Increased Risk of Stroke in Hypertensive Women Using Hormone Therapy: Title and subTitle BreakAnalyses Based on the Danish Nurse Study FREE

STROKE AFFECTS 1 per 1000 women per year aged 45 to 65 years.¹ Postmenopausal use of female steroid sex hormones, so-called hormone therapy (HT), may potentially affect the development of stroke; however, results from observational studies have provided conflicting results.² This may be explained by different HT regimens and different hormone compounds. Furthermore, strokes may be classified as fatal or nonfatal and as ischemic or hemorrhagic, yet studies lack consistency in stroke definition. In addition, HT users have been reported to be healthier than nonusers concerning cardiovascular risk factors,^{3 - 4} but definition of a "healthy user" effect may vary between studies and introduce confounding or selection bias in observational epidemiological studies. Finally, the mixed results may also stem from interactions between risk factors for stroke and HT, ie, different effects of HT among different subgroups of women.

Recently, the results from the Women's Health Initiative (WHI) trial showed an increased risk of stroke, especially nonfatal, in women treated with combined conjugated estrogen and medroxyprogesterone acetate.⁵ However, the risk associated with the traditional Scandinavian HT regimen based on estradiol-17β in combination with 19-norethisterone derivatives has not been examined.

The purpose of the present study was therefore to examine associations between the Scandinavian tradition of HT use and stroke in general, including total strokes, ischemic or hemorrhagic strokes, and fatal or nonfatal strokes, with a focus on effect modification by risk factors for stroke.

This study is based on data from the Danish Nurse Study on prevention of osteoporosis and atherosclerosis. In 1993, all female members of the Danish Nurses' Association older than 44 years (n = 23 178) received a comprehensive questionnaire concerning health, lifestyle, and reproductive conditions, including detailed questions on HT exposure. Of these, 19 898 women (85.8%) answered and returned the questionnaire.⁶

The study was approved by the scientific and ethical committees for Copenhagen and Frederiksberg, Denmark. The Danish Data Protection Agency was notified of the study, and the Danish National Board of Health gave permission for access to the national registries of hospital discharges and cause of deaths.

Premenopausal women (n = 5182) were excluded from the study population. A woman was defined as premenopausal if she reported having monthly bleeding without current HT use. Based on registry or self-reported information, an additional 214 women with previous myocardial infarction, 108 with previous stroke, and 43 with previous transient ischemic attack were excluded. Furthermore, 1529 with self-reported cancer and 285 with missing information on HT were excluded from the study base. Some women had more than 1 reason for exclusion. Included in the final analysis were 13 122 healthy postmenopausal women.

Information on HT exposure was self-reported and classified as current, past, or never use. For some analyses, current and past users were combined as ever users. Current users were further subdivided based on whether the type of regimen was unopposed estrogen or estrogen-progestin therapy.

Information on covariates was also obtained from the questionnaire. Age was used continuously, and other covariates were included in the analyses as categorical variables. These comprised smoking history (never, former, or current), alcohol consumption (0, 1-14, or >14 U/wk), body mass index calculated as weight in kilograms divided by the square of height in meters (<18.5, 18.5-25, >25-30, or >30), leisure time physical activity (>4 h/wk hard, >4 h/wk moderate, or sedentary), hypertension or hypertension-lowering medication (yes or no), angina medication (yes or no), diabetes mellitus (yes or no), and thyroid disease (yes or no).

Information on "first ever stroke" was retrieved from the national patient registry of hospital discharges, which through the personal identification number registers all hospital admissions in Denmark, or from the cause of deaths registry, which registers all causes of deaths in Denmark. A linkage to the central person registry, which registers all death dates, was also performed. These registries are based on International Classification of Diseases, Eighth Revision (ICD-8), Ninth Revision (ICD-9), and, since 1994, Tenth Revision (ICD-10) codes. Stroke cases were defined as ICD-8 codes 430 to 434 or 436 and as ICD-10 codes I60 to I65. For analyses estimating the risk of ischemic stroke, ICD-8 codes 432 to 434 or 436 and ICD-10 codes I63 to I65 were used, whereas ICD-8 codes 430 to 431 and ICD-10 codes I60 to I62 were used for hemorrhagic stroke.

Strokes were subdivided into fatal and nonfatal events. Fatal event information was obtained from the cause of deaths registry. If death occurred within 28 days after a nonfatal stroke diagnosis, as indicated by linkage to the central person registry, it was also considered a fatal event. A patient having a nonfatal stroke and later in the observation period having a fatal stroke was considered in the nonfatal and fatal stroke analyses, but with 2 different event dates.

The follow-up for fatal and nonfatal events was continued until December 31, 1998.

