The primary goal of this case–control study was to estimate the odds ratio for incident acute myocardial infarction in relation to current use of estrogen and estrogen-progestogen in a population with a high prevalence of estrogen and estrogen-progestogen use. We used the exposure odds ratio to estimate the degree of risk. The secondary goals of this study were to estimate the odds ratios for incident myocardial infarction in 1) current users of estrogen or estrogen-progestogen in relation to the duration of current use and 2) past users of estrogen or estrogen-progestogen. These topics have not been examined closely in previous studies. We also estimated the odds ratio for incident myocardial infarction in current users of estrogen or estrogen-progestogen, considering many potentially confounding variables.

Methods

Top Methods Results Discussion Author & Article Info References

All case-patients were women 45 to 74 years of age who were hospitalized for myocardial infarction in any of 10 medical centers of the Kaiser Permanente Medical Care Program (KPMCP), Northern California region, between November 1991 and November 1994. Start and finish dates differed by facility so that the number of months of case ascertainment was different at each facility (mean, 33.4 months). Sources used to identify case-patients included daily admission logs containing admitting diagnoses and information about visits to KPMCP emergency departments. We also regularly reviewed the discharge diagnoses from a computerized database of all overnight hospitalizations in KPMCP, Northern California, centers. The medical records of all case-patients were abstracted by medical abstracters and reviewed by the medically trained project coordinators (one was a registered nurse who had research experience in cardiovascular disease studies; the other was a physician), who consulted with one of the physician-investigators as needed. Case-patients who had a history of coronary heart disease before the date of symptom onset, as ascertained by medical record review done before the study interview, were excluded in an attempt to limit the study to persons with incident coronary heart disease. This was done because the presence of coronary heart disease might influence clinical decision making with regard to the use of estrogen or estrogen-progestogen. In addition, a personal history of coronary heart disease was obtained from all interviewed participants (case-patients and controls) and was included as an adjustment variable in the analysis.

Diagnostic criteria for myocardial infarction were adapted from those of the American Heart Association Council on Epidemiology [11]. These criteria use presence or absence of chest pain, results of cardiac enzyme tests, and electrocardiograms to classify events as definite myocardial infarction, probable myocardial infarction, suspected myocardial infarction, or not myocardial infarction. Like investigators in other epidemiologic studies of myocardial infarction defined by using these criteria, we included events categorized as definite or probable cases of myocardial infarction. According to these criteria, the presence of cardiac pain and abnormal enzyme test results are sufficient to categorize an event as a definite or probable myocardial infarction, making it unnecessary to have the electrocardiogram examined by an independent expert. For women who could not be categorized as having definite or probable myocardial infarction on the basis of the presence of chest pain and abnormal enzyme test results, electrocardiograms were sent for coding to the Minnesota ECG Coding Center at the University of Minnesota. This information was used to categorize events as definite myocardial infarction, probable myocardial infarction, suspected myocardial infarction, or not myocardial infarction.

For each case-patient, an attempt was made to interview one control, matched for year of birth and facility of usual care. To do this, three potential controls were selected at random from among all female members of the KPMCP. If the first potential control could not be located, declined to participate, or did not speak English or Spanish, an attempt was made to enroll the second potential control, and, if necessary, the third, in order to obtain one interviewed control per case-patient. Controls were identified for 98.2% of confirmed case-patients, and 79.6% of controls were the first of the three potential candidates.

Eligible case-patients and controls were interviewed in person by trained interviewers who used a standardized instrument. All participants were interviewed as soon as possible after the events in the case-patient occurred. The mean interval between the index date and the interview date was 71.8 days for case-patients and 78.5 days for controls. If a case-patient died or was unable to communicate verbally, an attempt was made to interview a proxy. Interview information from proxies and the corresponding controls was excluded from this analysis because 1) the prevalence of both current and past hormone replacement therapy ascertained by proxy interview was much lower than that ascertained in direct interviews and 2) a comparison of proxy responses with medical records showed substantial underreporting of documented hormone use. We did not use information from medical records for the analysis because use on a particular date and long-term past use could not be reliably ascertained from these records.

