusted (β = −0.123, 95% CI −0.211 to −0.036; p = 0.006) models. AgeAccelPheno associated with faster annual declines in TICS-m with minimal adjustment (β = −0.091, 95% CI −0.176 to −0.006; p = 0.035) but not with full adjustment. No other epigenetic age estimators associated with changes in 3MS or TICS-m. Higher values of DNAm-based surrogates of growth differentiation factor 15, beta-2 microglobulin, Cystatin C, tissue inhibitor metalloproteinase 1, and adrenomedullin associated with faster annual declines in 3MS and TICS-m. Higher DNAm log A1c associated with faster annual declines in TICS-m only. DunedinPACE associated with faster annual declines in 3MS among APOE ε4 carriers but not among noncarriers (p-interaction = 0.020).

Discussion

Higher DunedinPACE associated with faster declines in TICS-m and 3MS scores among APOE ε4 carriers. DunedinPACE may help identify older women at risk of future cognitive decline. Limitations include the ancillary studies that collected epigenetic data not designed to study epigenetics and cognitive function. We examined epigenetic age estimators with global cognitive function and not specific cognitive domains. Findings may not generalize to men and more diverse populations.

Introduction

The maintenance of cognitive function, which refers to mental abilities such as learning, thinking, reasoning, remembering, attention, understanding, memory, and decision making, is critical for independence in aging and is thus an increasingly important public health challenge given the growth of the aging population. Many aspects of cognitive function decline with advancing age, even in the absence of neurodegenerative disorders including Alzheimer disease.^1,2 There is substantial heterogeneity in rates of cognitive decline across individuals.^2,3 Numerous studies have shown that various modifiable and nonmodifiable risk factors including education, socioeconomic status, physical activity, cardiovascular diseases (CVD), CVD risk factors,^4,5 hearing ability, and genetic factors (e.g., APOE ε4)⁶ are associated with cognitive decline.⁷ However, the age-associated molecular mechanisms underlying cognitive decline with aging are not fully understood.

Epigenetic clocks are DNA methylation (DNAm)-based biomarkers that capture biological processes involved in aging.^8-12 Epigenetic age acceleration indicates whether an individual is aging faster or more slowly biologically relative to their chronological age.⁸ Studies have shown that epigenetic age acceleration is prospectively associated with higher risk of mortality, coronary heart disease (CHD), and diverse age-related phenotypes, including poorer physical function, frailty, and cancer.^8,13,14 However, the association of epigenetic age acceleration with cognitive outcomes is less clear. Several studies have examined its association with cognitive function, but most were cross-sectional, or, if longitudinal associations were examined, had few repeated measures of cognitive function or relatively short follow-up periods.^15-18 Therefore, more research on the associations of epigenetic age acceleration with cognitive trajectories over longer periods among older adults is warranted.

In this study, we examined longitudinal associations of 5 well-known and widely studied blood-based epigenetic age estimators, including extrinsic epigenetic age acceleration (EEAA), intrinsic epigenetic age acceleration (IEAA), AgeAccelGrim2, AgeAccelPheno, and Dunedin Pace of Aging Calculated From the Epigenome (DunedinPACE), with cognitive function trajectories up to 13 years in a diverse cohort of older women. We hypothesized that epigenetic age acceleration, indicating faster biological aging, would be associated with faster declines in cognitive function measured annually.

Methods

Study Population

The Women's Health Initiative (WHI) Memory Study (WHIMS) is an ancillary study of the WHI hormone therapy trials. WHIMS was designed to investigate the effects of hormone therapy on cognitive outcomes among 7,427 women aged 65–80 years who were cognitively unimpaired at randomization in 1995–1998.¹⁹ Details on the WHIMS design and protocols have been published.¹⁹ Briefly, WHIMS conducted annual face-to-face follow-up cognitive assessments using the Modified Mini-Mental State Examination (3MS) from baseline until 2007 and in 2008 transitioned into the WHIMS-Epidemiology of Cognitive Health Outcomes (WHIMS-ECHO) study (n = 2,900), with annual telephone-based cognitive assessments using the modified Telephone Interview for Cognitive Status (TICS-m) through 2021.^19,20 The WHIMS of Younger Women (WHIMS-Y; n ∼1,000) study assessed (from 2009 to 2016) the long-term impact of hormone therapy assignment on cognitive function among women aged 50–54 at randomization in the WHI hormone therapy trials and had the same assessment procedures as WHIMS-ECHO.

The present analysis included data from WHIMS and WHIMS-Y women who were also selected for 3 WHI ancillary studies: (1) Integrative Genomics and Risk of Coronary Heart Disease and Related Phenotypes (BAA23), (2) Epigenetic Mechanisms of Particulate Matter-mediated Cardiovascular Disease Risk (EMPC), and (3) controls from the Bladder Cancer and Leukocyte Methylation Study (AS311).

BAA23 consisted of a case-cohort of 2,098 women in WHI who were free of CHD at baseline.²¹ Cases were defined as incident physician-adjudicated myocardial infarction, angina, coronary revascularization, or CHD death. EMPC was designed to study the epigenetic mechanisms of the associations between particulate matter air pollution and CVD consisting of a stratified random sample of 2,200 WHI clinical trial participants representative of the 68,132 participants randomized to the hormone therapy, calcium and vitamin D supplementation, or dietary modification trials.²² AS311 was a matched case-control study of bladder cancer (n = 405 cases and n = 455 controls), with matching based on enrollment year, age at enrollment, follow-up time, and DNA extraction method.²³ Each ancillary study measured DNAm using the Illumina HumanMethylation450 BeadChip (Illumina, San Diego, CA) from blood samples collected at WHI baseline before case ascertainment in BAA23, EMPC, and AS311. The R minfi package was used to perform quality control and normalization. Quality control included excluding probes on the Y chromosome, probes with a detection p-value <0.01 for >1% of samples, probes with a bead count <3 in >10% of samples, and probes measuring non-cytosine-guanine site methylation. The DNAm data was normalized using β-mixture quantile normalization.

