Noncardiovascular Disease Outcomes During 6.8 Years of Hormone Therapy: Title and subTitle BreakHeart and Estrogen/Progestin Replacement Study Follow-up (HERS II) FREE

The Heart and Estrogen/progestin Replacement Study (HERS) was a randomized, blinded trial to determine the effects of estrogen plus progestin compared with placebo in older postmenopausal women with coronary disease. Disease surveillance continued in HERS II for an additional 2.7 years, during which many of the women randomized to hormones took open-label estrogen prescribed by their personal physicians but only a few of those assigned to placebo did.¹ During the 6.8 years of observation in HERS and HERS II combined, we detected no overall effect of hormone therapy on cardiovascular disease (CVD) event rates.¹

Hormone therapy after menopause can have effects on a variety of disease outcomes. We present data on non-CVD events over this extended period of hormone therapy. We examine whether the increase in risk of thromboembolic events observed in hormone-treated women in HERS² diminishes over time, as observational studies have suggested,^{3 - 4} and whether the increased rate of biliary tract surgery that appeared to be present in hormone-treated women in HERS⁵ is confirmed as additional events occur. More generally, we present data on the effect of hormone therapy on other disease outcomes thought to be associated with hormone therapy, including fractures, cancer, and total mortality.

The design and methods of HERS and HERS II have been described.^{1 ,6 - 7} Briefly, participants were postmenopausal women younger than 80 years at baseline with coronary artery disease and no prior hysterectomy. Among the reasons for exclusion were a history of deep vein thrombosis or pulmonary embolism, history of breast cancer, endometrial hyperplasia or cancer, abnormal Papanicolaou (Pap) result, any hormone use within the past 3 months, and disease judged likely to be fatal within 4 years.

During the baseline period of HERS we obtained information by questionnaire and interview, and participants underwent physical examination, including pelvic examination with Pap smear and endometrial evaluation and screening mammography. All baseline measures except demographics and health history were repeated at the final HERS visit, an average of 4 months before enrollment in HERS II.

During the HERS trial, women were randomly allocated to receive either 0.625 mg/d of conjugated estrogen plus 2.5 mg of medroxyprogesterone acetate or an identical placebo. After stopping blinded medications at the end of the HERS trial, decisions about whether to undertake open-label hormone therapy were left to the women and their personal physicians. During HERS II participants were called at 4-month intervals and asked about hormone therapy and about symptoms or clinical encounters for possible disease events¹ ; assessment of lipid-lowering drugs (but not bisphosphonates) was also continued during HERS II.

Disease events were ascertained and documented using the methods from the HERS trial.⁷ For in-hospital nonfatal events and deaths, we required the hospital discharge summary and the following information: for pulmonary embolism, clinical signs or symptoms and a positive ventilation/perfusion scan or imaging result; for deep vein thrombosis, documentation by venography, impedance plethysmography, or Doppler ultrasound; for biliary tract surgery, an operative report; for clinical fracture, symptoms and a definite fracture on radiography; and for cancer, report of a tissue diagnosis. For all deaths in which clinical documentation was insufficient, we obtained a death certificate. For out-of-hospital deaths we interviewed the physician and/or next of kin for a description of the terminal event. Data pertaining to suspected outcome events were independently reviewed and classified according to the prespecified criteria used in the HERS trial by 2 physicians at the University of California, San Francisco Coordinating Center who were blinded to original treatment assignment and to open-label hormone therapy.^{1 ,6 - 7}

Both the telephone contacts and the documentation of outcomes were carried out with similar efficiency and completeness in the 2 HERS randomized groups.¹ The proportions of all HERS II deaths for which we obtained clinical documentation beyond a death certificate were 81% among women originally randomized to hormone therapy and 82% among those randomized to placebo.

