American Journal of Epidemiology
Copyright C 1996 by The Johns HopWns University School of Hygiene and Public Health
All rights reserved
Vol. 144, No. 7
Printed In U.S-A.
Relation between Residence in the Southeast Region of the United States
and Stroke Incidence
The NHANES I Epidemiologic Followup Study
Richard F. Gillum and Deborah D. Ingram
For at least 50 years, stroke death rates have been higher in the southeast region of the United States than
in other US regions. To test the hypotheses that stroke incidence is higher in the Southeast than in other
regions and that higher levels of known stroke risk factors in the Southeast explain the difference in incidence,
data were analyzed from a nationally representative, longitudinal cohort study of a sample drawn from the US
population, the First National Hearth and Nutrition Examination Survey (NHANES I) Epidemiologic Followup
Study (1971-1987). In white men and women aged 45-74 years, the risk of stroke was significantly higher in
the Southeast than the Northeast or the West in men and the Midwest in women. In white men, this excess
risk could not be explained by regional differences in multiple stroke risk factors (Northeast vs. Southeast
risk-adjusted relative risk = 0.71,95% confidence interval 0.52-0.98). In white women, some of the excess risk
associated with residence in the Southeast compared with the Midwest could be explained by the regional
differences in risk factors measured in NHANES I (Midwest vs. Southeast risk-adjusted relative risk = 0.73,
95% confidence interval 0.53-1.00). In blacks, regional differences that were statistically significant could not
be demonstrated. However, a strong association of increased stroke risk with nonmetropolitan residence in
blacks was demonstrated that was independent of region or other stroke risk factors. Higher stroke incidence
rates in the Southeast contribute to the higher stroke mortality rates in that US region. Am J Epidemiol 1996;
144:665-73.
blacks; cerebrovascular disorders; geography; rural population; urban population; women
For at least 50 years, stroke death rates have been
higher in the southeast region of the United States than
in other US regions (1, 2). Analysis of Medicare
hospitalization data indicated that stroke hospitalization rates are also higher in the Southeast than in other
regions (3). However, little difference in case fatality
rates was found among regions, suggesting that variation in stroke incidence and/or recurrence might be
the most important determinants of variation in mortality (3). Although a number of descriptive and ecologic analyses have attempted to determine the reason
for the geographic pattern, no previous studies have
directly examined the hypotheses that stroke incidence
is higher in the Southeast than in other regions, and in
nonmetropolitan than in metropolitan areas, and that
higher levels of known stroke risk factors in the Southeast explain the difference in incidence. This report
presents results of tests of these hypotheses using data
from a nationally representative, longitudinal cohort
study of a sample drawn from the US population.
MATERIALS AND METHODS
Study population
The NHANES I Epidemiologic Followup Study
(NHEFS) is a longitudinal study of participants in the
First National Health and Nutrition Examination Survey (NHANES I) who were 25-74 years of age at the
time of the survey examination in 1971-1975 (4-7).
The personal interviews and physical and laboratory
examinations of NHANES I provided the baseline data
for the NHEFS. This analysis was based on three
waves of follow-up data collected during 1982-1984,
1986, and 1987. Data collected consisted of interviews, health care facility medical records for the
period between baseline and last follow-up, and death
certificates for all decedents. Information about the
Received for publication September 8, 1995, and accepted for
publication May 21, 1996.
Abbreviations: NHANES I, First National Health and Nutrition
Examination Survey; NHANES II, Second National Health and Nutrition Examination Survey; NHEFS, NHANES I Epidemiologic Followup Study; SMSA, Standard Metropolitan Statistical Area
From the National Center for Health Statistics, Centers for Disease Control and Prevention, Hyattsville, MD.
Reprint requests to Dr. Richard F. Gillum, Office of Analysis,
Epidemiology, and Health Promotion, National Center for Health
Statistics, Centers for Disease Control and Prevention, 6525 Belcrest Road, Hyattsville, MD 20782.
665
666
Gillum and Ingram
place and date of hospitalizations was obtained from
the follow-up interviews and/or death certificates. The
completeness of data collection has been reported (8,
9).
This analysis included white and black persons
45-74 years of age at baseline examined in NHANES
I. Of the 7,874 white and black persons 45-74 years of
age at baseline, 135 (1.7 percent) were lost to followup. Excluded from the analyses were 107 persons with
unknown baseline physical activity history, systolic
blood pressure, diabetes history, cigarette smoking
status, history of heart disease, educational attainment,
or alcohol consumption. Additionally excluded were
128 persons with no follow-up interview or death
certificate and 244 persons who had a positive history
of stroke at baseline, defined as those who had ever
been told by a doctor that they had a stroke prior to the
baseline interview. After all exclusions, 7,260 persons
remained for the analyses (6,198 white persons and
1,062 black persons). The mean length of follow-up
for stroke-free survivors for the mortality analyses was
12 years.
