Abstract
This paper will review available data bearing on the relationship of post-menopausal hormone replacement therapy to the risk of first or recurrent ischaemic stroke. Although experimental data from both human and animal studies will be briefly mentioned, the bulk of evidence is from observational epidemiological studies. As such, the limitations of observational studies, particularly as applied to the health effects of post-menopausal hormone replacement therapy, will be emphasized. We conclude that there is no compelling consistent evidence that post-menopausal hormone replacement therapy either decreases or increases stroke risk. There are, however, reasons to be concerned that this therapy may contribute to stroke risk.
The effect of post-menopausal hormone replacement therapy (HRT) on the risk of vascular disease is of major public health importance. It has been estimated that a 50-year-old woman has a 20% lifetime probability of developing a stroke and an 8% lifetime probability of dying from a stroke (1). This article will focus on ischaemic stroke, the predominant form of cerebrovascular disease, rather than subarachnoid or intracerebral haemorrhage.
The paper is structured as a critical review rather than a meta-analysis for the following reasons. In a well-conducted randomized trial, bias is eliminated through the process of randomization. A properly executed meta-analysis of clinical trial data can further assist by reducing the role of chance. This is accomplished by increasing both the precision of the estimate and the power to detect an association. However, at present, no data from randomized clinical trials with stroke as an end-point are available, only data from observational studies. In observational studies, the critical limitation is the potential for bias, not lack of precision or power. This is particularly true for the issue of post-menopausal hormone replacement therapy, for reasons to be detailed later.
Although this review will focus on ischaemic stroke, this topic cannot be fully appreciated in isolation from evidence relating to coronary heart disease and venous thromboembolic disease. The effect of post-menopausal HRT on these other vascular end-points sheds light on its effect on atherogenesis and thrombosis, which is highly relevant to its potential influence on stroke risk. Similarly, observational evidence from human population studies will be supplemented by experimental data from animal models and from human studies with intermediate end-points.
It is important to bear in mind that many of the earlier studies of HRT considered oestrogen replacement therapy (ERT or unopposed oestrogen) rather than combined hormone replacement therapy (PERT or combined progestin and oestrogen) which is presently recommended to lessen the risk of uterine cancer.
Epidemiological studies of HRT and stroke risk
Earlier clinical trials in men using oestrogen for the treatment of prostate cancer (2) or for the prevention of cardiovascular disease (3) found an increased risk of thrombotic events. However, the doses of oestrogen used (≈ 2.50 mg to 10 mg/day) were relatively high compared to doses in common use today (0.625 mg/day) and more recent clinical trial data bearing on the risk/benefit ratio of HRT for vascular disease are lacking.
There are numerous observational epidemiological studies of the issue (see Paganini-Hill (4) for the most comprehensive recent review). This section will not discuss uncontrolled cohort studies (where all cohort members are taking HRT and the comparison is with general population statistics) because of potential bias due to an inappropriate comparison group. Case-control studies will be excluded for the same reason and, secondarily, because of the potential for inaccuracy or recall bias and limited accuracy in the recording of confounding factors. Only three cohort studies with internal controls having at least 200 stroke events will be discussed. As will be detailed in a later section, the main potential source of bias in epidemiological studies of HRT and stroke risk is selection bias, which continues to be an important potential bias even in cohort studies with internal controls. Selection bias refers to the fact that women who are prescribed and continue to take HRT are different in many known and, presumably, many unknown ways from women who do not take HRT.
Incident vascular disease/primary prevention
In a national sample of 1910 white post-menopausal women from the National Health and Nutrition Examination Survey (5) 250 incident stroke events occurred during an average of 12 years of follow-up. Since unopposed conjugated oestrogens were the primary form of HRT in the 1960s and 1970s (6), ’ever use’ of HRT ascertained at the baseline examination occurring between 1971 and 1975 would have been predominantly ERT. Subarachnoid haemorrhage and intracerebral haemorrhage were included; transient cerebral ischaemia was excluded. Decreased age, systolic blood pressure and body mass index, and increased length of school education, and household income adjusted for family size, were associated with HRT. Relative risks were adjusted for age and categorically defined length of school education, adjusted household income, smoking status, body mass index, systolic blood pressure, history of hypertension, diabetes mellitus and myocardial infarction.