The Cox proportional hazards model for left-truncated and right-censored data was used in the modeling of the time-to-stroke outcomes.⁷ The nurse's age at study entry was considered the delayed entry variable in the analysis. The Cox proportional hazards model was also used to model other outcomes with each of the HT exposure variables. For every Cox model, the proportional hazards assumption was checked. The first step in the analysis was modeling the outcome of interest univariately with the HT exposure variable (unadjusted for confounders except for age, the delayed entry variable) and estimating the hazard ratios (HRs) and their 95% confidence intervals (CIs). The second step in the analysis was multivariate modeling of the stroke outcome, estimating HT exposure variable HRs with 95% CIs with HT exposure adjusted for the confounders (smoking history, alcohol consumption, body mass index, physical activity, hypertension, angina, diabetes mellitus, and metabolic disease). Stepwise selection was made for each outcome to identify the significant confounders. The HT variable was kept in the model as the main variable of interest. Hazard ratios with 95% CIs for HT exposure were assessed. The third step in the analysis was the effect modification analysis, in which each of the significant covariates in the models was tested for interaction with the HT variable. For significant interactions, the HRs with 95% CIs were presented stratified on the effect modifier, as well as estimates of significant covariates in the models. Missing values were excluded from analysis. The analysis was performed in Stata version 7.0 (Stata Corporation, College Station, Tex).

Among the 13 122 postmenopausal women, 28.0% were currently using HT, 14.3% were past users, and the remaining 57.7% never used HT (Table 1). The median duration of HT among current users was 6 years (range, 0-43 years). The HT regimen in 35.5% of the women was unopposed estrogen, and 59.2% used combined therapy. The remaining 5.3% had missing information on the HT regimen. The predominant compounds were estradiol-17β and norethisterone acetate. Only 0.4% used conjugated equine estrogens. For past users of HT, the median HT duration was 2 years (range, 0-40 years). Hormone therapy users (current or past) differed significantly from never users in baseline characteristics, as they smoked more, consumed more alcohol, had lower body mass index, and had less diabetes mellitus (Table 1).

From 1993 until the end of 1998, there were 144 incident stroke cases. Of these, 39 were immediately fatal or fatal within 28 days. Another 7 women with a nonfatal stroke initially had a fatal stroke more than 28 days later, so 46 fatal strokes and 105 nonfatal strokes were registered during the observation period.

Of the 144 strokes, 99 were classified as ischemic strokes. Of these, 77 were nonfatal. However, 4 women later had a fatal stroke, so 26 strokes were considered fatal ischemic strokes. Among 45 strokes classified as hemorrhagic, 28 were nonfatal, of whom 3 women later had a fatal stroke; consequently, there were 20 fatal hemorrhagic strokes.

Overall, ever or current HT exposure was only weakly associated with the various subtypes of stroke, except for nonfatal stroke. However, current use of estrogen-progestin treatment was associated with an increased risk for the various subtypes of stroke (Table 2 and Table 3). Further analyses revealed that this was based on a consistent interaction between HT and hypertension, implying that current use of HT among hypertensive women was associated with a significantly increased risk of stroke (Table 4). Compared with hypertensive never users, the risks were highest with use of estrogen-progestin therapy, being most pronounced for ischemic stroke in current users of combined therapy (HR, 5.01; 95% CI, 2.09-12.00). Normotensive women were not at increased risk for stroke, with or without HT (Figure 1). No other interactions were found with the other cardiovascular risk factors.

No increased HR for hemorrhagic stroke was found among ever users of HT compared with never users (HR, 0.56; 95% CI, 0.29-1.11).

Overall, we found no association between HT and risk of stroke. However, use of HT (particularly estrogen-progestin therapy) was associated with an increased risk of subtypes of stroke, including the nonfatal stroke and the ischemic stroke, especially the nonfatal. When assessing whether the presence of risk factors for stroke modified the associations, we found a consistent significantly increased risk of total stroke and various subtypes of stroke associated with use of HT among hypertensive nurses. This risk was most pronounced among nurses using estrogen-progestin therapy. Normotensive nurses using HT had no increased risk of stroke.

Most observational investigations have considered the risk of total stroke or ischemic stroke only, and some have distinguished between ischemic and hemorrhagic events.¹ One is Falkeborn et al,⁸ who in an uncontrolled cohort found that the risk of ischemic stroke and subarachnoid or intracerebral hemorrhage was unassociated with HT use. In addition, a Danish case-control study⁹ found no increased risk of thromboembolic stroke and intracerebral or subarachnoidal hemorrhage with unopposed or estrogen-progestin therapy, after adjustment for confounding factors. Our findings of an increased risk of ischemic stroke but not total stroke among nurses using combined HT agree with results from the American Nurses' Health Study,¹⁰ although subanalyses among hypertensive women were not reported.

