American Journal of Epidemiology
Copyright C 1996 by Tha Johns Hopkins University School of Hygiene and Public Health
All rights reserved
Vol. 144, No. 9
Printed In U.S.A
Association of Transient Ischemic Attack/Stroke Symptoms Assessed by
Standardized Questionnaire and Algorithm with Cerebrovascular Risk
Factors and Carotid Artery Wall Thickness
The ARIC Study, 1987-1989
Uoyd E. Chambless,1 Eyal Shahar,2 A. Richey Sharrett,3 Gerardo Heiss,4 Louis Wijnberg,1
Catherine C. Paton,4 Paul Soriie,3 and James F. Toole5
The baseline examination (1987-1989) for the Atherosclerosis Risk in Communities (ARIC) Study was
conducted in 15,792 free-living residents aged 45-64 years in four geographically dispersed US communities.
A questionnaire on symptoms of transient ischemic attack (TIA) and stroke was evaluated by computer
algorithm for 12,205 of these participants. Data were also collected on lipoprotein levels, hemostasis,
hematology, anthropometry, blood pressure, medical history, lifestyle, socioeconomic status, and medication
use. Noninvasive high resolution B-mode ultrasonographic imaging was used to determine carotid arterial
intimal-medial wall thickness (IMT). The cross-sectional relation between the prevalence of TIA/stroke symptoms and putative risk factors was assessed by logistic regression, controlling for age and community. Odds
ratios for TIA/stroke symptoms were significantly elevated (p < 0.01) for diabetes mellitus, current smoking,
hypertension, lower levels of education, income, and work activity, and higher levels of lipoprotein(a), IMT,
hemostasis factor VIII, and von Willebrand factor. However, the relations with education and carotid IMT were
not present for black Americans. In whites, the relations of TIA/stroke symptoms to IMT were nonlinear. Only
at extreme levels of IMT were symptoms substantially more frequent: For example, men with an IMT greater
than 1.17 mm or women with an IMT greater than 0.85 mm had approximately twice the odds of having
positive TIA/stroke symptoms as those with lower IMTs. The authors plan in future analyses to address the
issue prospectively, as well as to examine the relation with magnetic resonance imaging-defined outcomes
and clinically defined incident stroke. Am J Epidemiol 1996; 144:857-66.
atherosclerosis; cerebral ischemia, transient; cerebrovascular
groups; risk factors; ultrasonography
disorders; coronary
disease; ethnic
(ARIC) Study provides researchers with an opportunity to assess risk factor relations with TIA/stroke in a
large, population-based cross-sectional and prospective study of white and African-American men and
women, either in terms of self-reported history of a
physician's diagnosis or in terms of responses to an
extensive questionnaire on TIA/stroke symptoms. In
this paper, we address cross-sectional relations using a
diagnostic algorithm based on the symptom questionnaire. The ARIC Study also provides the opportunity
to assess the relation between TIA/stroke symptoms
and a marker for carotid atherosclerosis, intimalmedial arterial wall thickness (IMT), as measured by
B-mode ultrasound. In the near future, we plan to
address the same issue in a prospective mode after 6
years of follow-up, as well as the relation of positive
TIA/stroke symptoms to outcomes determined by
magnetic resonance imaging. Thereafter, we will assess the relations between the baseline TIA/stroke
symptoms and subsequent clinically defined stroke.
Editor's note: A companion article by Toole et al.
appears on page 849 of this issue.
Few studies have identified risk factors for transient
ischemic attack (TIA) in a population of free-living
individuals. The Atherosclerosis Risk in Communities
Received for publication November 2, 1995, and in final form
April 18, 1996.
Abbreviations: ARIC, Atherosclerosis Risk in Communities; HDL,
high density lipoprotein; IMT, irrtlmal-medial [wall] thickness; TLA,
transient Ischemic attack.
1
Department of Biostatistics, School of Public Health, University
of North Carolina, Chapel Hill, NC.
2
Division of Epidemiology, School of Public Health, University of
Minnesota, Minneapolis, MN.
3
National Heart, Lung, and Blood Institute, Bethesda, MD.
4
Department of Epidemiology, School of Public Health, University of North Carolina, Chapel Hill, NC.
5
Department of Neurology, Bowman Gray School of Medicine,
Wake Forest University, Winston-Salem, NC.
Reprint requests to Dr. Uoyd E. Chambless, Collaborative Studies Coordinating Center, University of North Carolina, 137 East
Franklin Street, Suite 203, Chapel Hill, NC 27514-4145.
857
858
Chambless et al.
Altogether, this program of analyses should allow us
to assess the utility of ARIC's TIA/stroke questionnaire for use in large-scale epidemiologic studies.
MATERIALS AND METHODS
The ARIC baseline examination was conducted between 1987 and 1989 in a cohort of 15,792 free-living
residents who comprised a probability sample of persons aged 45-64 years in four geographically dispersed US communities: Forsyth County, North Carolina; Jackson, Mississippi (an all-black sample);
selected suburbs of Minneapolis, Minnesota; and
Washington County, Maryland. The general design of
the ARIC Study has been more fully described elsewhere (1). Among whites, participation rates were 67
percent for men and 68 percent for women; rates were
similar across centers (2). Among blacks, participation
rates were 42 percent for men and 49 percent for
women and were similar between Jackson and Forsyth
County. Some health information on nonparticipants
was available from a question on general health status
(2): There was little difference between black participants and nonparticipants, but among whites, the participants reported better general health than the nonparticipants, and the difference was greater among
men than among women.
Examination and laboratory procedures (lipoproteins (3), hemostasis (4-7), anthropometry (8), blood
pressure (9), medication use (8), and ultrasound measurements of the carotid artery (10-13)) are fully
described in the ARIC Manual of Operations and other
publications. Data were also collected on hematology,
diabetes mellitus, smoking, exercise, occupation, education, and family income.