As described,¹ the primary analyses (using SAS version 8.2, SAS Inc, Cary, NC) compared the risk of events among women assigned to hormone therapy with the risk among women assigned to placebo using unadjusted, intention-to-treat Cox proportional hazards models for time to first event. We censored women at the last contact or at loss to follow-up. In the analyses of biliary tract surgery, we excluded those with a cholecystectomy prior to enrollment in HERS.

Mortality was assessed in all HERS participants throughout the 6.8 years of follow-up, but morbidity surveillance during HERS II was limited to the 93% of surviving women who enrolled. To control for confounding, we estimated the effects of hormone therapy in adjusted Cox models that included all predictors significant at P<.20 in multivariate analysis. The 1993-1994 baseline values were used for all variables except statin use, which was included as a time-dependent covariate. In as-treated adjusted analyses, women were censored 30 days after they become nonadherent to randomly assigned treatment, defined as taking less than 80% of their HERS medication or its equivalent during HERS II.¹

The number of women randomized in HERS was 1380 in the hormone therapy group and 1383 in the placebo group. Of these, 1156 and 1165 enrolled in HERS II, representing 93% of the 2485 surviving women. Vital status was known for 99.8% of these women at the end of HERS II, with final telephone contacts completed in 99.5% of known survivors (see on page 53 of the printed journal). The mean duration of disease event surveillance was 6.8 years for women who survived, which included 2.7 years in HERS II.

Table 1 presents risk factors for non-CVD outcomes using measurements made at the outset of HERS in 1993-1994. All the variables were equitably distributed between randomized groups for both the HERS and HERS II cohorts.

Among women randomized to estrogen plus progestin, the proportions reporting at least 80% adherence to hormone therapy during years 1 through 6 were 81%, 78%, 74%, 67%, 50%, and 45%; comparable proportions for women randomized to placebo were 0%, 2%, 3%, 3%, 4%, and 8%.

There was a 2- to 3-fold increase in incidence of both deep vein thrombosis and pulmonary embolism in the hormone group during HERS (Table 2). The relative hazard (RH) for deep vein thrombosis was considerably smaller (1.23) and no longer statistically significant during HERS II. There was no comparable reduction in RH for pulmonary embolism, although the number of events available to detect such a time trend was small. When risk for venous thromboembolism was examined by year of observation (Table 3), the RH declined after the first 2 years, but the time trend was not statistically significant (P = .08).

The RH for any venous thromboembolic event over the entire 6.8 years was 2.08 (95% confidence interval [CI], 1.28-3.40). Event rates were 5.9 per 1000 person-years in the hormone group and 2.8 per 1000 person-years in the placebo group, an excess of 3.1 per 1000 person-years (P = .003). The number needed to treat (NNT) for 5 years per excess thromboembolic event is 65 when estimated by intention-to-treat and 50 in the as-treated analysis. Seven of the 73 women with thromboembolism died within 30 days of the event, and 3 of these deaths were judged due to the event (all were pulmonary embolisms in women randomized to hormone therapy). Stratifying the overall findings by baseline aspirin use, the data are weakly consistent with the hypothesis that aspirin attenuates the adverse effect of hormone therapy on risk of thromboembolism (RH, 1.68; 95% CI, 0.96-2.92 for aspirin users; RH, 4.23; 95% CI, 1.41-12.7 for nonusers; interaction P = .14).

The RH for biliary tract surgery in the hormone group compared with placebo was 1.39 during HERS, 1.70 during HERS II, and 1.48 overall (95% CI, 1.12-1.95) (Table 2). The overall RH after adjustment for statin use, a statistically significant predictor of lower rates of biliary tract surgery in our study, was 1.44 (95% CI, 1.10-1.90; P = .01).

The rate of surgery was 19.1 per 1000 person-years in the hormone group, an excess of 6.2 per 1000 person-years over the placebo group (P = .002). The estimated NNT for 5 years per excess surgery was 32 (intention-to-treat) and 31 (as-treated). Six of the 211 women who had biliary tract surgery died within 30 days, and 1 of these deaths was judged a consequence of the surgery.