Outcome definitions
Incident stroke cases met at least one of the following criteria (10, 11): 1) a death certificate with underlying or nonunderlying cause of death coded 4 3 1 434.9, 436, or 437.0-437.1 using the International
Classification of Diseases, Ninth Revision; or 2) one
or more hospital and/or nursing home stays during the
follow-up period with any discharge diagnosis with
these codes using the International Classification of
Diseases, Clinical Modification. The date of incidence
was estimated as one of the following: 1) date of first
hospital admission with a stroke diagnosis or 2) date of
death for stroke deaths without any stroke hospital
records. Table 1 shows the number of incident cases of
stroke in the analysis cohort by race, sex, and region.
Baseline variables
The region and urbanization level of the place of
residence were determined from the household address
of the subject at baseline examination. Definitions of
the regions, shown in the Appendix, differ from those
of the US Bureau of the Census (4, 5). Urbanization
definitions were as follows:
a standard metropolitan statistical area (SMSA) [was]
basically a county or a group of contiguous counties
which contains at least one city of 50,000 inhabitants
or more, or "twin cities" with a combined population
of at least 50,000. In addition to the county or counties
containing such a city or cities, contiguous counties
were included in an SMSA if, according to the 1960
Census, they were socially and economically inte-
TABLE 1. Number of Incident cases of stroke by race, sw,
and region In persons aged 45-74 years at baseline In the
analysis cohort, NHEFS,' 1971-1987
Rate/
1,000
personyears
95% Cl*
2,899
644
694
721
840
9.13
11.55
13.27
9.45
6.91-11.36
9.14-13.96
10.70-15.85
7.48-11.43
329
71
62
109
87
3,299
775
791
839
894
7.53
6.35
10.74
7.96
5.78-9.28
4.77-7.93
8.72-12.76
6.29-9.64
155
17
21
89
28
1,062
147
211
517
10.17
8.68
15.77
14.04
5.34-15.01
4.97-12.39
12.49-19.04
8.84-19.23
Na
of
cases
No.
at
risk
White men
All
Northeast
Midwest
Southeast
West
343
65
88
102
88
White women
All
Northeast
Midwest
Southeast
West
Afl blacks
All
Northeast
Midwest
Southeast
West
Race
187
• NHEFS, First National Health and Nutrition Examination
Survey Epidemiologlc Followup Study; Cl, confidence interval.
grated with the central city. Each SMSA must include
at least one central city, and the complete title of an
SMSA identifies the central city or cities (4, p. 43).
Analysis levels were the following: 1, in SMSA, in
central city ("central city"); 2, in SMSA, not in central
city ("suburb"); 3, not in SMSA ("not SMSA").
The baseline medical history questionnaire provided
information about selected conditions (e.g., diabetes)
diagnosed by a doctor (4, 5). A positive history of
heart disease at baseline was defined as having ever
been told by a doctor that one had ever had a heart
attack or heart failure or having used any medicine,
drugs, or pills for a weak heart during the 6 months
prior to the baseline interview. Other baseline variables were measured as described elsewhere (4, 5,
12-15).
Statistical methods
Stroke incidence rates were calculated as the number of incident cases per 1,000 person-years at risk.
Estimates of the risk of stroke incidence for persons
residing in each region relative to those residing in the
southeast region were derived from Cox proportional
hazards regression models computed using the
PHREG procedure in the Statistical Analysis System
(SAS) (16). Persons who did not develop stroke were
censored at either the date of their last follow-up
interview or their date of death.
Am J Epidemiol
Vol. 144, No. 7, 1996
Southeast Region and Stroke
All models included the age at baseline in single
years as a covariate. Risk-adjusted models included
the following baseline variables: current cigarette
smoking (yes, no), systolic blood pressure (mmHg),
history of diabetes (yes, no), educational attainment
(<12 years, ^12 years), history of heart disease (yes,
no), alcohol consumption (yes, no), and nonrecreational physical activity level (low, medium, high).
Results were essentially the same using expanded
models (not shown) that included serum cholesterol
(<200, 200-239, >239 mg/dl), number of cigarettes
smoked (smokers only), and body mass index (weight
(kg)/height (m)2). Separate models were developed for
white men, white women, and blacks. Models for
blacks included sex as a covariate. Effects of urbanization were examined in separate models for the
Southeast and for all other regions combined because
of sample size constraints and the different pattern of
urbanization in the Southeast compared with other
regions.
To assess the effect of the complex survey design on
the results, Cox proportional hazards regression analyses were performed using PROC SURVIVAL in
SUDAAN to incorporate the sample weights and the
stratification and clustering in the analysis (17, 18).
The results were consistent with minimal effects of the
667
complex survey design on the main conclusions derived from the unweighted estimates. Therefore, the
estimates from the unweighted Cox regression models
are presented. However, weighted means and percentages are shown in table 2 to show results more nearly
in agreement with those previously published for
NHANESI.
RESULTS
Table 2 shows the levels of selected stroke risk
factors by region, sex, and race in the analysis cohort.
Among whites, the prevalence of current smoking,
abstinence from alcohol, low education, diabetes, history of heart disease, prevalence of hypertension, and
residence in a nonmetropolitan area were generally
higher in the Southeast than in other regions. The
prevalence of abstinence from alcohol, low education,
residence in a nonmetropolitan area, and mean blood
pressure also tended to be higher in the Southeast than
in other regions among blacks.