For assessment of symptoms of TIA or stroke, participants were asked in a standardized manner whether
they had ever experienced the sudden onset of any one
of six neurologic symptoms: loss of vision, double
vision, speech dysfunction, weakness or paralysis,
numbness or tingling, or dizziness. For each of the six
symptoms, questions on duration and concomitant
symptoms were asked, and for double vision, numbness, and dizziness, questions were also asked to ascertain a possible noncerebrovascular cause. Only
symptoms with a sudden onset and a duration of at
least 30 seconds were considered. Neither dizziness
alone, nor numbness or paralysis in one body part
alone, nor double vision which went away upon closing of one eye was sufficient to be termed a "positive
TLA/stroke symptom."
The questionnaire was also used in the Asymptomatic Carotid Atherosclerosis Study clinical trial (14).
An expert panel of Asymptomatic Carotid Atherosclerosis Study neurologists chose various combinations
of symptoms likely to be indicative of TLA or retinal or
cerebral infarct, thereby providing the basis for our
computer TIA/stroke symptom algorithm. This "diagnostic" profile was also further refined into separate
"TIA symptoms" or "stroke symptoms" variables,
based on the duration of the symptoms (<24 hours or
^24 hours), and into "location" categories: right carotid artery, left carotid artery, or vertebrobasilar artery. More details on the questionnaire, the algorithm
and its sensitivity and specificity, and the prevalence
of symptoms are presented in a companion paper (15).
Noninvasive high resolution B-mode ultrasonographic imaging was used to determine the extent of
extracranial carotid atherosclerosis. Arterial mean
IMT was measured at the common carotid artery, the
bifurcation, and the internal carotid arteries on the left
and right sides; values for any missing sites out of the
six were imputed by a maximum likelihood procedure
using BMDP5V (16), simultaneously adjusting for
differences between readers at the ultrasound reading
center and for drift over time. Then the mean IMT
over all sites was calculated.
Hypertension was defined by the average of the
second and third sitting blood pressure measurements
being ^140/90 mmHg or by the current use of antihypertensive medication. Persons were classified as
current cigarette smokers, ex-smokers, or never smokers from their responses to questions about smoking
habits. Body mass index was defined as weight (kg)
divided by height (m) squared. Diabetes mellitus was
defined by a nonfasting blood glucose level of at least
200 mg/dl, a fasting level of at least 140 mg/dl, a
previous physician diagnosis, or treatment for diabetes
during the previous 2 weeks. The Baecke questionnaire and scale (17) were used to quantify usual physical activity. The questionnaire included 16 items asking about usual levels of exertion, and indices ranging
from 1 (low) to 5 (high) were derived for physical
activity in work, leisure time, and sports. Educational
status and annual family income were ascertained
from questionnaire responses.
Because of strong age and field center relations with
the prevalence of TIA/stroke .symptoms, as noted in
the companion paper (15), we report here field centerand age-adjusted mean values for continuous risk variables and adjusted proportions for categorical variables, by TLA/stroke symptom status, as estimated by
linear regression (for continuous variables) or logistic
regression (for dichotomous variables), stratifying according to race/sex. For a summary test over all race/
sex categories of the age- and field center-adjusted
differences between positive and negative TIA/stroke
symptom status, the category-specific Z values (difference/standard error of the difference) were combined
Am J Epidemiol
Vol. 144, No. 9, 1996
TIA/Stroke Symptoms and Cerebrovascular Risk Factors
linearly with "weights" equal to the square roots of the
proportions of the entire sample in the four race/sex
strata, so that the resulting test statistic was approximately normally distributed under the null hypothesis
of no relation between the risk factor and TIA/stroke
symptoms. The test was most sensitive when differences pointed in the same direction in the various
strata. Site-specific, adjusted mean carotid arterial
IMT, as well as the mean value over all sites, was
assessed in a multivariate model via PROC MIXED of
the SAS software package (18) to account for a high
frequency of missing data.
The adjusted means or proportions of potential risk
factors that showed significant differences between
study participants with and without TIA/stroke symptoms—at a screening level of p < 0.1 in at least two
age/sex strata or in only one stratum if there was
consistency in the direction of the relation across all
four strata—are presented, along with data on several
variables of a priori interest. Variables that were considered but did not meet these criteria are listed.
Because physical activity level might have been
affected by a participant's perception of health status
as reflected in answers on the TLA/stroke questionnaire, and because of the potentially different interpretations of this variable between those who were currently employed versus those not employed, physical
activity at work was not considered for further multiple risk factor analysis. Because some of the TIA/
stroke symptoms considered could be related to diabetic neuropathy, analysis was stratified by diabetes
status, and because of small sample sizes only the
results for nondiabetics are presented. Family income,
educational level, and arterial IMT were not included
in multiple risk factor analysis, because the first two
are likely determinants of the other risk factors being
considered and because IMT may be an intermediate
variable in the relation between other risk factors and
TIA/stroke symptoms. Thus, for each of these four
variables, prevalence odds ratios for positive TLA/
stroke symptoms were obtained in a separate logistic
regression model, adjusting only for age and field
center. For the continuous variables, the odds ratio was
calculated for a 1-standard-deviation difference in the
risk factor level, where the standard deviation was
computed for all four strata combined.
Using the p < 0.1 screening rule given above, risk
factors other than education, income, LMT, and work
activity level were selected for processing in a sexand race-specific multiple risk factor logistic regression model excluding diabetics. This model was simplified to exclude potential risk factors showing no
significant effect in any of the race/sex groups and not
notably affecting the size of the coefficients of other
Am J Epidemiol
Vol. 144, No. 9, 1996
859
variables in the model. In all presentations of odds
ratios from logistic regression models, we include
results from a summary test over all race/sex strata
similar to the one described for means and proportions. Nonlinearity of the effect of the significant continuous variable was considered by replacing the continuous variables with categorical variables to
represent sex- and race-specific tertiles.