None of the differences between groups in cancer incidence was statistically significant (Table 2). The overall RH comparing the hormone and placebo groups was 1.27 (95% CI, 0.84-1.94) for the 88 breast cancer cases, 1.39 (95% CI, 0.84-2.28) for the 64 lung cancer cases, and 0.81 (95% CI, 0.46-1.45) for the 47 colon cancer cases. Death due to these cancers during the period of observation occurred in 3% of women with breast cancer, 61% of women with lung cancer, and 17% of women with colon cancer.

Other cancers that occurred in at least 5 women included endometrial cancer (2 women in the hormone group and 8 in placebo); malignant melanoma (3 hormone and 5 placebo); lymphoma (7 hormone and 1 placebo); and ovarian cancer (5 hormone and 2 placebo). The total number of women with any cancer was 159 in the hormone group vs 135 in the placebo group (RH, 1.19; 95% CI, 0.95-1.50).

Women randomized to hormone therapy had more hip fractures than women randomized to placebo; the overall RH during 6.8 years of observation was 1.61 (95% CI, 0.98-2.66; P = .06) (Table 2). The RH was 1.16 in HERS and 2.11 in HERS II, a difference that is not statistically significant.

The RH estimates for vertebral, wrist, and other fractures were close to unity. Based on the total of 452 clinical fractures during 6.8 years of observation, the RH for any fracture was 1.04 (95% CI, 0.87-1.25). Prevalence of bisphosphonate use was 2.6% at HERS closeout (in 1998) among women randomized to hormone therapy and 2.5% among those randomized to placebo.

Death rates were high and increasing in this population of older women with coronary disease. Total mortality in the placebo group was 22 per 1000 person-years during HERS and 38 per 1000 person-years during HERS II (Table 4). The RH for total mortality in the hormone vs placebo group was 1.06 during HERS, 1.14 during HERS II, and 1.10 overall (95% CI, 0.92-1.31).

During the entire 6.8 years of observation, there were 261 deaths in the hormone group and 239 in the placebo group. Overall, 61% of the deaths were classified as due to CVD, 19% due to cancer, and 20% due to other cause. Among the CVD deaths, 132 women in the hormone group and 122 in the placebo group died of CHD and the remainder died of stroke (23 and 20) and peripheral arterial disease (4 and 2). Among the cancer deaths, 3 were due to breast cancer (all in women randomized to hormone therapy). For lung cancer deaths 20 occurred in women randomized to hormone therapy and 19 in those randomized to placebo; for colon cancer deaths, 2 and 6; and for all other cancer deaths, 26 and 19. Among the non-CVD noncancer deaths, the numbers in the hormone and placebo groups were 24 and 14 for infectious diseases (including pneumonia and all forms of sepsis); 15 and 13 for respiratory failure (primarily chronic obstructive pulmonary disease, excluding pneumonia); 2 and 9 for traumatic causes, and 10 and 15 for all other causes.

In addition to the unadjusted intention-to-treat findings described above, we also estimated the effects of hormone therapy in Cox regression analyses that adjusted for covariates that were predictors of the outcome. The purpose was to adjust for imbalances that could have developed because some women declined to enroll in HERS II. None of the RH estimates was appreciably altered by the multivariate adjustment (Table 5).

We also carried out analyses restricted to women who remained adherent to randomly assigned treatment (Table 5). These as-treated RH estimates had wider CIs than the intention-to-treat estimates due to the smaller numbers of events (40%-73% of the total in the various models). The as-treated RH for venous thromboembolism was higher than the unadjusted intention-to-treat value (3.04 vs 2.08 for the overall study); the as-treated RH was 5.83 during HERS (95% CI, 2.23-15.3) and 0.70 during HERS II (95% CI, 0.14-3.64). Other as-treated RH estimates in Table 5 differed somewhat from those estimated by intention-to-treat, but the CIs largely overlapped.

