Estrogen-replacement therapy and Alzheimer's disease in the Italian Longitudinal Study on Aging

The long-term use of estrogen-replacement therapy (ERT) by postmenopausal women is increasing, as are the number of suggested indications.^1,2 Several studies have supported the hypothesis of potential benefits for preventing or delaying the onset of age-associated dementias,^3-7 whereas two papers did not confirm these associations.^8,9 Estrogen use during the postmenopausal period has also been reported to lower the risk of dementia in women with Parkinson's disease.¹⁰

Findings from basic neuroscience have encouraged research in this area, providing information on the effects of estrogen on brain morphology and chemistry that explain how this steroid may influence brain functions.^11,12 Animal models point out that certain neuronal elements of the CNS, analogous to those involved in Alzheimer's disease (AD), are dependent on estrogen for their survival and function.¹³ Moreover, estrogen may influence the metabolism of the amyloid precursor protein, thus reducing the deposition of beta-amyloid and the development of senile plaques.¹⁴ In postmenopausal women, estrogen administration increases systemic arterial blood flow, including cerebral blood flow.¹⁵ All these factors may affect the risk as well as the clinical expression of dementia and AD.

We investigated the association of ERT use and estrogen-related variables with the prevalence of AD among the elderly Italian women enrolled in the Italian Longitudinal Study on Aging (ILSA).

Methods. Study population and design. The ILSA is a multicenter, population-based cohort study that investigated the frequency and the determinants of several age-associated conditions in the Italian elderly. The ILSA design has been described in detail elsewhere.¹⁶ The study components germane to this paper are described below. The study population consisted of 2,816 women aged 65 to 84 years, both free dwelling and institutionalized, selected from the population registers of eight Italian municipalities and randomly stratified by 5-year age groups according to an equal allocation sampling methodology.

The cross-sectional survey was carried out between March 1992 and June 1993. The case finding strategy consisted of a two-phase procedure: all participants were administered an extensive risk-factor interview and screening tests; those who screened positive underwent a clinical evaluation by a trained neurologist in each study center to confirm the final diagnoses.

The Mini-Mental State Examination (MMSE)¹⁷ with the cutoff score of 23/24 was employed to screen for dementia, as well as a previous diagnosis reported by the proxy respondent. The MMSE had been previously validated in each of the eight study centers against the DSM-III-R¹⁸ clinical diagnosis of dementia; the cutoff point of 23/24 had a sensitivity of 95% and a specificity of 90%.¹⁹ The structured clinical assessment consisted of sections B and H of the Cambridge Mental Disorders of the Elderly Examination(CAMDEX),²⁰ the Pfeffer Functional Activities Questionnaire,²¹ the Hamilton Depression Rating Scale,²² a neurologic examination, and the review of clinical records. The final diagnoses had to meet the DSM-III-R criteria for dementia syndrome, the NINCDS-ADRDA criteria²³ for possible and probable AD, and the ICD-10 criteria²⁴ for the other dementing diseases. To ensure the reliability of the diagnoses, before the survey the clinical investigators participated in an interrater agreement study on the application of the above-mentioned clinical diagnostic criteria. The reproducibility of the clinical diagnoses proved substantial(kappa index = 0.82 for the diagnosis of dementia syndrome; kappa index= 0.80 for the diagnosis of AD).²⁵ Once the prevalence survey was completed, all the suspect cases, random samples of AD and vascular dementia (VaD) cases, as well as a random sample of false positives were extensively reviewed by a panel of senior clinicians. Neither the local neurologists nor the panel of clinicians were aware of information on ERT use at any point during the diagnostic process.

We assessed major estrogen-related variables as follows. A standard history of ERT use was obtained by directly interrogating each participant and by interrogating the proxy respondent for those who screened positive for dementia. If a woman had ever taken ERT, she was questioned concerning the age at which she began the treatment and the age at which she stopped. Moreover, every woman was asked about her age at menopause and how old she was at menarche. Women were also questioned concerning the number of children and miscarriages, as well as their body weight at age 50. During the home interview, their prescriptions and boxes of pills were examined to ascertain the current use of ERT.

Based on each participant's report of the highest year of school completed, educational level was evaluated as years of schooling. Based on self-report, selected lifestyle characteristics, i.e., smoking and alcohol habits, were categorized as "ever" or "never." Other medical conditions, such as diabetes mellitus and hypertension, were defined according to standardized clinical diagnostic criteria.^26,27

Statistical analysis. Age-specific comparisons were carried out using the analysis of variance for continuous variables and by the chi-square test for categorical variables.²⁸ Demographic and medical characteristics in women who did and did not use ERT were compared by means of Student's t-test²⁹ for continuous variables and by the chi-square test for categorical variables.