All interview questions were asked in relation to an index date, which was the date of symptom onset for each case-patient and the same date for each matched control. Information about the lifetime use of estrogen and estrogen-progestogen preparations was obtained by using the calendar method, in which information on use of hormone replacement therapy is gathered by structuring questions in relation to significant life events. Visual aids, consisting of actual pills for commonly used preparations and a color chart with pictures of all hormone replacement preparations ever marketed in the United States, was used to facilitate recall of the formulation of hormone replacement preparations used currently and in the past.

A study participant was defined as having hypertension, diabetes, or a high cholesterol level if she answered "yes" to questions about whether she was taking medication for these risk factors (she was also defined as having a high cholesterol level if she reported that a physician had told her that she had high cholesterol levels). A participant was defined as a nonsmoker if she answered "no" to the question, "Have you ever smoked cigarettes?" If she answered "yes" to this question, she was categorized as a current regular, occasional, or former smoker on the basis of her answer to the question, "On [index date], were you still smoking regularly?" (Regularly was defined as at least 5 cigarettes per week almost every week.) Body mass index (kg/m²) was determined by using participant-reported height and weight. A participant was defined as having a previous self-reported history of coronary heart disease if she stated that a physician had told her that she had had a heart attack or angina.

The analysis was restricted to postmenopausal women. A study participant was categorized as postmenopausal if she had undergone bilateral oophorectomy or if she had not had a hysterectomy and reported that she had ceased having menstrual periods. Women who have undergone both hysterectomy and unilateral oophorectomy do not know whether they are pre- or postmenopausal on the basis of menstrual function; thus, we excluded such women if they were younger than 55 years of age because they could not be classified with certainty as postmenopausal. These exclusions were applied to both case-patients and controls.

Among women who used replacement therapy, most who had had hysterectomy used estrogen (50.6% used estrogen; 1.2% used estrogen-progestogen) and most women who had not had hysterectomy used estrogen-progestogen (18.4% used estrogen-progestogen; 3.0% used estrogen). For this reason, estimates of the odds ratios for myocardial infarction in users of estrogen who did not have hysterectomy and the odds ratios for myocardial infarction in users of estrogen-progestogen who had had hysterectomy were highly unstable statistically. Thus, we excluded from this analysis women who had had hysterectomy and used estrogen-progestogen, women who had not had hysterectomy and used estrogen, and women of unknown hysterectomy status (n = 3) who used either estrogen-progestogen or estrogen. Women who used only progestogen were also excluded as both case-patients and controls because there were few women in this group and valid estimates of risk in relation to exposure to progestogen only could not be derived. These exclusion criteria were applied to both case-patients and controls.

Conditional logistic regression was used in the main analyses. For these analyses, we defined an exposure variable-current hormone use-as "yes" for women who had had hysterectomy and used estrogen and women who had not had hysterectomy and used estrogen-progestogen. Because of the way in which hormone exposure is defined, the odds ratios for hormone use in the main analysis pertains to current use of estrogen for women who had had hysterectomy and to current use of estrogen-progestogen for women who had not had hysterectomy. Duration (in years) of hormone replacement therapy in current hormone users was examined by replacing the main hormone replacement treatment variable (estrogen for women who had had hysterectomy; estrogen-progestogen for women who had not) with years of use of either therapy. In the multivariate analyses, we adjusted for major established risk factors for coronary heart disease, including cigarette smoking, hypertension, and diabetes. We also adjusted for education as a measure of socioeconomic status and for race and ethnicity because of racial and ethnic differences in the incidence of and mortality rate related to coronary heart disease [12]. We adjusted for a variable if an appreciable change was seen in the exposure coefficients with the addition of that variable; only body mass index was added to the final model on this basis. Adding high cholesterol levels to the model did not change the odds ratio for hormone use. Because of this and the theoretical possibility that the prevalence of high cholesterol levels in this study sample may have been affected by hormone use, we did not include high cholesterol levels in the final model. We also tested for an interaction between hysterectomy status and exposure to hormone therapy by adding to the final model a term for hysterectomy and an interaction term for hysterectomy status and exposure status.