In white men aged 65-74 years, the incidence rate of
stroke per 1,000 person-years was higher in men residing in the southeast region than in men residing in
other regions (table 3). At ages 45-64 years in white
TABLE 2. Laval* of stroke risk factors (means or percentages)* at baseline In the cohort by region, sex, and race, NHEFS.t
1971-1987
smoker
' '
Any
alcohol
Intake
(%)
Low
physical
activity
(%)$
<12
years of
education
(%)
0.26
0.26
0.25
0.27
45.3
48.8
48.1
37.8
84.7
89.3
64.0
70.4
12.1
18.4
13.3
5.9
74.5
74.1
895
52.1
147.9
144.0
159.7
151.1
0.29
0.29
0.29
0.29
44.3
23.8
28.8
29.1
60.0
58.6
42.7
48.4
13.0
7.7
18.4
20.6
715
62.4
87.6
61.2
16.3
12.3
6.1
9.3
9.1
9.9
57.3
56.6
57.3
56.7
140.2
1385
138.9
136.6
0.26
056
0.25
056
34.6
38.9
45.3
42.8
85.8
84.9
65.6
80.7
11.8
7.2
9.2
12.3
46.7
495
57.0
45.3
45
6.2
5.3
4.8
57.5
57.7
58.2
57.9
139.4
1395
1425
137.7
056
0.27
0.26
0.25
28.8
24.6
32.9
28.3
75.0
67.6
42.4
70.4
10.6
8.1
8.6
13.0
43.0
41.9
54.3
40.0
65
5.1
5.9
3.7
Sex, race,
and region
Age
(years)
Systolic
blood
pressure
(mmHg)
Body
mass
Index
(kg/m»)
Black men
Northeast
Midwest
Southeast
West
57.8
55.2
57.2
56.3
143.7
140.8
145.0
149.1
Black women
Northeast
Midwest
Southeast
West
55.5
55.9
57.3
57.1
White men
Northeast
Midwest
Southeast
West
White women
Northeast
Midwest
Southeast
West
Current
History
of
diabetes
History of
heart
disease
(%)
tension
( )§
3.1
13.7
20.4
12.1
11.3
68.4
71.3
63.5
75.9
62.5
28.1
9.1
64.0
69.0
76.1
78.0
2J
8.5
60.7
27.1
13.0
13.5
11.9
55.5
60.3
61.7
57.0
22.9
36.2
625
32.3
8.5
7.7
10.9
6.6
55.3
57.3
57.9
54.5
23.4
43.7
59.0
33.7
10.0
6.8
4.3
9.3
11.7
Hyper-
Residence
outside
SMSAt
(%)
4.6
9.1
• Weighted.
t NHEFS, First National Health and Nutrition Examination Survey Epidemiotogic FoUowup Study; SMSA, Standard Metropolitan Statistical
Area
% Nonrecreatjonal.
§ Systolic blood pressure, 2140 mmHg; dlastollc blood pressure, £90 mmHg; or on medication for high blood pressure.
Am J Epidemiol
Vol. 144, No. 7, 1996
668
Gillum and Ingram
TABLE 3. Relative risks for ttroke Incidence associated wtth region of residence In whites aged 45-74
years, NHEFS,* 1971-1987
Sax and
Unadjusted
and region
person-years
Men
45-64 years
Northeast
Midwest
Southeast
West
65-74 years
Northeast
Midwest
Southeast
West
45-74 years
Northeast
Midwest
Southeast
West
Women
45-64 years
Northeast
Midwest
Southeast
West
65-74 years
Northeast
Midwest
Southeast
West
45-74 years
Northeast
Midwest
Southeast
West
Age adjusted
Risk aCfustedt
RR*
95% Cl*
RR
95% Cl
3.82
5.80
5.74
3.83
0.66
1.07
1.00
0.66
0.35-1.23
0.62-1.84
0.74
1.32
1.00
0.89
0.39-1.43
0.75-2.34
16.76
20.59
23.15
17.49
0.75
0.88
1.00
0.76
0.52-1.07
0.63-1.24
0.68
0.90
1.00
0.72
0.47-0.98
0.63-1.27
9.13
11.55
13.27
9.45
0.72
0.92
1.00
0.73
0.53-0.99
0.69-1.23
0.71
0.99
1.00
0.76
0.52-0.98
0.74-1.33
2.78
^33
3.98
3.36
0.72
0.60
1.00
0.82
0.37-1.37
0.30-1.20
0.85
0.81
1.00
0.94
0.43-1.69
0.40-1.65
15.02
11.72
18.05
14.13
0.81
0.63
1.00
0.76
0.58-1.14
0.45-0.90
0.86
0.70
1.00
0.82
0.61-1.22
0.49-1.00
7.53
6.35
10.74
7.96
0.79
0.63
1.00
0.77
0.59-1.07
0.46-0.86
0.85
0.73
1.00
0.85
0.62-1.16
0.53-1.00
0.37-1.18
0.54-1.05
0.55-0.97
0.45-1.50
0.55-1.04
0.58-1.02
0.49-1.62
0.51-1.00
0.57-1.02
0.50-1.76
0.59-1.14
0.64-1.13
* NHEFS, First National Health and Nutrition Examination Survey Epfdernjologic Followup Study; RR, relative
risk; Cl, confidence Interval.
t Adjusted for baseline age, smoking, history of diabetes, history of heart disease, education, systolic blood
pressure, alcohol use, and physical activity.
men, the incidence rates were higher in the Midwest
and Southeast than in the Northeast and West
In white men aged 45-74 years, the risk of stroke
was significantly lower in the northeast and west regions compared with the southeast region after adjusting for age (table 3). Relative risks were not changed
when all risk variables were added to the model. These
results suggest that, in white men, regional differences
in stroke risk exist that cannot be explained by the
stroke risk factors measured in NHANES I.
