Potential conflict of interest: C.S.M. is an employee of the National Institutes of Health.
Black individuals younger than 75 years have more than twice the risk for stroke death than whites in the United States. Regardless of race, stroke death is approximately 50% greater in the “stroke belt” and “stroke buckle” states of the Southeastern United States. We assessed geographic and racial differences in estimated 10-year stroke risk.
The Reasons for Geographic and Racial Differences in Stroke study is a population-based cohort of men and women 45 years or older, recruited February 2003 to September 2007 at this report, with oversampling of stroke belt/buckle residents and blacks. Racial and regional differences in the Framingham Stroke Risk Score were studied in 23,940 participants without previous stroke or transient ischemic attack.
The mean age-, race-, and sex-adjusted 10-year predicted stroke probability differed slightly across regions: 10.7% in the belt, 10.4% in the buckle, and 10.1% elsewhere (p <0.001). Geographic differences were largest for the score components of diabetes and use of antihypertensive therapy. Blacks had a greater age- and sex-adjusted mean 10-year predicted stroke probability than whites: 11.3% versus 9.7%, respectively (p <0.001). Race differences were largest for the score components of hypertension, systolic blood pressure, diabetes, smoking, and left ventricular hypertrophy.
Although blacks had a greater predicted stroke probability than whites, regional differences were small. Results suggest that interventions to reduce racial disparities in stroke risk factors hold promise to reduce the racial disparity in stroke mortality. The same may not be true regarding geographic disparities in stroke mortality. Ann Neurol 2008;64:507–513
Stroke is the third leading cause of death in the United States, accounting for 1 of every 16 deaths in 2004.1 Residents of the Southeastern United States (“stroke belt”) have approximately 50% greater rates of stroke mortality than the remainder of the United States.2–4 These geographic differences have existed since at least 1940,2 with excess deaths demonstrated for men, women, whites, and blacks.4
The cause of the excess stroke mortality in the stroke belt is unknown.5, 6 Differences in stroke incidence may contribute to the excess mortality, but little national data on regional differences in incidence are available to address this hypothesis. For white men and women, data from the first National Health and Nutrition Examination Survey (NHANES I) suggest that stroke incidence was greater in the Southeast than the Northeast, but the pattern was inconsistent between the Southeast and other regions.7 In addition, although the stroke belt is at least as pronounced for blacks as for whites8–10 and the incidence of stroke was greater among blacks11–13 in some studies, in NHANES I, regional differences in stroke incidence among blacks were not as striking as those among whites.7 Lower rates of stroke hospitalization in the stroke belt could also explain underlying differences in mortality; however, among Medicare recipients, those in Southeastern states were more likely to be hospitalized for stroke than in other regions.14 Another potential cause of the stroke belt is differences in case fatality by region; however, little information is available to support this hypothesis.15 Similarly, differential case fatality according to race has not been clearly demonstrated.12, 13, 16 Hence, the available data suggest that geographic and racial variations in stroke mortality relate to differences in incidence, not case fatality. If this is the case, then disparities in stroke risk factors may underlie observed differences in stroke mortality.
The Framingham Heart Study investigators identified nine risk factors for stroke: age, sex, systolic blood pressure, antihypertensive therapy, diabetes, current smoking, prior cardiovascular disease, atrial fibrillation, and left ventricular hypertrophy (LVH).17 The Framingham Stroke Risk Score (FSRS), developed and validated in a primarily white population, predicts the 10-year probability of stroke based on these nine risk factors. To address the role of stroke risk factors on regional and racial differences in stroke mortality, we studied geographic and racial differences in the FSRS in a national cohort.
Subjects and Methods
The REasons for Geographic And Racial Differences in Stroke (REGARDS) Study is a national longitudinal study initiated in January 2003 to elucidate the causes of geographic and racial disparities in stroke mortality.18 REGARDS was designed to recruit 30,000 community-dwelling black and white participants (50% from each group) aged 45 years or older from the continental United States. By design, 20% of the cohort resided in the “buckle” of the stroke belt (coastal plain region of North Carolina, South Carolina, and Georgia), 30% from the stroke belt states (remainder of North Carolina, South Carolina, and Georgia plus Alabama, Mississippi, Tennessee, Arkansas, and Louisiana), and 50% from the other 40 contiguous states (referred to as “rest of the nation” here). Participants were recruited from those randomly selected from a commercially available nationwide list purchased from Genesys. They were contacted by mail, then telephone. Participants completed a standardized telephone interview including demographic information and medical history (including history of stroke or transient ischemic attack). An in-home examination was performed subsequently to obtain physical measures (blood pressure, height, weight), electrocardiogram, medication inventory, and fasting blood and urine samples. In the home, written informed consent was obtained using methods approved by the institutional review boards of all participating institutions. As of September 1, 2007, both the phone interview and in-home examination were completed for 29,185 participants. This analysis included 23,940 participants without a self-reported history of stroke or transient ischemic attack (n = 3,182), and who had data for all components of the FSRS (≥1 component missing on 2,063). Those with missing data did not differ in sociodemographic characteristics from included participants.