Because the use of estrogen or estrogen-progestogen as hormone replacement therapy was predominantly determined by hysterectomy status, we did an analysis to examine the odds ratio for myocardial infarction in hormone users after stratifying for hysterectomy status in order to provide separate estimates of the odds ratio for myocardial infarction associated with estrogen and estrogen-progestogen. This analysis required performing an unmatched analysis and used unconditional logistic regression, adjusting for age and the variables included in the matched analysis model.

Our results are reported as odds ratios. In the discussion section, we use the term "relative risk" to describe the results from other studies, which are a mix of cohort studies (which reported results as relative risks) and case–control studies (which reported results as odds ratios).

Results

Top Methods Results Discussion Author & Article Info References

A total of 735 events that were considered to be possible incident cases of myocardial infarction were identified by our surveillance during the period of case ascertainment. Of these events, 685 were confirmed cases of definite or probable myocardial infarction. Of the 685 women with confirmed cases, 1 was not eligible for interview because she did not speak English or Spanish. Interviews were completed for 636 of the remaining 684 eligible case-patients or their proxies (93.0%). Thirty-five women refused to be interviewed, 12 were not interviewed because of physician refusal, and 1 could not be located. Two case-patients were excluded because all three of their potential controls refused to participate or were ineligible.

Exclusion criteria were applied as shown in Table 1. After we excluded proxy respondents, women who were definitely or were possibly premenopausal, users of progestogen only, women who had had hysterectomy and used estrogen-progestogen, women who had not had hysterectomy and used estrogen, women with incomplete information on hormone use, and incomplete case-patient-control sets, 438 case-patient-control sets were available for analysis.

Table 2 shows the characteristics of case-patients and their matched controls and the odds ratios for each characteristic, with the matching variables taken into account. Overall, the prevalence of current use of either estrogen or estrogen-progestogen was 34.3% in controls. The odds ratios for myocardial infarction (taking into account only the matching variables) were elevated in relation to smoking, hypertension, diabetes, high cholesterol levels, having never used or occasionally using alcohol, high body mass index, low income, low level of education, and physical inactivity. A self-reported history of previous coronary heart disease was not associated with an increased risk for myocardial infarction, but case-patients who had a history of coronary heart disease at the time of their clinical event (ascertained by medical record review) were excluded from the study by design. The odds ratio for myocardial infarction in current users of estrogen compared with women who had never used estrogen or estrogen-progestogen was 0.85 (95% CI, 0.60 to 1.21). Among women who had not had hysterectomy, the odds ratio for myocardial infarction in current users of estrogen-progestogen compared with women who had never used estrogen or estrogen-progestogen was 0.59 (CI, 0.38 to 0.94).

Table 3 shows that the odds ratio for myocardial infarction in current users of estrogen or estrogen-progestogen compared with women who had never used estrogen or estrogen-progestogen, after adjustment for confounders, was 0.96 (CI, 0.66 to 1.40). Hysterectomy status was excluded from the final model because adding it to the model minimally decreased the estimate of the odds ratio for hormone use (from 0.96 to 0.92) and no significant interaction between hormone use and hysterectomy was seen (P > 0.2) from comparison of the log-likelihood of models with and without the interaction term. The adjusted odds ratio for myocardial infarction in past users of estrogen or estrogen-progestogen compared with women who had never used estrogen or estrogen-progestogen was 1.07 (CI, 0.72 to 1.58). The odds ratios for duration of hormone use among current users did not show a clear trend of decreasing risk with longer duration of use.

Table 4 shows data stratified by hysterectomy status. The odds ratios for current use of estrogen in women who had had hysterectomy (odds ratio, 0.95 [CI, 0.50 to 1.80]) and current use of estrogen-progestogen in women who had not had hysterectomy (odds ratio, 0.89 [CI, 0.52 to 1.53]) were consistent with the results of the main analysis.