There were 824 WHIMS women with DNAm data and cognitive assessments from WHIMS baseline (1995–1998) to 2007 and 403 (341 in WHIMS-ECHO, 62 in WHIMS-Y) with DNAm data and cognitive assessments from 2008 to 2021. We excluded data from 29 AS311 cases in WHIMS (of whom 10 participated in WHIMS-ECHO)²⁴ as treatment for cancer could affect cognitive function and for reasons related to joint model convergence (described in statistical methods). The final analytic sample consisted of 795 women for analyses of cognitive function measured from WHIMS baseline (1995–1998) to 2007 and 393 for analyses of cognitive function measured from 2008 to 2021 (including 331 WHIMS-ECHO and 62 WHIMS-Y participants; eFigure 1).

Cognitive Assessment

Details on the WHIMS protocol for measuring cognitive function are published elsewhere.¹⁹ Women completed the 3MS annually from baseline (1995–1998) until 2007. In 2008, the WHIMS protocol, which had annual, in-person global cognitive function assessments with the 3MS, was transitioned to remote annual assessments with the TICS-m in the WHIMS-ECHO study through 2021.²⁰ The 3MS is a widely used global cognitive function measure and consists of a 17-item questionnaire, with scores ranging between 0 and 100, and higher scores indicating better performance.²⁵ The TICS-m is a widely used measure of global cognitive function and consists of 14 items measuring memory, language, general knowledge, and attention, with scores ranging between 0 and 50; higher scores indicate better performance.^26,27

Epigenetic Age Estimator Calculations

We calculated 5 epigenetic age estimators: IEAA, EEAA, AgeAccelPheno, AgeAccelGrim2, and DunedinPACE.^8-11,28,29 We examined these 5 epigenetic age estimators as they are among the most well-known and widely studied epigenetic age estimators in the literature. IEAA, EEAA, AgeAccelPheno, and AgeAccelGrim2 were derived from the Horvath, Hannum, DNAm PhenoAge, and DNAm GrimAge2 epigenetic clocks, respectively, which were calculated with the online Horvath and Clock Foundation DNA Methylation Age Calculator.^8,29,30 IEAA and EEAA are first generation epigenetic age estimators.^8-10 IEAA captures cell-intrinsic age-related processes independent of blood cell composition, whereas EEAA captures age-related changes in blood cell composition.^8-10 AgeAccelPheno is a second-generation epigenetic age estimator that captures phenotypic age, consisting of age and 9 clinical biomarkers (e.g., creatinine), and therefore differences in life span and health span.¹¹ AgeAccelGrim2 is a composite biomarker (i.e., weighted linear combination) of DNAm-based surrogates of plasma proteins, a DNAm-based estimator of smoking pack-years, age, and sex.²⁸ In addition, we examined each DNAm-based component of AgeAccelGrim2 individually: smoking pack-years (DNAm Packyrs), adrenomedullin (DNAm ADM), beta-2 microglobulin (DNAm B2M), cystatin C (DNAm Cystatin C), growth differentiation factor 15 (DNAm GDF-15), leptin (DNAm Leptin), log-scale high sensitivity C-reactive protein (DNAm logCRP), log-hemoglobin A1C (DNAm logA1C), plasminogen activation inhibitor 1 (DNAm PAI-1), and tissue inhibitor metalloproteinase 1 (DNAm TIMP-1).²⁸

Both AgeAccelPheno and AgeAccelGrim2 have been shown to associate more strongly with age-related phenotypes and mortality risk, respectively, than first generation epigenetic age estimators.^8,11,28 IEAA, EEAA, AgeAccelPheno, and AgeAccelGrim2 were calculated as the residual from a linear regression model of the corresponding epigenetic clock on chronological age. Positive values suggest a higher epigenetic age relative to chronological age and thus faster biological aging, while negative values suggest slower biological aging.⁸

DunedinPACE was calculated using R code available at GitHub,³¹ and its development and derivation are distinct from IEAA, EEAA, AgeAccelPheno, and AgeAccelGrim2.²⁹ DunedinPACE was developed using longitudinal data and captures the pace of aging across several organ systems (e.g., cardiovascular, immune, pulmonary); it has been shown to be associated with cognitive function, morbidity, and disability.²⁹ DunedinPACE values >1 indicate faster biological aging while values <1 indicate slower biological aging.²⁹ All epigenetic age estimators and AgeAccelGrim2 component variables were standardized with a mean of 0 and SD of 1.

Covariates

Baseline questionnaires assessed age, race and ethnicity (Black, White, or Hispanic/Latina), education (≤high school equivalent, some college, college graduate), alcohol consumption (yes or no), and current smoking status (yes or no). Hypertension and diabetes were ascertained through participant self-report of physician diagnosis and medication use. Height and weight were measured with a stadiometer and balance beam scale, respectively, to calculate body mass index (BMI; kg/m²). Incident CHD consisted of clinical or definite silent myocardial infarction and was physician-adjudicated after review of participant medical records.³² APOE ε4 carrier status, defined as presence of at least 1 ε4 allele, was determined in a subset of 653 women (283 in WHIMS-ECHO, none in WHIMS-Y) based on 2 single nucleotide variants, rs429358 and rs7412, obtained from a genome-wide association study in a WHI core study (W63). Imputation was performed using the 1000 Genomes Project reference panel and the MaCH algorithm implemented in Minimac.³³ Both SNPs had high imputation quality (R² >0.97 for rs429358 and R² >0.97 for rs7412).³⁴ Hormone therapy treatment arm (not randomized, estrogen alone, estrogen control, estrogen plus progestin, or estrogen plus progestin control) and ancillary study (BAA23, EMPC, or AS311) were included as covariates in analyses.

Statistical Methods

We used R 4.2.2 for all analyses.³⁵ Means and SDs, or counts and proportions, were calculated for study variables. We calculated Pearson correlations between epigenetic age acceleration measures, DNAm-based components of AgeAccelGrim2, age, as well as 3MS and TICS-m. Lastly, we calculated Pearson correlations between (1) the first and second and (2) second to last and last 3MS and TICS-m scores as well as the last 3MS score and first TICS-m score to evaluate the relationships between cognitive function measures in the analytic sample.