In white men aged 45-74 years, the incidence rates
for stroke varied little across the levels of urbanization
of the place of residence (table 4). There was no
significant association seen within the Southeast or
within all other regions combined between stroke risk
and urbanization level (table 4).
In white women aged 65-74 years, the incidence
rate of stroke was highest among those residing in the
southeast region, intermediate among those residing in
the northeast and west regions, and lowest among
those residing in the midwest region (table 3). The
same pattern was seen at ages 45-64 years. In white
women aged 45-64, 65-74, and 45-74 years residing
in the southeast region, the age-adjusted stroke risk did
not differ significantly compared with the northeast or
west regions. However, in the Midwest, the ageadjusted risk was significantly lower for women aged
65-74 and 45-74 years than in the Southeast (table 3).
This association was diminished but remained borderline significant after controlling for other risk variables. These findings suggest that only part of the
regional differences in stroke incidence in white
women can be explained by regional differences in the
risk factors measured in NHANES I.
White women aged 45-74 years residing outside an
SMSA had only slightly higher incidence rates of
Am J Epidemiol
Vol. 144, No. 7, 1996
Southeast Region and Stroke
669
TABLE 4. Relative I1sks for stroke Incidence associated with SMSA* residence tn whites and blacks
aged 45-74 years by region, NHEFS,• 1971-1987
Baseline region
and race
BaaeDne
residence
.
"
Unadjusted
Age adjusted
Risk adjustedt
person-years
RR*
95%CI»
RR
95% Cl
Central city
Suburb
Not SMSA
10.89
13.94
18.98
1.00
1.61
1.66
0.68-3.81
1.01-2.72
1.00
1.47
1.66
0.61-3.53
0.99-2.79
Central city
Suburb
Not SMSA
14.08
12.85
13.29
1.00
1.07
1.18
0.56-2.06
0.68-2.11
1.00
1.19
1.02
0.61-2.31
0.56-1.86
Central city
Suburb
Not SMSA
11.54
4.90
12.88
1.00
0.51
1.40
0.25-1.05
0.85-2.32
1.00
0.55
1.31
0.26-1.14
0.77-2.21
Central city
Suburb
Not SMSA
9.24
11.10
21.63
1.00
1.46
2.21
0.65-3.27
1.22-3.99
1.00
1.27
2.10
0.56-2.90
1.13-3.89
Central city
Suburb
Not SMSA
10.43
9.54
10.10
1.00
1.05
0.94
0.76-1.45
0.69-1.27
1.00
1.15
0.99
0.82-1.59
0.73-1.34
Central city
Suburb
Not SMSA
7.33
5.67
8.48
1.00
0.91
1.18
0.63-1.32
0.86-1.61
1.00
0.93
1.24
0.64-1.35
0.90-1.70
Southeast
An blacks
White men
White women
All other regions
All blacks
White men
White women
• SMSA, Standard Metropolitan Statistical Area; NHEFS, First National Health and Nutrition Examination
Survey EpkJemlologlc Followup Study; RR, relative risk; Cl, confidence interval.
t Adjusted for baseline age, sex (blacks), smoking, history of diabetes, history of heart disease, education,
systolic blood pressure, alcohol use, and physical activity.
stroke than did those in an SMSA central city or in an
SMSA suburb, where the rate was lowest (table 4). A
similar pattern was seen at ages 45-64 and 65-74
years (not shown). Rates for ages 45-74 years combined are shown in table 4. In regression models, there
was no significant effect of urbanization on risk of
stroke within the Southeast or all other regions combined (table 4).
In blacks aged 45-64 and 65-74 years, the incidence
rates for stroke were slightly higher in the Southeast than
in all other regions combined (not shown). At ages 45-74
years, the stroke risk in all other regions combined did
not differ significantly from that in the Southeast in
blacks: age- and sex-adjusted relative risk = 0.78, 95
percent confidence interval 0.56-1.07; risk-adjusted relative risk = 0.87, 95 percent confidence interval 0.631.21. Incidence rates were higher for blacks residing
outside SMSAs than for those residing in central cities of
SMSAs (table 4). Residence outside SMSAs was significantly associated with higher age- and sex-adjusted and
withrisk-adjustedriskof stroke compared with residence
in a central city (table 4).
