|Home | About | Journals | Submit | Contact Us | Français|
Associations between postmenopausal hormone therapy (HT) and cognitive decline may depend on apolipoprotein E (APOE) status or timing of initiation.
We included 16,514 Nurses’ Health Study participants aged 70–81 years who were followed since 1976 and completed up to three telephone cognitive assessments (2 years apart), between 1995 and 2006. The tests assessed general cognition (Telephone Interview of Cognitive Status (TICS)), verbal memory, and category fluency. We used longitudinal analyses to estimate differences in cognitive decline across hormone groups. APOE genotype was available in 3697 participants.
Compared with never users, past or current HT users showed modest but statistically significant worse rates of decline in the TICS: the multivariable-adjusted difference in annual rate of decline in the TICS among current estrogen only users versus never users was −0.04 (95% CI = −0.07, −0.004); for current estrogen+progestin users, the mean difference was −0.05 (95% CI = −0.10, −0.002). These differences were equivalent to those observed in women who are 1–2 years apart in age. We observed no protective associations with early timing of hormone initiation. We found suggestive interactions with APOE e4 status (e.g., on TICS, p-interaction = 0.10), where the fastest rate of decline was observed among APOE e4 carriers who were current HT users.
Regardless of timing of initiation, HT may be associated with worse rates of decline in general cognition, especially among those with an APOE e4 allele.
Postmenopausal hormone therapy (HT) have long been hypothesized to protect against cognitive decline and dementia in aging women.(Gibbs and Gabor, 2003) However, the current epidemiologic data provide mixed support for this hypothesis. Results from observational studies have been inconsistent,(Lethaby et al., 2008) and the Women’s Health Initiative Memory Study (WHIMS), a randomized trial of women aged ≥65 years have reported detrimental effects of HT in relation to change in cognitive function,(Espeland et al., 2004; Rapp et al., 2003; Resnick et al., 2006; Shumaker et al., 2004; Shumaker et al., 2003) risk of dementia,(Shumaker et al., 2004; Shumaker et al., 2003) as well as brain volume on MRI.(Resnick et al., 2009b) In addition, there are important questions that remain. First, the ‘critical period hypothesis’ suggests that HT may be neuroprotective if started near menopause (within 5 years of menopause)(MacLennan et al., 2006; Maki and Sundermann, 2009), in contrast to the older ages of WHIMS participants (aged 65–79 years at enrollment). Biological evidence supports this hypothesis,(Gibbs and Gabor, 2003; Markowska and Savonenko, 2002) yet few epidemiologic studies have addressed whether the relation of HT to cognitive decline may depend on the timing of initiation. Second, apolipoprotein E e4 status, which increases the risk of cognitive decline, may also modulate the effect of estrogen on neuroprotection, and thus, it has been proposed that the efficacy of HT should be more carefully assessed by apolipoprotein E genotype.(MacLusky, 2004; Nathan et al., 2004)
Therefore, we used data from the Nurses’ Health Study (NHS), a large prospective cohort in which we collected detailed data on hormone use beginning at or near menopause, to investigate the relation of HT to cognitive decline, including the type of hormone, timing of initiation and interaction with APOE genotype. These analyses extend our previous work on the relation between HT and cognitive decline,(Kang et al., 2004) with double the follow-up time and greater ability to examine associations.
The Nurses’ Health Study (NHS) began in 1976, when 121,700 female registered nurses women, aged 30 to 55 years, living in 11 US States completed a mailed questionnaire on lifestyle and health. Every two years, follow-up questionnaires have been mailed to participants to update their information, and >90% follow-up of the total possible person-time has been maintained.
For the study of cognitive function, we selected participants aged 70 years and older, who were free of diagnosed stroke. From 1995–2001, we contacted 21,085 eligible women for a baseline telephone cognitive assessment and 19,415 (92%) completed the interview. For these analyses, we excluded women at baseline with no information on hormone use, leaving 16,514 participants for analysis. In addition, for analyses of timing of hormone initiation, we included only women with natural menopause or bilateral oophorectomy, since age at menopause is difficult to determine precisely in other groups (e.g., women in whom the uterus was removed premenopausally, but one or both ovaries remained intact). Every two years, follow-up cognitive assessments were conducted, with > 90% of eligible participants completing each assessment. There were 16,514 women who completed the baseline assessment, 14,700 women who completed the 2nd and 12,612 women who completed the 3rd. This study was approved by the Institutional Review Board of the Brigham and Women’s Hospital.
