American Journal of Epidemiology
Copyright O 1996 by The Johns Hopkins University School of Hygiene and Public Health
All rights reserved
Vol. 144, No. 3
Printed In U.SA.
Increase in Fasting Insulin and Glucose over Seven Years with Increasing
Weight and Inactivity of Young Adults
The CARDIA Study
Aaron R. Folsom,1 David R. Jacobs, Jr.,1 Lynne E. Wagenknecht,2 Susan P. Winkhart,1 Carla Yunis,1'3
Joan E. Hilner,4 Peter J. Savage,5 Delia E. Smith,6 and John M. Flack7
To characterize 7-year changes in fasting serum insulin and glucose concentrations, the authors analyzed
population-based data on 3,095 nondiabetic black and white men and women who were initially aged 18-30
years in the Coronary Artery Risk Development in Young Adults (CARDIA) Study. Measurements were made
of fasting insulin (by an assay with little cross-reactivity to proinsulin) and fasting glucose on frozen samples
from baseline and Year 7 examinations. Over the 7-year period, mean fasting insulin increased 10-25%, mean
fasting glucose increased 7-10%, and mean body mass increased 7-12% across the four race-, sex-groups.
The strongest predictor of both insulin increase and glucose increase was an increase in body mass over the
7 years. Adjusted for age and examination time period in race-, sex-specific repeated measures analyses,
fasting insulin increased longitudinally by approximately 5 p.U/mL per 5 kg/m 2 increase in body mass index
(p < 0.05). Adjusted for age and time period, fasting insulin increased over the 7 years by approximately 2.5
/xU/mL per 0.08 unit increase in waist/hip ratio (p < 0.05), although this association was much stronger
cross-sectionally. In a similar model, each 100 unit decrease in physical activity longitudinally predicted a
0.1-0.2 /xU/mL increase in fasting insulin (p < 0.05 in black men only); this association was stronger and
statistically significant in all race-, sex-groups cross-sectionally. Fasting insulin was not associated with
energy intake either cross-sectionally or longitudinally, but age- and time-adjusted associations of insulin
change with change in several nutrients (e.g., energy from fat) were statistically significant in whites. The
authors conclude that marked increase in weight in young adulthood adversely alters glucose and insulin
metabolism, and that, if not reversed, this may lead to harmful health consequences in later life. Am J
Epidemiol 1996; 144:235-46.
exercise; glucose; insulin; longitudinal studies; obesity
Considerable evidence indicates that insulin and/or
insulin resistance, and related metabolic consequences, play important roles in the pathogenesis of
non-insulin-dependent diabetes mellitus (NIDDM) ( 1 3). Hyperinsulinemia, a marker of insulin resistance in
nondiabetics (4), usually precedes by years the hyperglycemia and insulinopenia that accompany frank
NIDDM (2, 3). Hyperinsulinemia and insulin resistance also are associated with dyslipidemia (5, 6),
elevated blood pressure (5, 6), impaired fibrinolysis
(7), atherosclerosis (8), and increased incidence of
coronary heart disease (9-11), although the latter association is controversial (12-14).
Numerous population-based cross-sectional studies
(15-17) have reported that, besides family history of
NIDDM (18, 19), correlates of hyperinsulinemia and
glucose intolerance include obesity, abdominal fat preponderance, physical inactivity, and dietary composition. However, few population-based investigations
(2, 20) have characterized prospectively factors that
Received for publication June 30, 1995, and accepted for publication January 5, 1996.
. Abbreviations: CARDIA, Coronary Artery Risk Development in
Young Adults Study; NIDDM, non-insulin-dependent diabetes mellitus.
1
Division of Epidemiology, School of Public Health, University of
Minnesota, Minneapolis, MN.
2
Department of Public Health Sciences, Bowman Gray School of
Medicine, Winston-Salem, NC.
3
Division of General and Preventive Medicine, Department of
Medicine, Medical School, University of Minnesota, Minneapolis,
MN.
* CARDIA Coordinating Center, University of Alabama at Birmingham, Birmingham, Al_
s
National Institutes of Health, National Heart, Lung, and Blood
Institute, Division of Epidemiology and Clinical Applications, Bethesda, MD.
6
University of Alabama at Birmingham, Division of Preventive
Medicine, Birmingham, Al_
7
Bowman Gray School of Medicine, Hypertension Center, Winston-Salem, NC.
Reprint requests to Dr. Aaron R. Folsom, Division of Epidemiology, School of Public Hearth, University of Minnesota, Suite 300,
1300 South Second Street, Minneapolis, MN 55454-1015.
235
236
Folsom et al.
relate to changes in insulin and glucose concentrations. Further, limited data are available on men and
women of various ethnic groups. A study of young
adults is particularly relevant because young adulthood is characterized by important changes in diet,
weight, physical activity, and other likely contributors
to hyperinsulinemia. We therefore characterized
7-year changes in fasting serum insulin and glucose
concentrations in the Coronary Artery Risk Development in Young Adults (CARDIA) Study.