Am J Epidemiol
Vol. 144, No. 7, 1996
DISCUSSION
In white men aged 45-74 years, residence in the
Southeast was associated with a 39 percent increased
age-adjusted risk of stroke incidence compared with
the northeast region. This excess risk could not be
explained by regional differences in multiple stroke
risk factors. In white women aged 45-74 years, only
part of the 59 percent excess in age-adjusted risk
associated with residence in the Southeast compared
with the Midwest could be explained by the regional
differences in risk factors measured in NHANES I. In
blacks, statistically significant regional differences
could not be demonstrated, probably because of limited statistical power. However, a strong association of
increased stroke risk with residence outside SMSAs in
blacks was demonstrated that was independent of region or other stroke risk factors. This is the first
nationwide study that attempts to explain geographic
variation in stroke incidence or mortality using individual-level variables and a prospective cohort study
design rather than more limited ecologic methods.
670
Gillum and Ingram
A recent analysis of 1989 Medicare data showed
high rates of hospitalization for stroke in the Southeast
and low rates in Mountain and Northeast states, consistent with NHEFS data for incidence (3). Incidence
is one of the chief determinants of hospitalization rates
together with prevalence, comorbidity, and factors related to medical practice and access to care. The lack
of variation in case fatality rates calculated from 15
through 180 days after admission for stroke in Medicare data suggests that variation in stroke mortality is
likely related to variation in stroke incidence and/or
other factors influencing hospital utilization (3). In
whites in 1965, the Nationwide Cerebrovascular Disease Morbidity Study found a higher incidence of
hospitalized stroke in three areas with high stroke
death rates (Upper Coastal Plain, North Carolina; Pee
Dee, South Carolina; Savannah, Georgia) than in three
low-rate areas (Miami, Florida; Denver, Colorado;
Flint Hills, Kansas) (19). Stroke hospital case fatality
did not vary consistently among areas. No data on
stroke risk factors were available in these studies.
NHEFS data are consistent with these studies in showing a high incidence of stroke in the Southeast as being
an important factor in the high stroke mortality in that
region (1-3, 19).
A previous study of regional variation in ischemic
heart disease incidence showed a pattern consistent
with patterns of mortality from the same cause,
namely, low incidence in the West compared with
other regions (20). Differences in baseline risk factors
could explain none of the difference between the West
and all other regions combined.
Few studies have examined urbanization as well as
region in relation to stroke mortality, and none have
examined stroke incidence. In an unpublished analysis
of data from the National Center for Health Statistics
for 1968-1985, the age-adjusted rate of stroke death
showed little effect of urbanization in the Midwest and
West. However, in the Southeast and Northeast, stroke
death rates were higher in nonmetropolitan areas than
in suburbs or central cities in both whites and blacks.
Death rates were higher in the Southeast than in other
regions for residents of small metropolitan and nonmetropolitan areas. The current NHEFS analysis is not
inconsistent with this demonstration of independent
effects of urbanization and region on stroke mortality.
Similarity in the geographic distribution of stroke
mortality and of several major stroke risk factors has
led to the suggestion that these factors may explain the
variation in stroke mortality (1, 21, 22). However,
major inconsistencies in the patterns of risk factors and
stroke mortality are consistent with NHEFS findings
that standard risk factors explain at most only part of
the variation in stroke mortality. A few analyses have
examined the variation in prevalence of hypertension
among US regions (23, 24). An analysis of data from
the Second National Health and Nutrition Examination
Survey (NHANES IT) (1976-1980) found the prevalence of hypertension to be higher in the Southeast (44
percent) than in all other regions combined (34 percent) in black females; this difference was not found
for other sex and race groups (23). Further, hypertension control was lower for black females in the Southeast than in other regions. In NHANES I, the mean
systolic blood pressure was slightly higher in the
Southeast than in other regions for each sex and race
group for both the NHEFS cohort (table 2) and the
entire NHANES I sample (13). An analysis of National Health Interview Survey data for 1983-1987
found a higher prevalence of self-reported hypertension in the southeast region than in other regions in
both blacks and whites of both sexes, consistent with
earlier reports (24, 25). The prevalence was highest
among persons residing outside SMSAs in the Southeast and Midwest (24).
Consistent with the stroke incidence pattern, the
prevalence of cigarette smoking was highest in the
southeast region and lowest in the northeast region in
white men and highest in the Southeast and lowest in
the Midwest in white women in the NHEFS cohort
(table 2) and in previous reports of National Health
Interview Survey data (24). However, this pattern was
not seen for black men or women. The prevalence of
self-reported heart disease was also highest in the
Southeast compared with other regions among white
men and white women (24). The prevalence of selfreported diabetes was reported to be highest in the
southeastern region and in the states from the eastern
and north central regions (26), consistent with NHEFS
findings for white men. The prevalence of diabetes at
age 35-54 years was the only stroke risk factor consistently highest in Savannah, Georgia, intermediate in
Hagerstown, Maryland, and lowest in Pueblo, Colorado, duplicating the gradient of stroke mortality rates
in the Three-area Epidemiologic Study (27).
Ecologic studies have suggested a relation of low
socioeconomic status and high stroke mortality (28,
29), but such an association is inconsistently reported
in cohort studies. As shown in table 2 and reported
elsewhere, the education level in the Southeast was
consistently lower than in other regions (24). However, controlling for low educational attainment in
addition to other risk factors did not explain regional
differences in stroke risk.