The telephone battery included six tests: the Telephone Interview for Cognitive Status (TICS),(Brandt et al., 1988) a telephone adaptation of the Mini-Mental State Exam (mean=33.8, SD=2.7, range=8–41), immediate (mean score=9.4, SD=1.7, range=0–12) and delayed (mean score=9.0, SD=2.0, range=0–12) recalls of the East Boston Memory Test,(Albert et al., 1991) category fluency (naming animals, mean=16.9, SD=4.6, range=0–38),(Welsh et al., 1994) delayed recall of TICS 10 words list (mean=2.3, SD=2.0, range=0–10); and digit span backwards (mean=6.7, SD=2.4, range=0–12). Participation rates were identical across all cognitive tests.
We had four primary outcome measures for analyses: TICS, category fluency, digit backwards and verbal memory. As there is particular interest in the effect of HT on verbal memory,(Sherwin, 2003) we also calculated a verbal memory composite score, where we averaged the z-scores of four tests of verbal memory (immediate and delayed recalls of TICS 10-word list, immediate and delayed recalls of EBMT) among women who completed all of them. Because composite scores integrate information from multiple tests, they are stable measures of cognitive function.(Bennett et al., 2002; Bretsky et al., 2003) In addition, we calculated a global composite score that summarized the overall performance on our battery by averaging z scores from the six tests.
We found a correlation of 0.81 comparing the global scores of the telephone versus extensive in-person assessments in a validation of our battery among 61 educated women. We also found high reliability of test performance among 35 women who were given the TICS twice, 31 days apart (test-retest correlation = 0.7). Finally, among 88 older female health professionals, poor cognitive performance as determined by our telephone assessment was strongly associated with dementia diagnosis after three years; poor performance in the TICS and in verbal memory were both strongly associated with significantly increased risks of dementia.(Grodstein et al., 2007)
Every two years from 1976, women were asked about HT use. Information on duration and type of hormones was collected, as was age at menopause and type of menopause. Women’s reports of age at menopause(Willett et al., 1983) and type of menopause(Colditz et al., 1987) were shown to be valid when compared with medical records.
We defined initiation of hormones near menopause as use occurring within five years of the reported age at menopause, and first use at older ages was defined as initiation more than five years after menopause. As in our previous publication(Kang et al., 2004) we focused on estrogen only HT, as this formulation was available when the vast majority of the women in our cohort became postmenopausal. Also, in our previous publication,(Kang et al., 2004) we defined “initiation at older ages” as 65 years or older, to be consistent with the study population of the WHIMS,(Writing Group for the Women’s Health Initiative Investigators, 2002) and we defined “initiation near menopause” as within two years of menopause. Recently, the WHI has evaluated the role of timing of hormone initiation in relation to various outcomes,(Prentice et al., 2009) and they compared the effects of initiation of HT within 5 years of menopause with initiation of use 5 years after menopause; therefore, we present the associations with initiation within 5 years of menopause and initiation 5 years after menopause.
A buccal cell specimen was collected from willing participants in the Nurses’ Health Study in 2002. Of these, 3697 women were in the cognitive study, and had apolipoprotein E e4 genotyped; these women were representative of the cognitive cohort in terms of mean age and health characteristics (e.g., BMI, HT use, etc). Samples were processed using ReturPureGene DNA Isolation Kit (Gentra Systems Minneapolis MN) to extract genomic DNA from human cheek cells. Polymorphisms were genotyped using Taqman™ (Applied Biosystems, Foster City, CA) assays.
To analyze differences in change in cognition according to HT use, we used multivariable general linear models, where the dependent variables were the scores at each assessment, and the slopes of change over time were the estimates of interest. All models were fitted by maximum likelihood, incorporating the longitudinal correlation within study subjects using unstructured covariance structures and treatment assignment was modeled as fixed effects; for statistical testing, we used Wald tests.(Fitzmaurice et al., 2004) For all statistical analyses, Proc Mixed in SAS (SAS release 9.1, SAS Institute Inc., Cary, NC) was used.
In all analyses, data on hormone use and on potential confounders considered information provided through the questionnaire immediately prior to the baseline cognitive assessment. We included the following variables associated with both cognitive decline and hormone use, as well as variables predicting cognitive decline in the literature: age, education, diabetes, high blood pressure, vitamin E supplementation, body mass index, cigarette smoking, physical activity, socioeconomic status (represented by husband’s level of education), antidepressant use, alcohol intake, aspirin use, and other NSAID use. We also adjusted for the mental health index and energy-fatigue index from the Medical Outcomes Short Form-36; these indices combine items on questions about feeling hopeless, feeling downhearted (for mental health index) and being worn-out, or feeling tired (for the energy-fatigue index), such that higher scores indicate better mental health or vitality.