MATERIALS AND METHODS
The CARDIA Study is a multicenter longitudinal
study assessing coronary heart disease risk factors and
their evolution in 5,115 young black and white men
and women aged 18-30 years at a baseline examination (Year 0) in 1985-1986 (21). Participants were
recruited at random from community areas of Birmingham, Alabama; Chicago, Illinois; and Minneapolis, Minnesota; and from the Kaiser-Permanente
health plan in Oakland, California. Second (Year 2),
third (Year 5), and fourth (Year 7) examinations were
performed in 1987-1988, 1990-1991, and 19921993, respectively. A total of 4,086 (80 percent of the
baseline cohort) participated in the Year 7 examination.
In the CARDIA examinations, standardized and
consistent methods were used. Race and smoking status were assessed by questionnaire. Participants were
asked about the frequency of performance in the previous year of specified heavy- and moderate-intensity
activities. From the frequency and an assigned intensity of each activity, scores for heavy, moderate, and
total activity were computed (22).
In the Year 0 and Year 7 examinations, usual dietary
intake was assessed by the CARDIA dietary history,
an interviewer-administered quantitative food frequency questionnaire using the past month as a frame
of reference (23-25). Several nutrients possibly related to hyperinsulinemia (15, 17) were examined:
energy; total, saturated, monounsaturated, and polyunsaturated fatty acids (expressed as a percentage of
energy); cholesterol (per 1,000 kcal energy); protein
(as a percent of energy); carbohydrates (as a percent of
energy); sucrose (as a percent of energy); and magnesium (per 1,000 kcal energy). Usual intakes of wine,
beer, and hard liquor were assessed, and ethanol consumption was calculated (in grams) using the following equivalents: 5 oz of wine, 17.02 ml; 12 oz beer,
16.7 ml; and 1.5 oz liquor, 19.09 ml.
Participants were asked at all examinations to report
whether they had ever had diabetes mellitus, and
whether they were currently taking medication for this
condition. At the Year 0 and the Year 5 CARDIA
examinations, participants were also asked whether
they knew the health of their natural parents and, if so,
whether either had ever had diabetes; information on
parental history of diabetes from the Year 5 examination was used in analysis.
Body weight (in light clothing) was measured to the
nearest 0.5 pound (0.2 kg) with a calibrated scale.
Height (without shoes) was measured to the nearest
0.5 cm using a vertical ruler. Body mass index (kg/m2)
was computed as an index of relative weight. Waist
circumference was measured with the participant
standing, at the minimum abdominal girth, in duplicate, to the nearest 0.5 cm. Hip circumference was
measured at the maximal protrusion of the hips at the
level of the symphysis pubica, in duplicate, to the
nearest 0.5 cm. The average of the two measurements
of each circumference was used to calculate waist/hip
ratio.
Participants were asked to fast for 12 hours prior to
the CARDIA examination and to avoid smoking and
heavy physical activity for the preceding 2 hours.
Blood was drawn for insulin and glucose into vacuum
tubes containing no preservative, and within 60 minutes serum was separated by centrifugation at 4°C.
Serum was stored in cryovials (fluoride-containing for
glucose) and within 90 minutes of drawing was frozen
at -70°C until analysis at the laboratories.
Glucose was measured in each examination by the
hexokinase method. Blind analysis of split specimens
yielded a technical error of 2.0 percent of the mean
and an r = 0.99. Fasting insulin was measured originally during the Year 0 examination by a nonspecific
insulin assay, a modification of the immunoassay of
Herbert et al. (26). However, for this report, we
wanted to use a more specific insulin assay. To assure
comparability of insulin across visits, we not only
measured insulin in Year 7 participants on sera stored
one year from the Year 7 examination, we also used
the new assay on sera stored for 8 years from Year 0.
The new radioimmunoassay employed an overnight,
equilibrium incubation (27). The key feature of this
assay is the use of a unique antibody that has less than
0.2 percent cross-reactivity to human proinsulin and
its primary circulating split form Des 31,32 proinsulin.
Blind analysis of split serum samples yielded a technical error of 16.6 percent of the mean and an r =
0.98. The Pearson correlation of log insulin values for
Year 0 by the original (26) and subsequent (27) methods was 0.81.
Our analysis aimed to determine the longitudinal
predictors of fasting insulin level. From the 4,086
participants who attended both Year 0 and Year 7
examinations, we excluded hierarchically 127 at Year
0 and 363 at Year 7 who had not fasted for at least 10
Am J Epidemiol
Vol. 144, No. 3, 1996
Seven-Year Insulin and Glucose Trends
hours; 32 at Year 0 and 71 at Year 7 who were
pregnant or were unsure at the time of examination; 56
at Year 0 and 97 at Year 7 who reported diabetes; 238
at Year 0 and 2 at Year 7 who had missing insulin
values using the specific insulin assay; and five persons with extreme insulin values. This left a total of
3,095 for analyses. These 3,095 had mean values of
insulin (by our original method (26)), glucose, body
mass index, waist/hip ratio, alcohol intake, and age
similar (p Si 0.05) to all 4,842 nonpregnant, nondiabetic, fasting participants at Year 0. However, they
had a 3 percent lower prevalence of current smoking.
