Background Despite decades of evidence from observational studies, the use of hormone therapy for the prevention of cardiovascular disease (CVD) among postmenopausal women is controversial. The recent completion of several randomised clinical trials examining the effects of hormone therapy on CVD presents an opportunity to provide a more precise estimate of the cardiovascular risks of hormone therapy.
Objective To summarise the effects of hormone therapy on CVD in postmenopausal women.
Search strategy MEDLINE, EMBASE, the Cochrane Library, DARE and CENTRAL were searched for clinical trials reporting mortality and/or CVD outcomes in association with hormone therapy. Bibliographies and editorials were also reviewed. All studies were reviewed and rated for quality independently by two reviewers.
Selection criteria High quality, randomised placebo-controlled clinical trials of hormone therapy (duration greater than one year) in non-hospitalised postmenopausal women were included.
Data collection and analysis Summary relative risks were estimated for all-cause mortality, coronary heart disease (CHD) mortality, non-fatal acute myocardial infarction (AMI) and all stroke.
Main results Seven randomised clinical trials met the inclusion criteria. The use of hormone therapy had no significant effect on all-cause mortality, non-fatal AMI or CHD mortality, with relative risks of 1.02 [95% confidence interval (CI) 0.93–1.13], 1.00 (0.88–1.14) and 0.99 (0.82–1.21), respectively. For all stroke, the summary relative risk was 1.29 (1.13–1.48).
Author's conclusions This systematic review, incorporating the latest available trial data, shows that hormone therapy does not significantly change the risk of all-cause morality, CHD death or non-fatal AMI but increases the risk of stroke in postmenopausal women.
The last two decades have seen the accumulation of a vast amount of epidemiological data concerning the use of hormone therapy and cardiovascular disease (CVD). Despite this, the risks or benefits of this therapy for the prevention of CVD remain unclear.
There is evidence from a number of observational studies that women taking hormone therapy have better CVD risk factor profiles1 and reduced coronary heart disease (CHD)2–4 and stroke outcomes.5–8 Recently, however, data from several large randomised controlled clinical trials of hormone therapy have provided evidence suggesting that hormone therapy is not associated with any beneficial effect for CVD and may actually confer an increased risk.
The aim of this systematic review was to provide a more precise estimate of the cardiovascular effects of hormone therapy using meta-analyses.
Identification of relevant trials
MEDLINE (1955 to August 2004), CENTRAL, CINAHL and EMBASE (1966 to August 2004) databases were searched to identify randomised controlled trials of hormone therapy and CVD. MeSH search terms were combined using Boolean operators as follows: (‘hormone replacement therapy’ OR ‘oestrogen replacement therapy’ OR ‘estradiol’) AND ‘cardiovascular disease’. This search was limited to any clinical trial, controlled clinical trial or randomised clinical trial. Animal studies were excluded. Reference lists, letters, editorials and clinical trial databases were also reviewed. Consistent with changing practice, the term ‘hormone therapy’ rather than ‘hormone replacement therapy’ is used in the review.
The initial search strategy identified 1004 abstracts. To be included in the meta-analysis, studies had to satisfy the following inclusion criteria:
1randomised controlled trials of greater than one year duration;
2compared the use of hormone therapy to a placebo;
3recruited non-hospitalised, postmenopausal women as participants; and
4measured ‘hard’ cardiovascular outcomes.
Definition of outcomes
‘Hard’ outcomes were chosen, as they are generally consistent across trials and less susceptible to misclassification. Therefore, less robust outcomes such as transient ischaemic attack, angina pectoris and heart failure were not included. Cardiovascular outcomes of interest included any of the following:
(i)non-fatal acute myocardial infarction (AMI);
(iii)death due to CHD; and
Deaths due to CHD included documented fatal AMI, sudden death within 1 hour of onset of CHD symptoms, unwitnessed death occurring out of hospital in the absence of other known cause and death due to revascularisation or other coronary procedures. The ‘all stroke’ outcome included fatal and non-fatal strokes but excluded transient ischaemic attacks.
Studies examining surrogate cardiovascular outcomes such as progression of atherosclerosis, composite CVD risk scores or CVD risk factor profiles were specifically excluded.