Race was defined by self-report requesting participants to select their race from a list (white, black or African-American, Asian, native Hawaiian or other Pacific Islander, American Indian, Alaska native, or other). They were then asked whether they were also Hispanic or Latino. Only white or black non-Hispanic participants were eligible. Age, sex, use of antihypertensive therapy, and history of heart disease (myocardial infarction or heart attack, coronary artery bypass surgery, coronary angioplasty, or stenting) were defined based on self-report of a physician diagnosis. Systolic blood pressure was the average of two measurements taken by a trained technician after the participant was seated for 5 minutes, measured using a standard protocol and regularly tested aneroid sphygmomanometer. Diabetes was defined as fasting glucose greater than 6.99mmol/L (126mg/dl), nonfasting glucose greater than 11.1mmol/L (200mg/dl), or self-reported medication use for diabetes. Current smoking was classified for FSRS determination by the response to the question, “Do you smoke cigarettes now, even occasionally?” Atrial fibrillation was defined as either self-report of a diagnosis by a healthcare professional or by electrocardiogram. LVH was defined by electrocardiogram using the modified Cornell Index. Although the Cornell Index might be favored,19 this requires a 12-lead electrocardiogram, and the first 6,490 participants enrolled had only a 7-lead electrocardiogram. The FSRS was calculated for each participant using age, sex, systolic blood pressure, antihypertensive therapy, diabetes, smoking, prior cardiovascular disease, atrial fibrillation, and LVH.17, 20 Among the first 15,017 participants with a 12-lead electrocardiogram, use of the Cornell Index to define LVH had minimal impact on the FSRS (correlation coefficient of scores with Cornell Index and modified Cornell Index = 0.99).
Analyses were performed for both the mean FSRS and for each individual component of the FSRS. Because results were reported in sex/race strata, no weights were applied to the analysis. Analysis of variance was used to assess regional differences in factors contributing to the FSRS, after adjustment for age, sex, and race. All analyses were performed using SAS version 9.1.3 (SAS Institute, Cary, NC).
Levels of the risk factors that are part of the FSRS are shown for sex/race strata of the REGARDS population without stroke/TIA in Table 1. Table 2 provides the age-, race-,and sex-adjusted FSRS and its components by region. The mean 10-year age-, race-, and sex-adjusted predicted stroke risk in the stroke belt and buckle were 10.7 and 10.4%, respectively, only slightly greater than in the rest of the nation (10.1%). The difference in FSRS was driven primarily by regional differences in two risk factors, diabetes and use of antihypertensive medications, with the largest regional difference in the prevalence of diabetes. The prevalence of diabetes in the stroke belt was 3.5 percentage points greater than in the rest of the nation, and 5.1 percentage points greater in the stroke buckle than the rest of the nation. For antihypertensive medication use, these percentage differences were 4.5 and 5.2%, respectively.
Table 1. Framingham Stroke Risk Score Components at Baseline in the Reasons for Geographic and Racial Differences in Stroke (REGARDS) Cohort Participants without Stroke/TIA by Race and Sex, 2003–2007
Table 2. Mean Age-, Race-, and Sex-Adjusted Framingham Stroke Risk Score, and Percentage or Mean Values for Each Score Component, by Geographic Region in the Reasons for Geographic and Racial Differences in Stroke (REGARDS) Cohort Participants without Stroke/TIA, 2003–2007
Blacks had a greater age- and sex-adjusted mean (standard deviation) FSRS than whites, 11.3% (11.8) compared with 9.7% (10.1). This was true in all three regions (p < 0.001) and reflected a substantially worse risk factor profile for all FSRS components (p < 0.0001 for each) except history of heart disease and atrial fibrillation, which were less common in blacks. Table 3 shows geographic differences in FSRS and its components stratified according to race. Differences in the FSRS among regions were small but significant among whites. Among blacks, the FSRS was only slightly greater in the stroke belt than the rest of the nation, and it did not differ in the stroke buckle. Most regional differences in FSRS components comparing the stroke belt or buckle with the rest of the nation were similar in blacks and whites, except for greater prevalences of atrial fibrillation in the stroke buckle and heart disease in the stroke belt among whites but not blacks. Smoking was less common in the stroke buckle than other regions for blacks but not whites, and more common in the belt for whites but not for blacks (p for interaction between race and region < 0.0001). For both ethnicities, antihypertensive medication use was more common in the stroke belt or buckle compared with the rest of the nation, although systolic blood pressure was higher in the belt (but not buckle) among whites and lower in the buckle (but not belt) among blacks (p for interaction between race and region = 0.001). Diabetes was about twice as common among blacks than whites in all regions, but the difference in prevalence of diabetes in the stroke belt compared with the rest of the nation was greater for whites (p for interaction between race and region = 0.02). This was related to a similar prevalence of diabetes in the stroke belt compared with the rest of the nation among black men (Fig 1). In other race/sex groups in Figure 1, the prevalence of diabetes tended to increase from the rest of the nation to belt to buckle.