As described in the companion paper (15), there was
not a substantial overlap between positive TIA/stroke
symptoms and positive responses to the question, "Has
a physician ever told you that you've had a stroke,
TLA, or ministroke?" However, to check that the relations with risk factors were not entirely due to those
responding "Yes" to this question, we excluded all
such persons and repeated the final logistic regression
analyses. Furthermore, to check that the relations were
similar for stroke symptoms (those lasting at least 24
hours) and TLA symptoms, we fitted logistic regression models separately for symptoms lasting at least
24 hours and those lasting less.
RESULTS
Of the 15,792 ARIC participants, we excluded 412
without a TLA/stroke questionnaire and another 2,881
with only partial data, due almost entirely to the early
use of a discontinued version of the questionnaire. An
additional 217 participants whose diagnostic computer
algorithm outcome for TLA/stroke was missing were
excluded as well. Also excluded were 7 participants
whose age was outside of the 44- to 65-year range, 31
participants who were neither African-American nor
white, and the 39 nonwhite participants from the
Washington County and Minneapolis field centers,
which left 12,205 persons for this study. Race- and
sex-specific sample sizes and numbers of participants
with positive TLA/stroke symptoms are presented in
table 1.
TABLE 1. Numbers of participants with positive stroke or
TlA*/stroke symptoms, by race and sex: The ARIC* Study,
1987-1989
Race and sax
Stroke
symptoms
TIA or stroke
symptoms
Sample size
Black
Female
Male
55
122
33
55
White
Female
Male
107
89
343
216
4,685
4,187
Total
284
736
12,205
2,085
1,248
• TIA, transient tschemic attack; ARIC, Atherosclerosis Risk in
Communities.
860
Chambless et al.
TABLE 2. Age- and field canter-adjusted mean values for continuous potential csrsbravascular risk factors and intimal-fnedlal
wall thickness, by occurrence ofTIA*/8troke I symptoms, race, and ssx: The ARIC* Study, 1987-1989
Black
White
TIA or stroke
symptoms
Female
Male
Female
Male
White blood cell count
(thousands)
Yas
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
225.2
217.6
125.2
111.6
140.1
144.3
97.6
93.2
198.2
170.1
55.7
57.8
17.2
17.7
38.5
40.1
144.5
137.9
174.3
154.8
354.9
328.8
133.0
125.2
120.6
117.4
151.6
137.6
5.9
5.7
207.9
210.4
120.3
118.0
123.6
132.4
94.0
94.9
153.1
148.6
47.1
50.8
12.5
14.0
34.5
36.8
137.0
136.8
152.8
145.5
323.0
309.4
109.8
110.5
116.3
113.6
139.7
132.4
5.9
5.6
219.5
218.9
138.4
129.2
139.9
142.6
96.0
93.5
91.9
87.0
54.8
57.5
16.1
17.0
38.7
40.5
137.2
135.7
138.3
130.7
304.8
299.8
130.6
127.9
114.0
112.4
116.2
111.9
6.5
6.1
212.5
209.8
153.4
147.5
117.2
118.7
98.4
96.6
98.0
74.9
41.2
42.5
10.6
10.3
30.5
32.2
140.4
139.1
131.6
125.0
299.3
293.9
110.5
113.4
112.3
108.7
126.5
113.0
7.1
6.4
Systolic Wood pressure
(mmHg)
Yes
No
132.0
129.6
134.5
131.9
118.7
117.9
119.9
120.2
0.243
Diastolic blood pressure
(mmHg)
Yes
No
77.5
77.3
82.0
82.3
69.3
69.7
73.4
73.3
0.749
Body mass Index}
Yes
No
Yes
No
Yes
No
31.0
30.8
0.92
0.91
1.79
2.24
0.73
0.73
28.9
27.5
0.95
0.94
1.84
2^7
0.79
0.79
27.4
26.6
0.91
0.89
1.82
1.95
0.73
0.70
28.1
27.4
0.98
0.97
2.09
2.34
0.84
0.79
<0.001
Risk factor
Total cholesterol (mg/dl)
Triglyceride (mg/dl)
ApoQpoprotein Al (mg/dl)
Apolipoprotein B (mg/dl)
LJpoproteln(a) (ng/ml)
HDL* cholesterol (mg/dl)
HDLt cholesterol (mg/dl)
H D L J cholesterol (mg/dl)
LDL* cholesterol (mg/dl)
Factor VIII (%)
Rbrinogen (mg/dl)
Factor VI1 (%)
Anttthrombin III (%)
von WIDebrand factor (%)
Waisthlp ratio
Work activity lndex§
1 nil maJ-medial wall
thickness (mm)
Yes
No
pvaluet
0.173
0.008
0.009
0.041
0.004
<0.001
0.224
<0.001
0.150
<0.001
<0.001
0.292
0.002
<0.001
<0.001
<0.001
<0.001
<0.001
* TIA, transient Ischamlc attack; ARIC, Atherosclerosis Risk In Communities; HDL, high density Bpoprotein; LDL, low density llpoprotein.
t p value for test of the difference In mean values by TIA/stroke symptom status over all race/sex groups.
t Weight (kgVhelght (m)1.
§ Index from the Baecke questionnaire and scale (17).