Of current estrogen users in this study sample, 84% used conjugated estrogen. Of these, 78% used 0.625-mg doses of conjugated estrogen. Comparisons of odds ratios between women using conjugated estrogen and women using other types of estrogen and comparisons of odds ratios between women with different doses of estrogen had very low statistical power, and the results of these comparisons are not presented here.

Of current estrogen-progestogen users, 98% used medroxyprogesterone acetate as their progestogen. Medroxyprogesterone acetate was used with estrogen in several different regimens, including daily use of 2.5 or 5.0 mg and intermittent use of 5.0 or 10.0 mg for 7, 10, or 12 days of the month. The number of participants using each regimen was small, and we did not do analyses in each subgroup because the precision of point estimates of the odds ratio for any single regimen was low.

Discussion

Top Methods Results Discussion Author & Article Info References

In this study, no statistically significant increase or decrease in the odds ratio for myocardial infarction was seen in postmenopausal women who currently use hormone replacement therapy. Neither the odds ratio for myocardial infarction in women who had not had hysterectomy and were current users of estrogen-progestogen (0.89) nor the odds ratio for myocardial infarction in women who had had hysterectomy and were current users of estrogen (0.95) were significantly different from 1.0 after adjustment for confounders. However, despite the large number of case-patients and the relatively high prevalence of current estrogen and estrogen-progestogen use, the lower bound of the 95% CIs for our odds ratio estimates are compatible with a reduction in the relative risk for myocardial infarction in current users of estrogen-progestogen and estrogen similar to that reported in other studies. Thus, although our study does not establish a benefit for estrogen or estrogen-progestogen with respect to myocardial infarction, it does not rule out the possibility that estrogen-progestogen and estrogen may help prevent myocardial infarction in postmenopausal women.

In previous studies [5-10], estimates of the relative risk for myocardial infarction in users of estrogen-progestogen range from 0.33 for current users of estrogen-progestogen in the randomized trial by Nachtigall and colleagues [5] to 1.2 for women who had ever used estrogen-progestogen in the hospital-based case–control study by Rosenberg and coworkers [8]. Our results add to this contradictory and confusing literature.

Four published papers [1-4] have presented quantitative estimates of the relative risk for coronary heart disease in relation to estrogen use by pooling relative risk estimates from published studies. The pooled estimates are 0.55 [1], 0.50 [2], 0.65 [3], and 0.5 [4]. In each of these analyses, the upper bound of the 95% CI excluded 1.0, showing a statistically significant reduction in the risk for coronary heart disease in estrogen users. Not all of the studies included in pooled estimates of relative risk in previous quantitative overviews and meta-analyses adjusted for risk factors for coronary heart disease and other potentially confounding variables. More important, all of the analyses that derived a pooled estimate of the relative risk for coronary heart disease in estrogen users included studies based only on fatal coronary events. Of 11 such studies [13-23], 8 reported relative risks of less than 0.5. The pooled estimate of relative risk for fatal studies (taking into account the precision of the individual estimate) is 0.37 (CI, 0.28 to 0.49). There is no evidence of heterogeneity in these estimates (test for homogeneity, Q = 6.4; df = 10; P > 0.2). Fatal coronary heart disease is related to risk factors for the occurrence of the disease and risk factors for survival given disease. It is possible that studies that included only fatal coronary events did not adequately consider factors that affect survival, such as access to medical care and compliance with treatments that affect prognosis, which may differ between hormone users and nonusers. Furthermore, some studies of fatal coronary heart disease [16, 22] have relied on data from proxies. We found that proxies substantially underreport current and past hormone use. The discrepancy between our results and those of meta-analyses of studies of estrogen may be due to inclusion in the meta-analyses of studies of fatal events and studies that used proxy respondents.

Directly comparable studies of the effect of estrogen on myocardial infarction-those that examined current users of estrogen with both fatal and nonfatal coronary heart disease and adjusted for cardiovascular risk factors-are summarized in Table 5 [7-10, 24-27]. Relative risk estimates range from 0.60 in the most recent analysis of data from the Nurses' Health Study [10] to 0.97 in the 1976 hospital-based case–control study by Rosenberg and coworkers [24]. Our estimate of the odds ratio for myocardial infarction in current users of estrogen is closer to 1.0 than are many of the estimates from similar studies, but it is within the range of estimates found in previous studies.