This report examines non-CVD outcomes over a total of 6.8 years of observation during and following the HERS randomized trial of hormone therapy in postmenopausal women with coronary disease. We found an increased risk of venous thromboembolism and biliary tract surgery among women randomized to hormone therapy; rates of other important disease outcomes were not favorably affected.

HERS^{2 ,7} confirmed reports from observational studies^{3 ,8 - 9} that hormone therapy after menopause increases risk of venous thromboembolism. The estrogen component of HERS treatment is the likely cause because estrogen without progestin is associated with venous thromboembolism^{3 - 4} and selective estrogen receptor modulators also increase the risk.^{10 - 11} Risk factors for thromboembolism in HERS participants included lower extremity fracture, cancer, surgery, and nonsurgical hospitalization; use of aspirin or statins appeared to be protective.²

HERS participants represent a population at relatively high absolute risk of deep vein thrombosis or pulmonary embolism. The overall rate in the placebo group, 2.8 per 1000 person-years, is far higher than that observed in healthy young postmenopausal women but resembles rates in other populations of elderly women.^{3 ,12} Therefore our estimated NNT, 1 excess thromboembolic event among every 50 to 65 women taking hormones for 5 years, is probably much smaller than it would be for younger and healthier women.

The longer follow-up available in HERS II suggests that the relative risk for venous thromboembolic events may decrease after the second year of hormone therapy (P = .08). Similar decreases over time have been reported in observational studies of postmenopausal hormone^{3 - 4} and oral contraceptive¹³ use, although there is generally some residual excess risk. A decreasing risk is plausible, either through attrition of a susceptible subgroup¹⁴ or by developing tolerance, and the as-treated analysis suggests that it is not just due to decreased compliance with hormone therapy during HERS II. However, the decrease might partly reflect our decision in 1997 (after noting that venous thromboembolism was more common in hormone-treated women) to emphasize to study participants the need to stop HERS treatment in the event of fracture, immobilization, surgery, or cancer.¹⁵

Several decades ago the Coronary Drug Project randomized trial found that high-dose estrogen therapy caused gallbladder disease in men,¹⁶ probably due to alteration of the concentration of cholesterol in the bile,¹⁷ and observational studies of women receiving postmenopausal estrogen have had similar findings.⁵ We previously reported a 38% higher adjusted rate of biliary tract surgery in hormone-treated women (P = .09).⁵ The longer period of observation reported here has revealed the increased risk to be statistically significant. Gallbladder disease was 3 times more common than venous thromboembolism in HERS women, and the NNT for 5 years to cause 1 excess surgery was 31.

Cancer was 19% more common in the hormone therapy group, but the finding was not statistically significant, nor were there statistically significant differences in the rates of any specific cancer. The most common of these, breast cancer, occurred slightly more frequently in the hormone group; the second most common, lung cancer, also occurred slightly more frequently in the hormone group; and the third most common, colon cancer, occurred slightly less often in the hormone group. For each of these 3 cancers, statistically significant associations in the same direction have been found in observational studies and biological plausibility has been discussed.^{18 - 23} However, the wide CIs and limited duration of follow-up do not permit clear inferences from these observations of cancer occurrence.

Risk of endometrial cancer was 75% lower among women randomized to hormone therapy than among those assigned to placebo, but the difference was not statistically significant. The fact that risk is not increased provides assurance that the progestin component of HERS treatment prevents the endometrial hyperplasia and cancer resulting from prolonged use of estrogen.^{24 - 25}

Estrogen is widely believed to prevent osteoporotic fractures. Observational studies reveal 50% lower fracture rates among women taking hormones than in women who are not,^{26 - 27} and there is strong clinical trial evidence for a favorable effect of postmenopausal estrogen treatment, with or without progestin, on bone mineral density in various populations, including older women.^{28 - 29} However, the clinical trial evidence for an effect on fractures has been limited.^{30 - 31} Our earlier report from the HERS main trial revealed little difference between the hormone and placebo groups in risk of any type of fracture.³² Surprisingly, the additional follow-up experience from HERS II suggests a risk of hip fracture among women in the hormone therapy group that is higher than that in the placebo group. Chance may explain the finding, which does not meet the criteria for statistical significance, is considerably smaller in the as-treated analysis, and lacks biological plausibility.

