Our understanding of the etiology and risk factors for sudden cardiac death (SCD) is limited compared with other acute cardiovascular events such as myocardial infarction (MI) and stroke. It is also difficult to study because of varying definitions of SCD used in the published literature. Sudden cardiac death affects approximately 300,000 to 350,000 Americans each year ((1),(2),(3),(4),5) with an estimated annual incidence of 0.1 to 0.2% (6). Although SCD is less common than MI and stroke, best estimates suggest that SCD accounts for 13% of natural deaths (6) and one-half of all coronary deaths (7), the leading cause of death in Americans (8).

Post-menopausal women now have the greatest population burden of cardiovascular disease, including SCD (9). Yet the etiology of SCD in women is less clear, because women are underrepresented in studies of SCD, and they experience fewer SCD events than men. Several studies suggest that pre-existing coronary disease is less predictive in women, and other etiologies are more likely ((5),(10),(11),12). Sudden cardiac death trends over time also differ according to sex: SCD death rates decreased faster in men compared with women in all age groups between 1989 and 1998 (2). The objectives of this paper are to estimate the annual incidence rate of SCD in post-menopausal women; identify sociodemographic, physiological, traditional, medical, and behavioral risk factors for SCD in women; and determine population-attributable fractions for SCD risk factors.

The Women's Health Initiative (WHI) includes 161,808 study participants at 40 study sites across the United States. Women participating in the observational study (OS) numbered 93,676; and 68,132 participated in 1 or more of the 3 clinical trials (hormone replacement therapy, calcium and vitamin D, and dietary modification) (13). These interventions have been described in detail previously (14). Briefly, the calcium and vitamin D trial randomized women to receive calcium and vitamin D supplements or a placebo. The dietary modification trial randomized women to follow a low-fat, high fruit, vegetable, and grain diet or their usual eating pattern (control subjects). The hormone therapy trial randomized women to receive an estrogen-alone pill, an estrogen plus progestin pill, or placebo. All women were post-menopausal and 50 to 79 years of age at baseline. Women who did not provide written informed consent, did not plan to reside in the study recruitment area for at least 3 years, had medical conditions predictive of a survival time of <3 years, had characteristics inconsistent with study compliance (alcoholism, drug dependency, mental illness, dementia), or who were actively participating in another controlled trial were excluded from the OS and clinical trials. Other exclusion criteria were used for each of the trials as described previously (14). Women enrolled between 1993 and 1998 and were followed until August 2009. We included all WHI participants in the OS and in the control and intervention arms of the randomized trials in our analyses.

Sociodemographic variables and traditional coronary heart disease (CHD) risk factors were measured by self-report at baseline with standardized questionnaires (age, race, income, marital status, education, smoking status, family history of MI, physical activity [total metabolic equivalents/week], multivitamin use, lifetime hormone use, comorbidities/disease history, and medication use) as well as by trained certified staff at the baseline exam (height, weight, body mass index [BMI], waist-to-hip ratio, blood pressure, and pulse). After women sat quietly for 5 min, blood pressure was measured with a mercury manometer twice, 30 s apart, and the average was used in this analysis. At baseline, participants additionally reported their diet with a validated food frequency questionnaire (15) designed specifically for this population of post-menopausal women. White blood cell count (WBC) was measured via standardized automated technique on fresh samples at each local WHI site. We chose to categorize WBC with a cutpoint of 6.7 Kcell/ml on the basis of findings that a WBC above 6.7 Kcell/ml might distinguish individuals who are at a high risk of cardiovascular events (16). Drugs that prolong the QT-interval were defined according to Zipes et al. (6).

At baseline, participants were considered to have a medical condition if they self-reported a physician diagnosis at baseline and were also using drugs for that condition. Prior CHD (excluding MI) included participants with a history of the following at baseline: cardiac arrest, coronary bypass surgery (CABG), angioplasty of coronary arteries (PTCA), or angina. For our multivariable time-varying exposure models, prior CHD (excluding MI) included the aforementioned in addition to adjudicated CABG, adjudicated PTCA, or adjudicated angina during follow-up. We excluded MI from our definition of CHD, because after looking at the individual components of CHD (cardiac arrest, CABG, PTCA, angina, revascularization, and MI), MI seemed to be the most important component and therefore was interesting to consider on its own.

Carotid artery disease was defined as baseline self-report of physician diagnosis or adjudicated carotid artery disease during follow-up. Diabetes was defined as self-report of physician diagnosis and taking medication at baseline or self-report of treated diabetes during follow-up (nonadjudicated). Hypertension was defined as systolic blood pressure ≥140 mm Hg or diastolic blood pressure ≥90 mm Hg or using antihypertension medication during follow-up (self-reported). Participants were asked to bring all of their prescription medications to their baseline visit. These medications were entered into a pharmacy database (Master Drug Database, Medi-Span), and this database was used to identify participants using various medications.