All abstracts were evaluated independently by two reviewers (DJM, SLR); any disagreement was resolved by discussion. These reviewers then examined the full articles of studies that met the inclusion criteria. Final eligibility of studies was decided by consensus.
The quality of the trials examining hormone therapy for the prevention of CVD was assessed with consideration of several standard quality criteria including adequate randomisation, blinding of patients, reviewers and researchers, intention-to-treat analysis and follow up rates.23 Trials were allocated points if they met the criteria for a possible maximum of five points.23 Quality criteria were applied to the trials independently by two reviewers (DJM and SLR). Trials scoring four or above were considered high quality and included in the analysis.
All meta-analyses were performed using the Der Simonian and Laird random-effects model in Review Manager (RevMan) v4.2 (Update Software, Oxford, England). Heterogeneity between trials was assessed by using the Mantel–Haenszel χ2 test. Because not all the trials reported their results as incidence rates, instead reporting adjusted hazard ratios, authors were contacted to provide numbers of events and person-years for each outcome of interest. Pooled relative risks and 95% confidence intervals (95% CIs) were estimated for all trials as a whole and stratified into two groups: combination trials (trials using hormone therapy composed of oestrogen and progesterone) and oestrogen-only trials. Results are displayed in forest plots.
The effect of age on the four outcomes was assessed by stratified meta-analysis. Trials were divided into those where participants’ mean age at baseline was less than 65 years or 65 years and greater. The 65-year threshold was chosen arbitrarily but reflected the rapid increase of CVD in women after this age. It was not possible to further stratify the trials into other age categories, as data were not available.
The broad search strategy used in this review resulted in the identification of many studies that did not satisfy the inclusion criteria. Typically, studies were excluded because they were observational studies, represented multiple publications or subanalyses from an already included trial or were trials testing other hormones or addressing different or short term outcomes. Of the 1004 abstracts reviewed, nine trials were identified that warranted closer review. Of these, only seven met all the inclusion criteria, and two trials met all but one criterion (1Fig. 1). The included trials were the Estrogen Replacement and Atherosclerosis (ERA) Trial,9 the Oestrogen in the Prevention of Re-Infarction Trial (ESPRIT),10 the Heart and Estrogen Replacement Study (HERS),11–13 the Women's Angiographic Vitamin and Estrogen (WAVE) Trial,14 the Women's Estrogen for Stroke Trial (WEST),15 the Women's Health Initiative Estrogen and Progestin Trial (WHI-EP)16–18 and the WHI Estrogen Only (WHI-E) Trial.19 The characteristics of the seven included trials, which together comprised a total of more than 32,000 subjects, are shown in 1Table 1.
95.9% without CHD, 4.1% with prior MI or revascularisation
Two trials were excluded: the Women's International Study of Long Duration Estrogen after Menopause (WISDOM)20,21 and a study by Nachtigall et al.22 The recently terminated WISDOM trial was excluded because no results were available for analysis. Nachtigall et al.22 studied hormone therapy in long term hospitalised patients with co-morbidities and thus did not meet the inclusion criteria.
The two trials stemming from the Women's Health Initiative were treated as separate trials as they used different hormone treatments, had separate placebo groups and different patient populations. In this review, they are referred to as WHI-EP and WHI-E trials. Furthermore, while the WHI-EP trial was terminated in July 2002, the WHI-E trial continued and was terminated in February 2004.
The WEST, WAVE, ESPRIT and ERA trials included both women with and without a uterus. In contrast, women who had undergone hysterectomy were excluded from HERS and WHI-EP. All women in WHI-E had undergone a hysterectomy and were allocated oestrogen only or placebo.
The HERS trial underwent review after four years, the outcome of which was to reveal treatment allocation to allow patients to make an informed choice about their hormone therapy. Follow up of participants continued for a further 2.7 years in an open-label, unblinded study phase: HERS II. Because of the possible reduction in robustness of the data, these analyses only included data up to the end of HERS.
Assessment of quality
The quality of the trials included in this meta-analysis was very high with all seven trials receiving the maximum score using the quality criteria of Jadad et al.23 All trials concealed allocation via randomisation, were placebo-controlled and conducted analysis by intention-to-treat. Overall, few (3%) participants were lost to follow up and all trials adjudicated outcomes in a blinded fashion.
