Lifetime Estrogen Exposure and Cognition in Late Life

The Cache County Study

Joshua M. Matyi, MS; Gail B. Rattinger, PharmD, PhD; Sarah Schwartz, PhD; Mona Buhusi, MD, PhD; JoAnn T. Tschanz, PhD


Menopause. 2019;26(12):1366-1374. 

In This Article



This research used extant data from the Cache County Study on Memory in Aging (CCSMA; see Reference 21 for a detailed description), targeting women without dementia at the baseline (wave 1) visit. The majority (99%) of the participants were white. The Institutional Review Boards at Utah State University, Duke University, and Johns Hopkins University approved all study procedures.


Briefly, beginning in 1995 in its first wave, this study surveyed 5,092 residents (2,928 women) of Cache County, Utah, 65 years of age or older.[21] Three triennial waves of dementia ascertainment were subsequently performed over the course of 12 years to identify risk factors for AD. Demographic information including age and education, genotype at the apolipoprotein E (APOE) locus, and history of medical conditions, self-report of height and weight, medication use, family history of dementia, depression, activities of daily living, diet, and lifestyle factors including physical activity, smoking, and drinking were obtained in wave 1 and updated in subsequent waves.

A multistaged dementia screening and assessment protocol was followed in each wave. Cognitive screening was conducted at each wave using an adaptation of the 100-point modified Mini-Mental State Examination (3MS),[22] or if the participant was unable, dementia screening with a proxy using the Informant Questionnaire for Cognitive decline (IQCODE).[23] In waves 1 and 2, individuals who scored <87 on the 3MS or whose IQCODE was above 3.27 were sent to the next stage (Dementia Questionnaire [DQ]), which consisted of an interview with a knowledgeable informant to determine dementia symptomology. Participants whose interviews suggested either (1) the presence of significant cognitive impairment or (2) possible dementia as rated by a neuropsychologist or geropsychiatrist or (3) those who were members of a randomly selected panel to complete all stages of screening and assessment were selected to complete a clinical assessment. The clinical assessment included neuropsychological testing, medical and neurological evaluation, and a clinical interview with a knowledgeable informant. Results of the clinical assessment were reviewed by a study geropsychiatrist and neuropsychologist and preliminary diagnoses of a cognitive condition (if any) were assigned. Diagnoses of dementia followed the Diagnostic and Statistical Manual of Mental Disorders (DSM-III-R) criteria.[24] Differential diagnosis of AD followed criteria specified by the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association (NINCDS-ADRDA).[25] Diagnoses of other dementias followed standard protocol.[21] Persons who were not identified with dementia at a given wave were followed at the subsequent wave(s) using similar procedures, except for the elimination of the DQ stage in waves 3 and 4.

Between waves 1 and 2, all surviving female participants without dementia in wave 1 were contacted by either telephone (n = 2,099) or through an in-person interview (n = 48) and administered a women's health questionnaire (WHQ) regarding their reproductive history. Survey items included age at menarche and menopause, number of pregnancies and live births, cumulative months spent breastfeeding, use of HT, and history of hysterectomy or oophorectomy. Data from these 2,147 women who completed the interview formed the basis of the present analyses.


The 3MS[22] was used to measure cognitive status at each wave. The screening test assesses five factors including psychomotor skills, memory, identification and association, orientation, and concentration and calculation, implicating a broad measure of cognition.[26] The 3MS has shown high internal consistency (α = 0.87) and validity in identifying dementia (area under curve [AUC] = 0.94 [SE = 0.01]; Z = 5.38, P < 0.01).[27] In the CCSMA, minor adaptations were made to the 3MS,[22] for example, substituting easily verifiable items (recall of prominent current and past political figures) for those related to personal demographics. As reported previously, 3MS scores were adjusted for sensory/motor deficits, by excluding items affected by these issues, with the adjusted score calculated from the remaining percentage of correct points out of the new total points (multiplied by 100), thus retaining the original scale of 0 to 100 points.[21]

Estrogen Exposure

Lifetime endogenous estrogen exposure (EEE) was calculated after the methods of Fox, Berzuini, and Knapp[10] using the reproductive window (menopausal age minus age of menarche), minus total duration of breastfeeding. If no breastfeeding was reported for parous women, 1.5 months were subtracted per pregnancy to account for mean ovulatory regulation time.[28] Exogenous estrogen exposure was generated using duration of HT (independent of dosage) at each wave and treated as a time-varying variable. In addition, type of HT was coded (none; opposed [ie, estrogen compounds that include progesterone]; unopposed [estrogen alone]) as well as the timing of first use of HT expressed in years. HT timing relative to menopause was then transformed into a categorical variable a priori to include those who did not take HT (no HT; continuous or within 1 y of menopause; between 1 and 5 y of menopause; 6 y or more after menopause).

Additional Covariates

Covariates tested in statistical models were guided by previous research on variables that potentially affect both estrogen (or its effects) and cognition in late-life. These included age, level of formal education, APOE genotype (number of E4 alleles), body mass index (BMI),[29] physical exercise,[30] overall health,[3] and depression status.[31] BMI at wave 1 was based on self-reported height and weight and calculated using the formula kg/m2. Physical activity was categorized by the recommendations for older adults provided by American College of Sports Medicine and American Heart Association (ACSM; AHA)[32] using metabolic equivalent (MET) transformations of various physical activities.[33] Designations of "sedentary" (no physical activity reported), "light" (<450 MET-min/wk), "moderate" (450–750 MET-min/wk), and "vigorous" (>750 MET-min/wk) were assigned to the data depending on the frequency and duration of physical activity. Physical activity data were not collected in wave 2; therefore wave 1 data were carried forward to ensure that all available cases were included in analyses. Overall health was ascertained by asking participants to rate their health on the day of the interview as either "excellent," "good," "fair," or "poor." Depression status was assessed with the Diagnostic Interview Schedule (DIS) of the DSM-III-R[24] and was categorized as "no current depressive episode/no current depression medication use," "no depressive episode/with depression medication use," "minor depression," and "major depression."[34]

Statistical Analysis

In addition to providing descriptive statistics for the sample, comparisons for participants included versus excluded in the analyses were made using chi-square tests for categorical variables and independent samples t tests for continuous variables. A series of linear mixed effects models were used to investigate the proposed relationships between lifetime estrogen exposure and cognitive decline. The first model examined EEE; the second model examined EEE and duration of exogenous HT exposure (time varying); the third model examined EEE and type of HT (none, unopposed, opposed); and the fourth model examined EEE and timing of HT initiation relative to menopause. Finally, as a substantial number of participants discontinued HT use between waves 3 and 4 (coinciding with the FDA black box warning),[35] we examined in a subset of participants, any effects of stopping HT use on 3MS scores. For these analyses, a three-level categorical variable was calculated to capture (1) those who used HT continually through wave 3 but discontinued use at wave 4, (2) those who used HT continually through both wave 3 and wave 4, and (3) those who never used HT.

In addition to examining fixed effects, mixed effects models address within-participant correlation and random effects inherent in longitudinal data.[36] Furthermore, these models allow for missing data across time points and do not exclude cases on a list-wise basis. The inclusion of each variable/covariate was examined for improvement in model fit (P < 0.05) by comparing negative two log-likelihood values for nested models. All statistical models employed SPSS version 23.