Adjusted means and proportions by TIA/stroke
symptom status
In comparisons of the field center- and age-adjusted
mean values (table 2) and proportions (table 3) between study participants who did and did not experience TIA/stroke symptoms, overall statistically significant (p < 0.05) differences that were consistent in
direction in all four race/sex strata were found for
diabetes, hypertension, work activity index, family
income, factor Vm, von Willebrand factor, fibrinogen,
antithrombin m, current (vs. never) smoking, triglycerides, apolipoprotein A-I, lipoprotein(a), high density
lipoprotein (HDL) cholesterol, HDL3 cholesterol,
white blood cell count, body mass index, and waist:
Am J Epidemiol
Vol. 144, No. 9, 1996
TWStroke Symptoms and Cerebrovascular Risk Factors
861
TABLE 3. Age- and field center-adjusted proportions for dichotomous potential cerebrovascular risk factors, by occurrence of
TIA*/stroke symptoms, race, and isex: The ARIC* Study, 1987-198J)
Black
TIA or stroke
symptoms
FUsk factor
Hypertension
Family Income £$25,00Q/year
Educational level £ high school
Diabetes mellitus
Current smoking
Former smoking
White
pvaluef
Female
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Male
0.75
0.54
0.13
0.25
0.59
0.61
0.35
0.17
0.28
0.25
0.19
0.18
0.72
0.54
0.23
0.38
0.58
0.56
0.21
0.16
0.46
0.40
0.32
0.34
Female
Male
0.34
0.26
0.64
0.70
0.83
0.84
0.11
0.06
0.31
0.24
0.24
0.25
0.37
0.28
0.78
0.80
0.82
0.83
0.17
0.07
0.31
0.24
0.47
0.47
<0.001
<0.001
0.001
<0.001
<0.001
0.751
• TIA, transient ischemic attack; ARIC, Atherosclerosis Risk in Communities.
t p value for test of the difference in proportions by TIA/stroke symptom status over all race/sex groups.
hip ratio. The carotid IMT differences were statistically significant overall, but were nonzero only among
whites. The variables meeting the screening rule, as
well as low density lipoprotein cholesterol, apolipoprotein B, HDL^ cholesterol, total cholesterol, and
blood pressure, were included in tables 2 and 3. The
other variables not found to be important by this
screening (alcohol, leisure activity index, sports activity index, hematocrit, hemoglobin, activated partial
thromboplastin time, and protein C) are not listed in
tables 2 and 3 and were dropped from further consideration.
In analysis of site-specific IMT by TIA/stroke
symptom status (not shown), results for blacks were
inconsistent across carotid artery sites. For whites,
however, the group with TIA/stroke symptoms had
greater IMTs by 0.03-0.05 mm over the six sites (p =
0.0001 for females; p = 0.0008 for males), with dif-
ferences generally being larger at the bifurcation and
internal carotid arteries than at the common artery.
Age- and field center-adjusted prevalence odds
ratios
Among males, the prevalence odds ratios for a
5-year age increment were 1.3-1.4 and statistically
significant, but in females the odds ratios were close to
unity. There were strong field center differences, especially for blacks. Table 4 shows the odds ratios for
TIA/stroke symptoms at different levels of education,
income, diabetes, work activity level, and carotid artery IMT, controlling only for age and field center.
Results were similar in direction to those obtained in
the comparison of adjusted means and proportions—
positive for diabetes and carotid IMT and negative for
work activity, education, and income—and all yielded
TABLE 4. Odds ratios for TlAt/stroke symptoms according to a difference In a selected risk factor (or
IMTf), adjusting each factor only for field center and age, by race and sex: The ARIC* Study,
1987-1989
Black
Factor
Female
Male
Yesvs.no
2.3
1.4
1.7
2.5
0.94
0.6
0.6
0.9
0.7
0.18 mm
1.1
1.05
1.25
1.2
212 years vs. less
0.9
1.1
0.6
0.7
$25,000 vs. less
0.5
0.4
0.6
0.7
Diabetes meJIItus*
Work activity*,*
Carotid IMT*
Education*
White
Difference
Annual income*
Female
Male
* p < 0.001 for test over ail race/sex groups.
fTIA, transient Ischemic attack; IMT, intimal-medJai [wall] thickness; ARIC, Atherosclerosis Risk In Communities.
t Index from the Baecke questionnaire and scale (17).
Am J Epidemiol
Vol. 144, No. 9, 1996
862
Chambless et al.
overall statistical significance (p < 0.01). All were
consistent in direction among the four race/sex groups,
except that the relation with education was not present
for blacks.
Adjusted prevalence odds ratios for multiple risk
factor models
In logistic regression models containing the multiple
risk factors meeting the screening criteria and excluding diabetics (and the variables education, work activity, IMT, and income), only the variables reported in
table 5 remained statistically significant in at least one
race/sex group (p < 0.05). We retained body mass
index and hypertension in the table because of special
interest in these variables. The other variables were
dropped from further analysis, little changing the coefficients of the remaining variables in table 5. The
prevalence odds ratios for a 1-standard-deviation difference in lipoprotein(a) were elevated in males and
were statistically significant for white males. The odds
ratios for factor Vm were elevated in all groups, but
somewhat more so and statistically significantly in
whites. Von Willebrand factor was not included because of its close biologic and statistical relation to
factor Vm, but in models with von Willebrand factor
replacing factor VIQ, the odds ratios for von Willebrand factor were similar to those for factor VIII—
somewhat lower and not statistically significant for
white females, somewhat higher for white males. Odds
ratios for other variables in the model were little
changed when von Willebrand factor was included in
place of factor VIII, except for being notably increased
for hypertension among black men and women. The
odds ratios comparing current smokers with never
smokers were at or above 1.6 in all groups and were
statistically significant in three. They were still elevated in former smokers compared with never smokers
but not statistically significantly. The odds ratios for
hypertension were generally elevated (odds ratios s
1.20) but were not individually statistically significant
at the 0.05 level, although the p value for the overall
test was 0.003. The odds ratios for a 1-standarddeviation difference in body mass index were above
unity for all race/sex groups and the overall p value
was statistically significant, but odds ratios were not
statistically significant in individual groups.
When the models in table 5 were expanded to allow
for nonlinearity of the continuous risk factors being
considered or for cigarette-years of smoking, by replacing the continuous risk variables with tertile categorical variables, in no case was the third-to-second
tertile odds ratio statistically significantly different
from the second-to-first tertile odds ratio. Nonlinearity
of the IMT effect was similarly explored, adjusting
^ T- in n y- w
^3 O O O ^5 O
!