Parallel analyses were run for insulin and glucose,
but, because our main interest was insulin change,
only selected glucose findings are presented. Most
analyses were race- and sex-specific. Characteristics
of participants at the Year 0 and Year 7 examinations
were described by computing means and standard
errors or percentage distributions. We depicted bivariate associations by computing insulin change
(Year 7 to Year 0) within strata of various potential
predictors or change in these predictors. We also computed Pearson correlation coefficients.
Proc MIXED in SAS (SAS Institute, Cary, North
Carolina) was used to model the repeated measures for
the two examination periods, Year 0 and Year 7.
Various models for the dependent variable, fasting
insulin, included time-invariant characteristics of individuals (for example, family history of diabetes), timevarying characteristics (such as body mass index
(BMI)), and, possibly, interactions of predictors with
examination period. Age is a special predictor because
change in age between examinations is identical to,
and therefore completely confounded with, the change
in time between examinations. For ease of interpretation, we therefore decomposed the time-varying predictors, for each person, into an average level and
time-related deviations from that average. Because the
age deviation is totally confounded with the time period, we modeled age as only the fixed covariate, i.e.,
baseline age (age0). The time period effect can be
interpreted as the change in insulin (glucose) due to
aging if we assume there are no omitted confounders
and no secular trend in the insulin data. Models were
run sex- and race-specific. A typical model for insulin
in person i at period j , v,y, might be:
= V + /^period, + /3,age« +
X period;) + frCBMI.vg), + /^
+ «</,
with some models containing other average/deviation
pairs for time-varying predictors such as waist/hip
ratio and physical activity. Errors of measurements of
y are assumed to be correlated within a person at
Am J Epidemiol
Vol. 144, No. 3, 1996
237
different times, but assumed to be uncorrelated between people. We interpret /30 as the total 7-year
cohort change in insulin; when time period is coded 0
for examination 0 and 1 for examination 7, /3j is the
slope of baseline insulin on baseline age; / ^ is the
difference between slope of the Year 7 insulin on
baseline age and j8t; J33 is the slope of differences in
average insulin between persons of different average
body mass index; while /34 is the slope of 7-year
change in insulin contemporaneous with change in
body mass index within persons. f53 and /34 therefore
represent cross-sectional and longitudinal associations, respectively.
RESULTS
Univariate changes
As table 1 shows, fasting serum insulin increased
substantially over the 7 years in all race-, sex-groups
of nondiabetic young adults. The increase in mean
insulin concentration ranged from 10 percent in white
women to 25 percent in black men. Relative ranking of
insulin level was maintained to a degree, as evidenced
by correlations between Year 0 and Year 7 logarithmically-transformed fasting insulin ranging from r =
0.51 to 0.61 across race-sex groups. There were concomitant increases in all subgroups in fasting glucose,
ranging from 7 percent in white women to 10 percent
in black men, and in body mass index, ranging from 7
percent in white men to 12 percent in black women.
Waist circumference and waist/hip ratio increased,
while physical activity, alcohol intake, and smoking
prevalence tended to decrease (except for black men
for the latter two). Participant-reported parental history of diabetes ranged from 10.2 percent to 23.6
percent, and was higher in blacks than whites, consistent with the excess of diabetes among blacks found in
national surveys (28).
Bivariate changes
Depictions of the change in insulin with changes in
several other factors are provided in the figures.
Changes in average fasting insulin were strongly related to change in weight and were generally consistent across race-, sex-groups (figure 1). Persons who
lost weight had a fall in insulin levels; those whose
weight remained stable within 5 lbs (2.3 kg) had very
little insulin change, whereas those who gained weight
had increases in insulin levels. As shown in table 2,
participants with baseline waist/hip ratios above the
median tended to exhibit more change in fasting insulin with weight gain or loss than did those with waist/
hip ratios below the median. For example, in participants with waist/hip ratio above the median, loss of at
least 5 lbs (2.3 kg) resulted in declines of fasting
238
Folsom et a).