The association between estrogen-related variables and AD was measured by the odds ratios (ORs).²⁸ The ORs and the 95% confidence intervals (CIs) were calculated using standard techniques.²⁸ Multivariate regression analysis was used to estimate the risk for AD as a function of all the estrogen-related variables under study.³⁰ Analyses were processed using the 1995 release of the SPSS statistical software.³¹

All the statistical levels quoted (p values) are two-tailed.²⁸

Results. The sample attrition is illustrated in thefigure. From the initial sample of 2,816 women, we could assess the past or current use of ERT in 2,046 subjects (72.6%). In terms of age, the nonrespondent women did not significantly differ from the respondents (p = 0.212). Women were subsequently excluded from the risk factor analyses if they did not complete the screening tests (13% of the eligible subjects), if information on education was missing (4%), or if they did not undergo or did not complete the clinical assessment in the event they had screened positive for dementia (0.5%). Those 464 women were significantly older (p < 0.001) and more educated (p < 0.001) than the women included in the analyses, whereas no significant difference was found in regard to the distribution of ERT use between the two groups(11% for the excluded subjects versus 12% for the included ones). The final sample for the risk factor analyses included 1,568 women (seefigure).

• Download figure
• Open in new tab
• Download powerpoint

Figure. Attrition of the Italian Longitudinal Study on Aging population at each step of the survey.

Only 238 women (11.6%) reported that they had used ERT after their menopause. The prevalence rate of ERT use among women aged 65 and over, adjusted to the 1991 Italian female population, was 12.3%. Intable 1 we report the distribution by age group of ERT users and of never users, as well as the mean duration of ERT.

The use of ERT was significantly more common among women in the younger age groups than among those in the older age groups. The mean duration of treatment was 3.1 years (SD 4.3), without significant differences across the age groups.

We found 92 patients with AD. As shown in table 2, a history of ERT use was significantly less common among women who suffered from AD than among women who were free of the disease. Therefore, the risk of AD for women who had ever used ERT was reduced to three-fourths below that of women who had never used estrogen (OR, 0.24; 95% CI, 0.07 to 0.77). Only three AD patients had ever taken ERT, therefore the OR for different treatment durations could not be estimated. We also compared the distribution of other estrogen-related variables between AD cases and controls. The risk for AD increased with advanced age at menarche (OR, 2.04; 95% CI, 1.11 to 3.77), whereas age at the last menstrual period, the number of children, and the body weight at age 50 years were unrelated to the risk of the disease.

The distribution of potential confounding variables by ERT use is shown intable 3. ERT users were significantly younger and more educated than those who had never used ERT. ERT users were more often wine drinkers than were nonusers. Neither the distribution of variables related to endogenous estrogen nor the occurrence of hypertension and diabetes differed between the two groups.

When we used the multivariate analysis to adjust for age, education, age at menarche, age at menopause, smoking and alcohol habits, body weight at 50 years of age, and number of children, ERT use was still significantly related to AD (OR, 0.28; 95% CI, 0.08 to 0.98), as were age (OR, 1.19; 95% CI, 1.13 to 1.26) and education (OR, 0.82; 95% CI, 0.74 to 0.91).

The age at onset of the disease did not differ significantly between ERT users and nonusers (78.04 ± 7.03 years versus 74.85 ± 6.04 years; p = 0.959).

The age- and education-adjusted mean score of the MMSE was significantly lower for ERT nonusers than for ever users (24.2 ± 5.3 versus 25.9± 4.6; p < 0.001).

Discussion. Our results provide additional evidence concerning the negative association between postmenopausal ERT and AD, as well as estimates of the prevalence of ERT use in the Italian elderly.

Although the use of ERT has increased during the past decade, there are still many variations across different countries.³² A major advantage of this study is that it could measure directly, for the first time, the frequency of ERT use in Italian elderly women. Only 12.3% had ever taken that treatment after their menopause, for a mean treatment duration of approximately 3 years, thus indicating that the use of ERT in Italy is still sporadic and rather short-term, with many users discontinuing treatment after 1 year. International comparisons by Jolleys and Olesen³² ranked Italy and Spain lowest on the post-menopausal ERT consumption list. According to that paper, in Italy during the period 1991-1992, less than 1% of women between the ages of 45 and 70 took ERT, whereas in the United States that proportion was 20% and in the United Kingdom and the Scandinavian countries from 11 to 16%. Those estimates were obtained from pharmaceutical sales statistics, and the proportion of women taking ERT was calculated by dividing the total number of "women treatment years" by the eligible population. This methodology can be employed for cross-national comparisons but cannot provide reliable information about how many women have ever taken ERT in each country.