Our findings are also in agreement with the results from the recently published WHI trial,⁵ in which a significantly increased risk of nonfatal but not fatal stroke was found in the intervention group compared with control subjects. The WHI trial, however, also found an increased risk of total stroke, possibly related to the type of HT used: conjugated estrogens in combination with medroxyprogesterone acetate in the WHI trial vs estradiol-17β in combination with norethisterone acetate or levonorgestrel in the present study. Compared with our Danish nurses, women participating in the WHI trial were generally older (63.3 vs 58.8 years), more overweight (22 vs 13 had body mass index >25), and more frequently diabetic and hypertensive; only 10% of the WHI women smoked compared with 40% among the Danish nurses. The WHI trial reported "no noteworthy" interaction of HT with blood pressure. To our knowledge, no others have tested for interactions with hypertension.

The present findings indicate that hypertensive women are at an increased risk of stroke with use of oral contraceptives¹¹ and HT. Although the pathophysiologic mechanisms for an increased risk of stroke among hypertensive women using HT remain speculative, it is possible that hypertension induces endothelial dysfunction and that HT does not have a protective effect when the endothelium is damaged.¹² Hypertension also seems to exacerbate atherosclerosis,¹³ and HT might not have beneficial effects in the final stages of atherosclerosis.¹⁴

The present study is based on a large prospective cohort of postmenopausal nurses with a good response rate at baseline (85.8%) and a complete follow-up for fatal and nonfatal events through individual linkage to national registries. However, the self-reported use of HT may lead to bias. Therefore, the information on HT use was previously validated by comparing information from pharmacoepidemiological registries for approximately 15% of the nurses living in 2 Danish counties.¹⁵ This comparison showed a sensitivity of 78.4% and a specificity of 98.4% for HT exposure status, suggesting that the never HT users had a higher chance of being correctly classified than the current HT users. A nondifferential misclassification like this would tend to bias the risk estimate toward unity. Also validated was the self-reported type of HT in 1993, categorized as unopposed estrogen vs estrogen-progestin, with a sensitivity of 93.0% and a specificity of 86.3%.¹⁵ This allows us to believe that there is little chance of misclassification in the recording of the type of exposure. Finally, even though the questionnaire accurately captures information on HT, changes in HT status over time may provide another source of misclassification, as HT was recorded at baseline only. Such a misclassification would have resulted in conservative estimates too.

It may be argued that another potential weakness of our study is that information on hypertension is based on self-reports. However, nurses are likely to have better qualifications for answering questions about medication and illness compared with the general population. In addition, they are more likely to be diagnosed with hypertensive disorders because of the easy access to blood pressure monitoring. Finally, the fact that we were able to detect a highly significant interaction using this crude measure of self-reported hypertension suggests that our findings are conservative and that stronger associations would have resulted if hypertension had been more accurately measured.

Stroke diagnoses from registries were also validated with hospital case reports, which verified 89% of the ischemic strokes and 88% of the hemorrhagic strokes.⁹ This validation was based on ICD-8 coding, with a change to ICD-10 in 1994. Johnsen et al¹⁶ recently validated the ICD-10 system and found a predictive value of 87.6% for ischemic stroke but only 65.7% for hemorrhagic stroke in the Danish national patient registry compared with hospital case reports. These findings suggest that the results from the subgroup analyses on ischemic strokes may be the most valid.

In a large prospective cohort study, we found an increased risk of stroke among hypertensive but not normotensive women using HT, particularly among those using estrogen-progestin therapy. The results need further confirmation but may potentially have great clinical relevance in guiding practitioners when prescribing HT. The present study would suggest that HT should be avoided in hypertensive women.

Corresponding author: Ellen Løkkegaard, MD, PhD, Centre for Alcohol Research, National Institute of Public Health, Svanemøllevej 25, 2100 Copenhagen, Denmark (e-mail: loekkegaard@dadlnet.dk).

Accepted for publication February 5, 2003.

Author contributions: Study concept and design (Drs Løkkegaard, Ottesen, and Pedersen); acquisition of data (Drs Løkkegaard, Hundrup, Obel, and Pedersen); analysis and interpretation of data (Drs Løkkegaard, Heitmann, and Pedersen, Ms Jovanovic, and Mr Keiding); drafting of the manuscript (Drs Løkkegaard, Heitmann, and Pedersen); critical revision of the manuscript for important intellectual content (Drs Løkkegaard, Heitmann, Ottesen, and Pedersen, Ms Jovanovic, and Mr Keiding); statistical expertise (Drs Løkkegaard and Heitmann, Ms Jovanovic, and Mr Keiding ); obtained funding (Drs Løkkegaard, Ottesen, Hundrup, and Obel); administrative, technical, and material support (Drs Løkkegaard, Ottesen, and Pedersen); study supervision (Drs Heitmann, Ottesen, and Pedersen).

This study was funded by grants 99-1-F-22720, 99-2-F-22775, and 01-1-9-F9-22882 from the Danish Heart Foundation, Hauser Plads, Copenhagen. The study was part of the PhD project for Dr Løkkegaard.

The sponsors of this study were not involved in the study design, data collection, analyses, interpretation, or writing of the report.