TABLE 1. Race- and sex-specific mean values of selected risk factors at baseline and 7 years later and
the mean differences ± standard error (SE), the CARDIA Study, 1985-1986 to 1992-1993
Risk factor and
race and sex*
Fasting insulin (>xU/mL)
Black men
Black women
White men
White women
Fasting glucose (mg/dL)
Black men
Black women
White men
White women
Body mass index (kg/m2)
Black men
Black women
White men
White women
Waist (cm)
Black men
Black women
White men
White women
Waist/hip ratio
Black men
Black women
White men
White women
Physical activity score
Black men
Black women
White men
White women
Alcohol (ml/day)
Black men
Black women
White men
White women
Smoking (% current)
Black men
Black women
White men
White women
Parent with diabetesf (%)
Black men
Black women
White men
White women
Means
Baselne
7-Vfear
Mean ditterenca
±SE
11.6
13.0
10.5
10.2
14.5
15.0
12.4
11.2
2.9
2.0
1.9
1.0
± 0.3
± 0.3
±0.2
±0.2
83.4
79.4
84.4
80.5
91.7
86.1
91.9
86.1
8.3
6.7
7.5
5.6
±
±
±
±
24.6
25.6
27.0
2.4 ±0.1
28.7
24.2
23.1
25.9
24.9
3.1 ±0.1
1.7 ±0.1
1.8 ±0.1
80.6
75.9
82.7
71.7
87.6
83.4
88.1
76.1
7.0
7.5
5.4
4.4
0.815
0.737
0.837
0.723
0.841
0.762
0.858
0.733
537
280
509
406
466
225
414
306
-71
-56
-95
-100
16.7
5.1
18.5
8.8
19.4
4.9
15.3
6.5
2.9
-0.2
-3.3
-2.2
±
±
±
±
1.4
0.4
1.0
0.5
31.4
29.6
23.5
24.6
33.4
27.6
20.9
19.1
2.0
-2.0
-2.6
-5.5
±
±
±
±
1.4
1.2
1.0
1.2
±
±
±
±
0.6
0.4
0.4
0.3
0.3
0.3
0.2
0.3
0.026 ± 0.002
0.024 ± 0.002
0.020 ± 0.001
0.009 ± 0.001
± 15
±9
± 10
±9
18.6
23.6
10.2
13.5
* Numbers of participants were 612 black men, 770 black women, 874 white men, and 839 white women,
t Data from Y&ar 5 examination. Included in the denominator are those with unknown parental history: 21
percent of black men, 17 percent of black women, 8 percent of white men, and 8 percent of white women.
insulin ranging 0.5 to 3.8 /xU/mL across race-, sexgroups, and a gain of >25 lbs (>11.25 kg) resulted in
increases in insulin of 5.2 to 7.4 /i,U/mL. In contrast,
for those with waist/hip ratio below the median,
weight loss resulted in little change of insulin (from
- 2 . 3 to +1.1 /iU/mL) and 2=25 lb (> 11.25 kg) gain
resulted in smaller increases in insulin (from 2.9 to 5.8
jiU/mL). The same type of interaction was seen be-
tween baseline body mass index and weight gain, such
that the largest fasting insulin changes with weight
change were among those with higher body mass index.
Greater weight gain over the 7 years was also associated with greater rises in fasting glucose (figure 2),
and, as expected, greater rises in glucose were associated with greater increases in insulin (figure 3). Decreases in physical activity (figure 4) were associated
Am J Epidemiol
Vol. 144, No. 3, 1996
Seven-Year Insulin and Glucose Trends
239
Weight
Change
(lbs)
.8
3
OB
a
a
J3
u
a
a
Black
Men
Black
Women
White
Men
White
Women
FIGURE 1. Mean change In fasting Insulin in relation to weight change over a 7-year period, the CARDIA Study, 1985-1986 to 1992-1993
(1 Ib = 0.45 kg).
TABLE 2. Mean change in fasting serum insulin over 7 years in relation to weight gain and baseline
waist/hip ratio, the CARDIA Study, 1985-1986 to 1992-1993
Race and sex by
baseline waist/nb
rallo above or below
median*
Black men
Above median
Below median
Black women
Above median
Below median
White men
Above median
Below median
White women
Above median
Below median
Mean Insulin change (jjU/mL) by weight change group
iSbs
Lost 5 b 8
to gained
5lbs
Gained
5-15
bs
Gained
15-25
bs
Gained
225
bs
-1.8
1.1
2.1
-0.2
1.9
0.5
4.9
2.0
7.2
5.8
-3.8
-2.3
-0.9
-0.8
0.3
0.1
2.9
1.2
6.9
4.4
-0.5
-1.4
0.6
0.0
1.6
0.7
2.4
2.5
7.4
2.9
-2.4
-1.0
0.0
-0.4
1.3
0.1
2.2
2.0
5.2
4.2
Lost
* Medians were: men, 0.8255; women, 0.7247.
1 1 Ib - 0.45 kg.
with greater rises in fasting insulin levels between
examinations.
Repeated measures analysis
Table 3 presents results of sex- and race-specific
repeated measures analyses for insulin in relation to
the predictor variables. Each model included terms for
the average level and deviation of a single predictor
variable as well as terms for age and time period. Body
mass index was associated strongly and positively
with average fasting insulin, adjusting for age and time
period. Averaged across examinations, fasting insulin
Am J Epidemiol
Vol. 144, No. 3, 1996
cross-sectionally was about 4 pJJ/rnL higher per 5
kg/m2 in men and 3 /xU/mL higher per 5 kg/m2 in
women. The longitudinal relation, represented by the
body mass index deviation coefficient, was somewhat
stronger: fasting insulin increased about 5 /AU/mL per
5 kg/m2 increase in body mass index, slightly less (3.4
piU/mL) in white women.