To determine who developed a medical condition or disease during follow-up, participants reported emergency room visits, overnight stays in hospital, and outpatient coronary revascularization procedures every 6 months in the clinical trials and annually in the OS. Medical records for overnight hospital stays and outpatient coronary revascularization procedures were scrutinized for potential outcomes of interest. Centrally or locally trained physician adjudicators classified outcomes by reviewing medical records (17).

Medical records for all deaths were scrutinized for potential outcomes of interest. Deaths caused by coronary disease were confirmed on the basis of death certificates, autopsy reports, circumstances of death, electrocardiogram, laboratory test results, and reports from all relevant procedures. Centrally or locally trained physician adjudicators classified outcomes by reviewing medical records (17). Our primary endpoint was SCD, which was defined as death occurring within 1 h of symptom onset or within 1 h after the participant was last seen without symptoms, and death that occurred in the absence of a potentially lethal non-coronary disease process. The medical record or interview of witnesses had to document that patient collapse was directly observed, as by hospital notes and cardiopulmonary resuscitation records or by a relative or observer clearly reporting that the patient was found unresponsive within <60 min from previous direct observation of stable clinical status. We included in-hospital and out-of-hospital SCDs. We excluded cases of nonsudden coronary death (death that occurs more than 1 h after symptom onset) from our analyses (n = 1,517).

Means and proportions of traditional CHD risk factors and other baseline covariates were compared by SCD status to determine risk factors for SCD in the WHI cohort (Table 1). The p values for (Table 1) were calculated with the t test for continuous variables and the chi-square test for categorical variables. The list of potential risk factors was developed on the basis of previous knowledge and existing published reports. We used backward elimination (p = 0.05), due to the large number of risk factors, forcing age, race, prior CHD (excluding MI), prior HF, and atrial fibrillation into our Cox proportional hazard model to determine what covariates to use in our final multivariable model (Table 2). Variables were included in the selection model if they were a significant risk factor for SCD (crude hazard ratio [HR] parameter p value ≤0.2).

We checked multicollinearity with the variance inflation factors from a linear regression model including all covariates/predictors and with height as the dependent variable. No variance inflation factors were >1.3. The final model (Table 2) used a time-varying analysis where the following medical conditions/diseases were allowed to vary over time: CHD (excluding MI), heart failure, diabetes, MI, carotid artery disease, and hypertension (18). Because SCD is relatively rare, we combined the WHI clinical trial and observational cohorts to get stable estimates of the HRs of potential SCD risk factors. However, because the inclusion criteria were different for the clinical trial and observational cohorts, we used inverse probability weighting to account for potential selection bias when combining the clinical trial and OS into a single analytic cohort (18). We conducted a sensitivity analysis by excluding participants who reported a history of CHD (including MI) at baseline or developed CHD during follow-up (Table 3) to see whether traditional CHD risk factors have a different impact on risk of SCD in this subgroup.

We only reported SCD estimates for white and African-American races, because there were very few SCD events for the other race groups, which precluded calculating stable estimates ((Figure 31_gr1)A and Figure 31_gr1B). Age-standardized rates were calculated on the basis of the 2000 U.S. Census. Incidence rate confidence intervals (CIs) were based on the following formula: estimated rate = ±1.96 × (estimated rate ÷ √n cases) from the New York State Department of Health website (19). We calculated population-attributable fractions on the basis of the following formula: [P_e(RR − 1)/P_e(RR − 1) +1], where P_e is the prevalence of the exposure, and relative risk (RR) is estimated by the HR (20). We calculated p trend values by creating a new variable in the multivariable model that assigned the category median to each participant. The Wald p value was used for trend. We verified the proportional hazards assumption by including log(time) × exposure cross-product terms in our models (for covariates that we did not model to vary with time). All data analyses were conducted with SAS (version 9.2; SAS Institute, Inc., Cary, North Carolina). This study was approved by the institutional review boards of all collaborating institutions, and all subjects gave informed consent.

After a mean follow-up time of 10.8 (SD 2.8) years, 418 women experienced SCD. The incidence rate of SCD was 2.4/10,000 women/year in this cohort of post-menopausal women (95% CI: 2.2 to 2.7) (Figure 31_gr1A). The incidence rate of SCD was 4.3/10,000 women/year among African Americans (95% CI: 3.2 to 5.4) compared with 2.3 for whites (95% CI: 2.1 to 2.6). Although the incidence rate was higher among participants with a history of MI or heart failure, there did not seem to be any additive or multiplicative interaction for the effect of MI and heart failure combined ((Figure 31_gr1)A and Figure 31_gr1B). Rates were similar when adjusted for age (Figure 31_gr1B).