Q.
ooddpo
O
^ W O O y- O
s
85
O
T-
9! u)
O Q y- O
Q
(o a K
iiiiii
q oq a ID N <t
d d
d d d
^
i
cq
d
CO
O O O O i- O
O N N Q S N
T~
i-t O ^ CO i-t
5 ini
s
l
O iu
Am J Epidemiol
Vol. 144, No. 9, 1996
TIA/Stroke Symptoms and Cerebrovascular Risk Factors
only for age and center, first using tertiles and then,
because of special interest in this variable, using quintiles (blacks) and deciles (whites). For whites, the odds
ratios comparing the third tertile with the first tertile
were 1.49 (p = 0.045) for men and 1.23 (p = 0.159)
for women. The quintile/decile models showed a significant LMT effect only for whites and only at the
highest decile. Using the 90th percentile cutpoint for
white females, 0.85 mm, the stroke/TLA symptom
odds ratio comparing this top group with all others was
2.04 (p < 0.001). For white males, the IMT effect
began only at the 95th percentile (1.17 mm), with an
odds ratio of 2.10 (p = 0.007), comparing values
above the 95th percentile with those below it.
After exclusion of persons who had ever been told
by a physician that they had had a stroke or TLA, the
associations between risk factors and TLA/stroke
symptoms presented in table 5 were generally similar.
Considering only the results whose statistical significance (p < 0.05) changed, the odds ratio for current
smoking went from 1.88 (p < 0.01) to 1.67 {p =
0.04) in white men and from 1.86 (p = 0.03) to 1.56
(p — 0.16) in black women. Similarly, when considering the effect on the LMT relation of excluding
persons ever told by a physician that they had had a
TLA or stroke, adjusting only for age and center, there
was little change for white females, but the relation for
white males became weaker and statistically nonsignificant.
In considering whether the associations in table 5
were the same for symptoms of stroke (symptoms
lasting for >24 hours) or TLA alone (symptoms lasting
for <24 hours) as they were for TLA/stroke combined,
we were faced with a sample size problem: The numbers of nondiabetic participants with symptoms lasting
at least 24 hours were 37, 25, 94, and 75 for black
females and males and white females and males, respectively. Thus, one would expect fluctuations in
odds ratios when changing the outcome under consideration, especially among blacks, where sample sizes
were smaller. The odds ratio for a 1-standard-deviation difference in body mass index became smaller
among black males and females for stroke symptoms
but much larger for TLA symptoms (for black females,
odds ratio = 1.37, p < 0.01; for black males, odds
ratio = 1.63, p = 0.05), whereas for whites the body
mass index odds ratio remained near unity for <24hour and ^24-hour outcomes. The odds ratio for a
1-standard-deviation difference in age went to 1.35
(p = 0.01) for stroke symptoms in white males, stayed
nearly the same for black males, and stayed close to 1
for females. The most notable change was that for the
hypertension odds ratios, which were much larger for
stroke symptoms alone and were near unity for TLA
Am J Epidemiol
Vol. 144, No. 9, 1996
863
symptoms, except in white females. The stroke symptom/hypertension odds ratios were 3.16 (p < 0.01) for
black females, 1.53 (p = 0.3) for black males, 1.05
(p = 0.85) for white females, and 1.62 (p = 0.06) for
white males. The odds ratios for current smoking
remained elevated for stroke or TLA alone and were
sometimes higher, sometimes lower than those for the
combined TLA/stroke outcome. No consistent change
was seen between the stroke-symptoms-alone or TLAsymptoms-alone model and the combined outcome
model for lipoprotein(a), factor VLLL, or von Willebrand factor. Similarly, in considering the effects on
the LMT relations with stroke symptoms or TLA symptoms alone, adjusting only for age and center, the odds
ratios for white males barely changed; those for white
females were similar for TLA but decreased to 1.1 for
stroke. For blacks, the odds ratios remained statistically nonsignificant.
DISCUSSION
There have been few studies of the relation between
TIA symptoms and cerebrovascular risk factors.
Dennis et al. (19), in a study of incident TLA in
Oxfordshire, England, reported that incidence increased sharply with age and was similar in males and
females. Qizilbash et al. (20) considered risk factors in
a case-control study of cases who were referred because of a TLA or minor ischemic stroke to the Oxfordshire neurology clinic or the Oxfordshire community stroke project. Significant positive relations were
found between TIA or minor ischemic stroke and
fibrinogen and total cholesterol, and a negative relation was seen with HDL cholesterol, adjusting for
multiple risk factors. Schreiner et al. (21), in the ARIC
Study, found a positive relation between lipoprotein(a)
and a self-reported previous physician's diagnosis of
TLA or stroke, controlling for age and several covariates. Folsom et al. (22), also reporting from the ARIC
Study and using the same self-reported physicians'
diagnoses, found a positive relation with fibrinogen,
hemostasis factor VII, factor VLLI, von Willebrand
factor, and protein C, each in at least one sex.
In a prospective study of stroke and TLA combined
in Copenhagen women, approximately one seventh
TIA cases, Lindenstrom et al. (23) found relative risks
of 1.4 for cigarette smoking and 1.5 for lack of physical activity, with no significant effect of body mass
index or alcohol. For postmenopausal women, the
risks for smoking were influenced by hormone replacement therapy. Results for both sexes from the
same study (24) found significant relative risks of 1.3
for smoking, 1.2 for daily tranquilizer use, and 1.2 for
body mass index (per 5.3-kg/m2 increment). Physical
inactivity during leisure time and daily alcohol intake
864
Chambless et al.
were not significant overall. In a separate analysis
(25), these investigators found relative risks of 1.5 for
a 5-year increase in age, 1.3 for shorter versus longer
education, and 1.2 for low versus above-low income.