Average waist/hip ratio was related strongly and
positively with average fasting insulin, but change in
waist/hip ratio was less strongly related to change in
fasting insulin, adjusting for age and time period.
Average insulin was between 3.2 and 5.6
240
Folsom et al.
11
Weight
Change
(lbs)
I <5
Black
Men
Black
Women
White Men
White
I
-5to + 5
I
5.1 to 15
3
15.1 to 25
]
>25
Women
FIGURE 2. Mean change in fasting glucose in relation to weight change over a 7-year period, the CARDIA Study, 1985-1986 to 1992-1993
(1 Ib = 0.45 kg).
Glucose
Change
(mg/dl)
a
I -86to+l
OB
a
m
I +2 to 6
u
I +7 to 11
Black
Men
Black
Women
White Men
White
Women
FIGURE 3. Mean change in fasting Insulin In relation to glucose change over a 7-year period, the CARDIA Study, 1985-1986 to 1992-1993.
higher cross-sectionally per 0.08 unit greater waist/hip
ratio; this estimate was somewhat greater in blacks
than whites and in men than women. The longitudinal
increase in insulin was about 2.5 /i,U/mL per 0.08 unit
increase in waist/hip.
Average physical activity level was cross-sectionally associated strongly and inversely with average
fasting insulin but change in physical activity was
weakly, and generally not statistically significantly,
associated with change in fasting insulin, adjusting for
age and time period. Average insulin was about 0.3
fiU/mL lower per 100 exercise units, whereas longitudinally insulin decreased 0.1-0.2 /xU/mL per 100
exercise units.
As table 3 shows further, adjusted for age and time
period, fasting insulin was associated negatively with
Am J Epidemiol
Vol. 144, No. 3, 1996
Seven-Year Insulin and Glucose Trends
241
~ 6
i
Physical Activity
Score Change
1 5
Black Men
Black Women
White Men
White
•
-1491 to-235
•
-234to-65
B
-64 to+83
0
+84 to 1415
Women
FIGURE 4. Mean change in fasting insulin in relation to physical activity score change over a 7-year period, the CARDIA Study, 1985-1986
to 1992-1993.
alcohol intake, but consistently longitudinally and
cross-sectionally only for black women. Fasting insulin was seemingly unrelated to average level or change
in energy intake. In whites but not blacks, average
levels and deviation in fasting insulin were related
positively with average levels and deviation in the
percentage of energy derived from total, saturated, and
monounsaturated fatty acids. These associations were
not particularly strong, generally less than 1 p,U/mL of
fasting insulin increase per one standard deviation
increase in fat calories. Fasting insulin was not consistently associated with intake of polyunsaturated
fatty acid energy, cholesterol, or protein energy. In
several subgroups, however, average level and/or
change in insulin was associated negatively with
average/deviation in intake of carbohydrate energy
and magnesium, and was associated positively with
sucrose energy.
In all race-, sex-groups, fasting insulin was associated positively, but generally not statistically significantly, with parental history of diabetes, adjusted for
age and time period (data not shown). Compared with
those without a parental history of diabetes, participants with a family history had an approximately 1
/i,U/mL higher average fasting insulin concentration
and a 0.5 to 1.0 /xU/mL greater increase in insulin over
the 7 years.
Compared with those who never smoked, participants who smoked at an examination generally had
about 1 ju,U/mL lower age- and time period-adjusted
fasting insulin values (data not shown). Furthermore,
compared with those who did not change smoking/
Am J Epidemiol
Vol. 144, No. 3, 1996
nonsmoking status over the 7 years, the nonsmokers
who became smokers had a smaller rise in fasting
insulin and the smokers who became nonsmokers had
a greater rise in insulin. These smoking findings, while
sometimes obtaining statistical significance, were generally not consistent or large among the various race-,
sex-groups.
Murtivariate repeated measures analyses
We ran a number of multivariate models using combinations of variables in table 3. By far the most
important predictors of fasting insulin level were body
mass index and waist/hip ratio. When these variables,
along with physical activity, age, and time period,
were considered, there was little contribution of smoking, dietary intake, alcohol consumption, or family
history to insulin level. Table 4 shows the estimates
for the final model.
Body mass index was associated strongly and positively with fasting insulin, adjusting for waist/hip
ratio, physical activity, age, and time period (table 4).
Average insulin was about 3 jxU/mL higher per 5
kg/m2 in men, and 2 /xU/mL higher per 5 kg/m2
in women. The longitudinal relation was stronger:
insulin increased about 5 /lU/mL per 5 kg/m2 increase
in body mass index, somewhat less in white women
(3.0 /iU/mL per 5 kg/m2). In the same model, average
waist/hip ratio was related strongly and positively to
average fasting insulin, but change in waist/hip ratio
was weakly, and generally not statistically significantly, related to change in fasting insulin. Similarly,
242
Folsom et al.