As of the first cognitive assessment, 31% of women had never used HT, 35% were past users and 34% were current users. Among past and current HT users, the majority (63%) initiated hormone use within 5 years of menopause, while 37% had initiated therapy five years after. The large majority (74%) of current hormone users used estrogen alone. Of those currently using estrogen, 71% was oral conjugated estrogen, 7% oral estradiol, 5% oral piperazine estrone sulfate, 4% transdermal estrogen, and the remaining 12% various other estrogens.
In general, past and current hormone users were similar to non-users with respect to demographic, lifestyle, and health variables (Table 1). The estrogen only users were on HT for an average of 15 years and the estrogen plus progestin users were on HT for an average of 11 years. However, current hormone users (especially of estrogen plus progestin) had a lower prevalence of obesity and type 2 diabetes, and used more vitamin E supplements, non-steroidal anti-inflammatory drugs and had a higher level of education. Current hormone users were also more likely to show more decline in scores from the first assessment.
When we examined differences in mean rates of change in cognition, both past and current HT use were associated with modestly worse rates of decline in several tests (Table 2). Overall, the differences that we found between women who used hormones and those who had never used hormones were equivalent to the differences we observed between women 1–2 years apart in age in this cohort. That is, HT use, either estrogen only or estrogen and progestin, was associated with a cognitive aging effect of 1–2 years. For example, in the TICS, compared with women who never used HT, both past and current users had rates of decline that were significantly worse (multivariable-adjusted difference in rate = −0.04 points per year; 95% CI, −0.07, −0.006 for past users; difference in rate = − 0.04 points per year; 95% CI, −0.07, −0.004 for current users of estrogen only; difference in rate = − 0.05 points per year; 95% CI, −0.10, −0.002 for current users of estrogen and progestin use); given the difference in rate of change with being older by 1 year in this cohort, the estimates of associations are equivalent to users being older by 1–2 years. The same trends of worse rates of decline were observed in relation to verbal memory, digit span backwards and the global score; associations were less clear with the animal naming test. Long-term users of estrogen only (20+ years) or estrogen and progestin (10+years) had worse rates of decline, but these differences were not statistically significant for most outcomes, likely due to the smaller number of women using this regimen.
We also examined associations between timing of HT initiation and cognitive decline, separated by those who began taking HT near menopause (within 5 years of menopause)(Maki and Sundermann, 2009; Prentice et al., 2009) or after menopause (at least 5 years after menopause) (Table 3). Hormone use was largely of estrogen alone, and therefore, to explore the differences of rates of decline by timing of initiation, we restricted the analyses to estrogen only users. Of note, among those who initiated estrogen only HT within 5 years of menopause 98% initiated HT within 2 years of menopause. We observed that initiation of use within 5 years of menopause was associated with worse rates of decline for all tests, with some associations being significant. When we examined a smaller subgroup of women who initiated HT at least 10 years after menopause or within 5 years of their first cognitive assessment (data not shown in table), we did not observe significant associations with cognitive decline in any of the tests. When we evaluated the association with timing of estrogen only HT initiation among a small proportion of the women with bilateral oophorectomy (21%), we did not observe that early initiation showed protective associations with cognitive decline. In fact, we observed significant associations that were consistent with adverse effects in relation to cognitive decline (Table 3). In the statistical tests for interaction, we observed no significant effect modification by menopausal type (natural menopause or surgical menopause) for timing of HT initiation.
We examined the interaction of any apolipoprotein E e4 allele with current hormone use (Table 4). For all tests except for the digit span backwards, we found that much of the apparent negative association between current HT and cognition was focused in women with an e4 allele. For example, on the TICS, compared with women without the e4 allele who never used hormones, the rate of decline for women without e4 who were currently using HT was similar (multivariable-adjusted difference = 0.00; 95% CI, −0.07, 0.07). However, for women who were both carriers of the ε4 allele and were currently using HT, there were significantly worse rates of decline (multivariable-adjusted difference = −0.10; 95% CI −0.20, −0.01; p-interaction=0.10). The difference in rate of decline between ε4 allele carriers who were current HT users and non-carriers who never used HT was cognitively equivalent to the difference observed with being 6.6 years older. Similar trends were observed in relation to the other cognitive outcomes, with the exception of digit span backwards (Table 4).
Overall, we found significant yet modest adverse associations between HT use and decline over 4 years in general cognitive performance, as measured by the TICS, verbal memory and attention. These results were consistent with the findings of the WHI. Most women in our study who were current users of HT were estrogen only users who had initiated use close to the time of menopause; we observed that early initiation of hormone was associated with worse rates of decline. Thus, there was no support for the “window hypothesis” for cognitive benefits in our data. However, we observed a suggestion that adverse effects of HT may be focused among those with the apolipoprotein e4 allele that warrants further study.