The risk of stroke among women who had ever used HRT at baseline compared to the women who had never used HRT therapy was 0.7 (95% CI, 0.5–1.0). The risk of death from stroke among women who had used oestrogen therapy compared to those who had not was 0.4 (95% CI, 0.1–0.9). Of the 1910 participants, HRT status was determined using information from proxies in 436 and information from subjects only in 1474. When analyses were restricted to subjects’ responses only, the relative risk for all stroke was 0.8 (95% CI 0.5–1.5) and the mortality risk was 0.9 (0.3–3.7). These differences may be due to precision of estimates based on the smaller sample size or the inclusion of proxy information on HRT use may have biased the results.
The Copenhagen City Heart Study (7) estimated the effect of ever use of HRT in a cohort of 4717 post-menopausal women among whom 238 incidents of transient cerebral attacks (TIAs) and stroke events occurred during 12 years of follow-up. Ever use of HRT, ascertained at the baseline examination occurring between 1976 and 1978, would have been predominantly ERT. In the larger study, which included women as young as 35 years of age, there were 265 events composed of 76 ischaemic strokes, 11 intracerebral infarctions, 128 non-specified strokes, 48 TIAs. Subarachnoid haemorrhage was excluded from the stroke end-point. Age at menopause was not associated with risk of cerebrovascular disease.
Age, length of school education, household income, and leisure time physical activity, were associated with both HRT and stroke risk and were thus considered confounders. Relative risks were adjusted for age, length of school education, household income, physical inactivity at leisure time, daily alcohol intake, daily consumption of alcohol, daily consumption of tranquilizers, and body mass index. No adjustment was performed for hypertension or level of blood pressure. There was a significant interaction (
The Nurses Health Study estimated the effect of current use or past use of ERT and PERT on stroke in a cohort of 59 337 women who were 30–55 years at baseline (8, 9). Information on hormone use was obtained in 1976 and every 2 years through 1990 and stroke events were ascertained until 1992. There were 572 strokes composed of 285 ischaemic strokes, 155 subarachnoid haemorrhages, and 132 other or unspecified types. A person-year approach was taken and use of ERT accounted for 18.8% of the person-years while use of PERT accounted for 6.2 of the person-years. Current users of HRT generally had a better cardiovascular risk profile and were less likely to have a parental history of premature myocardial infarction, history of diabetes mellitus or use cigarettes, and were more likely to use multivitamins, vitamin E or aspirin than never users. In contrast, they were more likely to have a high cholesterol level.
Analyses were adjusted for categorically defined age, time, age at menopause, type of menopause, past oral contraceptive use, body-mass index, diabetes, high blood pressure, high cholesterol level, smoking status and amount smoked per day, parental history of premature myocardial infarction. Furthermore, the fact that the study cohort is relatively homogeneous in terms of education and occupation is advantageous because it tends to reduce unmeasured variation in health habits that could aggregate with HRT. Due to sample size, analyses were stratified by ERT/HRT use or all-type stroke/ischaemic stroke, but not by both factors.
For all-type stroke, current use of ERT was associated with a relative risk of 1.3 (95% CI, 0.95–1.7), while current use of PERT was associated with a risk of 1.1 (95% CI, 0.7–1.8).). There was a monotonic trend (
That current use of HRT was associated with a significantly increased risk of ischaemic stroke in the Nurses Health Study is noteworthy because, in the same cohort, current use of HRT was strongly associated with a decreased risk for major coronary disease (9). For coronary heart disease, ERT was associated with a relative risk of 0.6 (95% CI, 0.4–0.8) while PERT was associated with a relative risk of 0.4 (95% CI, 0.2–0.8). Since it is likely that the same selection biases related to HRT use would be operating for both the stroke and coronary heart disease end-point in the same cohort, these data strongly suggest a different biological effect of HRT on each end-point. This point of view is also supported by the aggregate of observational epidemiological data on the association between HRT and both coronary disease and stroke (4, 9).
The overwhelming majority of studies have shown that HRT is associated with a reduced risk of first myocardial infarction, while the results for incident stroke have been more mixed, with recent large case-control studies showing no significant effect of ERT or PERT on ischaemic stroke risk (Pedersen 1997, Pettiti 1998). A different biological effect of HRT on different clinical end-points is also supported by the consistent data from six recent studies, including the Nurses Health Study, showing that HRT is associated with an increased risk for deep venous thrombosis (12, 16). These findings are also consistent with experimental data suggesting that HRT may retard atherogenesis, while exerting a prothrombotic effect.