TABLE 3. Estimated change in fasting sorum insulin (jdVml) P*f specified differenca In related
variablas oonsldered Individually, the CARDIA Study, 1986-1986 to 1992-1993
Predfctor
variabtet
Bod/ mass index (5 kg/m2)
Average
Deviation
Waist/hip ratio (0.08 unit)
Average
Deviation
Physical activity (100 units)
Average
Deviation
Alcohol intake (10 ml/day)
Average
Deviation
Energy (1,700 kcal/day)
Average
Deviation
Total fat (6.0% of energy)
Average
Deviation
Saturated fat (3.0% of energy)
Average
Deviation
Monounsaturated fat (2.5% of energy)
Average
Deviation
Polyunsaturated fat (2.0% of energy)
Average
Deviation
Cholesterol (65 g/1,000 kcal)
Average
Deviation
Protein (2.5% of energy)
Average
Deviation
Carbohydrates (7.0% of energy)
Average
Deviation
Sucroae (6.0% of energy)
Average
Deviation
Magnesium} (50 g/1,000 kcal)
Average
Deviation
Black
men
Black
women
While
man
White
women
3.8*
4.9*
2.7*
4.8*
3.9*
5.8*
2.7*
3.4*
5.6*
2.7*
4.1*
2.5*
3.9*
2.8*
3.2*
2.8*
-0.3*
-0.2*
-0.3*
-0.2
-0.3*
-0.1
-0.4*
-0.1
-0.1
-0.5*
0.1
-0.6*
-0.1
-0.1*
-0.4*
-0.1
02.
0.0
-0.1
-0.0
0.5
0.3
0.9*
0.2
0.4
0.1
0.1
0.1
0.7*
0.3*
0.5*
0.3*
0.5*
0.5*
0.5*
0.4*
0.4
-0.2
-0.2
0.1
0.3
0.1
0.8*
0.5*
-0.0
0.2
0.2*
0.2
0.2
0.1
0.1
0.3
0.1
0.0
0.0
0.3
0.4
-0.9*
-0.1
-0.4
0.7*
0.3
0.5
0.3
0.1
0.3
0.4
-0.3
0.8*
-0.0
-0.1
-0.2
-0.1
0.1
0.1
-0.5*
-0.3*
-0.3
-0.3*
0.7
-0.0
0.2
0.1
0.8*
-0.2
-0.8*
-0.7*
-1.2*
-0.1
-0.4
-0.2
-1.2*
-O.3
-0.7*
-0.3*
0.1
• p < 0.05.
t Each race- and sax-specific repeated measures analysis included terms for the average level and deviation
of a single predictor variable, as well as age, time period, and their interaction. Specified differences in parentheses are approximately one standard deviation.
$ Includes supplements.
average physical activity level was associated strongly
and inversely to average fasting insulin, but change in
physical activity was weakly, and not statistically significantly, related to change in fasting insulin, adjusted
for body mass index, waist/hip ratio, age, and time
period.
The relation of age with fasting insulin in the model
described in table 4 was complex because of an age by
time interaction. That is, the multivariate adjusted
relation of fasting insulin with age depended on the
examination. The cross-sectional association of insulin
with age was negative at baseline, as previously reported (15), but much closer to flat at year 7 (data not
shown). The interaction also indicated that insulin
decreased longitudinally in the youngest members of
the 18-30-year-old CARDIA cohort, but remained
steady in the older participants.
An interaction term, for body mass index average by
its deviation, added to the model in table 4 was statistically significant (p < 0.01) in white men and
Am J Epidemiol
Vol. 144, No. 3, 1996
Seven-Year Insulin and Glucose Trends
243
TABLE 4. Estimated change in fasting scrum Insulin OiU/mL) par specified difference in related
variables considered simultaneously, the CARDIA Study, 1985-1988 to 1992-1993
Predictor
vaitaUet
Body mass index (5 kg/hi*)
Average
Deviation
Waist/hip ratio (0.08 unit)
Aye rage
Deviation
Physical activity (100 units)
Aye rage
Deviation
Black
men
Black
women
Whfie
man
While
women
3.0*
4.5*
2.2*
4.7*
3.3*
5.6*
22*
3.0*
3.2*
1.0
2.A*
0.2
1.4*
0.5
1.3*
1.1*
-0.20*
-0.14
-0.25*
-0.02
-0.30*
-O.04
-0.13*
0.00
• p < 0.05.
f Each race- and sex-specific repeated measures analysis included terms for average level and deviation of
aD variables, aa weO as age, time period, and their interaction.
white women only. A waist/hip average by body mass
index deviation term, instead, was statistically significant {p < 0.01) in all race-, sex-groups. Corresponding to table 2, this interaction indicated a greater
increase in insulin per increment change in body mass
index for those with higher average waist/hip than
those with lower average waist/hip.
