Summary of main results
The stimulus for updating this review was the publication of DOPS 2012, however this trial only contributed 3.5% and 19.5% weight to the all-cause mortality and cardiovascular mortality, respectively, in the treatment of a primary prevention population. In the trial population as a whole, there is no evidence that hormone therapy has a role in either the prevention or treatment of cardiovascular disease. There was no strong evidence that treatment with hormone therapy had an effect on overall death rates, cardiovascular disease-related death, non-fatal myocardial infarction, angina, or the number of patients undergoing revascularisation procedures. On the contrary, it is associated with an increased risk of stroke, venous thromboembolism and pulmonary embolism.
The excess risk of stroke in our analyses was observed in the primary prevention analysis (which includes those randomised to either oestrogen alone or oestrogen in combination with progestogen compared to placebo). These findings are based on the two largest trials, WHI I 2002 and WHI II 2004, with follow-up of 5.6 and 7.1 years, respectively. Whilst, no strong evidence of increased risk was observed in any of the secondary prevention trials, including the largest trial HERS I 1998, it is probable that the results from the primary prevention trials are applicable to secondary prevention populations, and that subgroup analyses of these trials were underpowered (due to small trial sizes, low event rates and shorter length of follow-up) to detect any statistically significant differences in stroke rates between hormone therapy and placebo treatment arms. In both WHI I 2002 and WHI II 2004, the excess risk of stroke observed with hormone therapy use was driven by an excess of ischaemic rather than haemorrhagic stroke, with 79.8% and 80.3% of strokes, respectively, observed within the trials being ischaemic (Hendrix 2006; Wassertheil-Smoller 2003). In the same two trials, an increased risk of stroke was apparent after two years of treatment in women taking combination hormone therapy, and after four years for women randomised to oestrogen alone (Hendrix 2006). In both trials, the hazard ratios for ischaemic stroke did not differ significantly in subgroups based on age, years since menopause, prior cardiovascular disease, hypertension or diabetes mellitus status, body mass index, or statin or aspirin use at baseline (Hendrix 2006; Wassertheil-Smoller 2003).
The finding of increased risk for both venous thromboembolism and pulmonary embolism within the overall trial populations appears in our analyses to be driven largely by the excess risk observed in combination hormone therapy (oestrogen combined with progestogen) trials. The greatest risk in primary prevention populations was shown in WISDOM 2007 and WHI I 2002, both testing combination hormone therapy. In secondary prevention populations, the greatest risk was demonstrated in HERS I 1998, also assessing combination hormone therapy. Subgroup analysis of WHI I 2002 for the outcome of venous thromboembolism demonstrated the greatest risk with combination hormone therapy in the first year of treatment (HR 4.01), with lower risk in subsequent years (Cushman 2004). WHI II 2004 also demonstrated a tendency to higher risk early on with only modest increased risk after two years of treatment with oestrogen alone (Curb 2006). When comparing the two studies (combination hormone therapy compared to oestrogen alone), the difference in risk was most apparent after year two when the higher risk in combination hormone therapy was most noticeable (Curb 2006).
Both WHI I 2002 and WHI II 2004 undertook further prespecified subgroup analyses to evaluate the association between participant baseline characteristics and venous thromboembolism and pulmonary embolism risk. Not surprisingly, given the fact that no excess risk was observed within the trial, WHI II 2004 investigators found no strong evidence of interactions between oestrogen alone use and age, body mass index, or most other venous thromboembolism risk factors. The authors did however note, that hazard ratios for combination therapy in WHI II 2004 were significantly higher than those for oestrogen alone, even after adjusting for venous thromboembolism risk factors (Curb 2006). In WHI I 2002, increasing age, being overweight and obese, and having a factor V Leiden mutation (a blood coagulation disorder) were associated with a higher risk of venous thromboembolism compared to placebo (Cushman 2004). Both WHI I 2002 and WHI II 2004 undertook prespecified subgroup analyses to evaluate whether any clinical characteristics of the trial populations may potentially moderate the effects of hormone therapy. The potential predictor variables examined included: age, time since menopause, presence or absence of vasomotor symptoms, prior hormone use, coronary heart disease risk factor status and presence or absence of preexisting cardiovascular disease (Hsia 2006; Manson 2003). None of these variables significantly affected results, although we observed a non-significant trend for a reduction in coronary heart disease risk for women who initiated hormone therapy use within ten years of menopause.
Subgroup analysis of time of treatment commencement in relation to the menopause, found a benefit in overall survival and coronary heart disease (composite of death from cardiovascular causes and non-fatal myocardial infarction) in the hormone therapy group in those who started less than 10 years after their menopause (or before the age of 60). This is similar to the trend in coronary heart disease events in WHI II 2004 shown in Hsia 2006. There was no strong evidence of effect on stroke. There was, however, strong evidence of increased risk of venous thromboembolism, whether started before or after the age of 60.