Graffagnino et al. (26), in a case-control study of 85
patients (one third with TIA and two thirds with
atherothrombotic stroke) in Ontario that focused primarily on family history of vascular disease, hypertension and hyperlipidemia were significant risk factors in multivariate analysis, but not smoking,
although smoking was significant in univariate analysis.
In the many published papers on risk factors for
stroke (focusing on cerebral infarcts if there was a
separate analysis), statistically significant positive relations have been found for cigarette smoking (23, 24,
27-29), fibrinogen (30-32), lack of physical activity
(23, 28), hypertension or blood pressure (26, 28, 29,
31, 33-36), cholesterol (29), triglycerides (29), waist:
hip ratio (31), lower educational level (25), body mass
index (24, 37), diabetes mellitus (33), and lower income (25). On the other hand, in some of these same
studies, factors considered but not statistically significantly related to stroke were cigarette smoking (26,
35), physical inactivity (24), cholesterol (28, 34, 35),
triglycerides (28, 38), body mass index (23, 28, 29),
educational level (29), and diabetes (29).
The strong and positive relation between hypertension and TIA/stroke symptoms in this study contrasts
with the lack of a relation between blood pressure and
TIA/stroke symptoms. In further analysis (not shown),
cross-classifying by self-reported use of antihypertension medication within the previous 2 weeks versus
three levels of blood pressure (< 140/90 mmHg, 140/
90-160/95 mmHg, > 160/95 mmHg), black females
and whites of both sexes who took medication and had
normal blood pressure had strongly and statistically
significantly higher odds of symptoms than those with
normal blood pressure who were not using medication;
for black women, the odds were also significantly
elevated for those who took medication and had blood
pressure above 160/95 mmHg. Other categories for
these three groups and for black men overall failed to
show significantly more frequent TIA/stroke symptoms than for persons with normal blood pressure who
were not using medication. Because of sample size
considerations, analyses by type of pharmacologic
treatment were not undertaken. It is possible that even
treated, controlled hypertension carries an excess burden of TLA/stroke symptoms, and also that certain
patterns or types of antihypertensive medication are
associated with the observed excess of TIA/stroke
symptoms. Because these are cross-sectional data
without accompanying records of the participants'
medical history of hypertension—such as the duration
of hypertension, its severity, and compliance with
antihypertension regimens—it is not fruitful to speculate on the several possible interpretations of these
findings.
The relation between TLA/stroke symptoms and carotid artery IMT, a potential marker for atherosclerosis, was positive and statistically significant for whites
but not for blacks, either when IMT was considered as
a linear term in the logistic model or when variables
representing quintiles of IMT were entered. Further
investigation for whites leads to the conclusion that in
these data the relation appears nonlinear (on the logistic scale), and indeed only for extreme IMT do symptoms appear significantly more frequently. Whites
with extreme carotid arterial IMT had approximately
twice the odds of experiencing strokeATIA symptoms
as those without extreme IMT. "Extreme" is defined
here as >0.85 mm for women (90th percentile) and
>1.17 mm for men (95th percentile). The mechanisms
linking carotid IMT with stroke/TLA symptoms remain
speculative. Thickening of the extracranial carotid arteries, as assessed from relatively short arterial segments, may correlate with having a carotid plaque (i.e.,
a source of emboli to the brain) or may be a marker of
diffuse intracerebral atherosclerotic disease. Either of
these correlated factors may account for the observed
association between carotid IMT and stroke/TLA
symptoms.
We considered the age- and field center-adjusted
relations of two socioeconomic factors to the history
of TIA/stroke symptoms, using the variables education
(at least high school vs. less) and annual family income (at least $25,000 vs. less). Family income was
strongly and consistently negatively related to symptoms, but educational level was so related only for
whites. These variables probably contribute strongly
to the levels of the other risk factors being considered,
so it is not surprising that the relations between the
socioeconomic variables and TIA/stroke symptoms
were sometimes weakened when the other risk factors
were included.
The ARIC Study is especially important for its
ability to assess the relation of these TIA/stroke
symptoms to risk factors in a large African-American population. The multiple risk factor-adjusted
associations were positive and quite similar between
black and white males for lipoprotein(a) and hypertension and between blacks and whites of both sexes
for factor VIII and smoking. The odds ratios for
black males were marginally statistically significant
for body mass index and age and were higher than
those for the other sex/race groups. With respect to
the relation with body mass index, it is interesting
Am J Epidemiol
Vol. 144, No. 9, 1996
TIA/Stroke Symptoms and Cerebrovascular Risk Factors
that the ARIC age- and center-adjusted findings for
TIA/stroke are exactly parallel to the age-adjusted
findings of the Evans County Study of incident
stroke risk factors (39), conducted more than two
decades ago: namely, that body mass index was
related to cerebrovascular disease for white men and
women and for black men, but not for black women.
The odds ratio for hypertension was highest among
the four ARIC groups for black females, statistically
significant for all participants when adjusted only
for age and center, and marginally statistically significant (except for black males) with adjustment
for the other factors. The age- and field centeradjusted odds ratios for arterial IMT were close to
unity for blacks, in comparison with a significant
positive relation for whites. This difference between
whites and blacks is perhaps related to the observation (40) that, compared with whites, stroke in
blacks is much more frequently attributable to intracranial lesions than to lesions in the extracranial
carotid arteries.
The ARIC Study investigators are able to study the
relations between putative risk factors and symptoms
of TLA and stroke in a free-living population with the
diagnosis of symptoms being implemented in a standardized manner. Thus, the well-known problems with
the repeatability of TLA diagnosis are reduced, but just
as important is the reduction in differential information bias with respect to reported history of physiciandiagnosed TLA or stroke. Certainly, the screen used
here to select potential risk factors for further consideration was broad, and one must be cautious in interpreting "statistically significant" results when so many
tests are done. Nevertheless, selection for multiple risk
factor assessment did include consideration of consistency across different race/sex groups. Moreover, this
report lays the foundation for future consideration in a
prospective mode of the relations described here from
cross-sectional analysis. In a prospective analysis of
risk factor relations, persons without TLA/stroke
symptoms at baseline would be followed to see
whether prior baseline risk factor levels were related
to the incidence of TLA/stroke symptoms. In our crosssectional study, the possibility that the symptoms affected the risk factor levels cannot be ruled out.
