Department of Family and Preventive Medicine, University of California, San Diego, La Jolla, California 92093-0607
Address all correspondence and requests for reprints to: Elizabeth Barrett-Connor, M.D., Division of Epidemiology 0607, University of California, San Diego, 9500 Gilman Drive, La Jolla, California 92093-0607. E-mail: ebarrettconnor{at}ucsd.edu
There are many reasons to believe that estrogen is cardioprotective: the universally higher coronary heart disease (CHD) rates in men than in women, the reported increased risk after premature menopause, the in vivo and in vitro studies showing at least 10 estrogen effects that would be expected to prevent or delay CHD, the many epidemiological studies that almost universally report a lower risk of CHD in postmenopausal women who take estrogen alone or with a progestin compared to that in women who do not, and clinical trials in animals, including nonhuman primates (1, 2, 3). Nevertheless, the only completed large randomized, double blind, clinical trial with CHD as the primary outcome, the Heart and Estrogen/Progestin Replacement Study (HERS) (4), showed no reduced risk of CHD and, in fact, showed an increased risk during the first year of treatment. In this paper we review the evidence for cardioprotection and the HERS results and their implications.
Gender differences (1)
At the end of the 20th century, CHD remains the most common fatal disease of both women and men in most of the Westernized world. The 2.5- to 4.5-fold increased relative risk of CHD death in men compared to women is seen in countries with high or low rates of heart disease. This virtually universal male excess of CHD in diverse populations with very divergent lifestyles and disease rates is only compatible with an intrinsic female advantage or a male disadvantage. The female advantage is usually attributed to estrogen.
On the average, women have heart disease about 10 yr later than men. One misleading half-truth is that after the menopause CHD rates accelerate in women to catch up with rates in men. In fact, women do not catch up. CHD rates do become more similar in men and women as they age, but this reflects a deceleration in male rates during middle age rather than an acceleration in womens rates postmenopause.
The menopause (1)
Until the 20th century, only half of all women survived to age 50 yr, whereas the average age at menopause (4951 yr) has not changed since recorded history. The increasing number of postmenopausal women has drawn attention to the menopause and the consequences of estrogen deficiency. Because CHD is also uncommon in women before age 50 yr, an association has been drawn with menopause.
Despite the expectation that low postmenopausal estrogen levels cause heart disease, vital statistics show that CHD death rates in women increase exponentially with age beginning at age 20 or 30, with no change in the (log) slope of CHD rates between age 45 and 55 yr. Thus, the often repeated statement that CHD increases after menopause is also a half-truth; CHD rates increase with age, as does the probability that a woman is postmenopausal.
Many years ago, autopsy studies described "excess" coronary artery atherosclerosis in relatively young women who had been oophorectomized. More recent studies have yielded inconsistent results (5). In the largest study, the Nurses Health Study (6), bilateral oophorectomy, but not natural menopause, was associated with an increased risk of CHD; no increased risk was observed in oophorectomized women who had been treated with estrogen.
Studies of premature menopause are confounded by uncertainties about the factors leading to early menopause, which might themselves be related to CHD risk, and the roles of stress and other social and emotional consequences of menopause before age 40 yr. In addition, even the best studies have not controlled for important known confounders such as cigarette smoking or social class.
There have been many studies of CHD risk factor differences in pre- and postmenopausal women. For example, age-specific cross-sectional data from Framingham suggested that low density lipoprotein (LDL) cholesterol increases linearly with age beginning in the early 30s, long before menopausal estrogen deficiency (7). In contrast, the prospective Pittsburgh Healthy Womens Study found a striking increase in LDL cholesterol at the time of menopause (8) along with a small decrease in high density lipoprotein (HDL) cholesterol and no changes in blood pressure, blood glucose, or insulin. Results from a multicenter prospective study of pre- and perimenopausal women, the Study of Women across the Nation, are expected this year.
Endogenous estrogen
Estradiol levels in postmenopausal women are reduced by about 75% compared to premenopausal levels. Small prospective studies of natural menopause have shown that estrogen levels do not fall gradually, but become lower within the year before the last menstrual period and remain stable thereafter (9).
Cross-sectional studies have generally failed to find an association between endogenous estrogen levels and CHD risk factors in women (10, 11). One prospective study (The Healthy Womens Study) (12) found a nonlinear association between change in estradiol levels and change in HDL-2 cholesterol, but the association was not independent of confounders.