Despite apparent ecologic correlations of stroke occurrence and risk factors, NHEFS analyses of individuallevel data revealed that multiple stroke risk factors failed
to explain regional differences in stroke incidence in
Am J Epidemiol
Vol. 144, No. 7, 1996
Southeast Region and Stroke
white men and only partly explained differences in white
women. It is possible that risk factors other than those
included in the present analysis might be responsible for
observed regional variations. Other possible stroke risk
factors not considered in the present analysis that seem
unlikely to explain regional variation include sickle cell
disease or trait, oral contraceptive use, elevated hematocrit, climate, and syphilis (1).
It has been suggested that the diet in the Southeast
may resemble the diet associated with high stroke
mortality in Japan in being lower in animal protein and
higher in grain-derived complex carbohydrates and
sodium than diets of other US regions (1). It may also
be lower in potassium and calcium, which also have
been suggested as nutrients inversely associated with
stroke occurrence. These patterns might be especially
pronounced in the nonmetropolitan areas of the Southeast and among blacks.
In the Nationwide Food Consumption Survey of
1977-1978, 3-day food records (24-hour recall plus
2-day food record) for individuals showed only
slightly lower meat group intake and similar fish intake in the Southeast and in other regions, but much
lower fruit intake in the Southeast than in other regions
(30). About half the protein intake was from the meat
group in all regions, but a lower portion was from the
milk group in the South. Average nutritive values were
lower in the Southeast compared with other regions for
food energy, protein, calcium, and all other nutrients
except iron. However, protein intake in the Southeast
was still 161 percent of the Recommended Daily Allowance, but calcium intake was only 78 percent of
this measure. Expressed as the intake per 1,000 kcal,
protein intake was similar and calcium intake lower in
the Southeast compared with those intakes in other
regions.
Higher stroke risk in the Southeast in whites and in
nonmetropolitan areas in blacks could also be related
to a lower access to and utilization of preventive health
services in these areas that are not fully controlled for
by baseline risk factor levels. For example, a longer
duration of undetected and/or uncontrolled hypertension in blacks in nonmetropolitan areas than in SMSAs
may not be fully reflected in baseline blood pressure
differences. The use of preventive services tended to
be lower in the southeast region and in nonmetropolitan areas in national surveys in 1973 and 1982 (31,
32). The finding that none of the association with
SMSA in blacks in the Southeast was explained by
other stroke risk factors is consistent with this interpretation (table 4).
Limitations of the study include possible bias arising from loss to follow-up, missing data on baseline
risk variables, and, for the incidence analysis, missing
Am J Epidemiol
Vol. 144, No. 7, 1996
671
hospital data at follow-up. A further source of possible
bias was misclassification due to inaccuracy of diagnoses coded on death certificates and hospital records.
Inaccuracy of baseline history of stroke could result in
cases being identified during follow-up that were actually diagnosed prior to baseline. Further, for stroke
incidence the year of onset was taken as the year of
first hospitalization, but this could be in error if an
earlier hospitalization for stroke was not reported or
the records not found.
Errors in measurement of baseline variables could
also result in bias, but this seems unlikely given the
standardized nature of the NHANES I examination.
One exception was baseline smoking data, which were
derived or imputed from follow-up interview questions for about half the cohort. However, this approach
has been judged valid for analyses of mortality (15).
Controlling for baseline history of hypertension in
addition to baseline systolic blood pressure did not
alter the results. The relatively long interval between
the baseline examination and the occurrence of stroke
events could produce bias toward the null because of
unmeasured changes in residence prior to and after
baseline. Of particular concern is the migration of
healthy elderly from the Northeast to Florida (33).
This also would tend to produce bias toward the null
by lowering rates for the Southeast and raising them
for the Northeast. Persons not changing residence may
still have experienced a change in urbanization level.
Further, misclassification of urbanization may have
occurred because data from the 1960 Census were
used for the baseline NHANES I classification. The
impact of this was likely minimal for comparisons of
metropolitan versus nonmetropolitan areas. Comparisons of the distribution of residence in SMSAs among
the NHEFS sample and the 1960 and 1970 Census
data revealed that the weighted percentage not in
SMSAs shown in table 2 was lower than for all ages in
the census for the Southeast, perhaps because regions
are defined differently by the census (34). Examination surveys of the National Center for Health Statistics have found lower response rates in large metropolitan areas but higher response rates in
nonmetropolitan areas in the Southeast compared with
other regions (35, 36). Therefore, results for the analyses of urbanization should be interpreted with caution, even though the use of separate models for the
Southeast and "not Southeast" should minimize potential bias.
The location of the "Stroke Belt" may have shifted
from the southeastern coastal plain to the Mississippi
River valley (2, 37, 38). However, this would not
affect comparisons by regions as defined here.
NHANES I examination locations in the Southeast
672
Gillum and Ingram
included both coastal plain counties and other counties
(4, 5); hence, persons with and without "Stroke Belt"
risk factor levels and other exposures at baseline were
captured in this analysis.