To examine the longitudinal associations of baseline epigenetic age estimators with changes in global cognitive function scores during follow-up, we used multivariable, linear mixed-effects models with random intercepts and slopes. To determine whether epigenetic age estimators were associated with cognitive decline, we tested interaction terms between epigenetic age estimators and time (time between DNAm measurement and 3MS or TICS-m measurement) using a log-likelihood ratio test. Models included a quadratic term for time to allow for nonlinear changes in cognitive function. The linear mixed effects model equation was as follows:

$${\mathrm{Y}}_{\text{ij}}={\beta }_0+{\beta }_1\times \text{epigenetic}\hspace{0.17em}\text{age}\hspace{0.17em}{\text{estimator}}_{\mathrm{i}}+{\beta }_2\times {\text{time}}_{\text{ij}}+{\beta }_3\times (\text{epigenetic}\hspace{0.17em}\text{age}\hspace{0.17em}{\text{estimator}}_{\mathrm{i}}\times {\text{time}}_{\text{ij}})+{\beta }_4\times {\text{time}}_{\text{ij}}^2+{{\displaystyle \sum }}_{\mathrm{k}=1}^{\mathrm{k}}{\beta }_{\mathrm{k}}\times {\text{covariate}}_{\text{ki}}+{\mathrm{b}}_{0\mathrm{i}}+{\mathrm{b}}_{1\mathrm{i}}\times \text{time}\hspace{0.17em}{ϵ}_{\text{ij}}$$

where Y_ij is the 3MS or TICS-m score for participant i at time j, β₀ is the fixed intercept, β₁ is the fixed effect for epigenetic age estimator, β₂ is the fixed effect for time, β₃ is the fixed effect for the interaction between epigenetic age estimator and time, β₄ is the fixed effect for the quadratic effect of time, β_k is the fixed effect for covariate k, b_0i is the random intercept for participant i, b_1i is the random slope for participant i, and ϵ_ij is the error for participant i at time j.

We specified separate models for each epigenetic age estimator and for each cognitive function measure (3MS and TICS-m). Model 1, the minimally adjusted model, adjusted for age, education, race, and ethnicity, as published studies have shown differences in cognitive test scores by race and ethnicity.³⁴ Model 2, the fully adjusted model, additionally adjusted for potential confounders selected from the literature, including alcohol, smoking, hypertension, diabetes, BMI, hormone therapy treatment arm, ancillary study, and the first 3MS or TICS-m score at entry into WHIMS or WHIMS-ECHO to account for baseline cognitive function.^16,18,36,37 We also adjusted models for incident CHD, given that BAA23 selected participants according to incident CHD and that epigenetic aging as well as cognitive function are associated with CVD.^14,38 Models were not additionally adjusted for depressive symptoms as it did not appreciably change the study results. There were missing data for the following covariates (<1.9%): education, smoking, alcohol use, hypertension, diabetes, and BMI; thus, we proceeded with complete case analysis.

In sensitivity analyses, we evaluated effect modification by APOE ε4 carrier status using a 3-way crossproduct interaction term between epigenetic age estimators, APOE ε4, and time in the fully adjusted model and performed stratified analysis for any significant interactions. To account for death during follow-up, we conducted further sensitivity analyses using a joint modeling approach that consisted of a linear mixed-effects submodel as described above and a Cox proportional hazards submodel.^39-43 The Cox submodels had all-cause mortality as the outcome of interest and contained the same covariates as the linear mixed-effects models. Follow-up time was set as time from baseline to death, loss to follow-up, or end of follow-up, whichever came first. The Cox proportional hazards equation was as follows:

$$\mathrm{h}(\mathrm{t}|{\mathrm{X}}_{\mathrm{i}}\hspace{0.17em})=\mathrm{h}0\left(\mathrm{t}\right)\times \exp (\gamma \times {\text{epigenetic}\hspace{0.17em}\text{age}\hspace{0.17em}\text{estimator}}_{\mathrm{i}}\hspace{0.17em}+{{\displaystyle \sum }}_{\mathrm{k}=1}^{\mathrm{k}}{\alpha }_{\mathrm{k}}\hspace{0.17em}\times {\text{covariate}}_{\text{ki}}\hspace{0.17em})$$

where h(t|X_i) is the all-cause mortality hazard for individual i at time t given their covariate values, h0(t) is the baseline hazard, γ is the effect of epigenetic age estimators on the hazard, and α_k is the effect of the kth covariate on the hazard.

To account for selective attrition and differences in study characteristics between WHIMS and WHIMS-ECHO participants, we calculated stabilized inverse probability of attrition weights and applied them in the linear mixed effects models. We used logistic regression to calculate the inverse of propensity scores to predict participation in WHIMS-ECHO conditional on age, race, ethnicity, education, alcohol use, smoking, hypertension, diabetes, CHD, BMI, hormone therapy treatment arm, ancillary study, and the first 3MS score at entry into WHIMS. These weights were stabilized by the probability of WHIMS-ECHO participation conditional on age, race, ethnicity, and education and truncated to the 5th and 95th percentiles.

As the present study was descriptive and was not designed to test a specific hypothesis, we present nominal p-values for all analyses and did not apply a Bonferroni-corrected threshold to account for multiple testing; thus, findings should be interpreted with caution.

Standard Protocol Approvals, Registrations, and Participant Consents

The present study was approved by the WHI Clinical Coordinating Center (Fred Hutchinson Cancer Research Center, Seattle, WA), and all participants provided written informed consent.

Data Availability

The anonymized data supporting the present study findings are available upon reasonable request to the WHI program in accordance with publications and presentations policies. Information can be found at whi.org/get-started.

Results

The mean (SD) baseline age for participants was 70.8 (3.9) years; 10.7% were Black, 3.9% were Hispanic/Latina, and 85.4% were White (Table 1). In WHIMS-ECHO/WHIMS-Y, the mean baseline (SD) age was 67.5 (7.3) years; 12.2% were Black, 5.3% were Hispanic/Latina, and 82.4% were White (Table 1). The median follow-up (interquartile range for k cognitive assessments per participant) in WHIMS and WHIMS-ECHO/WHIMS-Y was 8.9 years (7–10 cognitive assessments/participant) and 7.9 years (3–9 cognitive assessments/participant), respectively. Pearson correlations between epigenetic age estimators, age, and the first 3MS or TICS-m score at the start of WHIMS or WHIMS-ECHO/WHIMS-Y are shown in eFigure 2. There was a stronger correlation between age and the TICS-m (r = −0.30) than between age and the 3MS (r = −0.08). Further, Pearson correlations were r = 0.64 between the first and second 3MS scores, r = 0.83 between the second to last and last 3MS scores, 0.64 between the first and second TICS-m scores, r = 0.70 between the second to last and last TICS-m scores, and r = 0.37 between the last 3MS score and first TICS-m score.