The model in table 4 was also run using the fasting
insulin/glucose ratio, another index of insulin resistance, as the dependent variable (not shown). It
yielded qualitatively similar conclusions to those for
table 4.
Table 5 shows race-, sex-specific models for fasting
glucose using the same predictors. Average body mass
index, its deviation, and average waist/hip ratio were
strongly positively associated with glucose level. Notably, neither average nor deviation in physical activity was associated with fasting glucose.
Race, sex, and fasting insulin
To examine the race and sex relations with fasting
insulin, we constructed a repeated measures model
that included all subjects and variables to indicate
race, sex, race by time period, and sex by time period
interactions. After adjustment for age and time period,
blacks were found to have average fasting insulin
values 2.0 jiU/mL higher than whites {p < 0.01), and
insulin increased 1.1 ptU/mL more (p < 0.01) in
blacks than whites over the 7-year period. Men had
average fasting insulin levels 0.5 ^xU/mL lower than
women (p = 0.02) but levels increased 0.9 /xU/mL
more in men than women (p < 0.01) over the 7-year
period. In a model that also adjusted for average and
deviation in body mass index, waist/hip ratio, and
physical activity, blacks still had 1.1 /xU/mL higher
average fasting insulin levels (p < 0.01), but the
change in insulin was no longer different between
blacks and whites.
DISCUSSION
The CARDIA Study has previously documented
substantial 7-year declines in mean levels of physical
activity (29) and increases in caloric intake and weight
(30) in its cohort of adults initially aged 18-30 years.
These findings are consistent with the nationwide
trend toward higher body weight over the past two
TABLE 5. Estimated change in fasting serum glucose (mg/dL) per specified difference in related
variables, the CARDIA Study, 1985-1986 to 1992-1993
Predictor
variaHet
Body mass index (5 kg/m2)
Average
Deviation
Waist/hip ratio (0.08 unit)
Average
Deviation
Physical activity (100 units)
Average
Deviation
Black
men
1.5*
3.7*
Black
women
Whfie
1.5*
2.7*
2.6*
3.8*
1.6* .
2.4*
1.0*
1.5*
men
3.2*
1.3*
1.5*
1.4
0.7
0.2
0.1
-0.1
0.0
-0.2
0.0
-O.1
Whfie
women
0.0
0.0
* p < 0.05.
t Each race- and sex-specific repeated measures analysis included terms for average level and deviation of
all variables, as well as age, time period, and their interaction.
Am J Epidemiol
Vol. 144, No. 3, 1996
244
Folsom et al.
decades (31). More importantly, they demonstrate that
young adulthood in the United States is accompanied
by a significant deterioration in these indicators of
energy balance.
To date, the impact of sedentary life-style and obesity on serum insulin and glucose concentrations, or on
insulin sensitivity, has been demonstrated primarily
from cross-sectional epidemiologic studies (15-17)
and short-term weight loss or exercise trials (32-35).
The few extant longitudinal natural history studies of
weight change and insulin have examined Pima Indians (2) and women undergoing menopause (20), but
relatively few data are available from ethnically mixed
populations of young adults. It is especially important
to understand the evolution of hyperinsulinemia and
hyperglycemia in young adulthood, because of the
roles diat these metabolic abnormalities may play in
the development of NIDDM (1-3) and atherosclerotic
vascular disease (8-11).
We chose to measure fasting insulin rather than
other markers of insulin sensitivity for pragmatic reasons. Among persons with normal glucose tolerance,
fasting insulin has a moderately high correlation (typically r > 0.6) with insulin resistance measured by the
hyperinsulinemic, euglycemic clamp or by the intravenous glucose tolerance test with minimal modeling
(4, 36). We also chose to remeasure Year 0 insulin
levels on stored serum samples along with the Year 7
samples, incorporating a highly specific insulin radioimmunoassay (26). It is possible that the Year 0 samples had deteriorated somewhat, causing the insulin
changes to be over- or underestimated. However, prior
short-term data suggest that insulin is stable at — 70°C
(37). There was a correlation of 0.81 between original
and remeasured Year 0 insulin concentrations, using
different methods, suggesting that at least rank ordering of insulin levels was well preserved.
These CARDIA results demonstrate in all four racesex groups that an individual's level and 7-year
changes in fasting insulin and glucose concentrations
were linked strongly and positively to level and
change in body mass index. This corroborates previous longitudinal findings in other, quite different populations (2, 20).
Although the link of adiposity with hyperinsulinemia and hyperglycemia is incompletely understood,
the leading theory is that obesity promotes insulin
resistance through increased levels of free fatty acids
and lipid oxidation in muscle at the expense of glycogen synthesis and utilization. Consistent with shortterm weight reduction trials (32, 33), weight maintenance or loss in these free-living CARDIA participants
was associated, on average, with no increase or with a
decrease in insulin concentrations. This suggests that
hyperinsulinemia is preventable or reversible, which
may be particularly important in young adulthood
when it has persisted only a few years.
