Hormone Replacement Therapy and Heart Disease Prevention: Title and subTitle BreakExperimentation Trumps Observation

The true method of knowledge is experiment.—William Blake, 1788

Three different meta-analyses^{1 - 3} have concluded that estrogen replacement therapy (ERT) decreases the risk of coronary heart disease (CHD) by 35% to 50%. The predicted increase in life expectancy in hormone users, based on estimates of the risk of CHD in users of ERT derived from observational studies, is 2 to 3 years.^{2 ,4} The effects of ERT and combined estrogen-progestin replacement therapy (HRT) on lipids and fibrinolysis are strongly beneficial for both ERT and HRT.⁵ This has led to extrapolation of the results of observational studies of ERT to HRT and to promotion of the use of both ERT and HRT in women with CHD.^{2 ,6} The results of the Heart and Estrogen/progestin Replacement Study (HERS), which are reported by Hulley et al in this issue of JAMA,⁷ add critically important data on the effects of HRT. In this randomized controlled trial of 2763 postmenopausal women with established coronary disease, treatment with estrogen plus progestin did not reduce the rate of CHD events (eg, nonfatal myocardial infarction or CHD-related death). These findings are a sobering reminder of the limitations of observational research, the incompleteness of current understanding of the mechanisms of vascular disease, and the dangers of extrapolation.

Observational studies of ERT and CHD are numerous. Most of them, including at least 5 studies that have examined outcomes of ERT use in women with established coronary disease,^{8 - 12} show a lower risk of reinfarction, CHD-related death, and coronary artery restenosis in users of ERT. Studies of HRT for primary prevention of CHD (there appear to be no specific studies of HRT in women with established coronary disease) are not as numerous as studies of ERT,^{13 - 18} but these studies show a reduction in the risk of CHD that is in the same direction and is of the same magnitude as the reduction in the risk of CHD in users of ERT. The seemingly overwhelming nature of the observational data showing a lower risk of CHD in hormone users raised concerns about the ethics of randomization of women to hormone therapy, where some subjects would receive a placebo. Can these observational studies be wrong?

Reasons to view cautiously the observational results for CHD in users of ERT and HRT have always existed. Women with healthy behaviors, such as those who follow a low-fat diet and exercise regularly, may selectively use postmenopausal hormones. These differences in behavior may not be taken into account in the analysis of observational studies because they are not measured, are poorly measured, or are unmeasurable. Estimates of the relative risk of CHD in hormone users will then be biased toward finding a protective effect of hormone use for CHD. Barrett-Connor¹⁹ dubbed this prevention bias and demonstrated empirically that it exists. In 2 different randomized trials of drugs to prevent CHD, the subjects who faithfully took their placebo had a lower risk of CHD than subjects who were noncompliant with taking placebo.^{20 - 21} The relative risk of CHD in subjects compliant with taking placebo compared with noncompliant subjects was 0.7 in the Coronary Drug Project²⁰ and 0.4 in the Beta-Blocker Heart Attack Trial.²¹ A similar reduction in the relative risk of CHD was found in compliant subjects compared with noncompliant subjects taking active drug. Women categorized as users of ERT or HRT in observational studies are, by definition, compliant. Compliance bias is large enough to explain entirely reductions in the relative risk of CHD between users and nonusers of ERT and HRT of the magnitude found in observational studies.²² Consistent biases would produce consistent results.

The Coronary Primary Prevention Trial,²³ the West of Scotland Coronary Prevention Study,²⁴ the Helsinki Heart Study,²⁵ the Scandinavian Simvastatin Survival Study,²⁶ and the AFCAPS/TexCAPS²⁷ are randomized trials that provide incontrovertible evidence for the ability of some lipid-lowering drugs to prevent CHD events in persons free of coronary disease (primary prevention), as well as those who have coronary disease (secondary prevention). The consistency of these observations has led some to believe that lipid lowering can predict reductions in the risk of CHD and that lipid lowering can be used as a surrogate end point for CHD. In the arms of the Coronary Drug Project that involved randomization of men with coronary disease to estrogen or placebo, neither high-dose (5.0 mg) nor low-dose (2.5 mg) conjugated equine estrogen reduced the risk of recurrent CHD events^{28 - 29} despite beneficial effects of estrogen on lipid levels in men that were of the same magnitude as those in women. For female hormones, beneficial lipid effects do not directly translate to reductions in disease end points, at least in persons with coronary disease.