Table 1.

Baseline Characteristics of WHIMS Participants

Characteristics	WHIMS (n = 795)	WHIMS-ECHO/WHIMS-Y (n = 393)
Age, y, mean (SD)	70.8 (3.9)	67.5 (7.3)
Race and ethnicity, n (%)
Black	85 (10.7)	48 (12.2)
Hispanic/Latina	31 (3.9)	21 (5.3)
White	679 (85.4)	324 (82.4)
Highest education level, n (%)
High school/GED or less	238 (30.1)	102 (26)
Some college	330 (41.7)	158 (40.2)
College graduate	224 (28.3)	133 (33.8)
Hormone therapy trial arm, n (%)
Estrogen alone intervention	183 (23.0)	84 (21.4)
Estrogen alone control	163 (20.5)	82 (20.9)
Estrogen plus progestin intervention	224 (28.2)	112 (28.5)
Estrogen plus progestin control	225 (28.3)	115 (29.3)
Health behavior/status
Current smoker, n (%)	51 (6.5)	24 (6.2)
Alcohol use in past 3 mo, n (%)
Non-user	112 (14.3)	52 (13.4)
Former user	183 (23.4)	88 (22.7)
Current user	488 (62.3)	247 (63.8)
BMI, kg/m2, mean (SD)	28.7 (5.4)	29.0 (5.3)
APOE ε4 carrier,a n (%)	143 (21.9)	54 (19.1)
CVD risk factors
Hypertension, n (%)	345 (43.7)	143 (36.7)
Diabetes, n (%)	70 (8.8)	30 (7.7)
Incident CHD, n (%)	558 (70.2)	289 (73.5)
First cognitive function score,b mean (SD)	−0.01 (1.01)	0.01 (1.01)
Days between DNA methylation measurement and first cognitive function assessment, mean (SD)	70.0 (58.2)	4,561.7 (437.2)
Epigenetic age acceleration measures, mean (SD)
EEAA	−0.11 (6.6)	−0.61 (6.9)
IEAA	−0.13 (5.1)	−0.34 (5.3)
AgeAccelPheno	0.01 (6.8)	−0.70 (6.7)
AgeAccelGrim2	−0.01 (4.4)	−0.65 (4.3)
DunedinPACE	1.03 (0.12)	1.01 (0.12)

Abbreviations: BMI = body mass index; CHD = coronary heart disease; CVD = cardiovascular disease; EEAA = extrinsic epigenetic age acceleration; IEAA = intrinsic epigenetic age acceleration; WHIMS = Women's Health Initiative Memory Study; WHIMS-ECHO = Women's Health Initiative Memory Study Epidemiology of Cognitive Health Outcomes; WHIMS-Y = Women's Health Initiative Memory Study of Younger Women.

^aAPOE ε4 carrier status defined as presence of at least 1 ε4 allele.

^bFirst cognitive function score refers to the first cognitive function score (i.e., baseline) using the Modified Mini-Mental State Examination in WHIMS and the first Telephone Interview for Cognitive Status-modified score in WHIMS-ECHO/WHIMS-Y. Scores were standardized for comparison across the 2 measures.

Epigenetic age estimators were not significantly associated with changes in 3MS scores in the minimally adjusted or fully adjusted models (all p > 0.299; Table 2). A 1-SD increment in DunedinPACE (i.e., 0.12 units), indicating faster pace of aging, was associated with faster annual declines in TICS-m scores in the minimally adjusted (β = −0.118, 95% CI −0.202 to −0.034; p = 0.006) and fully adjusted (β = −0.123, 95% CI −0.211 to −0.036; p = 0.006) models, as shown in Table 2 and illustrated in Figure 1 using predictive margins from the linear, mixed-effects model. A 1-SD increment in AgeAccelPheno (i.e., 6.74 years) was associated with faster annual declines in TICS-m scores in the minimally adjusted model (β = −0.091, 95% CI −0.176 to −0.006; p = 0.035); however, this association was not significant in the fully adjusted model (β = −0.078, 95% CI −0.169 to 0.013; p = 0.093; Table 2). EEAA, IEAA, or AgeAccelGrim2 were not significantly associated with changes in TICS-m scores in any model (Table 2).

Table 2.

Longitudinal Associations of Epigenetic Age Estimators in 1 SD Increments With 3MS and TICS-m Scores Among Older Women Enrolled in the WHIMS 1995–2021