Those who experienced SCD were older, more likely to be African American, had a lower income, smoked, and were less educated (Table 1). Those who experienced SCD also had a higher waist-to-hip ratio, BMI, resting pulse, WBC, and were less active. The following medical conditions were more common among participants who experienced SCD: diabetes, hypercholesterolemia, hypertension, stroke, MI, CHD, heart failure, CABG, PTCA, atrial fibrillation, angina, and peripheral arterial disease. Several drugs were used more often among those who experienced SCD: aspirin, lipid-lowering medications, blood pressure medications, and diabetes medications. Current hormone use was less common among those who experienced SCD compared with those who did not.

In our multivariable analysis (Table 2), the following were independently associated with increased risk of SCD in post-menopausal women: older age, African-American race, smoking, higher resting pulse, higher waist-to-hip ratio, WBC above 6.7 Kcell/ml (top 25% of the population), and a history of the following medical conditions/diseases: heart failure, diabetes, MI, carotid artery disease, and hypertension. Lower income, higher BMI, and history of atrial fibrillation were also associated with an increased risk of SCD; however, these associations were not statistically significant. Population-attributable fractions were 22% for hypertension, 15% for waist-to-hip ratio ≥0.87, 14% for MI, 11% for WBC above 6.7 Kcell/ml, 8% for smoking, 8% for heart failure, and 4% for diabetes (Table 4).

We conducted a sensitivity analysis by excluding participants with a history of CHD (including MI) at any time before SCD (Table 3). The results were very similar, except that African-American race was no longer a statistically significant independent risk factor, and resting pulse no longer seemed to be associated with risk of SCD. The remaining risk factors had a slightly larger magnitude of effect.

We additionally tested the effect of the 3 WHI interventions on risk of SCD in the clinical trial cohorts. None of the 3 interventions was associated with risk of SCD: HR: 1.04, 95% CI: 0.69 to 1.56 for the calcium and vitamin D supplement intervention versus control; HR: 0.94, 95% CI: 0.65 to 1.37 for the low-fat diet intervention versus control; HR: 0.90, 95% CI: 0.51 to 1.60 for the estrogen-alone intervention versus control; and HR: 0.61, 95% CI: 0.32 to 1.14 for the estrogen plus progestin intervention versus control. However, the CIs are wide and do not rule out the possibility of effects in 1 or both directions, especially for the estrogen plus progesterone analysis in which there were only 16 cases of SCD in the intervention group and 25 in the control group.

We also ran a model that included all 43 potential risk factors that were crudely associated with risk of sudden cardiac death (p value ≤ 0.2). We additionally included—besides factors included in Tables (Table 2) and (Table 3)—alcohol; multivitamin use; physical activity; maternal history of MI; marital status; education; systolic blood pressure; hormone use; time of menopause; having a current care provider; optimism construct score; hostility construct score; whether or not the participant was using the following drugs: aspirin, lipid-lowering drugs, steroids, antidepressants, any antihypertensive agents, beta-blockers, diuretics, angiotensin-converting enzyme inhibitors/angiotensin receptor blockers, or calcium channel blockers; and participant history of the following medical conditions: high cholesterol, stroke/transient ischemic attack, cardiac catheterization, peripheral artery disease, rheumatoid arthritis, deep vein thrombosis/pulmonary embolism, lupus, and migraines. None of these risk factors reached statistical significance.

Our results suggest that MI, heart failure, current cigarette smoking, diabetes, hypertension, and abdominal obesity might be major risk factors for SCD in women. African-American women had a 60% increased risk of SCD in this study population even after all known risk factors and sociodemographic factors were accounted for. Although MI was the strongest risk factor for SCD, hypertension was associated with a larger population-attributable fraction, likely because of its high prevalence in this population (Table 4). We found that approximately one-half of SCD victims were not diagnosed with CHD before death, which is consistent with previous research (9).

Our estimated rate of SCD in post-menopausal women (2.4/10,000/year) is much lower than that reported by Zipes et al. (7) (10 to 30/10,000/year). This might be a result of healthy cohort bias where women in the WHI might be healthier on average than post-menopausal women in general. However, our estimate might be more accurate and specific than prior estimates, which would result in a lower estimated rate (21). Zipes et al. (7) defined SCDs as deaths from ischemic heart disease that occurred at home or in the emergency room. This definition is more sensitive/inclusive, which might partly explain the higher estimated rate.

