Relation between hormone replacement therapy and ischaemic heart disease in women: prospective observational study

Exclusion criteria

Because estimates were stable whatever definition of postmenopausal status was given, we excluded only premenopausal women (n=5182). These women were considered premenopausal because they reported monthly bleeding without current use of hormone replacement therapy.

From register based or self reported information we also excluded 336 women with previous ischaemic heart disease, 176 with previous stroke, 1157 with cancer, and 285 with missing information on hormone replacement therapy. Some women had more than one reason for exclusion. We included 13 084 healthy postmenopausal women in the analyses.

Use of hormone replacement therapy and confounders

Information on use of hormone replacement therapy was self reported and classified as current, past, or never use. A further subdivision for current users was based on type of regimen: unopposed oestrogen or combined therapy.

We obtained information on covariates from the questionnaire. Apart from age, which was used continuously, we included other covariates as categorical variables: diabetes (yes or no), other metabolic disease (yes or no), hypertension or hypertension lowering drugs (yes or no), drugs for angina (yes or no), familial predisposition of women to myocardial infarction (yes or no), smoking (never, former, current), alcohol consumption (0.1-14 or >14 units/week), body mass index (<18.5, 18.5-25, 25-30, >30 kg/m²), leisure time physical activity (heavy >4 hours/week, moderate >4 hours/week, inactive), and self rated health (bad, medium, or good).

End points

We retrieved information on a first episode of ischaemic heart disease from the National Patient Registry of Hospital Discharges, which registers all hospital admissions in Denmark, or from the Cause of Death Register, which registers all causes of death in Denmark. We also linked the data to the Central Person Register, which registers the dates of all deaths. These registers are based on diagnoses from the international classification of diseases system, which changed its coding from the eighth revised version to that of the 10th version in January 1994. Cases of ischaemic heart disease were defined as myocardial infarction, other acute or chronic ischaemic heart disease, angina, or electrocardiographically diagnosed heart disease (ICD codes 410–414 in ICD-8 and codes I20-I25 in ICD-10). For analyses estimating risk of myocardial infarction we used ICD codes 410 and I21-I23. The observation time was until end of 1998.

Statistical methods

We modelled the time to outcomes for ischaemic heart disease by using the Cox proportional hazards model for left truncated and right censored data. We used nurse's age as the underlying time where nurse's age at entry into the study is considered as delayed entry time in the analysis. We used the Cox model to analyse each of the three outcomes of death, ischaemic heart disease, and myocardial infarction, with each of the three variables for use of hormone replacement therapy. For every Cox model we checked the proportional hazard assumption graphically and with Schoenfield residuals. We used univariate analysis to construct a model of the outcome of interest, and we estimated the hazard ratios with 95% confidence intervals for use of hormone replacement therapy, unadjusted for confounders (except age, which is the delayed entry variable). We used multivariate analysis to construct a model of the outcomes, and we estimated the hazards ratios with 95% confidence intervals for use of hormone replacement therapy, adjusted for confounders (familial predisposition, smoking, alcohol consumption, body mass index, physical activity, hypertension, angina, diabetes, thyroid disease, and self rated health). We made a stepwise selection for each outcome to identify the significant confounders where the variable for hormone replacement therapy was kept in the model as the main variable of interest. Thirdly, we analysed the modifying effect of variables by testing the significant covariates in the models for interaction with use of hormone replacement therapy. For significant interactions we present the results stratified on the effect of the variable modifying the effect. We report hazard ratios with 95% confidence intervals for use of hormone replacement therapy at the strata as well as estimates of significant covariates in the model. We chose never users of hormone replacement therapy who had diabetes as the reference to clarify the significance of the modifying effect of diabetes. We excluded missing values from analysis. We performed analyses with SAS version 8.2 and Stata version 7.0.

Mortality

In three multivariate analyses, ever use or current use of hormone replacement therapy and current use of combined therapy was not associated with mortality (table 2). Despite this neutral association, diabetes modified the effect of treatment. Current users with diabetes had a significantly increased risk of death (hazard ratio 3.2, 1.4 to 7.5) compared with never users with diabetes (table 3), whereas the risk for ever users with diabetes was not significantly increased (2.5, 0.7 to 3.4) compared with never users with diabetes. In women without diabetes there was no increased risk of death associated with use of hormone replacement therapy. These increased risks were present even after adjustment for factors associated with an increased risk of death—smoking, hypertension, low body mass index, lower self rated health, angina, no or high alcohol consumption, and low physical activity (table 3).

Ischaemic heart disease

In multivariate analyses, ever users of hormone replacement therapy had a marginally increased risk of ischaemic heart disease compared with never users. The association became insignificant when ever use was subdivided into current use and past use (table 2). The effect of treatment was modified by diabetes, so the overall increased risk of ischaemic heart disease was due to a significantly increased hazard ratio (2.9, 1.1-7.9) among ever users with diabetes at baseline compared with never users with diabetes, whereas there was no increased risk with treatment in women without diabetes (table 3, figure). The hazard ratio was increased further (4.2, 1.4 to 12.5) when current users with diabetes at baseline were compared with never users with diabetes (table 3, figure). These findings were present even after adjustment for smoking, angina or hypertension, low self rated health, and high body mass index—all factors that predicted ischaemic heart disease (table 3). No other significant interactions were found between hormone replacement therapy and the risk factors.