Model	Effect	WHIMS (3MS scores)			WHIMS-ECHO/WHIMS-Y (TICS-m scores)
		SD	Β (95% CI)	p Value	SD	Β (95% CI)	p Value
EEAA
1	EEAA	6.58	−0.112 (−0.389 to 0.164)	0.425	6.94	0.156 (−0.324 to 0.636)	0.522
	EEAA × time		0.014 (−0.035 to 0.064)	0.566		0.007 (−0.079 to 0.093)	0.876
2	EEAA		0.053 (−0.181 to 0.287)	0.656		0.135 (−0.271 to 0.541)	0.514
	EEAA × time		0.020 (−0.033 to 0.073)	0.458		0.026 (−0.065 to 0.116)	0.576
IEAA
1	IEAA	5.06	−0.025 (−0.299 to 0.249)	0.858	5.27	−0.062 (−0.530 to 0.405)	0.794
	IEAA × time		0.026 (−0.023 to 0.075)	0.299		−0.024 (−0.107 to 0.058)	0.563
2	IEAA		0.031 (−0.202 to 0.263)	0.795		0.057 (−0.343 to 0.456)	0.780
	IEAA × time		0.023 (−0.030 to 0.075)	0.399		0.008 (−0.079 to 0.095)	0.859
AgeAccelPheno
1	AgeAccelPheno	6.82	−0.245 (−0.520 to 0.029)	0.080	6.74	−0.190 (−0.665 to 0.284)	0.430
	AgeAccelPheno × time		−0.008 (−0.057 to 0.041)	0.762		−0.091 (−0.176 to −0.006)	0.035
2	AgeAccelPheno		−0.204 (−0.439 to 0.031)	0.088		−0.251 (−0.665 to 0.162)	0.233
	AgeAccelPheno × time		−0.019 (−0.072 to 0.034)	0.483		−0.078 (−0.169 to 0.013)	0.093
AgeAccelGrim
1	AgeAccelGrim	4.44	−0.147 (−0.428 to 0.134)	0.306	4.33	0.296 (−0.187 to 0.778)	0.229
	AgeAccelGrim × time		0.002 (−0.049 to 0.053)	0.931		−0.033 (−0.121 to 0.055)	0.457
2	AgeAccelGrim		−0.099 (−0.348 to 0.149)	0.432		0.000 (−0.423 to 0.422)	0.999
	AgeAccelGrim × time		−0.002 (−0.057 to 0.053)	0.948		−0.043 (−0.135 to 0.048)	0.353
DunedinPACE
1	DunedinPACE	0.12	−0.156 (−0.442 to 0.130)	0.285	0.12	0.056 (−0.436 to 0.548)	0.822
	DunedinPACE × time		0.010 (−0.041 to 0.060)	0.706		−0.118 (−0.202 to −0.034)	0.006
2	DunedinPACE		−0.061 (−0.305 to 0.183)	0.624		−0.409 (−0.823 to 0.006)	0.053
	DunedinPACE × time		0.008 (−0.045 to 0.062)	0.760		−0.123 (−0.211 to −0.036)	0.006

Abbreviations: 3MS = Modified Mini-Mental State Examination; EEAA = extrinsic epigenetic age acceleration; IEAA = intrinsic epigenetic age acceleration; TICS-m = Telephone Interview for Cognitive Status-modified; WHIMS = Women's Health Initiative Memory Study; WHIMS-ECHO = Women's Health Initiative Memory Study-Epidemiology of Cognitive Health Outcomes; WHIMS-Y = Women's Health Initiative Memory Study of Younger Women.

Model 1 (n = 792 and 6,616 observations for WHIMS and n = 393 and 2,358 observations for WHIMS-ECHO/WHIMS-Y) adjusted for age, education, and race and ethnicity. Model 2 (n = 758 and 6,315 observations for WHIMS and n = 376 and 2,271 observations for WHIMS-ECHO/WHIMS-Y) additionally contained alcohol use, smoking, hypertension, diabetes, incident coronary heart disease status, body mass index, hormone therapy treatment arm, ancillary study, and the first recorded 3MS or TICS-m score at the start of WHIMS or WHIMS-ECHO/WHIMS-Y to account for baseline cognitive function. Both models included a quadratic term for time.

Figure. Estimated TICS-m Scores Over Follow-Up Across 3 Levels of DunedinPACE Scores From a Linear Mixed-Effects Model Among WHIMS-ECHO and WHIMS-Y Participants, 1995–2021.

The linear mixed-effects model contained random intercepts and slopes and adjusted for age, education, race, ethnicity, alcohol use, smoking, diabetes, hypertension, incident coronary heart disease status, hormone therapy treatment arm, ancillary study, a quadratic term for time, and the first TICS-m score to account for WHIMS-ECHO/WHIMS-Y baseline cognitive function. An interaction between DunedinPACE and time was included. Time was mean-centered. Results are for z-transformed DunedinPACE set to mean (−0.04) and 1 SD below (−1.01) and above (0.93) the mean. Scores above mean are suggestive of faster pace of aging. TICS-m = Telephone Interview for Cognitive Status-modified; WHIMS = Women's Health Initiative Memory Study; WHIMS-ECHO = Women's Health Initiative Memory Study-Epidemiology of Cognitive Health Outcomes; WHIMS-Y = Women's Health Initiative Memory Study of Younger Women.

When examining AgeAccelGrim2 components, higher DNAm GDF15, DNAm B2M, DNAm Cystatin C, DNAm TIMP1, and DNAm ADM were associated with faster annual declines in both 3MS and TICS-m scores in the minimally adjusted and fully adjusted models (Table 3). Among these, the strongest magnitudes of association were observed for DNAm B2M (3MS scores for fully adjusted model: β = −0.091, 95% CI −0.143 to −0.038; p = 0.001), DNAm GDF15 (3MS scores for fully adjusted model: β = −0.082, 95% CI −0.136 to −0.028; p = 0.003), and DNAm TIMP1 (TICS-m scores for fully adjusted model: β = −0.320, 95% CI −0.411 to −0.230; p < 0.001; Table 3). Higher DNAm log CRP was associated with faster annual declines in TICS-m scores in the minimally adjusted but not fully adjusted model; further, it was not significantly associated with changes in 3MS scores (Table 3). Higher DNAm log A1C was associated with faster annual declines in TICS-m scores in both models, but not with 3MS scores (Table 3). DNAm PAI1 or DNAm Packyrs were not associated with changes in 3MS or TICS-m scores overall (Table 3). DNAm leptin was associated with faster decline in 3MS but not TICS-m.

Table 3.

Longitudinal Associations of DNAm-Based AgeAccelGrim2 Components in 1 SD Increments With 3MS and TICS-m Scores Among Older Women Enrolled in the WHIMS 1995–2021