This study is an extended longitudinal follow-up study to a previous longitudinal investigation(Kang et al., 2004). It is worth noting the differences in the findings between the original publication and this follow-up study. The first study had 2 years of follow-up with two assessments, whereas this study had 4 years of follow-up with 3 assessments; differences in the results could be either due to methodological issues such as statistical variation, measurement error (i.e., more errors are inherent with more assessments), learning bias (i.e., the first study was likely more affected by learning than this study) or true biological changes in effects that are reflected in changes in associations with time. However, there are important similarities in the results of these two studies, and these results should be emphasized as they are more robust. First, neither the earlier study nor this study observed that the associations with postmenopausal hormone use in relation to cognitive decline were protective in nature. In fact, the directions of most associations were consistent with adverse relations: the first study observed that the adverse associations were significant with current use with long-term duration, whereas this present study found significant adverse association with any current use. Secondly, neither the earlier study nor this study provide support for the hypothesis that current use of postmenopausal hormones with early initiation may be beneficial for delaying cognitive decline. The first study found a significant increased risk of cognitive decline among women who were current users who initiated at older ages and a null association with current use with initiation early in the menopausal transition, whereas the present study observed significant adverse associations with early initiation. Finally, neither the earlier study nor this study found significant interactions between current hormone therapy use and apolipoprotein e4 allele on risk. However, this study found that current hormone users who were also apolipoprotein e4 positive had the worst rate of decline in the TICS and the rate was significantly worse than never users who were non-carriers, which merits further study in a larger study.
Our study has unique strengths. The majority of existing observational studies began their collection of hormone data in participants aged 65 years and older. Thus, there may be greater opportunity for misclassification HT use history, especially use near menopause, since women would have to recall hormone use from 15 or more years in the past. In particular, recall may be especially biased in those with latent dementia. In contrast, our detailed and prospective collection of data on hormones from 1976, beginning at or near menopause, minimizes the possibility for differential misclassification of HT use over time. In addition, the prospective data collection since menopause would minimize random misclassification, and enhance our overall ability to detect effects. Moreover, in our population of health professionals, all with access to healthcare and health knowledge, opportunities for confounding by HT may be minimized. Indeed, we found relatively few differences in characteristics of women who used and did not use HT. In addition, for other outcomes we have examined in the Nurses’ Health Study cohort (e.g., stroke, pulmonary embolism, heart disease, breast cancer and colon cancer), we have found relations with HT consistent with those reported in randomized clinical trials,(Grodstein et al., 2003) suggesting lack of important confounding.
Although experimental data suggest that estrogen positively affects neurotransmitter systems, improves synaptic properties of neurons, and increases blood flow and activation in the brain,(Fillit, 2002; Henderson, 2000) HT also modestly increases risk of stroke in our cohort(Grodstein et al., 2008) and in the WHI,(Wassertheil-Smoller et al., 2003) and elevated levels of C-reactive protein,(Cushman et al., 1999) an inflammatory marker (both stroke and inflammation are associated with cognitive decline). In addition, the WHI reported that HT decreased brain volume.(Resnick et al., 2009b)
Many large-scale observational epidemiologic studies have examined HT use in relation to cognitive decline, and the results have been inconsistent. Several clinical trials of 2–3 years duration in various populations of women have reported no effects of HT on cognitive function (Almeida et al., 2006; Pefanco et al., 2007; Resnick et al., 2009a; Viscoli et al., 2005; Yaffe et al., 2006). In a study of 1,800 women(Carlson et al., 2001), HT was related to significantly less decline on the MMSE over three years compared with non-use (difference in rate of change for hormone users vs. non-users=0·75, p<0·05), after adjustment for confounding; findings were similar for past and current use. In several studies, particularly population-based studies with great variation in socioeconomic status,(Fillenbaum et al., 2001; Yaffe et al., 2000) adjustment for confounding reversed or substantially attenuated any apparent cognitive benefits of hormone use. Consistent with our studies, in the Rotterdam Scan Study,(den Heijer et al., 2003) where most women were not on HT, higher levels of endogenous plasma estradiol were related to significantly smaller hippocampal volume, as well as poorer memory performance.