Currently, it is difficult to find a valid estimate of the rate of SCD in post-menopausal women, because of a lack of good quality data and varied definitions. Our definition of SCD (also called sudden cardiac arrest) is close to that used by the Nurses' Health Study (22), which also uses a 1-h limit but excludes non-arrhythmic deaths. Other studies, including the Cardiovascular Health Study, do not use a specific time frame and instead define sudden cardiac death as “sudden pulseless condition of cardiac origin in a previously stable individual that occurred out of the hospital or in the emergency department” (23).

Waist-to-hip ratio, a measure of body fat distribution, seemed to be a more important risk factor for SCD than BMI in this cohort of post-menopausal women. Because abdominal fat has a greater influence on inflammation than fat stored in other areas of the body, this might mean that inflammation plays a role in the etiology of SCD. Indeed, we saw a positive association between WBC, a broad measure of inflammation, and risk of SCD. Furthermore, smoking, diabetes, and hypertension were associated with increased risk of SCD, and these conditions are all associated with chronic inflammation (24). Some previous research also suggests a role of inflammation (24), where decreasing levels of albumin and increasing WBC was associated with an increased risk of SCD. Future studies should explore the association between other measures of inflammation such as C-reactive protein and other comorbid inflammatory diseases such as metabolic syndrome and gout, and risk of SCD.

Previous studies have also found that hypertension ((10),(25),(26),(27),28), smoking ((10),(25),(26),(27),28), heart rate ((10),(29),30), and African-American race ((31),32) are risk factors for SCD in women. Although atrial fibrillation was not a statistically significant independent risk factor in our study, unlike others ((25),27), we did see that it was associated with increased risk of SCD. The lack of statistical significance of this association could be due to measurement error, because participants self-reported a history of atrial fibrillation, and this information was collected at baseline only. Many studies also report that diabetes or glucose intolerance is a risk factor ((25),(28),(33),(34),(35),36), yet other large longitudinal cohorts do not find this association (1), and interestingly, studies limited to men also do not find an association with diabetes or glucose intolerance ((37),38). Albert et al. (28) similarly found a strong effect of MI, but most studies do not separate CHD from MI. The fact that CHD excluding MI is no longer an independent risk factor might mean that MI is the key component of CHD that is associated with an increased risk of SCD.

Unlike some studies, ours did not find that regular exercise ((39),(40),41), family history of CHD ((27),(28),42), hypercholesterolemia ((26),27), obesity ((25),28), or use of drugs that prolong the QT interval (35) were risk factors for SCD. This is likely because other risk factors considered capture the same pathway, for example waist-to-hip ratio and BMI. We also found that several risk factors for acute cardiac events were not independently associated with risk of SCD in this cohort, including dyslipidemia, aspirin or antidepressant use, hostility, and history of stroke, deep vein thrombosis, or pulmonary embolism. Differences between our study findings and others might be due to differences in underlying type of SCD. For example a higher proportion of SCDs caused by MI in 1 study versus another might result in a different set of risk factors that are important. Unfortunately, we do not have information on the underlying cause of SCD (arrhythmia, ventricular tachycardia, MI, stroke, and so forth), which makes it difficult to study the underlying mechanism.

Ours is 1 of the largest studies of SCD in women drawn from multiple ethnic groups and multiple geographic sites in the United States. Other strengths are its prospective design; physician-adjudicated outcomes including SCD, MI, CABG, PTCA, angina, and carotid artery disease; and excellent data on measured blood pressure and medication use. We also have comprehensive data on several sociodemographic, physiological, traditional, medical, and behavioral risk factors. Finally, we are the first to compare population-attributable fractions between major risk factors and to report an incidence rate for SCD in this population on the basis of a prospective study, with a specific definition, and with adjudicated cases of SCD.

The major limitation of our study is its observational nature, which can lead to residual confounding. Another important limitation is our lack of data on subtype of SCD, which would give us more insight about underlying mechanisms. However, our strict interpretation of the 60-min limit and requirement for documentation of timing within the case records provides additional rigor in our assessment of SCD. Some deaths coded as SCDs might actually be due to other causes, for example, cerebral hemorrhage, acute pulmonary embolism or aortic rupture, but these cases will also be present in any study of sudden death in a population.

We found that traditional CHD risk factors likely play an important role in the etiology of SCD in post-menopausal women. We also found evidence that the incidence rate of SCD in post-menopausal women might be lower than previously estimated. Future research should look into why African-American race was associated with a 61% increased risk of SCD even after adjusting for age, income, disease status, and the like. Finally, inflammation, diet, genes, and gene–environment interaction are all important areas for future research in discovering the etiology of SCD.