Rethinking Estrogen and the Brain

Those who share the current enthusiasm for a neuroprotective effect of estrogen1 may be surprised by the report by Yaffe et al.2 in this issue of the Journal that women with high endogenous estrone levels performed significantly more poorly on two of three cognitive function tests than women who had lower levels of estrone and that estradiol levels were unrelated to change in test performance. Thus, this first large study of endogenous estrogen and change in cognitive function fails to support the hypothesis that estrogen preserves brain function. How good is the evidence that exogenous estrogen is neuroprotective? The literature on this subject reviewed recently by Yaffe and colleagues3 includes a meta-analysis of 10 observational studies of estrogen and dementia. The metaanalysis results showing an overall small but significant protective effect of estrogen use may obscure design problems and the heterogeniety of the results. Biologic plausibility is estrogen's strongest suit. There are almost innumerable reasons why estrogen should be protective against both “benign memory loss” and senile dementia. As reviewed elsewhere,4 estrogen (or androgens aromatizable to estrogen) can produce permanent sexual dimorphism in nuclear volume, neuronal number, dendric length, neuronal membrane organization, synaptic formation, and neuronal connectivity. Some effects of sex steroids on the adult brain such as estrogen's facilitatory effect on synapse formation are thought to be impermanent, and decrements in hormone levels logically might be expected to impair behavioral and cognitive functions. In addition, estrogen has a stimulatory effect on multiple neurotransmitters, including serotonin, acetylcholine, and dopamine.5,6 Estrogen also enhances cerebral blood flow7 and acts as an antioxidant,8 which might reduce the oxidative damage caused by neurotoxic amyloid B protein in patients with Alzheimer's dementia. Strength of association is denoted by the magnitude of the risk estimate and its 95% confidence intervals. Most age-adjusted comparisons of estrogen use and dementia have wide confidence intervals that include one and show no statistically significant effect of estrogen.3 However, two of the most recent studies9,10 found a statistically significant 50% reduced risk of dementia in women who had ever taken estrogen. Further details regarding one of these studies11 suggest that a large proportion of women may not have recalled estrogen use reliably because nearly all use was remote.11 For editorial comment, see p 816 Consistency of associations in different studies has not been found. This is true for similarly designed studies in similar locations.3 For example, two California community-based studies of estrogen use and dementia-related death reported opposite results, one showing a nearly two-fold (statistically significant) increased risk12 and the other showing a nearly 30% (nonsignificant) decreased risk in women who used estrogen.13 Specificity of effect is definitely not a characteristic of studies of estrogen use. Women who take estrogen appear to be protected from multiple common chronic diseases, including cardiovascular disease, diabetes, osteoporosis, arthritis, and some cancers. This nearly universal benefit suggests that the observed associations could reflect selection factors (see below) rather than preventive or therapeutic effects.14 Stepwise graded associations showing decreasing risk with increasing dose or duration of therapy strengthen the likelihood that associations are causal. The only study of dementia risk by estrogen dosage found no graded association.13 Each of the three studies reporting risk related to duration of estrogen had different results: Paganini-Hill and Henderson13 reported a risk estimate of 0.49 after use for 7 or more years, whereas Tang et al.10 reported an odds ratio of 0.47 after use for less than 1 year (average use 4 months), and Kawas and colleagues9 found that risk reduction was not associated with duration of therapy. Temporality means that the exposure (estrogen) must precede the outcome (memory loss) for an association to be causal. This time sequence is difficult to ascertain in case-control studies of dementia because information on past hormone use by demented patients is usually obtained from a surrogate informant, who may not know when (or if) estrogen treatment was begun. Even in prospective studies, recall may be poor when estrogen use was remote from baseline, and it may be particularly poor in women with early memory loss.11 In addition, information about discontinuation or initiation of estrogen is usually limited during the follow-up interval. Recent publicity about estrogen's putative neuroprotective effects now increases the possibility that estrogen treatment was prescribed for early symptoms of dementia, such as forgetfulness and difficulty concentrating. Data on temporality are either suspect or incomplete in most studies. Prevention or reversibility of memory loss has not been shown in randomized double-blind clinical