Methods

Top Methods Results Discussion Author & Article Info References

As previously described [3], we recruited participants for a population-based case–control study of risk factors for subarachnoid hemorrhage in King County, Washington, between 1 July 1987 and 30 June 1989. The Human Subjects Committee at the University of Washington approved the study.

Study Population

Eligible case-patients had to be 18 years of age or older and to have had a spontaneous subarachnoid hemorrhage. Patients whose bleedings were found to originate from a source other than an intracranial aneurysm—including primary intraparenchymal hemorrhages, arteriovenous malformations, trauma, and neoplasms—were excluded, but those in whom an aneurysm could not be identified were included. Because of the eligibility criteria for controls (described below), patients who did not have a telephone or who did not speak English were excluded.

Case-patients were identified through a surveillance system that relied most heavily on the physicians and hospitals but also included the emergency medical systems serving the area and the King County medical examiner. Eligible patients were invited by their treating physician to participate in the study. Whenever possible, this invitation was directed to the patient, but because of death or disability, it often had to be directed to a family member or friend who was asked to act as the patient's surrogate or proxy.

We identified controls using random-digit telephone dialing [7, 8]. They were matched to each case-patient by sex and age within 5 years. Once an appropriately matched control was identified, he or she was invited to participate in the study and to identify a proxy who would also be willing to participate. Two matched controls and their proxy respondents were identified for each case-patient. Potential controls who did not speak English were excluded.

Data Collection

Information on previous exposures was collected during a structured in-person interview. The interview was usually conducted in the participant's home and was completed in about 1 hour. Some of the questions were linked to a reference time. For the case-patients, the reference was when the initial subarachnoid hemorrhage occurred. For the controls, the reference was the same time of day and day of week as that of their matched case-patient but in the week before the interview of the control.

As described previously [3], participants were questioned about their lifetime use of cigarettes, about their use of alcohol and stimulant drugs in the year before the reference date, and whether a physician had ever told them that they had hypertension. Detailed information was also collected from women on their lifetime use of medications containing estrogens and progestins, time from their last menstruation to the reference date, and weight and height. A personalized calendar was used to record major life events to assist recall. Photographs of all oral contraceptives and other hormone preparations ever marketed in the United States were also shown to all respondents. Whenever possible, the case-patient was interviewed, but the proxy for the case-patient, the two matched controls, and the proxies for the controls were always interviewed.

Data Analyses

Analyses in this report were limited to women who had had a subarachnoid hemorrhage and their age- and sex-matched controls. When the case-patient was able to provide information on exposure, interviews obtained directly with the two matched controls were used for comparison. When this was not possible, the proxy for the case-patient and the proxies for the two matched controls were used. The effects of exposures were assessed using conditional logistic regression [9], which takes into account the matching on age and respondent type (index compared with proxy); odds ratios and their 95% confidence intervals were also derived. Potential effect-modifying or confounding variables were evaluated using conditional logistic regression. Effect modification was assessed by examining stratum-specific odds ratios and the statistical significance of interaction terms. The extent of confounding was assessed by comparing the crude odds ratio derived from a model that did not contain the potential confounding variable with the adjusted odds ratio derived from the model that did contain it.

The effect of using information provided by proxy respondents for the primary exposures of interest was examined by computing odds ratios separately for the two strata: self-reported data and proxy-derived data. In accordance with standard methods for analyzing mixed index-proxy data sets [10], the difference in odds ratios between the two respondent strata was tested by including an interaction term between respondent type and the exposure. In none of the resulting models assessing main exposures were interaction terms for respondent type statistically different.

Finally, the reliability of proxy information was evaluated using the degree of agreement between controls and their proxy respondents. These analyses excluded the 15 pairs of controls and their proxies who were interviewed in each others' presence. The statistic and intraclass correlation coefficient were used as measures of reliability [11, 12].