Model	Effect	WHIMS (3MS scores)			WHIMS-ECHO/WHIMS-Y (TICS-m scores)
		SD	β (95% CI)	p Value	SD	β (95% CI)	p Value
DNAm GDF15
1	DNAm GDF15	108	−0.492 (−0.790 to −0.194)	0.001	129	−0.244 (−0.845 to 0.357)	0.425
	DNAm GDF15 × time		−0.070 (−0.120 to −0.020)	0.006		−0.140 (−0.228 to −0.051)	0.002
2	DNAm GDF15		−0.433 (−0.682 to −0.183)	0.001		−0.792 (−1.241 to −0.344)	0.001
	DNAm GDF15 × time		−0.082 (−0.136 to −0.028)	0.003		−0.168 (−0.258 to −0.078)	<0.001
DNAm B2M
1	DNAm B2M	1.16E5	−0.545 (−0.846 to −0.243)	<0.001	1.36E5	−0.253 (−0.852 to 0.346)	0.407
	DNAm B2M × time		−0.088 (−0.138 to −0.039)	<0.001		−0.153 (−0.240 to −0.066)	0.001
2	DNAm B2M		−0.424 (−0.674 to −0.175)	0.001		−0.633 (−1.094 to −0.172)	0.007
	DNAm B2M × time		−0.091 (−0.143 to −0.038)	0.001		−0.181 (−0.269 to −0.092)	<0.001
DNAm Cystatin C
1	DNAm Cystatin C	2.61E4	−0.218 (−0.516 to 0.080)	0.152	3.22E4	0.046 (−0.573 to 0.666)	0.883
	DNAm Cystatin C × time		−0.052 (−0.102 to −0.003)	0.038		−0.175 (−0.261 to −0.089)	<0.001
2	DNAm Cystatin C		−0.313 (−0.560 to −0.066)	0.013		−0.556 (−1.014 to −0.099)	0.017
	DNAm Cystatin C × time		−0.059 (−0.111 to −0.007)	0.027		−0.190 (−0.277 to −0.102)	<0.001
DNAm TIMP1
1	DNAm TIMP1	826	−0.443 (−0.788 to −0.098)	0.012	1.18E3	−0.222 (−1.153 to 0.709)	0.640
	DNAm TIMP1 × time		−0.082 (−0.132 to −0.032)	0.001		−0.297 (−0.386 to −0.208)	<0.001
2	DNAm TIMP1		−0.361 (−0.625 to −0.096)	0.008		−0.765 (−1.360 to −0.171)	0.012
	DNAm TIMP1 × time		−0.082 (−0.135 to −0.028)	0.003		−0.320 (−0.411 to −0.230)	<0.001
DNAm ADM
1	DNAm ADM	18.3	−0.331 (−0.621 to −0.041)	0.025	19.5	−0.177 (−0.707 to 0.353)	0.512
	DNAm ADM × time		−0.073 (−0.123 to −0.022)	0.005		−0.101 (−0.183 to −0.019)	0.016
2	DNAm ADM		−0.395 (−0.644 to −0.147)	0.002		−0.237 (−0.673 to 0.199)	0.285
	DNAm ADM × time		−0.076 (−0.129 to −0.023)	0.005		−0.097 (−0.183 to −0.011)	0.027
DNAm PAI1
1	DNAm PAI1	2.67E3	−0.281 (−0.558 to −0.004)	0.047	2.68E3	−0.036 (−0.514 to 0.442)	0.881
	DNAm PAI1 × time		−0.049 (−0.099 to 0.001)	0.053		−0.013 (−0.098 to 0.072)	0.765
2	DNAm PAI1		−0.191 (−0.431 to 0.048)	0.118		0.006 (−0.404 to 0.415)	0.978
	DNAm PAI1 × time		−0.046 (−0.099 to 0.007)	0.088		−0.008 (−0.097 to 0.080)	0.853
DNAm Leptin
1	DNAm Leptin	2.17E3	−0.389 (−0.663 to −0.116)	0.005	2.05E3	0.157 (−0.307 to 0.621)	0.505
	DNAm Leptin × time		−0.066 (−0.115 to −0.018)	0.007		0.038 (−0.043 to 0.118)	0.357
2	DNAm Leptin		−0.371 (−0.604 to −0.138)	0.002		0.403 (0.013 to 0.794)	0.043
	DNAm Leptin × time		−0.068 (−0.120 to −0.017)	0.009		0.069 (−0.015 to 0.153)	0.105
DNAm Packyrs
1	DNAm Packyrs	10.4	−0.223 (−0.507 to 0.061)	0.123	10.1	0.186 (−0.285 to 0.656)	0.438
	DNAm Packyrs × time		−0.032 (−0.084 to 0.020)	0.226		0.019 (−0.068 to 0.106)	0.671
2	DNAm Packyrs		−0.159 (−0.420 to 0.101)	0.229		0.065 (−0.364 to 0.494)	0.767
	DNAm Packyrs × time		−0.040 (−0.097 to 0.016)	0.159		−0.020 (−0.110 to 0.070)	0.668
DNAm log CRP
1	DNAm log CRP	0.45	−0.168 (−0.444 to 0.108)	0.233	0.45	0.216 (−0.261 to 0.692)	0.374
	DNAm log CRP × time		−0.003 (−0.053 to 0.046)	0.897		−0.089 (−0.171 to −0.008)	0.032
2	DNAm log CRP		−0.138 (−0.378 to 0.101)	0.257		−0.177 (−0.599 to 0.246)	0.411
	DNAm log CRP × time		−0.007 (−0.060 to 0.046)	0.790		−0.073 (−0.161 to 0.016)	0.106
DNAm log A1C
1	DNAm log A1C	0.03	−0.142 (−0.419 to 0.134)	0.313	0.03	−0.404 (−0.892 to 0.084)	0.104
	DNAm log A1C × time		−0.030 (−0.079 to 0.020)	0.247		−0.094 (−0.183 to −0.005)	0.039
2	DNAm log A1C		−0.068 (−0.309 to 0.173)	0.579		−0.330 (−0.756 to 0.095)	0.128
	DNAm log A1C × time		−0.019 (−0.071 to 0.034)	0.487		−0.095 (−0.187 to −0.003)	0.043

Abbreviations: 3MS = Modified Mini-Mental State Examination; A1C = hemoglobin A1C; ADM = adrenomedullin; B2M = beta-2-microglobulin; CRP = C-reactive protein; DNAm = DNA methylation; GDF15 = growth differentiation factor 15; PAI1 = plasminogen activation inhibitor 1; TICS-m = Telephone Interview for Cognitive Status-modified; TIMP1 = tissue inhibitor metalloproteinase 1; WHIMS = Women's Health Initiative Memory Study; WHIMS-ECHO = Women's Health Initiative Memory Study-Epidemiology of Cognitive Health Outcomes; WHIMS-Y = Women's Health Initiative Memory Study of Younger Women.