Table 3. Mean Age- and Sex-Adjusted Framingham Stroke Risk Score, and Percentage or Mean Values for Each Score Component in the Reasons for Geographic and Racial Differences in Stroke (REGARDS) Cohort Participants without Stroke/TIA by Geographic Region and Race, 2003–2007
Figure 2 presents geographic differences in the age-adjusted FSRS stratified by race and sex. Although differences between race and sex groups were clear and highly significant, differences by region were small. The average scores for those living in the stroke belt and buckle compared with the rest of the nation were significantly different at the p <0.05 level for only the belt and buckle regions among white men, and the belt region among black and white women. There were no regional differences in scores among black men.
Knowledge of racial and geographic patterns in stroke risk factors may provide useful information for public health interventions. Ultimately, the goal is to identify targets for interventions in specific population groups that may reduce the disparities in stroke incidence and consequently the burden of stroke in all populations. The FSRS is a summary index that incorporates “traditional” stroke risk factors and estimates the predicted stroke incidence over 10 years. In this national study, geographic differences in the average FSRS across regions were small. This suggests that only a small portion of the 50% greater stroke mortality in these regions is driven by the “traditional” risk factors comprising the FSRS. Because hypertension is the strongest stroke risk factor, we expected that hypertension would play a major role in differences in stroke fatality. However, geographic differences in diabetes were substantially larger than differences in hypertension (defined here by medication use). Findings suggest that interventions to reduce geographic disparities in diabetes, including optimizing prevention, diagnosis, and treatment, may hold promise for reducing geographic disparities in stroke mortality.
Although geographic differences in diabetes prevalence were the largest differences seen among stroke risk factors, this association differed in race/sex strata, all of whom share the greater stroke mortality of the stroke belt and buckle. There were small differences in the prevalence of diabetes in white women, diabetes was not more common in the stroke belt than in the rest of the nation among black men, and differences in diabetes prevalence by region were largest in black women. In the Behavioral Risk Factor Surveillance System (BRFSS), self-reported diabetes was greater in the southeastern US states in 2001, with the stroke belt states of Alabama ranking 1st (10.5%), Mississippi 2nd (10.3%), South Carolina 4th (9.4%), Arkansas 6th (8.9%), Louisiana 7th (8.5%), Tennessee 11th (8.3%), Georgia 22nd (7.7%), and North Carolina 23rd (7.6%).21
Most articles on causes of the stroke belt have largely assumed that the cause is greater stroke incidence in this region.5, 6 Our findings of relatively small differences in FSRS by region, and of similar FSRS in the “higher risk” stroke buckle than stroke belt raise a hypothesis that stroke case fatality may play a substantial role. Although components of the FSRS may influence case fatality, other factors not addressed in our analysis, such as poverty, access to care, and nontraditional risk factors also need to be considered. Studies such as REGARDS, which follow a large geographically dispersed cohort, will help to answer these questions.
Although geographic differences in the FSRS were relatively small and influenced by selected component risk factors, differences in the FSRS between whites and blacks were substantially larger, with differences in many of the component risk factors, in particular, diabetes and hypertension. These racial disparities have been described previously,22, 23 including higher systolic blood pressure in blacks than whites (by approximately 5mm Hg), even in the setting of greater prevalence of antihypertensive medication use.24 It has been suggested that greater prevalence rates of diabetes, hypertension, and other risk factors contribute to racial disparities in stroke risk,22, 25 but few prospective studies are available with sufficient numbers of black and white participants to address this question. Follow-up in the REGARDS cohort will help to clarify this issue.