Our findings contrast sharply with the most recent analysis of data from the Nurses' Health Study [10], which contains more participants than any other previously published study of hormone replacement therapy and coronary heart disease. In the Nurses' Health Study analysis, after adjustment for risk factors for coronary heart disease and menopausal status, the relative risks for major coronary artery disease were 0.39 (CI, 0.19 to 0.78) in current users of estrogen-progestogen and 0.60 (CI, 0.43 to 0.83) in current users of estrogen. The Nurses' Health Study differed from our study by including angioplasty and coronary artery bypass surgery as well as nonfatal myocardial infarction and fatal coronary heart disease. In addition, the Nurses' Health Study analysis did not and we did adjust for variables that measure socioeconomic status (although we found that removing education from our final model reduced the odds ratios for hormone use only slightly, from 0.96 to 0.92).

It is believed that estrogen prevents coronary heart disease because it favorably alters levels of high- and low-density lipoproteins, favorably alters the balance between coagulation and fibrinolysis, and relaxes arterial smooth muscle in the heart [4]. Some of these effects might slow the process of atherosclerosis. If this is the case, the effect of estrogen on the risk for coronary heart disease might be expected to persist even after discontinuation of therapy. Only three studies [7, 8, 10] have examined risk for myocardial infarction in relation to past use of estrogen; their estimates of relative risk are 0.69 [7], 0.9 [8], and 0.85 [10]. Grodstein and associates [10] reported that the effect of hormone therapy diminished after discontinuation. Our study and the previous studies [7, 8, 10] provide evidence to show that the protective effect of estrogen, if it exists, does not persist forever.

If estrogen has an effect on atherosclerosis, it is possible that longer durations of estrogen use would be associated with larger reductions in risk for coronary heart disease. We found no clear trend toward lower risk with longer duration of use. Grodstein and associates [10] also reported no trend toward decreasing risk with increasing duration of use for current users. In contrast, Rosenberg and coworkers [8] found a trend toward decreasing risk with increasing duration of estrogen use among current users (estimated relative risk, 1.5 for <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).

The data we present are not derived from an experiment, and bias can arise in many ways in nonexperimental studies. Recall bias is of concern in any case–control study that is based on interviews. Nonresponse rates were higher among controls than among case-patients. To assess the possibility of response bias, we examined the medical records of nonrespondent controls. The percentage of nonrespondents who used hormones in the 2 years before the index date was 27%, a percentage that is lower than the percentage of respondent controls (39%) who used hormones in the 2 years before the index date, as determined from medical record review. If this finding was reflected in current use patterns, it would tend to increase the odds ratio for myocardial infarction associated with hormone use.

Another limitation of our study arises from the loss of information from nonrespondent case-patients and the necessity of excluding information obtained from proxy respondents. Examination of information from a review of the medical records of nonrespondents or proxies and direct respondents showed that the percentage of women who used hormone replacement therapy in the 2 years before the index date was almost identical in the two groups-32% for nonrespondents or proxies and 31% for direct respondents. On the basis of this information, we do not believe that loss of information from nonrespondent case-patients and exclusion of proxy responses biased our results.

The effect of hormone use on coronary heart disease dominates the overall risk–benefit Equation foruse [3, 4]. Our results, although inconclusive, add to the growing body of evidence on this subject. Our findings suggest that if postmenopausal hormone use protects against myocardial infarction, the magnitude of the benefit may not be as large as has been estimated in some quantitative overviews that include studies of fatal coronary events only. Our study also underlines the importance of randomized clinical trials designed to address the question of the relation of hormone use to myocardial infarction. The completion of the Women's Health Initiative and other randomized trials will provide unconfounded, definitive estimates of the relative risk for acute myocardial infarction and other coronary events in current users of estrogen and estrogen-progestogen.

Dr. Petitti: Research and Evaluation. 100 South Robles Avenue. 2nd Floor, Pasadena, CA 91101.