Compliance and cross-over
Treatment compliance rates varied among trials. In the ERA trial, women in hormone therapy and placebo groups took 79% and 86% of prescribed medications, respectively. However, in the WAVE trial, compliance was lower, with women in the hormone therapy and placebo groups taking only 67% and 70% of tablets, respectively.
In the WEST trial, of the women who were allocated hormone therapy, compliance was 90%. Compliance rates in HERS were also high. After year one, the proportion of women who reported taking treatment in the hormone group was 82% and in the placebo group was 91%. These proportions had declined by the end of the third year to 75% and 81%, respectively. However, in ESPRIT, compliance was very poor and by the end of the study, 57% and 37% of those assigned estradiol valerate and placebo, respectively, were not compliant. Similarly, compliance was low in the WHI-E trial where, at the termination of the study, 53.8% of women had ceased taking study medication. In the WHI-EP trial, 42% and 38% ceased medication in the treatment and placebo groups, respectively.
Cross-over rates in the included trials were also low. The number of women who crossed over from taking placebo to open-label oestrogen in the WAVE and ERA trials were 4.2% and 4.8%, respectively. In the WHI-EP trial, 6.2% of women in the oestrogen plus progestin group and 10.7% in the placebo group had initiated hormone use through their own clinicians by the sixth year of study. Similarly, in the WHI-E trial, 5.7% and 9.1% of women in the treatment and placebo groups, respectively, had commenced hormone use by the sixth year of the study. Among women who discontinued study treatment in WEST, 1% and 2% of the estradiol and placebo group reported using open-label oestrogen, respectively. Among women who stopped medications in the HERS trial, 8% of those assigned to hormone therapy and 3% assigned to placebo reported taking open-label or transdermal hormone preparations. Crossover data were not specified in ESPRIT.
Effect of hormone therapy on all-cause mortality
When the seven included trials were analysed together, no significant differences between the hormone therapy and placebo groups for all-cause mortality rates were observed. The summary relative risk for the effect of hormone therapy on all-cause mortality was 1.02 (95% CI 0.93–1.13) (2Fig. 2). There was no significant heterogeneity between trials (P= 0.76). No significant differences were observed between hormone therapy and placebo when trials were stratified into combination therapy or oestrogen-only trials, consistent with no reduction in all-cause mortality over 2.0–6.8 years of treatment (Fig. 2).
Effect of hormone therapy on stroke
Only six of the seven included trials reported stroke in a manner consistent with the definition adopted here (fatal or non-fatal, ischaemic or haemorrhagic stroke, excluding transient ischaemic attacks). In the overall analyses, the summary relative risk for all stroke was 1.29 (95% CI 1.13–1.48) indicating that hormone therapy was associated with a significantly increased risk of stroke (3Fig. 3). There was no significant heterogeneity (P= 0.84). The risk of stroke was also significant in each meta-analysis of combination therapy trials; 1.29 (95% CI 1.06–1.56) and oestrogen-only trials; 1.30 (95% CI 1.07–1.57). There were insufficient data reported in the trials to analyse ischaemic and haemorrhagic strokes separately.
Effect of hormone therapy on CHD deaths
All trials had point estimates suggesting that hormone therapy had little effect on CHD deaths. Meta-analysis of the seven trials reporting this outcome produced a summary relative risk of 0.99 (95% CI 0.82–1.21; 4Fig. 4). There was no significant heterogeneity (P= 0.75). The results remained non-significant when meta-analyses were performed on combination therapy and oestrogen-only trials, separately.
Effect of hormone therapy on non-fatal AMI
All included trials had point estimates suggesting that hormone therapy had little effect on non-fatal AMI. Not surprisingly, meta-analysis of these data produced a summary relative risk of 1.00 (95% CI 0.88–1.14; 5Fig. 5). No significant heterogeneity was found (P= 0.44). The summary relative risks were also not significant when data were analysed in strata of combination therapy and oestrogen only.
Effect of age on hormone therapy and CVD
To determine whether the age at initiation of hormone therapy affected any outcome, a stratified analysis was performed which categorised trials by mean age of participants at baseline (2Table 2). The younger group comprised three trials with a mean age below 65 years while the older group comprised four trials with a mean age of 65 or greater (refer Table 1). Although the risk of stroke associated with hormone therapy was significant in both age groups, the risk was higher in trials of hormone therapy in younger women than older women (Table 2). There were no significant differences between the groups for the other outcomes. Furthermore, there was no significant heterogeneity observed.