The prevalence of ERT use was higher among younger and better-educated women in our survey. A possible explanation for these differences is the degree of awareness concerning menopause, the ERT, and the potential benefits of this therapy. Although we adjusted for major covariates of dementia, such as age and education, we cannot exclude the possibility that ERT use might reflect lifestyle characteristics that account for the observed protective effect.

Our data indicate a strong negative association between ERT use and the prevalence of AD. Our study differs from previous investigations of AD and ERT in that cases and noncases are from an unbiased sample of the Italian population. Moreover, the final diagnoses of dementia and dementia subtypes were made by trained neurologists using a structured clinical assessment.

On the other hand, these data come from an observational cross-sectional study, therefore they are prone to various potential problems. If women with cognitive impairment are more likely to be prescribed ERT by their physicians, this bias would lead to an underestimation of the association supporting the hypothesis of a protective effect. Furthermore, studies in the elderly are easily confounded by differential survival. Women using ERT who are cognitively intact may have better survival^33-35 and would more likely be alive and willing to participate in the ILSA survey. This type of bias could overestimate the association supporting the hypothesis of the protective effect. Another potential problem in the study is that recall bias regarding the use of ERT may be more severe in women with cognitive impairments, although we also questioned the next of kin for all the women who screened positive for dementia. Because of our study methodology, we were forced to rely on different sources of information about the exposure between cases and noncases. This would bias the results in favor of the protective effect.

Finally, it is reasonable to assume that dementia cases might be overrepresented in the nonrespondent group, and this represents another potential bias.

Data from Paganini-Hill and Henderson^3,6 supported the hypothesis that ERT use may prevent or delay the onset of AD. Those nested case-control studies, within the Leisure World Cohort study, identify both cases and controls on death certificates, thus making it difficult to disentangle the effect on incidence from that on case fatality.

The case-control studies of Henderson et al.³⁶ and Lerner et al.³⁷ found a significant negative association between ERT use and AD (ORs, 0.33 for the former and 0.49 for the latter). Brenner et al.⁸ could not find an association with ERT, assessed by computerized pharmacy data from a health maintenance organization; approximately one-half of both cases and controls did receive ERT (OR, 1.1). However, the risk was lower in women who had taken ERT most frequently (OR, 0.8; 95% CI, 0.3 to 1.8) and most recently (OR, 0.6; 95% CI, 0.3 to 1.2). Moreover, this study design could not evaluate whether ERT use might delay the onset of AD.

According to the results of Mortel and Meyer,³⁸ ERT may play a protective role in the risk not only for AD (OR, 0.55), but also for ischemic VaD (OR, 0.50).

Prospective investigations significantly minimize many of the bias that usually affects case-control investigations. In the recent paper by Kawas et al.⁴, a population-based cohort of 472 women was extensively followed for up to 16 years. This study concludes that ERT may yield a 54% reduction in risk for the disease. The results from the cohort study by Tang et al.⁵ favor the hypothesis that ERT protects against AD, in particular the age at onset was significantly later in women who had taken estrogen.

A recent population-based case-control study in Rochester, Minnesota,⁷ confirms the inverse association between estrogen therapy and AD, even adjusting for education and age at menopause, and finds a trend of decreasing ORs with increasing duration of therapy and with increasing total cumulative dose.

Previous studies examining the relationship of ERT with cognitive functions, in individuals with and without AD, have yielded conflicting results.^39-42 The extensive review by Birge¹³ concludes that ERT may have only minor effects on cognitive functions, moreover these effects are limited to selected cognitive domains, in particular verbal memory. Most of these studies enrolled women of relatively young ages, e.g., mean age ranging from 36.6 to 57 years,^39,42 therefore the study populations could have been too young to experience substantial cognitive impairment, and the duration of treatment may have been too short to have a detectable effect on cognitive functions.

In conclusion, both epidemiologic data and basic scientific findings currently available provide convincing evidence for the role of ERT in the treatment and prevention of AD. Any strategy that could reduce the risk of AD or delay its onset or the need for institutionalization assumes enormous importance given the economic, social, and familial burden of the disease. We therefore believe that a prospective clinical trial of estrogen is justified so that women and their physicians will be able to adequately weigh the risks and benefits of ERT for the treatment and prevention of dementia.

Acknowledgment

The authors thank Ms. Maria Elena Della Santa for typing the manuscript.

Appendix