Chance also could play a role in the larger question as to why we did not observe any reduction in risk of all fractures in the hormone group, although the confidence interval makes it unlikely that we missed a large benefit. The absence of routine spine radiographs limited our ability to detect vertebral fractures. Women studied in HERS were not selected for osteoporosis and are therefore not well suited to revealing the effects of fracture-prevention treatments. Clinical trials of bisphosphonates have found an effect on the risk of fracture in women with osteoporosis, but not in women with normal bone density.^{33 - 34}

We recorded 261 deaths in the hormone group and 239 in the placebo group. The absence of mortality benefit contrasts with the finding in observational studies of lower mortality rates among women who use postmenopausal hormones compared with nonusers.^{35 - 36} Population differences could underlie this disparity, but we believe that the lower mortality rate among hormone users in observational studies is primarily due to confounding; women who seek hormone therapy and remain compliant tend to be healthier and wealthier than those who do not.^{37 - 39} Because these characteristics cannot be measured precisely, their influence cannot be adequately addressed by statistical adjustment in observational studies.⁴⁰

Clinical trials have shown that short-term hormone therapy after menopause has favorable effects on surrogate markers for disease, such as blood lipid levels and bone mineral density, and that it relieves menopausal symptoms such as hot flushes and insomnia,^{41 - 42} but the effects of prolonged hormone therapy in preventing clinical events have not been established. HERS is the first randomized trial to provide substantial information on the common disease outcomes that hormones may influence.

HERS II increases the precision of the estimated RHs by adding events that reflect carryover effects from the randomized treatment phase as well as the effects of continuation of the originally assigned treatment. About 50% of the hormone group used open-label treatment during HERS II compared with less than 10% of the placebo group. Those women who did not continue their randomly assigned treatment (crossovers) diminish the power to observe effects of the randomized treatment but do not alter the fundamental value of randomization. To take advantage of this value, our primary analytic approach was an intention-to-treat comparison of outcomes measured over the entire 6.8 years. However, we also examined the findings with as-treated analyses to compensate for the effects of crossing over and with adjusted analyses to compensate for baseline differences resulting from the 7% of women who did not enroll in HERS II. The as-treated and adjusted findings differ somewhat from those of intention-to-treat, but the overall conclusions are not altered.

Inferences about the effects of randomized treatments are also contingent on avoiding unintended interventions applied disproportionately to one randomized group. Randomized assignment was no longer blinded in HERS II, so at the HERS closeout visit and in subsequent telephone calls we provided a neutral message to all women and left advice on hormone use and other preventive treatments to their personal physicians. We also took steps to prevent bias in the ascertainment of outcomes by choosing disease events that were objective and by maintaining the HERS systems for obtaining records and for blinded adjudication. The success of efforts to avoid between-group bias is supported by the comparability in the timing and completeness with which the telephone contacts were made and clinical event data collected.¹

Important limitations of HERS stem from the older age of HERS participants, who averaged 67 years at baseline and 74 years at the end of HERS II, the presence of coronary disease on entry, and the particular estrogen and progestin that we chose to study. These characteristics limit generalizability, and the effects of other hormones in younger, healthier postmenopausal women may be different. Further information on the effects of hormone therapy on disease outcomes in healthy postmenopausal women will be available at the conclusion of the Women's Health Initiative randomized trial.⁴³

Treatment with estrogen plus progestin in older women with coronary disease increased the rates of venous thromboembolism and biliary tract surgery and did not produce favorable trends in overall rates of CVD,¹ fracture, or death. Postmenopausal hormone therapy should be limited to indications that are supported by randomized trial evidence that beneficial clinical outcomes outweigh harmful ones.