Recurrent vascular disease/secondary prevention
There are no data so far concerning the effects of HRT in women who have already had a stroke, either in their youth or after menopause. Such data now exist for coronary heart disease with the publication of the HERS (Heart and Estrogen/Progestin Replacement Study) (17). HERS is a randomized, double blinded, placebo-controlled trial of continuous PERT in the secondary prevention of coronary heart disease (CHD); 2763 post-menopausal women, aged less than 80, with CHD, received conjugated equine oestrogen (0.625 mg) and medroxyprogesterone acetate (2.5 mg) in one tablet daily (
These clinical trial results for secondary prevention were unexpected since the weight of both observational studies and animal experiments strongly suggested that ERT and PERT would be protective. One possible explanation is that the effect of selection bias on observational studies has been underestimated. Alternatively, as proposed by the HERS authors, the early increase in CHD events may be a consequence of an immediate prothrombotic effect of oestrogen treatment which is gradually outweighed by a beneficial effect on the underlying progression of atherosclerosis, perhaps as a result of beneficial changes in lipoproteins. This would imply that the benefit of PERT on CHD would increase with longer periods of follow-up. Since thrombotic mechanisms probably play a larger role in stroke compared to CHD, it would also suggest that there may be less benefit of PERT for the prevention of stroke than CHD
Atherogenesis
Properly conducted an interpreted animal model can shed light into the potential mechanisms of human disease. In the case of HRT, they offer the potential of avoiding selection and compliance bias as well as better control of other confounding factors such as diet. Randomized experiments in an ovariectomized monkey model (18) show that both ERT and PERT with cyclically administered progestin were effective in the primary prevention of coronary atherosclerosis but had no effect on established plaque. Favourable changes in plasma lipoprotein profiles accounted for only 20% of the beneficial effect on coronary artery atherosclerosis; major effects were attributed to a diminished rate of LDL accumulation in coronary arteries which was independent of lipid levels. More recent data from the same research group suggest that PERT with continuously administered progestin antagonizes the atheroprotective effect of unopposed oestrogen (19).
Observational epidemiological studies of carotid artery intimal-medial thickness or stenosis suggest a similar effect but are limited by selection bias. Cross-sectional analyses of data from the Cardiovascular Health Study (20) showed that use of HRT (predominantly ERT or unopposed oestrogen) was associated with a lesser extent of intimal-medial thickness and a lower frequency of haemodynamically significant carotid stenosis. The lesser degree of intimal-medial thickness remained after adjustment for multiple other vascular risk factors, but was moderated by additionally adjusting for HDL and LDL cholesterol, suggesting that a beneficial effect of HRT on lipids may have partly mediated the effect on the carotid artery.
The Asymptomatic Carotid Atherosclerotic Progression Study (ACAPS), a factorial randomized clinical trial of Lovastatin and warfarin among persons with increased intimal-medial thickness and elevated LDL-cholesterol, provided an opportunity to observe the association of HRT with progression of intimal-medial thickness (21). No difference between users and non-users of HRT was found in the groups randomized to Lovastatin or warfarin. However, in the placebo group, intimal-medial thickness increased significantly more in the 64 non-users of HRT than the 27 HRT users, an effect that was independent of differences in baseline lipoprotein levels. Again, it should be emphasized that HRT was not randomized in this study.
The PEPI Trial randomized 875 healthy post-menopausal women to either placebo, unopposed oestrogen, and each of three oestrogen/progestin regimens. HDL cholesterol levels rose significantly more in women assigned to unopposed oestrogen than to oestrogen plus cyclic or continuous progestins (22).
Thrombosis and haemostasis
Because of the dynamic nature of thrombosis and haemostasis mechanisms, it is much more difficult to develop experimental animal models that can be confidently generalized to human disease mechanisms. For example, in post-menopausal monkeys fed an atherogenic diet, HRT increased the incidence of thrombosis after a standardized injury to the carotid artery from 40% to 54%. Given the small sample size of the study, this increase was not statistically significant and the authors conclude that the results provide some reassurance that current HRT regimens do not adversely affect the risk of arterial thrombosis in women (23).
It is also difficult to generalize to clinical end-points using even data from human randomized studies. In the PEPI trial (22), placebo was associated with a significantly greater increase in mean fibrinogen than any active treatment (0.10 g/l compared with −0.02–0.06 g/l); differences among active treatments were not significant. These differences were small on an absolute scale There is some evidence to suggest that effects on blood coagulation and fibrinolysis may depend on the route of administration, with transdermal administration having a more favourable effect (24). Other sections of this supplement will address these issues in more detail.