In cross-sectional studies of postmenopausal women referred for angiography, estrogen levels were unrelated to the amount of coronary atherosclerosis (13, 14). In the first prospective study, the risk of CHD was not associated with estrone or estradiol levels in community-dwelling women (15).
The absent association of endogenous estrogen with heart disease or heart disease risk factors may mean that premenopausal estrogen levels are cardioprotective, whereas postmenopausal estrogen levels are below the threshold necessary for an association with risk factors and CHD, that a single hormone assay does not adequately characterize the usual hormone level in individual women, that the wrong risk factors were studied, or that the estrogen-CHD hypothesis is wrong.
Experimental studies of hormone therapy and CHD risk factors
The Postmenopausal Estrogen/Progestins Intervention (PEPI) trial (16) was the first large randomized placebo-controlled trial of estrogen and common heart disease risk factors; 875 relatively young and healthy postmenopausal women were randomly assigned to placebo or to 1 of 4 active treatment regimens [placebo, daily conjugated equine estrogen (CEE), CEE plus cyclic medroxyprogesterone acetate (MPA), CEE plus daily MPA, or CEE plus cyclic micronized progesterone (MP)]. (The investigators considered using a transdermal estradiol, but this idea was abandoned because of the wide intraindividual differences in absorption and common allergic reactions to the available patch prep- aration.)
PEPI results were reported in 1995. All active treatments significantly reduced LDL cholesterol and increased HDL cholesterol compared to placebo, but CEE alone or with MP raised HDL significantly more than either CEE plus MPA regimen. Women assigned to each active treatment regimen had similar increases in triglycerides and absent increases in fibrinogen compared to those given placebo. Active treatment had no effect on blood pressure or insulin levels. Only MPA regimens raised plasma glucose levels. Results were similar in analyses restricted to women who took at least 80% of their assigned medication for 3 yr, except that unopposed CEE caused a significantly greater increase in HDL than that seen with any estrogen-progestin, including CEE plus MP (17).
Estrogen has multiple other effects that would be expected to be cardioprotective, including favorable changes in apolipoproteins, lipoprotein(a), plasminogen activator inhibitor-1, antithrombin III, homocysteine, endothelial function, vascular reactivity, and blood flow, as well as antioxidant and calcium channel-blocking activities (18, 19). Most studies of estrogens biological activity have limitations, however. In vitro and in vivo studies usually used pharmacological doses of estrogen and were short studies of acute effects. Nearly all in vivo studies used oral estrogen, which increases HDL cholesterol levels by a first pass liver effect, equivalent increases in HDL cholesterol are not seen after transdermal estrogen (20). It is therefore possible that a continuous postmenopausal oral estrogen regimen may not have the same effects as chronic exposure to endogenous estrogen with its cyclic changes, as seen in premenopausal women.
Observational studies of hormone therapy and heart disease
Since the 1970s, more than 30 case-control and prospective studies have reported less CHD in women who used estrogen. In a 1997 meta-analysis of published studies (2), the summary relative risk for CHD was 0.70 (30% protection) in women who ever used estrogen (presumed to be primarily unopposed estrogen) and 0.66 (34% protection) in women who used estrogen plus a progestin. The benefits appeared to be even greater in women with known CHD.
The Heart and Estrogen/Progestin Replacement Study (HERS) (4)
Background. The HERS was the first large, randomized, placebo-controlled clinical trial of estrogen and CHD in women. Before the initiation of HERS, the research community was divided about its importance. Some believed that the consistency of observational data and its biological plausibility were so convincing that a clinical trial was unnecessary, even unethical. Others favored a trial because nearly all known biases in observational studies would be expected to exaggerate any cardioprotective effect of estrogen, and only a randomized trial can control for both known and unknown differences in women who do or do not elect to take medication.
The design. The clinical trial advocates proposed a secondary prevention trial, i.e. a study of women who already had known CHD. This is usually the first protocol for clinical trials studying CHD outcomes. It is used because patients with known CHD are at much higher risk of a new event than persons without CHD, making it possible to show statistical differences with a smaller, shorter, and less costly study. Further, the limited observational data had suggested that estrogen would have even more benefit in women with CHD (21).