Aging, diabetes, smoking, hypertension, and obesity
have all been implicated in the pathogenesis of stroke
(37) and, not surprisingly, were found here to be
associated with TLA/stroke symptoms. Most plasma
lipids are generally not considered important risk factors for stroke, and, as might be expected, they were
not retained in our multivariate model, with the exception of lipoprotein(a). Lipoprotein(a) is believed to
have thrombogenic as well as atherogenic properties
Am J Epidemiol
Vol. 144, No. 9, 1996
865
and has been previously associated with stroke (21,
41). Among the hemostatic factors, factor VLLI (or its
correlate, von Willebrand factor) was retained in the
final model in this analysis. Both factors play important roles in hemostasis: factor VLU is involved in the
intrinsic coagulation pathway and von Willebrand factor regulates platelet adhesion to the vessel wall (42).
Higher plasma levels of either factor may signify
enhanced thrombogenic potential and an increased risk
of cerebrovascular disease (22). Since the two factors
were strongly correlated in ARIC (r = 0.7), it is
difficult to determine whether one or both of them may
be pathogenically important.
Stroke is a heterogenous clinicopathologic entity
ranging from a transient neurologic deficit due to
short-lasting ischemia to completed brain infarction or
intraparenchymal hemorrhage. Clearly, a symptombased algorithm encompasses not only this clinicopathologic spectrum but also symptoms of nonvascular etiology. Nevertheless, the associations reported
here most likely relate to the pathogenesis of ischemic
events, which are far more common than brain hemorrhage in the general US population.
ACKNOWLEDGMENTS
This research was supported by National Institutes of
Health contract NO1-HC-55015.
The authors acknowledge Ernestine Bland and La Sonya
Goode for assistance with manuscript preparation.
The authors recognize the active contributions of the
personnel at the following ARIC centers: University of
North Carolina, Chapel Hill, North Carolina—Phylh's
Johnson and Marilyn Knowles; University of North Carolina, Forsyth County, North Carolina—Jeannette Bensen,
Kay Burke, Carol Smith, and Pamela Williams; University
of Mississippi Medical Center, Jackson, Mississippi—Cora
L. Walls, Betty S. Warren, Dorothy P. Washington, and
Mattye L. Watson; University of Minnesota, Minneapolis,
Minnesota—Karen Birkholz, Charlene Bogden, Dot
Buckingham, and Carolyne Campbell; The Johns Hopkins
University, Baltimore, Maryland—Sunny Harrell, Patricia
Hawbaker, Joel G. Hill, and Mary Hurt; The University of
Texas School of Medicine, Houston, Texas—Pam Pfile,
Hoang Pham, Valerie Stinson, and Teri Trevino; Atherosclerosis Clinical Laboratory, The Methodist Hospital,
Houston, Texas—Wanda R. Alexander, Doris J. Harper,
Charles E. Rhodes, and Selma M. Soyal; Ultrasound Reading Center, Bowman Gray School of Medicine, WinstonSalem, North Carolina—Linda Allred, Suzanne Pillsbury,
Tiffany Robertson, and Anne Safrit; Coordinating Center,
University of North Carolina, Chapel Hill, North Carolina—Ken Kaufman, Ho Kim, Charmaine Marquis, and
Alison Meyer.
866
Chambless et al.
REFERENCES
1. The ARIC Investigators. The Atherosclerosis Risk in Communities (ARIC) Study: design and objectives. Am J Epidemiol 1989;129:687-702.
2. Jackson R, Chambless LE, Yang K, et al. Differences between
respondents and nonrespondents in a multi-center communitybased study vary by gender and ethnicity. J Clin Epidemiol (in
press).
3. The ARIC Investigators. ARIC manual of operations, no. 8:
lipid and lipoprotein determinations. Version 1.0. Chapel Hill,
NC: University of North Carolina, 1987.
4. Sharrett RA. The Atherosclerosis Risk in Communities
(ARIC) Study: introduction and objectives of the hemostasis
component. Ann Epidemiol 1992;2:467-9.
5. Papp AC, Hatzakis H, Bracey A, et al. ARIC hemostasis
study. I. Development of a blood collection and processing
system suitable for multicenter hemostatic studies. Thromb
Haemost 1989;61:15-19.
6. Wu KK, Papp AC, Patsch W, et al. ARIC hemostasis study, n.
Organizational plan and feasibility study. Thromb Haemost
1990;64:521-5.
7. The ARIC Investigators. ARIC manual of operations, no. 9:
hemostasis determinations. Version 1.0. Chapel Hill, NC:
University of North Carolina, 1987.
8. The ARIC Investigators. ARIC manual of operations, no. 2:
cohort component procedures. Version 1.0. Chapel Hill, NC:
University of North Carolina, 1987.
9. The ARIC Investigators. ARIC manual of operations, no. 11:
sitting blood pressure. Version 1.0. Chapel Hill, NC: University of North Carolina, 1987.
10. Bond MG, Barnes RW, Riley WA, et al. High-resolution
B-mode ultrasound scanning methods in the Atherosclerosis
Risk in Communities Study (ARIC). Neuroimaging 1991;1:
68-73.
11. Riley WA, Barnes RW, Bond MG, et al. High-resolution
B-mode ultrasound reading methods in the Atherosclerosis
Risk in Communities (ARIC) cohort. Neuroimaging 1991 ;1:
168-72.