There was no strong evidence of effect on death or coronary heart disease for the group who started treatment at 10 years or more after the menopause, however there was an increased risk of stroke and venous thromboembolism. We did not analyse any other outcomes, as these analyses relied significantly on subgroup reporting from WHI I 2002 and WHI II 2004, which only reported these outcomes according to time since the menopause, or age that treatment was started. There were insufficient data from other trials to make reliable analyses of other outcomes.
It is worth noting that the benefit seen in survival and coronary heart disease for the group starting treatment less than 10 years after the menopause is from combining five trials all performed in primary prevention populations and all with quite long follow-up, ranging from 3.4 years to 10.1 years. Looking at the event rates in these individual trials it can be seen that the greatest benefit is in those trials with the longest follow-up. It is possible that this could be due to an interaction with time on treatment, whereby coronary heart disease events occur in predisposed individuals early as opposed to later on with hormone therapy treatment, and therefore any risk reduction is observed in the later stages of treatment. This is consistent with the hazard ratio for coronary heart disease for the one-year intervals of follow-up observed in WHI I 2002 (Manson 2003). Therefore, it is not possible to say if short duration hormone therapy is beneficial in this population, only that hormone therapy taken for between 3.4 to 10 years is beneficial in this population.
In analysis of death according to year of treatment, there was no strong evidence of difference between treatment groups by individual year of treatment. There was also no strong evidence of difference in survival comparing cumulative years of treatment, until ten years of treatment, where there was a small survival benefit in the hormone therapy group. However, this was based on two relatively small primary prevention trials, where treatment was started shortly after the menopause (DOPS 2012; ERT II 1979). One of the trials had poor methodology (ERT II 1979). It is possible that there are other explanations for the benefit seen in this analysis, other than the duration of treatment, such as the timing of commencing treatment.
Overall completeness and applicability of evidence
There are a number of limitations to the evidence base reviewed. Firstly, it should be highlighted that the results are based on those obtained in 19 RCTs, with the majority of statistically significant findings derived from the results of the three largest trials, HERS I 1998, WHI I 2002 and WHI II 2004, which dominate the results. These three trials all evaluated oral conjugated equine oestrogen 0.625 mg, with or without continuous medroxyprogesterone (MPA) 2.5 mg. Other trials evaluating different types of hormone therapy tended to be much smaller with a shorter duration of follow-up, and reported few if any major clinical events. There is some debate regarding the external validity of the findings of WHI I 2002 and WHI II 2004, and the degree to which they apply to any type of hormone therapy, other than continuous combined oral conjugated equine oestrogen 0.625 mg with or without MPA 2.5 mg. The effects of hormone therapy may vary with different oestrogens and progestogens, different doses and routes of administration. However, in order to pool the results of different studies statistically, we had to make assumptions regarding a ‘class effect’ of hormone therapy, which may not be warranted.
The clinical outcomes of interest in the review were secondary outcomes in five of the trials (DOPS 2012; EPAT 2001; ERA 2000; ESPRIT 2002; WAVE 2002) and reported as adverse events in five more (ERT II 1979; Greenspan 2005; STOP IT 2001; WELL-HART 2003; WHISP 2006). It can therefore be postulated that these trials may not have been sufficiently powered in order to detect differences in clinical treatment effects between the hormone therapy and placebo arms, as this was not the primary aim of these trials. Furthermore, as previously highlighted, seven of the trials were stopped early (DOPS 2012; EAGAR 2006; EPHT 2006; EVTET 2000; WISDOM 2007; WHI I 2002; WHI II 2004), either as other trial results were published showing no beneficial effects on cardiovascular disease outcomes for hormone therapy relative to placebo, or observation of a detrimental effect either on cardiovascular disease outcomes or adverse events was shown. The mean length of trial follow-up therefore ranged considerably from seven months to 10.1 years, with a mean duration of follow-up of 3.6 (median of three) years across the trials. The early stopping of the trials has implications both for the power to detect differences in treatment effects between the hormone therapy and placebo arms (as the sample size will have been predicated based on the original proposed length of follow-up, and assumptions regarding the number of events observed), and also for limiting the availability of evidence on the longer-term treatment effects of hormone therapy compared to placebo. A further limitation of the evidence base reviewed relates to the impact of patient medication compliance, which ranged considerably between the trials. A high proportion of women in the trials did not receive the treatment to which they were randomised. Overall, the number of women who discontinued their medication or took less than 80% was disproportionately high in the hormone therapy trial arms, presumably due to medication side effects. The authors of WHI I 2002 noted that if discontinuation of treatment and initiation of non-study treatment occurred independently of risk factors for clinical outcomes, their intention-to-treat analysis underestimates both the harms and benefits of hormone therapy among women who adhere to treatment.