In HERS, continuous combined CEE and MPA was chosen as the active treatment regimen, in part because MPA was the commonly prescribed progestin added to estrogen therapy in the United States for postmenopausal women with a uterus and in part to avoid cyclic bleeding, aiding in maintenance of the double blind design. (The investigators wanted to study unopposed estrogen in women without a uterus, but funds were not available for this proposal.) The primary outcome was fatal and nonfatal CHD combined.
The results. The HERS results were reported in 1998. Overall 2763 women had been randomly assigned to placebo or continuous CEE plus MPA. Their average age was 66.7 yr. All had documented CHD; more than half were status-post coronary bypass or angioplasty, and more than 75% were taking aspirin, antihypertensive, or lipid-lowering therapy. At study closure, there was 100% follow-up. Analysis was by intention to treat.
After 4.1 yr of treatment, CHD rates did not differ in women assigned to active vs. placebo treatment; 12.5% of women assigned to CEE plus MPA had CHD death or nonfatal myocardial infarction compared to 12.7% of women assigned to placebo. There were nonsignificantly more CHD deaths in women assigned to CEE plus MPA (71 vs. 58) and nonsignificantly more nonfatal myocardial infarctions in women assigned to placebo (129 vs. 116). The frequency of all causes of mortality and other cardiovascular end points, including revascularization procedures, also did not differ by treatment assignment. This absent benefit occurred despite the expected favorable effects on LDL (11% decrease) and HDL (10% increase) cholesterol.
An unexpected HERS finding was a 50% increase in CHD risk (nominal P = 0.05) the first year after randomization to CEE plus MPA, which decreased over time to a nonsignificantly reduced risk by 4 yr. In a post-hoc analysis this pattern toward better prognosis with longer therapy showed a statistically significant test for trend.
Interpreting HERS
Power. Some have claimed that HERS had inadequate statistical power, because both compliance and the incidence of new CHD events were lower than predicted. This is incorrect. The true power of a trial is the observed power, not the expected power. The 95% confidence intervals show the range of risk or benefit likely to be found in 95 of 100 studies. HERS had 95% confidence intervals of 0.801.22, adequate power to detect a greater than 20% reduction or increase in CHD risk with hormone therapy.
Confounding. In HERS, women assigned to active treatment or placebo were well matched for multiple baseline variables, as would be expected in a large randomized trial. It is therefore not surprising that the 95% confidence intervals were not changed in analyses adjusted for baseline characteristics, nor were they changed after adjusting for the initiation of lipid-lowering medication during the trial. More than 80 other analyses were performed, attempting to find a subset of HERS women who might have benefited or been harmed by hormone therapy, such as a differential response by baseline medical therapy or revascularization procedure; all failed to show any significant interactions.
Inconsistency with meta-analyses. As reviewed previously (22), pooled risk estimates in meta-analyses of observational studies can be wrong when the studies share the same biases. A consistent bias (or biases) in observational studies can yield a consistent, but incorrect, result for either protection or harm. The observational studies of hormone therapy and CHD were subject to several biases that would falsely elevate the apparent benefit of estrogen, as reviewed below and previously (2).
Women taking estrogen tend to be younger, wealthier, more educated, and healthier than untreated women. This selection bias is clearly illustrated in the Pittsburgh Healthy Womens Study (23), in which women who elected to take estrogen after the menopause had more favorable levels of HDL cholesterol, fasting insulin, blood pressure, physical activity, and alcohol intake before the menopause than women who chose not to use estrogen. Thus, some of estrogens putative benefits might be spurious, in that women who are less likely to develop CHD are more likely to take estrogen.
Less than 20% of postmenopausal women take prescribed hormone replacement therapy for as long as 1 yr, but good compliance is a health-promoting behavior independent of therapy. Compliance bias has been shown most convincingly in clinical trials, where good compliance with placebo reduced the risk of CHD events by 4060% (24, 25), comparable to the reduction attributed to estrogen in observational studies.
Women taking estrogen are more likely to have electrocardiographic ST-segment elevation on exercise tolerance testing, possibly a digitalis-like effect of estrogen (26); ST segment changes are the commonest reason for referral to angiography. Thus, diagnostic detection bias could send more healthy estrogen-using women to angiography and account for the findings of published retrospective angiographic studies showing less coronary disease in users of estrogen.