Results

Top Methods Results Discussion Author & Article Info References

Study Participants

During the 2-year study, 171 incident cases of subarachnoid hemorrhage were identified in King County. The annual incidence per 100 000 was 12.2 new cases in women per year, 6.5 new cases in men, and 9.4 new cases overall [3]. Detailed information on the evaluation, treatment, and outcome of these patients is described elsewhere [13]. One hundred forty-nine case-patients were enrolled after 11 were excluded and 11 refused to participate. Of the 466 eligible controls, 313 (67.1%) agreed to participate in the study and identified a willing proxy. Detailed evaluation of cigarette smoking, use of alcohol, and history of hypertension in the entire group is presented elsewhere [3]. In these analyses, sex was not an effect modifier. This report concentrates on the 103 female case-patients, who represent 69% of all 149 case-patients. Forty-eight case-patients themselves could be interviewed; the proxies were interviewed for the remaining 55 case-patients because 38 case-patients had died and 17 had become disabled.

The mean ages of the 103 female case-patients and their 206 age- and sex-matched controls were 57.6 ±17.4 years and 57.4 ±16.9 years, respectively. Case-patients and controls were predominantly white and were similar with respect to marital status. Case-patients had received less formal education than had controls: 12.6 ±2.4 years compared with 13.5 ±2.5 years, respectively.

Risk Factors

Even after conditional logistic regression was used to control for potential residual confounding by age, premenopausal women were found to be at a lower risk for subarachnoid hemorrhage than postmenopausal women (odds ratio, 0.24; 95% CI, 0.09 to 0.68) (Table 1). Women who formerly or, even more so, currently received postmenopausal hormone replacement therapy were at a lower risk for subarachnoid hemorrhage than were women who had never received therapy (Table 1). Longer duration of use and more recent use were associated with lower risk (for trend, P < 0.002 for both).

The use of oral contraceptives was also associated with a reduced risk for subarachnoid hemorrhage that was not statistically different (Table 1). Case-patients who had never used oral contraceptives and those who formerly used them were combined as the reference group; in women who currently received oral contraceptives, the odds ratio was 1.5 (CI, 0.4 to 5.6). Findings were similar for oral contraceptive use between 1960 and 1970, when estrogen doses were higher, and oral contraceptive use between 1971 and 1990; the reduced risk was not significantly related to duration of use or time since last use (data not shown).

Body mass index was calculated as weight in kilograms divided by height in meters squared. As shown in Table 1, the relation between body mass index and the risk for subarachnoid hemorrhage was U-shaped but was not significantly different as described previously [14, 15].

Hypertension, smoking, and age did significantly modify the effect of some hormonal factors (Table 2). Being premenopausal had a stronger negative association among women without a history of hypertension than among those with a history of hypertension. The reduced risk associated with being premenopausal was also much greater among those who had never smoked than those who had smoked. Interestingly, the reduced risk associated with hormone replacement therapy was evident only among women who had ever smoked, not among those who had never smoked, and was greatest in women aged 45 to 64 years compared with those younger or older.

Age, race, education, cigarette smoking, alcohol use, and hypertensive status were not found to be important confounding variables for the other hormonal factors. In addition, the effects of hormone replacement therapy and oral contraceptive use were not substantially altered by controlling for menopausal status.

Twenty-eight women had a subarachnoid hemorrhage before menopause. The time of the last menstrual period was known in 23 of these patients. The times of the subarachnoid hemorrhages were not evenly distributed throughout the menstrual cycles. Most of the subarachnoid hemorrhages (74%) occurred 21 days or more after the start of the last menstrual period or during menstruation. Assuming an average menstrual cycle of 28 days, a menstrual period of 4 days, and an even distribution of subarachnoid hemorrhages throughout the cycle, 43% of hemorrhages would have been expected rather than 74% (difference, 31%; CI, 4% to 58%).

Reliability

By interviewing index participants and their proxy respondents, we were able to assess the agreement among the respondents for the exposures being studied. Some results of the reliability analyses for the controls are presented in Table 3. In general, agreement was high. The reliability of the data collected in this study is detailed in another report [16].

Proxy respondents were unable to provide information on specific preparations used for 82% of the instances in which oral contraceptive use was mentioned and for 67% of the instances in which hormone replacement therapy was mentioned. Forty-two percent of the index respondents could not provide information about oral contraceptive use, and 24% could not do so about hormone replacement therapy. This degree of missing information precluded analyses focusing on specific types or doses of hormone preparations.

Discussion

Top Methods Results Discussion Author & Article Info References

In our population-based case–control study, we found that several hormonal factors in women altered the risk for subarachnoid hemorrhage. Factors leading to a state of relative estrogen deficiency, such as menopause, the immediate perimenstrual period, and cigarette smoking [17], were associated with an increased risk. Factors leading to maintenance of estrogen levels, such as the use of postmenopausal hormone replacement therapy, were associated with a decreased risk.