12. The ARIC Investigators. ARIC manual of operations, no. 6A:
ultrasound assessment—part A, ultrasound scanning. Version
1.0. Chapel Hill, NC: University of North Carolina, 1987.
13. The ARIC Investigators. ARIC manual of operations, no. 6B:
ultrasound assessment—part A, ultrasound reading. Version
1.0. Chapel Hill, NC: University of North Carolina, 1987.
14. Executive Committee for the Asymptomatic Carotid Atherosclerosis Study. Endarterectomy for asymptomatic carotid artery stenosis. JAMA 1995;273:1421-8.
15. Toole JF, Lefkowitz DS, Chambless LE, et al. Self-reported
transient ischemic attack and stroke symptoms: methods and
baseline prevalence. The ARIC Study, 1987-1989. Am J
Epidemiol 1996; 144:849-56.
16. Dixon WJ, Brown MB, Engelman L, et al. BMDP statistical
software manual. Berkeley, CA: University of California
Press, 1990.
17. Baecke JA, Burema J, Frijters JE. A short questionnaire for
the measurement of habitual physical activity in epidemiological studies. Am J Clin Nutr 1982;36:936-42.
18. SAS Institute, Inc. SAS/STAT software: changes and enhancements. Release 6.10. Cary, NC: SAS Institute, Inc, 1994.
19. Dennis MS, Bamford JM, Sandercock PAG, et al. Incidence of
transient ischemic attacks in Oxfordshire, England. Stroke
1989;20:333-9.
20. Qizilbash N, Jones L, Warlow C, et al. Fibrinogen and lipid
concentrations as risk factors for transient ischaemic attacks
and minor ischaemic strokes. BMJ 1991;303:605-9.
21. Schreiner PJ, Chambless LE, Brown SA, et al. Lipoprotein(a)
as a correlate of stroke and transient ischemic attack prevalence in a biracial cohort The ARIC Study. Ann Epidemiol
1994;4:351-9.
22. Folsom AR, Wu KK, Shahar E, et al. Association of hemostatic variables with prevalent cardiovascular disease and
asymptomatic carotid artery atherosclerosis. Arterioscler
Thromb 1993; 13:1829-36.
23. Lindenstrom E, Boysen G, Nyboe J. Lifestyle factors and risk
of cerebrovascular disease in women: The Copenhagen City
Heart Study. Stroke 1993;24:1468-72.
24. Lindenstrom E, Boysen G, Nyboe J. Risk factors for stroke in
Copenhagen, Denmark. U. Life-style factors. Neuroepidemiology 1993;12:43-50.
25. Lindenstrom E, Boysen G, Nyboe J. Risk factors for stroke in
Copenhagen, Denmark. I. Basic demographic and social factors. Neuroepidemiology 1993;12:37-42.
26. Graffagnino C, Gasecki AP, Doig GS, et al. The importance of
family history in cerebrovascular disease. Stroke 1994;25:
1599-604.
27. Robbins AS, Manson JE, Lee I-M, et al. Cigarette smoking
and stroke in a cohort of U.S. male physicians. Ann Intern
Med 1994;120:458-62.
28. Haheim LL, Holme I, Hjermann I, et al. Risk factors of stroke
incidence and mortality: a 12-year follow-up of the Oslo
Study. Stroke 1993;24:1484-9.
29. Salonen JT, Puska P. Relation of serum cholesterol and triglycerides to the risk of acute myocardial infarction, cerebral
stroke, and death in eastern Finnish male population. Int J
Epidemiol 1983;12:26-31.
30. Wilhelmsen L, Svardsudd K, Korsan-Bengtsen K, et al. Fibrinogen as a risk factor for stroke and myocardial infarction.
N Engl J Med 1984;311:501-5.
31. Welin L, Svordsudd K, Wilhelmsen L, et al. Analysis of risk
factors for stroke in a cohort of men born in 1913. N Engl
J Med 1987;317:521-6.
32. Lee AJ, Lowe GD, Woodward M, et al. Fibrinogen in relation
to personal history of prevalent hypertension, diabetes, stroke,
intermittent claudication, coronary heart disease, and family
history: The Scottish Heart Health Study. Br Heart J 1993;69:
338-42.
33. Davis PH, Dambrosia JM, Schoenberg BS, et al. Risk factors
for ischemic stroke: a prospective study in Rochester, Minnesota. Ann Neurol 1987;22:319-27.
34. Szatrowski TP, Peterson AV Jr. Shimizu Y, et al. Serum
cholesterol, other risk factors, and cardiovascular disease in a
Japanese cohort. J Chronic Dis 1984;37:569-84.
35. Tanaka H, Ueda Y, Hayashi M, et al. Risk factors for cerebral
hemorrhage and cerebral infarction in a Japanese rural community. Stroke 1982;13:62-73.
36. MacMahon S, Peto R, Cutler J, et al. Blood pressure, stroke,
and coronary heart disease. Part 1. Prolonged differences in
blood pressure: prospective observational studies corrected for
the regression dilution bias. Lancet 1990;335:765-74.
37. Folsom AR, Prineas RJ, Kaye SA, et al. Incidence of hypertension and stroke in relation to body fat distribution and other
risk factors in older women. Stroke 1990;21:701-6.
38. Rhoads GG, Feinleib M. Serum triglyceride and risk of coronary heart disease, stroke, and total mortality in JapaneseAmerican men. Arteriosclerosis 1983;3:316-22.
39. Heyrnan A, Karp HR, Heyden S, et al. Cerebrovascular disease in the bi-racial population of Evans County, Georgia.
Stroke 1971;2:509-18.
40. Caplan LR, Gorelick PB, Hier DB. Race, sex, and occlusive
cerebrovascular disease: a review. Stroke 1986; 17:648 -55.
41. Jurgens G, Koltringer P. Lipoprotein(a) in ischemic cerebrovascular disease: a new approach to the assessment of risk for
stroke. Neurology 1987;37:513-15.
42. Ogston D. The physiology of hemostasis. Cambridge, MA:
Harvard University Press, 1983.
Am J Epidemiol
Vol. 144, No. 9, 1996