Model 1 (n = 792 and 6,616 observations for WHIMS and n = 393 and 2,358 observations for WHIMS-ECHO) adjusted for age, education, and race and ethnicity. Model 2 (n = 758 and 6,315 observations for WHIMS and n = 376 and 2,271 observations for WHIMS-ECHO) additionally contained alcohol use, smoking, hypertension, diabetes, incident coronary heart disease status, body mass index, hormone therapy treatment arm, ancillary study, and the first recorded 3MS or TICS-m score at the start of WHIMS or WHIMS-ECHO/WHIMS-Y to account for baseline cognitive function. Both models included a quadratic term for time.

The interaction between DunedinPACE and time in relation to 3MS scores varied significantly by APOE ε4 carrier status (p-interaction = 0.020) as did the interaction between DNAm PAI1 and time in relation to 3MS scores (p-interaction = 0.030). In stratified analysis, a 1-SD increment in DunedinPACE was associated with faster declines in 3MS scores among APOE ε4 carriers (β = −0.132, 95% CI −0.263 to −0.001; p = 0.048), but not among non-carriers (β = 0.039, 95% CI −0.025 to 0.103; p = 0.237), as illustrated in eFigure 3. Likewise, a 1-SD increment in DNAm PAI1 was associated with faster annual declines in 3MS scores among APOE ε4 carriers (β = −0.157, 95% CI −0.291 to −0.022; p = 0.022) but not among noncarriers (β = 0.003, 95% CI −0.054 to 0.061; p = 0.910). Otherwise, interaction terms between each epigenetic age estimator, APOE ε4 carrier status, and time were not significant (all p-interactions >0.05).

Results from sensitivity analyses using joint modeling to account for death during follow-up were generally consistent with those from the linear mixed effects models (eTable 1). Results from sensitivity analyses applying inverse probability weights in the main models to account for different characteristics between WHIMS and WHIMS-ECHO participants were generally consistent in direction to those from the main analyses (eTable 2).

Discussion

In this study of older women, a 1-SD increment in DunedinPACE was associated with faster annual decline in global cognitive function as measured by TICS-m, independent of sociodemographic factors, health-related behaviors, and comorbidities. EEAA, IEAA, AgeAccelPheno, or AgeAccelGrim2 were not significantly associated with changes in TICS-m. We did not observe any significant associations overall between epigenetic age estimators and changes in 3MS scores. However, we found that DunedinPACE was associated with faster annual declines in 3MS scores among women who were APOE ε4 carriers, but not among noncarriers.

These varying associations may be explained by the divergent methods used to develop each epigenetic age estimator and the corresponding age-related processes they capture. DunedinPACE was developed using data collected from the Dunedin Study birth cohort between ages 26 and 45 years as a DNAm-based marker of the Pace of Aging metric, which is composed of several biomarkers that capture changes in organ system function, including blood pressure and respiratory measures that are related to cognitive decline.^5,29,44 Notably, Pace of Aging does not include biomarkers of brain function or cognitive function.²⁹

Our findings are consistent with a few published studies that have examined DunedinPACE in relation to cognitive functioning. In the Dunedin Study, higher DunedinPACE was associated with greater intelligence quotient decline.²⁹ Among 48 participants in the Adult Health and Behavior Project (mean age = 45 years, 56% women, 16-year follow-up), participants who had the most cognitive decline had higher DunedinPACE.⁴⁵ In the Framingham Heart Study Offspring Cohort (N = 2,264, mean age = 65, 54% women, 14-year follow-up), a 1-SD increment (i.e., 0.12-unit increment) in DunedinPACE was associated with 1.27-fold higher incident dementia risk.⁴⁶ The present study extends the literature by showing that higher DunedinPACE may predict future cognitive decline among older women.

In our study, higher AgeAccelGrim2 was not associated with changes in 3MS or TICS-m scores. However, higher levels of several DNAm-based surrogates of plasma proteins contained in AgeAccelGrim2, including DNAm GDF15, DNAm B2M, DNAm Cystatin C, DNAm TIMP1, and DNAm ADM, were associated with faster declines in both scores, and higher DNAm log A1C was associated with faster annual declines in TICS-m scores. In the Multidomain Alzheimer's Preventive Trial (n = 1,096, age 69–94 years), higher GDF15 was associated with greater global cognitive function declines over 4 years.⁴⁷ In the Chinese Alzheimer's Biomarker and Lifestyle cohort, higher plasma B2M was cross-sectionally associated with lower Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA) scores.⁴⁸ In the Health and Retirement Study and China Health and Retirement Longitudinal Study, higher cystatin C was associated with faster cognitive decline.⁴⁹ Among participants with ischemic stroke, higher serum TIMP1 was associated with higher odds of cognitive impairment defined by low MMSE and MoCA scores; however, associations among individuals without ischemic stroke were unclear.⁵⁰ Higher ADM has been shown to be inversely correlated with lower Word Fluency Test scores and has been proposed as a therapeutic option for vascular cognitive impairment.^51,52 The nonsignificant association between AgeAccelGrim2 and cognitive function in the present study could be because of the inclusion of several DNAm-based plasma protein surrogates that are not associated with cognitive function. Future clocks that capture primarily cognitive function-related plasma proteins may be needed for risk stratification of cognitive outcomes. Larger studies with plasma protein measures, longer follow-up, and longitudinal cognitive function measures are needed to validate these findings.