Prevalence-incidence bias, which includes survivor bias, could explain the unexpected excess risk of CHD events observed early in HERS. In a 1993 study, Sturgeon and colleagues (27) reported 30% lower death rates in women using estrogen compared to those in women who never used estrogen. Additional analyses, however, revealed higher mortality rates from all causes and from circulatory diseases in women who had recently stopped taking estrogen compared to those in women who never took estrogen. The researchers wrote that this healthy hormone-user survivor effect was evidence that "nonexperimental studies are susceptible to overestimating the benefits of menopausal estrogen use, particularly current use, on mortality."
Limitations of a single trial. As a general rule, the results of a single trial are never completely convincing, and this is particularly true when the results are contrary to conventional wisdom. The HERS results are not unique, however, and actually parallel the only two other reports based on a large number of subjects in clinical trials of estrogen therapy and CHD, which also noted an early excess risk of CHD after starting estrogen.
In the 1960s, a large, randomized, placebo-controlled clinical trial of estrogen was conducted in men with known heart disease (The Coronary Drug Project) (28). Two of the treatment arms were CEE at doses of 2.5 or 5.0 mg daily. Estrogen was discontinued early in the trial, when an increased rate of myocardial infarction and thromboembolic events was observed. The untoward early effect has been attributed to the thrombogenicity of high dose estrogen and raises the possibility that the HERS estrogen dose, the usual replacement dose, is too high for women.
In 1997, Hemminki and McPherson (29) reported untoward results based on a pooled analysis of 22 randomized clinical trials that included 2859 women. Most of the trials were of less than 2-yr duration, so vascular outcomes, which were recorded as reasons for dropouts or adverse events, would have had to occur early. In the pooled analysis, the increased risk for women taking hormones compared to that for women not taking hormones was 1.39 for cardiovascular events and 1.64 for heart disease and venous thromboembolic events combined. The researchers noted the statistical improbability of this result if estrogen is cardioprotective.
Mechanisms. The HERS results have been challenged based on the paucity of mechanisms for harm. Because the increased risk of CHD events in HERS occurred only in the first year after randomization, any postulated mechanism should be compatible with this observation. The authors of the HERS paper postulated that the early mortality might be an arterial manifestation of the same factors leading to the excess risk of venous thrombosis. The increased risk of venous thromboembolic disease was observed consistently during each year of the HERS, whereas the excess risk of coronary artery thrombosis was seen only during the first year after randomization. It is possible that coronary artery thrombosis was less likely after estrogen-mediated lipid lowering had stabilized atherosclerotic plaques. The late benefit is plausible if estrogens cardioprotection is due mainly to its lipid-lowering effects, because trials of other lowering medications have shown a 1- to 2-yr lag between initiation of treatment and reduced risk of CHD.
The HERS results showing an early increased risk followed by a decreased risk are compatible with the attrition of women susceptible to an early untoward estrogen effect, leaving those who might be benefited in the trial. This explanation would imply that the immediate beneficial estrogen effects on endothelial function, vasodilation, and blood flow, posited by many investigators to be more important than the lipoprotein changes, either did not occur in HERS women or were completely masked by other effects leading to increased risk. Other possible acute untoward effects could include increased inflammation, blood viscosity, cardiac arrythmias, or changes in the arterial wall.
Finally, only the overall null result addresses the original primary HERS hypothesis. Because opposite results by duration of treatment were not one of the original hypotheses, the post-hoc analyses suggesting early risk and later benefit could be due to chance.
Generalizability. Most prevention trials are conducted in volunteers, who tend to be healthier, more advantaged, and more compliant than the general population. This should not make the HERS results less generalizable, because it is unlikely that estrogen would benefit only sicker or less educated compliant women.
Because HERS was a secondary prevention trial, the results do not exclude the possibility that hormone therapy may be useful for primary prevention, but not for secondary prevention, of CHD. This seems unlikely for two reasons. First, the evidence for primary protection in women at this time is based solely on epidemiological studies, which are subject to the biases described above. Second, and more importantly, in nearly all medication-CHD outcome trials, the benefit observed in secondary prevention studies has been confirmed in primary prevention studies (although it was less cost-effective). Nevertheless, it is possible that this "rule" predicting benefit does not apply to risk.