Table 2. Relative risk for CHD mortality, all stroke, non-fatal AMI and all-cause mortality in younger and older women associated with hormone therapy
Pooled RR (95% CI)
Mean age, less than 65 years
0.98 (0.75, 1.30)
Mean age, 65 years or over
1.00 (0.77, 1.31)
Mean age, less than 65 years
1.35 (1.14, 1.60)
Mean age, 65 years or over
1.20 (0.95, 1.51)
Mean age, less than 65 years
1.04 (0.79, 1.38)
Mean age, 65 years or over
0.94 (0.75, 1.17)
Mean age, less than 65 years
1.02 (0.90, 1.15)
Mean age, 65 years or over
1.03 (0.86, 1.24)
In postmenopausal women similar to those enrolled in these seven large trials, our systematic review suggested that hormone therapy increases the risk of stroke but has no statistically significant effect on risk of non-fatal AMI, mortality from CHD or from all causes. These findings differ from those seen in many observational studies, which suggested that hormone therapy was associated with reduced risks of CHD events2–4,24 and possibly stroke.5–8
The most salient finding of this review was the increased risk of stroke associated with hormone therapy use which persisted when analyses were stratified into trials using combination therapy and oestrogen only. Although some observational and interventional studies have previously reported a reduction in the risk of stroke associated with oestrogen use,5,6,8 the results presented here are not without support. The risk of stroke was higher among generally healthy, postmenopausal women who were receiving oestrogen in the Framingham Heart Study25,26 and in more recent follow up of the Nurses Health Study.4,27 More importantly, these results are consistent with clinical trial data from WHI16 and with two other meta-analyses of stroke from hormone therapy using a variety of study designs28 and one using only trials.29
The mechanism by which exogenous oestrogen may affect the risk of stroke has been postulated to involve many factors within the coagulation pathway. Estrogen regulates synthesis of several coagulation factors and fibrinolytic proteins.30 It increases levels of plasminogen, protein C and factor VII31 and reduces levels of anti-thrombin III,32 fibrinogen33 and plasminogen-activator inhibitor type 1.34 These effects are dependent on dose, type and duration of oestrogen.30 Similarly, hormone therapy has been shown to increase the sensitivity of neurones to ischaemia35 and may mediate the pro-inflammatory potential of oestrogen.36 In contrast, studies in postmenopausal women with atherosclerosis have shown that oestrogen enhances vasodilation and decreases platelet aggregation.37 Furthermore, short term trials have suggested that hormone therapy improves the vascular risk factor profile.1 The net impact of these various effects on CHD and stroke risk is unclear. Further research is required in this area.
There are many possible explanations to account for the disparate results between observational and clinical trial data of hormone therapy and CVD. The most important relate to specific inherent biases arising within observational studies. It has been suggested that women who routinely use hormone therapy may have lifestyles that afford them a lower risk of CVD. In general, these women are more affluent, leaner, exercise more, consume greater quantities of alcohol, are more highly educated, more likely to comply with treatment, have better access to health care and are more likely to be treated for conditions that, if not managed, would place them at high CVD risk.38,39 It is likely that these biases, at least in part, account for the disparity between data from observational and experimental studies.
It has been postulated that age differences are a major reason explaining the discrepancies between observational and experimental data.40 Other research has suggested that hormone therapy could be beneficial in younger women with minimal atherosclerosis, whereas hormone therapy in older women may destabilise existing plaque leading to plaque rupture and complications.40 It was estimated that in the key cohort studies (the Nurses’ Health Study,3 the Iowa Women's Health Study41 and others)24,42,43 the weighted mean age of participants was 57 years. In the trials included in this meta-analysis, the estimated weighted mean age was 64 years (data not shown). This difference suggests that age may at least partly explain the reported difference in the findings from observational and experimental data.
Furthermore, when the trials were stratified into younger and older age groups, the summary relative risks for CHD mortality, non-fatal AMI and all-cause mortality remained non-significant. However, hormone therapy was associated with a higher risk of stroke in the trials in which women had a mean age less than 65 years than the older group. This was not surprising given the relative lack of competing risk factors for stroke in younger women compared with older women. In the present analysis, the effect of age on the relationship between hormone therapy and CVD was limited as individual patient data were not available.
