Association between hormone replacement therapy and subsequent stroke: a meta-analysis

Quantitative data synthesis

We analysed data using Stata (version 7) and Cochrane Review Manager (version 4.2). We assessed the effect of hormone replacement therapy on dichotomous outcomes from the odds ratio calculated with a random effects model since we expected the trials to be heterogeneous.

We used pre-specified sensitivity analyses to explain any heterogeneity, including phase of stroke prevention (primary or secondary), type of hormone replacement therapy (oestrogen only or oestrogen plus progesterone), type of oestrogen (estradiol or conjugated equine oestrogen), size of trial (£ 5000 or > 5000 patients), length of follow up (£ 3 years or > 3 years), sex, and quality of trial (high (5) or low (< 5)). We assessed interactions between subgroups and treatment. We examined publication bias using Eggers test.12

Study characteristics

We identified 28 trials with 39 769 subjects for inclusion in our study (fig 1, table A on bmj.com).w1-w28 The trials varied in size between 59 subjectsw17 and 16 608.w26 Fifteen trials investigated primary prevention of stroke,w5-w7 w9-w12 w14 w17-w19 w22 w24 w26 w28 and 12 studied patients with prior vascular events (stroke,w1-w3 w21 ischaemic heart disease,w4 w8 w15 w16 w20 w23 w25 and venous thromboembolism w13). The average age of patients varied from 55 to 71. Three trials included men, with one trial of men exclusively.w1-w3 Three trials required that women should not have had a hysterectomy.w9 w20 w26 Follow up varied from 0.7 to 6.8 years. Twelve trials studied hormone replacement therapy with oestrogen alone, and 16 studied oestrogen plus progesterone. All trials, apart from five,w6 w14-w16 w27 were placebo controlled. Eleven trials, all small, did not record any stroke events.

Fig 1

Results of literature search for randomised controlled trials of hormone replacement therapy (HRT) that reported stroke events

• Download figure
• Open in new tab
• Download powerpoint

We excluded 12 trials (fig 1, table B on bmj.com), eight because they did not report vascular events, 30-w35 w37 w38 two because they did not distinguish between stroke and transient ischaemic attacks (total n = 685), 36 w39 and one because it did not have a control group. 40 The women's international study of long duration oestrogen after the menopause (WISDOM, n = 5664) was closed early after the release of data from the dual therapy arms of the women's health initiative trial 29; its data are yet to be published.

Data quality

Trials varied in their quality score11 from 2 to 5, median 5 (maximum score). All trials included were randomised, and 96% of trials gave adequate details of withdrawals.

Quantitative data synthesis

Stroke occurred in 2% of the participants randomised to no hormone replacement therapy, and this rate was significantly increased by a third (number needed to harm 147) in those randomised to hormone replacement therapy (table 1, fig 2). This increase in stroke resulted from an excess of ischaemic strokes but not primary intracerebral haemorrhage, as was seen in the women's health initiative dual trial. 26 An early increase in stroke occurred during the first six months of treatment in Viscoli et al's trial of secondary stroke prevention, 21 analogous to the early increase in coronary events seen in a trial of secondary prevention of coronary heart disease. 20 Both therapy arms of the women's health initiative trial (oestrogen alone and in combination with progesterone) were stopped prematurely because of therapy being associated with hazard. w26 w28

Fig 2

Effects of hormone replacement therapy (HRT) on stroke events

• Download figure
• Open in new tab
• Download powerpoint

A poor outcome after stroke, judged as combined death and dependency, was increased by half with hormone replacement therapy; we also found a non-significant increase in fatal stroke. This relation between hormone replacement therapy and severe stroke was present individually in three trials. 20 w21 w26 Hormone replacement therapy did not alter the rate of transient ischaemic attack (table 1). We found no statistical heterogeneity for any of the stroke outcomes.

Table 2 shows the results of our sensitivity analyses on several prognostic factors for the total stroke outcome. These results seem to be driven by the large women's health initiative trial, with significant effects being seen for the subgroups that included this study. However, we found no significant heterogeneity between trials examining primary versus secondary stroke prevention, oestrogen alone versus combination hormone replacement therapy, conjugated equine oestrogen versus estradiol, shorter versus longer follow up, smaller versus larger trials, those including exclusively men versus women alone, and high versus lower quality. We found no significant publication bias for the “all stroke” outcome (Eggers test P = 0.19).

Possible explanations for results

Why hormone replacement therapy should increase ischaemic stroke and its severity when biological plausibility and some previous observational studies suggested it might protect against cerebrovascular events remains unclear. Firstly, it is possible that the results of the studies we reviewed are wrong, but this is unlikely since in none of the studies did hormone replacement therapy reduce stroke—25 trials had neutral results, and three found that therapy increased stroke risk.

Secondly, the trials used either oestrogen alone or combination hormone replacement therapy. Although long term therapy with oestrogen alone can promote uterine cancer, this would not explain an increase in stroke. In contrast, adding a progestogen could have had detrimental effects since progestogens can promote atherogenesis and vasoconstriction.13 This is particularly relevant for medroxyprogesterone acetate, which was used in most of the trials of combination therapy. However, we found no heterogeneity between trials of oestrogen alone and combination therapy, suggesting that oestrogen itself, given as estradiol or conjugated equine oestrogen, might be the culprit.

Thirdly, within-class differences in hormone replacement therapy may mean that the most appropriate type of oestrogen has yet to be tested adequately; the trials assessed either conjugated equine oestrogens or estradiol but not other types such as phytoestrogen.14 However, there was no evidence for statistical heterogeneity between the trials with respect to type of oestrogen.

Fourthly, the dose of oestrogen (and progestogen if present) may have been too high. The usual starting doses of conjugated equine oestrogen and estradiol in Britain in postmenopausal women are 0.625 mg and 1 mg respectively, although the dose may be titrated up if menopausal vasomotor symptoms persist. These doses are below those used in several of the trials.

Fifthly, the delivery route may be important since substantial pharmacological differences exist between oral and transdermal administration of oestrogen, especially regarding hepatic first pass metabolism.

Lastly, several of the trials may have been too short, with a median length of less than three years, contrasting with the earlier observational studies. Of note, the trials by Simon et al and Viscoli et al showed an early vascular hazard that disappeared later. 20 w21 The hazard during the first year of treatment seems to reflect the development of a thrombophilic state that may not persist. This raises the possibility that an extended follow up would have revealed long term benefit. An analogous situation exists with statin treatment, whereby benefit was found in trials with longer rather than shorter follow up.15 Nevertheless, the women's health initiative trial, the largest of the trials we reviewed, had follow up for more than five years and yet found no beneficial effect on stroke risk in either of its therapy arms.

Conclusions

We have found that the use of hormone replacement therapy is associated with an increased risk of stroke, typically ischaemic in type and severe in nature. Hormone replacement therapy cannot be recommended for the primary or secondary prevention of stroke.