Results from published studies examining first- and second-generation epigenetic age estimators (IEAA, EEAA, and AgeAccelPheno) in relation to cognitive functioning have been inconsistent. This could be partly attributable to study population differences and examination of specific cognitive domain measures as opposed to global cognitive functioning. In the Healthy Aging in Neighborhoods of Diversity Across the Life Span Study (n = 199, mean age = 56 years, 48.9% women, 60.3% Black), higher EEAA was associated with faster annual declines in the Benton Visual Retention Test and the Trail-Making Test.¹⁸ Among middle-aged Black participants in the Atherosclerosis Risk in Communities cohort (N = 2,157), the Horvath or Hannum epigenetic age acceleration measures were not associated with 6-year changes in the Delayed Word Recall Test, Digital Symbol Substitution Test, or Word Fluency Test.¹⁶ Among older adults in the Lothian Birth Cohort 1936, the Horvath epigenetic age acceleration measure was not associated with change in general fluid-type intelligence derived from the Wechsler Adult Intelligence Scale-III over 6 years.¹⁵ These epigenetic age estimators were developed to capture cellular aging processes (IEAA and EAA), phenotypic age (AgeAccelPheno), or estimate mortality risk (AgeAccelGrim) and thus may not have adequately captured cognitive aging.^8,11,13,28

Our study's strengths include longitudinal measures of cognitive functioning over up to 13 years of follow-up and the examination of 5 epigenetic age estimators, including the relatively recent AgeAccelGrim2 and DunedinPACE measures. In addition, the sociodemographic, lifestyle, and health information collected in the WHI allowed for comprehensive adjustment for confounders. It is of importance that unlike most prior studies, we performed sensitivity analysis using joint modeling to account for death during follow-up.

We also note the limitations of the present study. In the WHI, the epigenetic age data came from several ancillary studies. Thus, the sample was not specifically designed to investigate epigenetic aging in relation to changes in cognitive functioning. A limitation of the 3MS and TICS-m is that they capture global cognitive function. Thus, future studies should examine associations of epigenetic age estimators with trajectories of specific domains of cognitive function. In addition, the correlation between 3MS and TICS-m scores was weaker compared with correlations between repeated measures of 3MS and TICS-m, suggesting heterogeneity in intraindividual responses when the cognitive assessment mode changed from in-person to telephone from WHIMS to WHIMS-ECHO and that 3MS and TICS-m measured global cognitive decline differently in our sample given the stronger correlation between age and the TICS-m vs the 3MS.⁵³ Selection bias may also have influenced our findings, as women had to have survived to 65 years or older to be enrolled in WHIMS and later WHIMS-ECHO; however, in sensitivity analyses applying inverse probability weighting to account for differences in study characteristics between WHIMS and WHIMS-ECHO participants, results were similar. The results of the present study may not generalize to men. Finally, because we examined multiple epigenetic age estimators, findings should be interpreted with caution.

Among older women, we found that higher DunedinPACE was associated with faster annual decline in TICS-m scores overall and in 3MS scores among APOE ε4 carriers. Larger longitudinal studies in more racially and ethnically diverse study populations inclusive of men are warranted to further generalize these findings. In addition, future studies should consider deriving epigenetic age estimators trained on cognitive outcomes to further investigate whether DNAm has applications in risk stratification for future cognitive decline.

Glossary

3MS

Modified Mini-Mental State Examination

A1C

hemoglobin A1C

ADM

adrenomedullin

AS311

Bladder Cancer and Leukocyte Methylation Study

BAA23

Integrative Genomics and Risk of Coronary Heart Disease and Related Phenotypes

B2M

beta-2 microglobulin

BMI

body mass index

CHD

coronary heart disease

CRP

C-reactive protein

CVD

cardiovascular disease

cystatin C

cystatin C

DNAm

DNA methylation

DunedinPACE

Dunedin Pace of Aging Calculated From the Epigenome

EEAA

extrinsic epigenetic age acceleration

EMPC

Epigenetic Mechanisms of Particulate Matter-mediated Cardiovascular Disease Risk

GDF-15

growth differentiation factor 15

IEAA

intrinsic epigenetic age acceleration

MMSE

Mini-Mental State Examination

MoCA

Montreal Cognitive Assessment

Packyrs

smoking pack-years

PAI-1

plasminogen activation inhibitor 1

TICS-m

modified Telephone Interview for Cognitive Status

TIMP-1

tissue inhibitor metalloproteinase 1

WHIMS

Women's Health Initiative Memory Study

WHIMS-ECHO

Women's Health Initiative Memory Study Epidemiology of Cognitive Health Outcomes

WHIMS-Y

Women's Health Initiative Memory Study of Younger Women

Appendix. Authors

Name	Location	Contribution
Steve Nguyen, PhD	Division of Epidemiology, Herbert Wertheim School of Public Health and Human Longevity Science, University of California San Diego, La Jolla	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data
Linda K. McEvoy, PhD	Division of Epidemiology, Herbert Wertheim School of Public Health and Human Longevity Science, University of California San Diego, La Jolla; Kaiser Permanente Washington Health Research Institute, Seattle, WA	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data
Mark A. Espeland, PhD	Departments of Internal Medicine and Biostatistics and Data Science, Wake Forest University School of Medicine, Winston-Salem, NC	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data
Eric A. Whitsel, MD, MPH	Department of Epidemiology, Gillings School of Global Public Health; Department of Medicine, School of Medicine, University of North Carolina, Chapel Hill	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data
Ake Lu, PhD	Altos Labs, San Diego, CA	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data
Steve Horvath, PhD	Altos Labs, San Diego, CA; Department of Epidemiology, UCLA Fielding School of Public Health, Los Angeles, CA	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data
Joann E. Manson, MD, DrPH	Division of Preventive Medicine, Brigham and Women's Hospital, Harvard Medical School, Boston, MA	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data
Stephen R. Rapp, PhD	Department of Psychiatry & Behavioral Medicine, Wake Forest School of Medicine, Winston-Salem, NC	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data
Aladdin H. Shadyab, PhD	Division of Epidemiology, Herbert Wertheim School of Public Health and Human Longevity Science, and Division of Geriatrics, Gerontology, and Palliative Care, Department of Medicine, University of California, San Diego, La Jolla	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data

Study Funding

The WHI program is funded by the National Heart, Lung, and Blood Institute, National Institutes of Health, U.S. Department of Health and Human Services through 75N92021D00001, 75N92021D00002, 75N92021D00003, 75N92021D00004, and 75N92021D00005. A.H. Shadyab and L.K. McEvoy were supported by grant RF1AG074345 from the National Institute on Aging. S. Nguyen is supported by grants K99AG082863 and 5T32AG058529 from the National Institute on Aging. A.H. Shadyab and S. Nguyen were supported by funds from a program made possible by residual class settlement funds in the matter of April Krueger v. Wyeth, Inc., Case No. 03-cv-2496 (US District Court, SD of Calif.).

Disclosure

The authors report no relevant disclosures. Go to Neurology.org/N for full disclosures.