Hazard ratios for ischaemic heart disease associated with use of hormone replacement therapy stratified on diabetic status

• Download figure
• Open in new tab
• Download powerpoint

Myocardial infarction

Diabetes also modified the effect of use of hormone replacement therapy in the subset of patients with ischaemic heart disease classified as myocardial infarction (table 3); no overall effect of treatment was found (table 2). Compared with never users with diabetes, there was an increased risk of myocardial infarction among women with diabetes who were current users at baseline (9.2, 2.0 to 41.4; table 3), but for ever users with diabetes the risk was not significantly increased. Otherwise only smoking and hypertension predicted myocardial infarction (table 3).

Harmful effect

Our finding of a harmful effect of hormone replacement therapy among women with diabetes is indicated by another study, where the cardioprotection associated with being female was only seen in women without, not with, diabetes. 1 13 The harmful effect among women with diabetes could be explained by an influence on glycaemic control, although the current literature is inconclusive.14^–18 Although the biological mechanism remains speculative, Koh and coworkers found that oestrogen did not improve the endothelium dependent vasodilation in women with type 2 diabetes despite its favourable influence on lipoprotein concentrations. 19 20 One explanation is that treatment does not benefit the impaired endothelium and that the procoagulative effects of treatment may dominate.

No other observational studies to our knowledge have examined the interaction between hormone replacement therapy and diabetes.21 Generally, studies have been in healthy postmenopausal women who have had no previous myocardial infarction, diabetes, hypertension, or angina.

To test for a neutral association between hormone replacement therapy and ischaemic heart disease in our sample, we re-examined the data after exclusion of women with diabetes, other metabolic disease, angina, and hypertension. We found no association before and after these exclusions (current users versus never users: 0.82, 0.40 to 1.70 for myocardial infarction and 1.30, 0.92 to 1.85 for ischaemic heart disease).

Our finding of a non-protective effect on ischaemic heart disease agrees with the Women's Health Initiative trial, despite its finding of a significantly increased risk of non-fatal coronary heart disease associated with hormone replacement therapy.8 The type of treatment used in this trial was, however, different from that used by most of the women in our study. Participants in the Women's Health Initiative trial were randomised to conjugated oestrogens combined with medroxyprogesterone acetate, whereas the Danish nurses mainly used oestradiol-17β combined with norethisterone acetate or levonorgestrel. Twice as many women (4.4%) in the Women's Health Initiative trial were treated for diabetes than in our study. Furthermore, the trial reported no important interaction with background variables, including diabetes, indicating that their study, unlike ours, was unable to show an increased risk associated with hormone replacement therapy among women with diabetes. This may have been due to the sample size.

Our finding of a neutral association between hormone replacement therapy and ischaemic heart disease contrasts with most other observational studies, where there was a protective effect of treatment on ischaemic heart disease. 3 4 One explanation is that, unlike other cohorts, the Danish nurses were not healthy users.4 We previously studied whether nurses in this cohort using hormone replacement therapy differed from non-users and found that current users could not be classed as healthy users as they drank more alcohol at weekends, smoked more, were more sedentary, used the healthcare system more often, and had a lower self rated health than non-users.22 These findings are similar for Danish women from the general population using hormone replacement therapy. Reasons why Danish women who use hormone replacement therapy cannot be classed as healthy users include free access to medical care and treatment subsidised by the public health system, prompting most menopausal women to seek medical assistance.

Two large Swedish studies found that hormone replacement therapy protected against myocardial infarction; in one, medium dose treatment offered better protection than low dose treatment. 23 24 Both studies were, however, based on prescription registries and were unable to correct for confounders. This prevents a comparison with our results. Also, the Swedish women who used hormone replacement therapy were healthy users.5

Study limitations

Our results were based on self reports and may therefore have introduced bias. However, in an earlier study we reported a sensitivity of 78.4% and specificity of 98.4% for use of hormone replacement therapy, showing that never users are correctly classified more often than current users. A non-differential misclassification like this tends to bias the risk estimate towards unity. Therefore we may have overlooked a small protective effect of hormone replacement therapy on ischaemic heart disease. It is unlikely, however, that it would influence the interaction between hormone replacement therapy and diabetes on ischaemic heart disease.

The information on diabetes was also self reported and did not distinguish between type 1 and type 2 diabetes. However, despite the risk that diagnosis may have been misclassified, we were able to detect a consistent significant interaction between hormone replacement therapy, diabetes, and risk of death from all causes and morbidity from coronary disease. We expect that nurses would give better answers to questions about treatment and illness than women in the general population. Also, the diagnosis is more likely to be correct because nurses have access to methods for measuring glycosuria. We do not believe that Danish doctors considered diabetes an indication for hormone replacement therapy, because the proportion of current users with diabetes at baseline was low.