There has been much debate regarding whether the oestrogen or oestrogen and progesterone combination hormone therapy are responsible for any possible adverse effects of hormone therapy. In this study, an increased risk of stroke was observed in the overall analyses including all trial data and stratified analyses of combination therapy or oestrogen-only trials. For the other outcomes, data from stratified analyses were not significantly different to the overall results. However, with due caution given to the limitation of stratified analyses (small numbers of trials in each stratum), it is tempting to note that in both outcomes of ‘CHD death’ and ‘non-fatal myocardial infarction’, the summary estimate in the oestrogen only group indicates that there may be a trend towards oestrogen having a protective effect. Additional studies should be undertaken to explore this further.
It is also possible that the incomplete capture of early clinical events in cohort studies and the differences in hormone regimens, duration of treatment and characteristics of study populations may have contributed to the discrepancies between observational and experimental studies.44
Strengths and weaknesses
There are several limitations of this systematic review. Participants in these trials were predominantly Caucasian women recruited from within the USA. This affects the generalisability of results to other populations and ethnic minorities.
The value of meta-analysis of any clinical question rests on whether it is sensible to pool the results of studies within the particular area of interest. In this review, meta-analysis was justified on the basis that the quality of the included trials was consistently high, there was no statistical evidence of significant heterogeneity between trials and all included trials were more similar than they were different. However, it must be borne in mind that small differences in trial characteristics such as the dose and formulation of hormone therapy, compliance and cross-over rates may still have influenced the results.
The choice of endpoints analysed in this review was limited by the endpoints reported by each trial. This was only partly overcome by contacting authors to obtain data pertaining to other outcomes. Consequently, the meta-analysis of a composite CVD outcome was not possible.
Although a thorough systematic search was conducted, it is possible that unpublished studies may have been missed. Therefore, the possibility that the systematic review could be affected by publication bias cannot be ruled out. Furthermore, a funnel plot was not meaningful due to the small number of trials.
There is a suggestion that hormone therapy may particularly increase the risk of cardiovascular events early after initiation. In this systematic review, discrimination between early and late events was not possible. All trials analysed their data using Cox proportional hazards regression, incorporating time-to-event and adjusting for covariates; however, the specific combinations of covariates varied widely among trials, limiting the use of the published hazard ratios in the analyses. Thus, the pooled relative risks estimated in this meta-analysis represent the crude relationship between hormone therapy use and CVD and may be influenced by confounding.
The overall strength of this review was underpinned by the inclusion of trials with high quality methodology and comparable characteristics. All trials had very similar end-point definitions, were blinded, placebo-controlled and were appropriately randomised.
This meta-analysis includes trials not included before and is more comprehensive than previous meta-analyses of the effect of hormone therapy on CVD.45 Other published meta-analyses on hormone therapy and CVD46,47 included data from observational (case-control and cohort) but very few experimental studies. Moreover, the experimental studies that were included in previous analyses were conducted some time ago, were methodologically flawed24 and did not meet our inclusion criteria.
The trials included in these analyses were conducted primarily in women of Caucasian background, thus research may be required in other populations. Given the results of this review, research should also focus on subgroups of women at particularly high risk of CVD, such as women with diabetes or hypercoagulable conditions. Other gaps in current understanding include the relationship between CVD and different hormone combinations, doses and prep+arations, the potential interactions between hormone therapy use and homocysteine and potential effect modification by aspirin.
Estimates of the effects of hormone therapy should be interpreted and compared cautiously. It is very difficult to directly compare the benefit of hormone therapy in preventing menopausal symptoms, osteoporosis, and improving quality of life to a small increased risk of stroke. The balance of risk and benefit must be weighed up for each individual patient.
When given for periods ranging from three to seven years, hormone therapy increases the risk of stroke, but has no significant effect on the risk of non-fatal AMI, or mortality from CHD or all causes in postmenopausal women. The use of hormone therapy for the reduction of risk or prevention of CVD is not supported by this systematic review and meta-analysis of experimental studies.
Competing interest notification
All authors have nothing to declare.
DJM was supported by a National Heart Foundation of Australia Post-Graduate Research Fellowship. SLR was supported by a Monash University Small Grants Award.▮