The MPA hypothesis. The most popular explanation for the negative HERS results has been an untoward effect of MPA. This progestin had the least favorable HDL and glucose effects in PEPI women (16) and, more important for an acute negative effect, has been shown to reduce estrogens favorable effects on endothelial function as well as atherosclerosis in several animal models. The most persuasive data come from studies of nonhuman primates in which estrogen alone (usually estradiol) protected against coronary artery vasospasm, whereas CEE and MPA did not (3, 30, 31). In addition, most of the observational studies suggesting cardioprotection in women taking combined therapy described women who were using a progestin other than (and more androgenic than) MPA. It is also possible that the HERS results were caused by the continuous MPA regimen, which yields a larger monthly dose of progestin than the cyclic monthly regimen.
Despite the internal consistency of the MPA hypothesis, it is premature to attribute HERS results to MPA. Similar early untoward results were observed in the Coronary Drug Project trial in men who received unopposed estrogen (28) and in the small trials reviewed and pooled by Hemminki and McPherson, where most early cardiovascular events occurred in women who were not treated with MPA (29).
If MPA is blamed for the early excess event rate in HERS, the cause could be either thrombogenic or impaired vasomotor function. Estrogen is more likely than MPA to be the thrombogenic component, in that a similar increased risk of venous thromboembolic disease also occurs with raloxifene, a selective estrogen receptor modulator that has estrogen agonist and antagonist properties and is taken without a progestin (32).
Studies of brachial arterial reactivity in women are not entirely helpful because no head to head comparison of CEE with and without MPA has been reported. One short cross-over study showed that the addition of vaginal micronized progesterone to transdermal estradiol did not modify estrogens endothelium-dependent effect on vasodilation (33). In a longer study, oral estradiol plus norethisterone had no effect on vasomotor function, at the very least suggesting that MPA is not the only "bad" progestin (34).
The Womens Health Initiative (WHI) (35)
These speculations about primary vs. secondary prevention and unopposed CEE vs. CEE plus MPA will be addressed by the WHI, a primary prevention trial being conducted in more than 27,000 postmenopausal women, only a small proportion of whom had CHD at baseline. Women with a uterus were randomized to daily CEE plus MPA or placebo (the HERS protocol), and those without a uterus were randomized to CEE alone or placebo. If this study is not closed early (for benefit or risk), the results are expected in the year 2005. If WHI results show no protection in women without CHD and without MPA, bias will be the most likely explanation for the benefit reported in observational studies of women who were unselected for CHD and were using primarily unopposed estrogen.
Unfortunately, no large studies using other types or doses of estrogen or progestin are in progress. A large clinical trial, Raloxifene Use for The Heart, has just begun, and others are in the planning process to see whether selective estrogen receptor modulators will be cardioprotective (32).
The clinical bottom line (at this writing)
Before HERS, the real controversy over hormone therapy was whether all postmenopausal women were likely to derive cardioprotective benefit and should therefore be encouraged to use hormone replacement therapy. The answer to this question is totally dependent on the presumption of a protective effect of hormone therapy on the risk of CHD. If HERS results are valid, and we think they are, then they demonstrate that CEE and continuous MPA, the most commonly used combined regimen in the United States and the United Kingdom, does not prevent recurrent CHD and may not prevent incident CHD either.
At present we have no trial data showing that the risk of CHD is reduced by estrogen alone or in combination with a progestin. In contrast, the cholesterol-lowering statins (3-hydroxy-3-methylglutaryl-coenyzme A reductase inhibitors) have been shown in clinical trials to reduce the risk of CHD in women with or without known heart disease (36, 37, 38). Therefore, in our opinion, the initiation of estrogen alone or in combination with a progestin for the primary purpose of preventing heart disease in women with or without heart disease is unjustified at this time, except in the clinical trial setting.
Summary
Despite the nearly universal finding from observational studies that postmenopausal estrogen therapy reduces the risk of CHD and the multiple plausible mechanisms by which estrogen might reduce the risk of CHD, hormone therapy had no benefit in the only large randomized clinical trial to date. Although it is possible that estrogen taken over the long term actually reduces CHD risk, it is not reasonable to begin the regimen used in HERS to prevent new or recurrent heart disease, given the observed excess early risk. Given the possible long term benefit, women who are already taking hormone replacement therapy may elect to remain on it. Women who are undecided should be asked to consider participation in clinical trials. The HERS has dramatically illustrated the need for them.
Received January 12, 1999.
Revised February 23, 1999.
Accepted February 23, 1999.
References