Selection bias
There are several reasons to be concerned that selection factors may be playing an important role in observational epidemiological studies of HRT and stroke. First, there is ample evidence that users and non-users of HRT differ in education, socioeconomic status and risk factor profile (4). Furthermore, among women with an intact uterus who are started on ERT or PERT, a substantial percentage (55% for ERT) do not continue to take this medication, thus raising the further potential for selection bias (22). Second, epidemiological studies show an almost uniform beneficial association with all health end-points. Compared to Swedish women in general, a 23346-member, population-based cohort of Swedish women who were prescribed HRT for an average of 8.6 years had a death rate which was 12% to 86% lower than expected in 11 of 12 major mortality categories (25). Similarly, prospective studies of HRT showing the largest reductions in cardiovascular disease also tended to show the largest reductions in cancer risk. This is not biologically plausible and suggests a healthy cohort effect (26). Third, in contrast to expectations based on our understanding of biological mechanisms, observational studies tend to show a greater benefit for PERT than for ERT. For example, in the Swedish cohort described above (25), compared to population data, the relative risk for ischaemic stroke was estimated to be 0.78 for ERT users and 0.57 for PERT users. Similarly, in the Nurses Health Study, compared to never use, the relative risks for ERT/PERT, respectively, were 0.6/.4 for coronary heart disease and 1.3/1.1 for stroke.
Thus, there is overwhelming evidence that selection bias plays an important role in observational studies of HRT and disease end-points. The only controversial issue is the magnitude of this bias and whether statistical adjustment is adequate to achieve an unbiased estimate of the effect of HRT. Adequate adjustment is difficult for a variety of reasons. First, known important confounding risk factors may not be measured in observational studies. Second, it has recently been demonstrated that pre-menopausal women who subsequently elected to use ERT during menopause had a better cardiovascular risk factor profile prior to use than did non-users (27). Adjustment for these differences is problematic because few observational studies have measures of the biological attributes of women prior to the use of HRT (28). Third, the use of categorized risk factor data also can introduce bias due to inadequate adjustment. Fourth, due to measurement error (29), it is quite difficult to adequately adjust for important differences in many risk factors, such as blood pressure, even if measured on a continuous scale at the appropriate time. Fifth, even if measurement on a continuous scale were perfect for known risk factors, it is highly likely that there are many other differences between users and non-users which are unknown. It is for this reason that a very high degree of emphasis was placed on the data from the Nurses Health Study. Contrasting results for the ischaemic stroke and venous thromboembolic disease compared to myocardial infarction in the same study is unlikely to be due to selection bias and probably suggests important biological differences in the effect of HRT on these end-points.
Summary and conclusions
There is no compelling consistent evidence at this time that ERT or PERT either increases or decreases stroke risk. However, there are reasons to be concerned that HRT may contribute to stroke risk. All observational studies are likely to be biased to an unknown degree in favour of better health outcomes for HRT users. Despite this bias and in contrast to their findings for coronary artery disease, the Nurses Health Study (8) showed a significant increase in ischaemic stroke associated with HRT use and a graded dose-response between oestrogen dose and stroke risk. The consistent evidence that HRT increases risk of deep venous thrombosis supports the possibility of a prothrombotic diathesis associated with HRT use.
New data from randomized clinical trials will be needed to resolve this uncertainty. The Women's Health Initiative (30), with an expected completion date of 2007, is a large primary prevention trial which is randomizing 27000 women, ages 50–79, to the HRT arm with follow-up for an average of 9 years. Active treatment is unopposed oestrogen for women without a uterus and oestrogen with continuously administered progestin for women with an intact uterus. The Women's Oestrogen for Stroke Trial (31), with a similar planned completion date, is a study of unopposed oestrogen for the secondary prevention of stroke. Until these studies are completed, neither primary nor secondary prevention of stroke can be considered an indication for HRT. There is, however, no reason to contraindicate hormone replacement therapy in migrainous women since migraine has not been shown to be a risk factor for stroke in post-menopausal women (32) and there are no data on the relationship between hormone replacement therapy, migraine and stroke.
Footnotes
Acknowledgements%
Dr Kittner was supported by the Geriatric Research Education and Clinical Center (GRECC) at the Baltimore Veterans Administration Medical Center.