trials. As reviewed recently,3 clinical trials of estrogen in women with dementia have been small, usually uncontrolled and unblinded, and unconvincing. Although randomized controlled trials conducted in recently menopausal women suggest rapid improvement in cognitive function, this may reflect relief of menopausal symptoms rather than an effect on cognition per se. One of the few placebo-controlled trials conducted in older women found no effect of transdermal estrogen on cognitive function test results.15 Age is an obvious confounder because memory loss increases with age whereas the use of estrogen decreases with age. Several studies suggesting benefit failed to control adequately for age differences; matching for age within 5 to 10 years is not adequate. Confounding by indication is a common problem in observational studies of drug efficacy in which an apparent link between therapy and benefit is attributable to one or more characteristics of those prescribed the medication, not to the medication itself. Studies of populations in which postmenopausal estrogen use is uncommon may be more susceptible to such prescription bias than studies in which use rates are higher. Women who take estrogen are apt to be healthier than women who do not.16 In one prospective study,17 for example, women who took estrogen after the menopause had significantly more favorable heart disease risk factors before the menopause than women who did not begin estrogen. The relationship of general health to memory loss is uncertain, but one study18 reported that coronary heart disease was a risk factor for dementia in women. Higher education (or ability associated with education) and social class are each independently associated with better performance on cognitive function tests.19 More educated women, who also tend to be of higher social class, are more often prescribed postmenopausal estrogen.17,20 No studies have adjusted for both education and social class. Depression can impair cognitive function, and perception of memory loss can cause depression.21 Some of estrogen's effect on cognition may be an effect on depressed mood.22 Very few studies have controlled for depression. A majority of women prescribed postmenopausal hormones are noncompliant.23 In the NHANES I Follow-up Study, 45% of a representative cohort of US women had used estrogen for at least 1 month, but only 20% continued use for 5 or more years.24 Compliance with hormone therapy requires particularly motivated women who are willing to make regular physician visits, take at least one daily medication, and tolerate uterine bleeding and other side effects. For unknown reasons, compliance per se offers health advantages. For example, participants in clinical trials who took most of their placebo have a significantly reduced risk of cardiovascular disease compared with less compliant subjects.25 As noted above, misclassification of exposure is likely for several reasons. Reliable information on estrogen cannot always be obtained from a cognitively impaired subject or surrogate. The history of hormone use for controls is often also obtained from surrogate informants in order to correct for information bias, but it seems likely that cognitively intact women take medication unknown to their partners. Prospective studies frequently rely on self-report of estrogen, particularly of remote use, and pharmacy records do not necessarily mean that the prescribed medication was taken. Misclassification of outcome occurs because the definite diagnosis of Alzheimer's type dementia requires an autopsy. Only one estrogen-dementia study26 has reported autopsy-confirmed diagnoses (in approximately half the cases). Senile dementia of the Alzheimer's type and multi-infarct dementia cannot be distinguished completely on clinical grounds, and the two conditions may co-exist.27 Estrogen could prove to be protective for one but not the other type of dementia. There are different problems when the outcome is performance on cognitive function tests. The best tests are those that predict future dementia, but it is uncertain which test is best. Estrogen may alter performance only on specific tests and delay “benign memory loss” but not dementia. In different studies estrogen has been associated with better performance on different tests of cognitive function. Few studies use the same roster of cognitive function tests, and many have used an undisclosed number of cognitive function tests. Only a few studies report positive test results after correcting for multiple comparisons.20 In conclusion, despite the multiple mechanisms whereby estrogen could prevent or delay memory loss or dementia, enthusiasm for the neuroprotective effects of estrogen appear to have outrun the actual data. All observational studies have limitations, and clinical trials are few and small. The unexpected results reported by Yaffe et al.2 in this issue of the Journal add another intriguing piece to this unsolved puzzle and highlight the need for some restraint before we have the results from bigger and better clinical trials.