The lipid hypothesis has dominated thinking about CHD for at least 4 decades. There is growing recognition that thrombotic phenomena play an important role in acute coronary syndromes.³⁰ Both ERT and HRT increase the risk of venous thrombosis.^{7 ,31 - 34} Estrogen administration has been shown to have general "procoagulant" effects.³⁵ In light of these data, it will be tempting to embrace the hypothesis that the HERS authors put forth to explain their findings. They speculate that the HERS results are a consequence of domination of the antiatherogenic effects of HRT by its early thrombogenic effects. The implication is that HRT use is more protective for CHD in women with longer durations of use and that a net benefit of HRT for CHD would have been observed with longer observation.

Recent studies of combined estrogen-progestin oral contraceptives show that the increase in the relative risk of venous thromboembolism is more elevated in the first year after initiation of use than later.³⁶ The relative risk of venous thromboembolism in users of ERT was somewhat more elevated in the first year of use than in later years in some^{31 ,34} but not all studies³³ that addressed this topic. Both a Nurses' Health Study analysis of death from CHD by Grodstein et al¹³ and a study of first myocardial infarction (MI) by Sidney et al¹⁷ demonstrated no evidence of a larger protective effect of ERT or HRT in women with longer durations of current use, and there was no higher risk of CHD in the first years after initiation of ERT or HRT use. Heckbert et al³⁷ reported a trend of declining risk of CHD with increasing duration of ERT use in current users, but the risk of CHD in the first 2 years after use was not elevated (estimated relative risk, 0.91). Only Rosenberg et al¹⁴ reported a pattern of relative risks of MI in hormone users that is the same as in HERS and is consistent with the speculation that there is an early thrombogenic effect of hormones that is counterbalanced by an antiatherogenic effect with longer duration of use. In the analysis of data by Rosenberg et al¹⁴ from a large case-control study, the adjusted relative risk of first MI in recent users of ERT was 1.5 for less than 1 year of use, 1.2 for 1 to 4 years of use, 0.6 for 5 to 9 years of use, and 0.5 for 10 years of use or longer. Thus, relevant data on CHD end points from published observational studies of postmenopausal hormone use, all from studies of ERT and HRT for primary prevention, are mixed.

The prior failure of extrapolation from the effects of hormones on lipid levels to effects on CHD end points and of observational data on ERT used for primary prevention to HRT used for secondary prevention dictates that any extrapolation from HERS results be judicious. Although the effects on lipid levels, fibrinolysis, blood pressure, glucose levels, and insulin levels are the same for continuous combined estrogen-progestin and intermittent estrogen-progestin hormonal regimens,⁵ it is not certain that the studied effects encompass all the physiologic parameters that might differ between continuous and intermittent regimens. For combination estrogen-progestin oral contraceptives, minor changes in the structure of the progestin molecule that resulted in more favorable lipid effects were associated with an increased risk of venous thromboembolism,^{38 - 39} showing that progestins influence importantly the thrombogenicity of combined regimens. The results of HERS may not apply to ERT, to different regimens of HRT, or to different progestins in women with coronary disease.

There remains an impressive body of observational data on ERT and HRT used for primary prevention. Women with coronary disease are likely to have unmodifiable risk factors, such as diabetes and obesity, that influence the tendency to thrombosis. If so, any procoagulant effect of hormones would be greatest in women with coronary disease. The HERS results should not be immediately extrapolated to ERT and HRT used for primary prevention.

The HERS results are important for women to consider in making decisions about hormone replacement. Physicians need to review carefully the HERS findings, but no woman, including those with coronary disease, should abruptly cease use of ERT or HRT because of the HERS results. Most women use hormones for reasons other than prevention of CHD. The beneficial effects of ERT and HRT on bone and menopausal symptoms have been established clearly in randomized trials. Discussions of the HERS findings between women and their physicians do not need to occur today, this week, or even this month. HERS identifies no new risks, and there is no emergency.

Fortunately, other randomized trials of ERT and HRT are underway. The Women's Health Initiative, a randomized trial of ERT and HRT for primary prevention, is scheduled to yield results by 2005.⁴⁰ WELL-HART, one of several randomized trials of hormones in women with established coronary artery disease examining angiography end points, will be completed in 2000 (Howard Hodis, MD, oral communication, June 1998). HERS should remove any perceived ethical barriers to randomization of women to ERT and HRT, thereby allowing more randomized trials to be conducted and perhaps improving recruitment in trials already started.

When exposures cannot be assigned at random, reliance on observational studies is obligatory. When an exposure can be assigned at random, it should be assigned randomly. Commitment to randomized trials as the standard of proof must be especially strong when the public health implications are so great.