Quality of the evidence
A summary of the findings and strength of evidence can be found in Summary of findings for the main comparison; Summary of findings 2; Summary of findings 3 and Summary of findings 4. In the primary prevention population, the quality of evidence for death and cardiovascular disease was high. We downgraded the quality of the evidence by one level for venous thromboembolism and pulmonary embolism due to inconsistency of effect across the study results. When HT considered as a secondary prevention strategy, the quality of the evidence was also high for death and venous thromboembolism. The confidence intervals for the estimated effect on stroke and pulmonary embolism could not exclude small decreases or large increases in risk. For the subgroup of studies addressing the effects of HT started less than 10 years since the menopause the quality of evidence was downgraded one level for the outcomes of mortality and coronary heart disease as the results of the analysis were dominated by the results of a few large trials. Overall study quality was high (Figure 4). The vast majority of trials had adequate generation of randomised sequences (15 out of 19), 17 out of 19 were double-blinded and 13 out of 19 were analysed on an intention-to-treat basis. Participants lost to follow-up were generally low, except in two trials: 14.9% in STOP IT 2001 and 19% in WHISP 2006, though these provided relatively low weight to the analysis. Only two out of 19 trials were at risk of selective outcome reporting.
Potential biases in the review process
There are a number of potential biases in the review process, although we made attempts to limit these. The bias of most concern is that of patient selection bias which limits external validity. Nearly all of the included trials had a mean participant age of over 60 years at baseline, and only one trial (DOPS 2012) focused on women who were either peri-menopausal or around the time of the menopause. Whilst these inclusion criteria reflected the aims of the trials, it does not reflect usual clinical practice, in which hormone therapy is prescribed for the relief of vasomotor symptoms at the time of menopause.
Despite extensive searches it is possible that we failed to identify all relevant studies. However, given the dominance of WHI I 2002 and WHI II 2004 on the results of the review, it is unlikely that we missed any trials large enough to impact substantially on the results. Additionally, as already indicated, assumptions had to be made in the analyses regarding the effects of different HT preparations in order to undertake meta-analyses. These assumptions may not be warranted, as it is as yet unclear how different preparations and doses may differ.
Our assessment of the timing hypothesis could be considered a post-hoc change to the original protocol for this review. The data for events since menopause (stratified in decades) were only available for two studies (WHI I 2002; WHI II 2004) and for the remaining studies baseline characteristics for eleven studies (DOPS 2012; EPHT 2006; ERA 2000; ERT II 1979; ESPRIT 2002; HALL 1998; HERS I 1998; WELL-HART 2003; WEST 2001; WHISP 2006; WISDOM 2007) were available for us to allocate the study population as a whole to either commencing treatment less than 10 years since the menopause or 10 years or more since the menopause. For six studies (EAGAR 2006; EPAT 2001; EVTET 2000; Greenspan 2005; STOP IT 2001; WAVE 2002) these data were not available. However, age was reported and therefore whole study populations were allocated accordingly to those less than 60 years of age or those 60 years of age or older when they commenced treatment. It is highly likely that trial populations were distributed across a range of ages and time since menopause, and it is therefore likely that a proportion of study populations were incorrectly allocated. This will be more of a problem in study samples with large standard deviations for time since menopause, or age, and also in those who have a mean or median age or time since menopause close to the cut-off (10 years since the menopause and 60 years of age). Although we remain confident that the subgroups are broadly representative of the study populations of interest, subgroup level data for each study or individual participant data would represent more robust approaches to testing the timing hypothesis.
Agreements and disagreements with other studies or reviews
Magliano 2006 pooled results from seven of the trials included in the current review (ERA 2000; ESPRIT 2002; HERS I 1998; WAVE 2002; WEST 2001; WHI I 2002; WHI II 2004), and concluded that there was no impact of hormone therapy compared to placebo on total mortality or non-fatal myocardial infarction, but strong evidence of an increased risk in the number of strokes (RR 1.29, 95% CI 1.13 to 1.48) observed with hormone therapy use. Likewise, a meta-analysis by Bath 2005, pooling 28 RCTs, reported hormone therapy was associated with an increase in the risk of stroke, particularly ischaemic stroke. Furthermore, those participants who had a stroke in the hormone therapy groups appeared to have a worse outcome. However, it is unclear to what degree the results of this review are applicable to post-menopausal women, as the review had very broad inclusion criteria, and pooled a wide range of trials which used different types of hormone therapy for a range of indications, some of which included male participants.
Salpeter 2006, in a meta-analysis aimed to examine the effect of hormone therapy on coronary heart disease events in younger and older post-menopausal women (defined as participants with a mean time from menopause of less than or greater than ten years, or mean age less than or greater than 60 years). The analyses of 23 trials (ten trials with younger women and 11 trials with older women), included the relevant Women's Health Initiative age-specific subgroup data in one or the other group as though they had originated from separate RCTs. The results showed that hormone therapy reduced coronary heart disease events in younger women, but not in older women. This is comparable with our findings in this review.
Miller 2002 performed a meta-analysis of venous thromboembolic outcomes in post-menopausal women using oestrogen replacement. The review included both oestrogen alone and combination therapy and RCTs, case-control studies and a cohort study. They found an increased risk with hormone therapy (RR 2.14, 95% CI 1.64 to 2.81). The analysis was published in 2002 and did not include either Women's Health Initiative studies, which contributed the largest portion of our included population, but the risk is comparable with our findings. They also found that the risk was highest in the first year, but in the majority of studies remained elevated for the duration of follow-up.