Confounding by variables not measured cannot be
excluded. Statistical power was relatively low for
blacks and did not permit sex-specific analyses. The
association of urbanization with stroke incidence may
be confounded by within-region geographic differences. For example, in the Southeast, many urban
centers lie disproportionately outside the coastal plain
of North Carolina, South Carolina, and Georgia (one
definition of the historical "Stroke Belt"), while nonmetropolitan counties with large numbers of blacks
occurred in the coastal plain and other areas (37-39).
In summary, among persons aged 65 years and over,
stroke incidence was higher in the Southeast than in
other regions. These differences were statistically significant in whites and could not be explained by controlling for multiple risk factors in white men. Risk
factors explained only part of the difference in white
women. Blacks living in nonmetropolitan areas were
at increased risk of stroke independent of region or
risk factors. The higher stroke mortality observed in
the Southeast appears to be associated with higher
stroke incidence. Further research is needed to determine the likely causes of this excess morbidity and
mortality.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
ACKNOWLEDGMENTS
The NHANES I Epidemiologic Followup Study has been
developed and funded by these agencies: National Center
for Health Statistics; National Institute on Aging; National
Cancer Institute; National Center for Chronic Disease Prevention and Health Promotion; National Institute of Child
Health and Human Development; National Heart, Lung, and
Blood Institute; National Institute on Alcohol Abuse and
Alcoholism; National Institute of Mental Health; National
Institute of Diabetes and Digestive and Kidney Diseases;
National Institute of Arthritis and Musculoskeletal and Skin
Diseases; National Institute of Allergy and Infectious Diseases; National Institute of Neurological and Communicative Disorders and Stroke; and the US Department of Agriculture. The field work was conducted by Westat, Inc.
14.
15.
16.
17.
18.
19.
20.
REFERENCES
1. Lanska DJ. Geographic distribution of stroke mortality in the
United States: 1939-41 to 1979-1981. Neurology 1993;43:
1839-51.
2. Casper ML, Wing S, Anda RF, et al. The shifting Stroke Belt:
changes in geographic pattern of stroke mortality in the United
States, 1962 to 1988. Stroke 1995;26:755-60.
3. Lanska DJ, Kryscio R. Geographic distribution of hospitaliza-
21.
22.
23.
24.
tion rates, case fatality, and mortality from stroke in the United
States. Neurology 1994;44:1541-50.
Miller HW. Plan and operation of the Health and Nutrition
Examination Survey, United States, 1971-1973. Rockville,
MD: National Center for Health Statistics, 1973. (Vital and
health statistics, Series 1, no. 10a) (DHEW publication no.
(HSM) 73-1310).
Engel A, Murphy RS, Maurer K, et al. Plan and operation of
the HANES I augmentation survey of adults 25-74 years,
United States, 1974-1975. Rockville, MD: National Center
for Health Statistics, 1978. (Vital and health statistics, Series
1, no. 14) (DHEW publication no. (PHS) 78-1314).
Cohen BB, Barbano HE, Cox CS, et al. Plan and operation of
the NHANES I Epidemiologic Followup Study: 1982-84.
Hyattsville, MD: National Center for Health Statistics, 1987.
(Vital and health statistics, Series 1, no. 22) (DHHS publication no. (PHS) 87-1324).
Cox CS, Rothwell ST, Madans JH, et al. Plan and operation of
the NHANES I Epidemiologic Followup Study: 1987. Hyattsville, MD: National Center for Health Statistics, 1992. (Vital
and health statistics, Series 1, no. 27) (DHHS publication no.
(PHS) 92-1303).
Madans JH, Kleinman JC, Cox CS, et al. Ten years after
NHANES I: report of initial follow-up, 1982-84. Public
Health Rep 1986; 101:465-73.
Madans JH, Cox CS, Kleinman JC, et al. Ten years after
NHANES I: mortality experience at initial follow-up,
1982-84. Public Health Rep 1986;101:474-81.
Finucane FF, Madans JH, Bush TL, et al. Decreased risk of
stroke among postmenopausal hormone users: results from a
national cohort Arch Intern Med 1993;153:73-9.
White LR, Losonczy KG, Wolf PA. Cerebrovascular disease.
In: Cornoni-Huntley JC, Huntley RR, Feldman JJ, eds. Health
status and well-being of the elderly. New York: Oxford University Press, 1990:115-35.
National Center for Health Statistics. HANES I hematology
and clinical chemistry procedures developed or utilized by the
Center for Disease Control, Bureau of Laboratories,
1971-1975. Washington, DC: US GPO, 1979.
Roberts J, Maurer K. Blood pressure levels of persons 6-74
years, United States, 1971-74. Rockville, MD: National Center for Health Statistics, 1977. (Vital and health statistics,
Series 11, no. 203) (DHEW publication no. (HRA) 78-1648).
McLaughlin JK, Dietz MS, Mehl ES, et al. Reliability of
surrogate information on cigarette smoking by type of informant Am J Epidemiol 1987;126:144-6.
Machlin SR, Kleinman JC, Madans JH. Validity of mortality
analysis based on retrospective smoking information. Stat
Med 1989;8:997-1009.
SAS Institute, Inc. SAS/STAT software: changes and enhancements, release 6.07. Cary, NC: SAS Institute, Inc, 1992:
433-80. (SAS technical report P-229).