In recent years, it has been proposed that the timing of initiation of HT might be important, and that there may be a ‘critical window of opportunity’ for HT initiation for beneficial effects of estrogen in the brain. There are several lines of suggestive evidence from animal studies, observational studies and trials for this hypothesis. Early initiation of estradiol improved the cognitive performance of female rats after ovariectomy, but not when there was delay.(Gibbs and Gabor, 2003; Markowska and Savonenko, 2002) In humans, the Cache Study found that when HT was initiated near menopause, use was inversely associated with risk of Alzheimer’s disease, but current hormone use initiated within the recent 10 years past was associated with elevated risk(Zandi et al., 2002); however, these data were based on very few women who reported initiation near menopause. In contrast, Petitti et al,(Petitti et al., 2008) in an observational study utilizing a prescription database, found no suggestion that HT initiation near menopause was related to a decreased risk of dementia at older ages. Additionally, in the WHI data, there was no significant interaction by timing of initiation (p=0.31).(Espeland et al., 2004) Although we observed a suggestion of increased risk of substantial decline, which appeared focused among late initiators, in our previous report of cognitive decline over two years,(Kang et al., 2004) in the present analyses, with longer follow-up and greater power, these relations were no longer observed. In addition, neither the Nurses’ Health Study, nor the Women’s Health Initiative, have reported that the timing of HT initiation influences risk of stroke.(Grodstein et al., 2008, Prentice et al., 2009)
Interestingly, there are several lines of evidence indicating that apolipoprotein E e4 allele may interact with estrogen status. For example, in apolipoprotein E e4 transgenic mice, females demonstrated greater cognitive impairment than males.(Raber et al., 1998) Estrogen is known to modulate the expression of apolipoprotein E, and estradiol promotes synaptogenesis in response to injury via an apolipoprotein E -dependent mechanism.(Stone et al., 1998) However, estradiol had no such effect in the absence of apolipoprotein E or in the presence of apolipoprotein E e4.(Struble et al., 2008) In humans, the Cardiovascular Health Study found a significant interaction (p=0.04) between apolipoprotein E e4, estrogen use and cognitive decline, where estrogen use was significantly inversely associated with cognitive decline in women without apolipoprotein E e4, (RR = 0.59; 95% CI, 0.36, 0.99) but was non-significantly adversely associated among women positive for e4 (RR = 1.33; 95% CI, 0.74, 2.42).(Yaffe et al., 2000) In the same study, when internal carotid wall thickness, which was a significant predictor of cognitive decline, was examined, a similar interaction was observed, supporting the biologic plausibility of this interaction.(Yaffe et al., 2000) Our results also suggested the presence of such effect modification by apolipoprotein E e4, and this interaction clearly deserves further investigation.
Limitations of our study should be considered. Hormone use was self-reported. However, we believe any errors to be minimal since validation studies of many self-reported exposures in this cohort of registered nurses have proven their reports to be highly accurate, (Colditz et al., 1987; Willett et al., 1983) and since detailed information on hormone use was collected from participants every two years beginning at or near menopause. Misclassification of HT would likely have been random given the generally high cognitive functioning and may have led us to slightly underestimate associations. Because this was an ongoing follow-up study, we were able to assess the associations with cognitive change over 3–4 years by HT use status and initiation. However, we were unable to assess acute changes in cognition related to HT use. However, we were unable to assess the acute or short term changes in cognition. Our relatively homogenous population of aging female nurses comprises a select group of highly educated women and thus our findings should be interpreted in the context of our population characteristics. Finally, loss-to-follow-up is a potential source of bias in prospective studies; yet we maintained 92% follow-up, which did not vary by hormone use, making it unlikely that there was major bias.
Overall, HT was related to modestly worse general cognitive function and verbal memory, irrespective of the timing of initiation. However, any adverse effects of hormones on cognition appeared to be focused on women with an apolipoprotein E e4 allele, and this interaction clearly warrants further investigation.
Support: Research grant from Wyeth Pharmaceuticals (0712X1-4416) and NIH grants AG15424, AG13482 and CA87969.
Study sponsorship or funding:
Research grant from Wyeth Pharmaceuticals (0712X1-4416) and NIH grants AG15424, AG13483 and CA87969
We would like to acknowledge the investigators, staff and participants of the Nurses’ Health Study for help in this work. The work presented in this manuscript was supported by research grants AG15424, AG13483 and CA87969 from the National Institutes of Health (NIH) and Wyeth Pharmaceuticals (grant number 0712X1-4416). The sponsor Wyeth was not involved in the conduct of this study (study design, data collection, data analysis and interpretation or the manuscript preparation).
Disclosures of all authors’ financial relationships:
Dr. Kang received research support as a principal investigator for this 1- year study (September 2008 to September 2009) from Wyeth Pharmaceuticals.
Dr. Grodstein reports no disclosures.