One of the strongest associations was for menopausal status. After controlling for differences in age, we found that women still menstruating were at significantly reduced risk for subarachnoid hemorrhage compared with those who had ceased menstruating. Premenopausal women who had never smoked were at the most reduced risk. For patients who were still menstruating, the subarachnoid hemorrhage seemed to occur late in the menstrual cycle or during menstruation. Heyman and associates [18] also found an uneven distribution of bleeding during the menstrual cycle, with most of the bleeding occurring soon after the start of a menstrual period. The perimenstrual period of the menstrual cycle is characterized by many hormonal changes, including a decrease in estrogen and progestin levels immediately before menstruation [19]. The effect of those changes on the endometrium is to promote bleeding; perhaps a similar effect could occur with cerebral vessels.

Stober and colleagues [20] observed that subarachnoid hemorrhage was common in women aged 50 to 59 years and proposed that decreased estrogen levels after menopause might contribute to aneurysm formation [20]. Results from epidemiologic studies of hormone replacement therapy and subarachnoid hemorrhage have been inconclusive [21-24]. One study supported a reduced risk in women who currently receive hormone replacement therapy but not in those who formerly received therapy [23]. In our study, patients' risk for bleeding was significantly reduced with the use of hormone replacement therapy. The existence of a strong dose response for both the duration and recency of use support a causal association with subarachnoid hemorrhage. Hormone replacement therapy was most protective in women who were current or former smokers.

Previous studies have suggested that oral contraceptives increase the risk for subarachnoid hemorrhage, with an estimated relative risk of approximately 2.0 for women currently receiving them and those who formerly received them [25, 26]. More recent studies, including ours, have not found an increased risk in these women [27, 28]. Use of older preparations with larger amounts of estrogens [29, 30] may have led to increased blood pressure [26, 31, 32] and increased risk for subarachnoid hemorrhage more so than do modern preparations. More information on the risk of current use should be forthcoming from ongoing studies [33].

By what mechanism might sex hormones, especially estrogens, affect the risk for subarachnoid hemorrhage? Handa and associates [34] describe three factors important in the formation of cerebral aneurysms in experimental animals: altered hemodynamic stress in the circle of Willis, hypertension, and increased vessel fragility. Estrogens could potentially affect all of these factors to influence the risk for aneurysm formation and rupture. Estrogens have been found to improve lipid profiles [35, 36] and thus may reduce the risk for atherosclerosis. This has been used to explain the reduced risk for all atherosclerotic cardiovascular diseases in women who receive hormone replacement therapy [26]. Atherosclerosis has been considered a risk factor for aneurysm formation because of altered hemodynamics. In addition, low-dose estrogens are associated with a reduced blood pressure, as are hormone replacement therapy [37] and perhaps more recent formulations of oral contraceptives. A decreased blood pressure would be associated with a reduced risk for aneurysm formation and rupture.

Finally, estrogens may have a direct effect on vessel fragility. In other tissues, such as the skin, estrogen deficiency can lead to wasting of collagen and elastin [38, 39]. Perhaps a similar process occurs in blood vessels. Recent studies that used transcranial Doppler imaging showed the effect of estrogens on cerebral blood vessels. Investigators described a decrease in the pulsatility index of postmenopausal women after they began receiving hormone replacement therapy [40]. These findings on vessel compliance suggest that estrogens in this setting could make the vessel less stiff and possibly less likely to form aneurysms. Other studies also have suggested a direct effect of estrogens on blood vessels [41-45].

Our study could not assess dose or levels of estrogens, progestins, or both; however, the data do suggest that relative deficiencies of endogenous estrogens increase the risk for spontaneous subarachnoid hemorrhage. Exogenous estrogens seem to reduce the risk. The woman at greatest risk for subarachnoid hemorrhage emerging from this study is one who smokes, is younger than 65 years, has had early menopause, and has not received hormone replacement therapy. Estrogen deficiency would not explain the higher incidence in women compared with that in men, however. Perhaps a change in estrogen levels is more important than the absolute level, or a more complex etiologic model may be needed that includes other exposures that differ by sex. Future epidemiologic studies should bring us closer to understanding why subarachnoid hemorrhage so often affects women and how best to prevent it.