Our findings extend previous reports on geographic variation in hypertension from the NHANES.26–28 In NHANES III,26, 27 there was a slightly greater prevalence of hypertension in the Southeast for white men (27 vs 24%), white women (22 vs 21%), and black men (35 vs 33%), with a more substantial difference for black women (35 vs 28%). In further analysis assessing age groups (40–59 and 60–79 years), hypertension was more prevalent in the Southeast in seven of eight age/race/sex strata; however, these differences were statistically significant only among black men aged 40 to 59 and white men aged 40 to 59 years.28 We are not aware of other studies providing a true regional comparison of hypertension prevalence. Because the focus of this article was overall differences in the FSRS that consider use of antihypertensive medications and the average systolic blood pressure, we focused on these parameters rather than hypertension per se. Antihypertensive medication use was more common in the stroke belt than the rest of the nation for both whites and blacks. However, the pattern of blood pressure was not consistent, with whites having a higher blood pressure in the stroke belt (but not buckle) than the rest of the nation, and blacks having lower systolic blood pressure in the stroke buckle (but not belt) than the rest of the nation. Thus, there is a mixed picture regarding geographic variations in hypertension and blood pressure, with NHANES showing trends for higher blood pressure in the Southeast, and REGARDS showing greater use of antihypertensive medications but an inconsistent pattern in measured blood pressure. The NHANES reports used a broader definition of Southeast, which may explain some of these differences. We have reported elsewhere24 detailed analyses of correlates of regional and racial differences in hypertension.
Few population-based data are available on prevalence of atrial fibrillation and LVH by region and race. Case–control studies and hospital series suggest that atrial fibrillation is less prevalent and may be less strongly associated with stroke risk among blacks than whites in the United States.29, 30 In one study of blacks (2.7% with previous stroke; 68% with hypertension), 49% of participants had LVH by echocardiogram, and this was associated with an increased risk for magnetic resonance imaging–documented stroke and white matter disease.31
The strengths and weaknesses of this study merit discussion. Study examinations were conducted in participants' homes by a large number of examiners rather than in a limited number of field sites as in most similar studies. Although substantial training and standardization efforts were undertaken,18 these cannot replicate the quality of field center–based epidemiological studies. This shortcoming is offset by the advantages of a large nationally based sample with oversampling of blacks and residents of the stroke belt and buckle, enabling optimal assessment of geographic and racial disparities. As with all epidemiological studies, participation rates are a concern. We conservatively estimated a 40% participation rate in REGARDS, which compares favorably with other epidemiological studies. This analysis included 23,940 participants without stroke or TIA, with 41% black participants. Thus, these results are more precise than other studies that included far fewer subjects.7, 28 Unlike the BRFSS,21 REGARDS measured most components of the FSRS rather than relying on self-reported data, providing better reliability for the FSRS. About 8% of participants were missing at least one component of the FSRS, leading to exclusion of these individuals. Although we relied on self-reported cerebrovascular disease as a selection criterion for this analysis, any impact of potential misclassification on findings would be small given the sample size and rarity of disease. As in other population-based studies, REGARDS is limited to noninstitutionalized individuals who had a telephone, which may restrict generalizability. Finally, the FSRS has not been validated in blacks; therefore, it may not be an optimal tool for assessing stroke mortality risk in this group.
In summary, we observed only modest differences in the FSRS between the stroke belt or stroke buckle compared with the rest of the United States, and differences were mostly driven by geographic differences in diabetes and hypertension. Interventions to reduce geographic disparities in these factors may be promising for reducing geographic disparities in stroke mortality. However, given the small regional differences in the FSRS, it appears unlikely that the increased stroke mortality in the Southeast is due solely to differences in traditional risk factors that comprise the score. Follow-up data from this study will examine these questions and the role of other risk factors. Conversely, we observed the expected racial differences in hypertension, diabetes, atrial fibrillation, and LVH that could contribute to racial differences in stroke risk. These disparities offer hypotheses for interventions that might reduce racial disparities in stroke mortality.
This research is supported by the NIH (National Institute of Neurological Disorders and Stroke, U01 NS041588 G.H.).
We acknowledge the participating investigators and institutions: University of Alabama at Birmingham (Study PI, Data Coordination Center, Survey Research Unit): G. Howard, L. McClure, V. Howard, L. Wagner, V. Wadley, R. Go; University of Vermont (Central Laboratory): M. Cushman; Wake Forest University (ECG Reading Center): R. Prineas; Alabama Neurological Institute (Stroke Validation Center, Medical Monitoring): C. Gomez, D. Rhodes, S. Bowling, S. Orr; University of Arkansas for Medical Sciences (Survey Research): L. Pulley; University of Cincinnati (Clinical Neuro-Epidemiology Unit): B. Kissela, D. Kleindorfer; Examination Management Services Incorporated (In-Home Visits): A. Graham; National Institute of Neurological Disorders and Stroke: C. Moy.