Software for survey design analysis (SUDAAN), version 5.30.
Research Triangle Park, NC: Research Triangle Institute,
1992.
Ingram D, Makuc D. Statistical issues in analyzing the
NHANES I Epidemiologic Followup Study. Hyattsville, MD:
National Center for Health Statistics, 1994. (Vital and health
statistics, Series 2, no. 121).
Kuller L, Anderson H, Peterson D, et al. Nationwide cerebrovascular disease morbidity study. Stroke 1970;l:86-99.
Garg R, Madans JH, Kleinman JC. Regional variation in
ischemic heart disease incidence. J Clin Epidemiol 1992;45:
149-56.
Gillum RF. Cerebrovascular disease morbidity in the United
States, 1970-1983. Age, sex, region, and vascular surgery.
Stroke 1986;17:656-61.
Gillum RF. Stroke in blacks. Stroke 1988; 19:1-9.
Roccella EJ, Lenfant C. Regional and racial differences
among stroke victims in the United States. Clin Cardiol 1989;
12(suppl 4):IV18-lV22.
Gillum RF. Prevalence of cardiovascular and pulmonary disAm J Epidemiol
Vol. 144, No. 7, 1996
Southeast Region and Stroke
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
eases and risk factors by region and urbanization in the United
States. J Natl Med Assoc 1994;86:105-12.
Wilder CS. Prevalence of chronic circulatory conditions,
United States, 1972. Rockville, MD: National Center for
Health Statistics, 1974. (Vital and health statistics, Series 11,
no. 94) (DHEW publication no. (HRA) 75-1521).
Regional variation in diabetes mellitus prevalence—United
States, 1988 and 1989. MMWR Morb Mortal Wkly Rep
1990;39:805-9.
Stolley PD, Kuller LH, Nefzger MD, et al. Three-area Epidemiological Study of geographical differences in stroke mortality. H. Results. Stroke 1977;8:551-7.
Casper M, Wing S, Strogatz D. Variation in the magnitude of
black-white differences in stroke mortality by community
occupational structure. J Epidemiol Community Health 1991;
45:302-6.
Siegel PZ, Deeb LC, Wolfe LE, et al. Stroke mortality and its
socioeconomic, racial, and behavioral correlates in Florida.
Public Health Rep 1993;108:454-8.
US Department of Agriculture. Food and nutrient intakes:
individuals in four regions, year 1977-78. Nationwide Food
Consumption Survey 1977-78, report 1-3. Washington, DC:
US GPO, 1983.
Bloom B. Use of selected preventive care procedures, United
States, 1982. Hyattsville, MD: National Center for Health
Statistics, 1986. (Vital and health statistics, Series 10, no. 157)
(DHHS publication no. (PHS) 86-1585).
Moss AJ, Wilder MH. Use of selected medical procedures
associated with preventive care, United States, 1973. Hyattsville, MD: National Center for Health Statistics, 1977. (Vital
and health statistics, Series 10, no. 110) (DHEW publication
no. (HRA) 77-1538).
Lanska DJ, Peterson PM. Effects of interstate migration on the
geographic distribution of stroke mortality in the United
States. Stroke 1995;26:554-61.
US Bureau of the Census. Historical statistics of the United
States, colonial times to 1970, part 1. Washington, DC: US
Am J Epidemiol
Vol. 144, No. 7, 1996
673
GPO, 1975.
35. Gordon T, Miller HW. Cycle I of the Health Examination
Survey: sample and response, United States, 1960-1962.
Washington, DC: National Center for Health Statistics, 1974.
(Vital and health statistics, Series 11, no. 1) (DHEW publication no. (PHS) 1000).
36. Forthofer RN. Investigation of nonresponse bias in NHANES
n. Am J Epidemiol 1983;117:507-15.
37. Howard G, Evans GW, Pearce K, et al. Is the Stroke Belt
disappearing? Stroke 1995;26:1153-8.
38. Howard G, Howard VJ. The end of the Stroke Belt? It may be
too early to declare victory. Stroke 1995;26:1150-2.
39. Lanska DJ, Kuller LH. The geography of stroke mortality in
the United States and the concept of a Stroke Belt, Stroke
1995;26:1145-9.
APPENDIX
As used in this study, the regions of the United States are
defined as follows: Northeast Maine, Vermont, New
Hampshire, Massachusetts, Connecticut, Rhode Island,
New York, New Jersey, and Pennsylvania; Midwest Ohio,
Illinois, Indiana, Michigan, Wisconsin, Minnesota, Iowa,
and Missouri; Southeast Delaware, Maryland, District of
Columbia, West Virginia, Virginia, Kentucky, Tennessee,
North Carolina, South Carolina, Georgia, Florida, Alabama,
Mississippi, and Louisiana; and West: Washington, Oregon,
California, Nevada, Texas, New Mexico, Arizona, Idaho,
Utah, North Dakota, South Dakota, Colorado, Montana,
Wyoming, Alaska, Kansas, Nebraska, Oklahoma, Arkansas,
and Hawaii.
These definitions differ from those of the US Bureau of
the Census.