We also found that CARDIA participants with a
higher average waist/hip ratio or body mass index
exhibited a greater change in insulin with weight
change than did those with lower levels. This suggests
that weight gain may particularly disrupt insulin sensitivity in those with greater adiposity, especially excess visceral adiposity. This is consistent with the very
high risk of diabetes among those with both a high
body mass index and a high waist/hip ratio (38, 39).
While waist/hip ratio was a good predictor of insulin
level, change in waist/hip ratio did not predict change
in insulin very well. The shortcoming of this ratio as a
measure of change in visceral adiposity has been recognized previously (40).
The fact that levels and changes in body mass were
the strongest correlates of fasting insulin and glucose
change does not negate the importance of levels and
changes in physical activity and energy intake. The
balance of energy intake and expenditure determines
body weight. Insulin (but not glucose) was associated
significantly with the average level of physical activity
in the multivariate model but only weakly with change
in physical activity. Insulin (and glucose) were not
materially associated with energy intake. Physical activity and energy intake, however, are measured imprecisely by epidemiologic questionnaires (25), and
thus we may have underestimated their relations with
insulin levels. Measurement of change in physical
activity in a largely sedentary population may be especially difficult, thereby obscuring longitudinal associations with insulin. Alternatively, the discrepancy
between the longitudinal and cross-sectional associations for physical activity may reflect different contributions of short-term versus lifelong physical activity,
respectively. Exercise trials provide stronger evidence
of an independent contribution of physical activity and
fitness to insulin sensitivity and glucose tolerance (34,
35).
Dietary feeding studies have demonstrated clearly
that short-term alterations of carbohydrate content (41)
and starvation (42) can change fasting insulin concentrations. Whether the nutrient composition of the habitual diet contributes to hyperinsulinemia and insulin
resistance is uncertain. Total and saturated fat seem to
decrease and polyunsaturated fat to increase the fluidity of cell membranes, the number and binding of
insulin receptors, and the rate of insulin-stimulated
glucose transport (43). Several investigators (15, 17),
including the CARDIA investigators (15), have shown
cross-sectionally in humans that total fat, saturated fat,
or sucrose may be associated positively, whereas polyAm J Epidemiol
Vol. 144, No. 3, 1996
Seven-Year Insulin and Glucose Trends
unsaturated fat, magnesium, and alcohol may be associated negatively with insulin level. In our bivariate
longitudinal models, fasting insulin tended to be associated negatively but weakly with alcohol intake.
Among whites but not blacks, level and change in
fasting insulin also tended to be associated positively
with level and deviation in intake of total, saturated,
and monounsaturated fatty acids, and negatively with
carbohydrates and magnesium. These associations
were weak, potentially as a result of the low accuracy
of dietary questionnaires; accuracy was lower in
CARDIA blacks than whites (25). Nevertheless, the
data suggest that a higher dietary fat to carbohydrate
ratio may contribute to greater obesity, with concomitant insulin resistance.
The association of insulin with the behavioral factors examined, other than diet, tended to be similar for
blacks and whites. Insulin concentrations were higher
at Year 0 among blacks than whites, and especially
higher among black women, who also had the highest
average body mass index. Adjustment for body mass
index, waist/hip ratio, and physical activity cut the
black-white difference in fasting insulin in half but did
not eliminate it. Adjustment for current body mass
index, however, may not fully account for possible
different durations of obesity in blacks versus whites.
The increase in insulin concentration over the 7 years
was also greater in blacks; however, adjustment for
body mass index, waist/hip ratio, and physical activity
eliminated this difference. While our data cannot rule
out a true difference in insulin sensitivity between
blacks and whites, a large part of the apparent difference seems to be explained by differences in adiposity
and energy balance.
We expected parental history of NIDDM to be associated positively with insulin and glucose concentrations (17-19). It was in several race-, sex-groups,
but the estimate of the effect was not striking. Furthermore, there was not a greater impact of weight gain on
insulin in those with versus without a family history of
diabetes (data not shown). We did not validate parental history through medical records, so its effect may
have been underestimated through misclassification.
Alternatively, many of the parents of CARDIA participants may still be too young to have expressed a
diabetic phenotype.
Cigarette smoking, although not consistent among
race-, sex-groups, seemed to be associated with lower
insulin concentrations. This contrasts with previous
reports that smoking may exacerbate insulin resistance
(44). The negative association that we observed might
be explained by the fact that smokers tend to be
thinner than nonsmokers (45). Indeed, smoking was
not materially associated with insulin levels once we
Am J Epidemiol Vol. 144, No. 3, 1996
245
accounted for body mass index, waist/hip ratio, physical activity, age, and time period.
In this 18-30-year-old cohort, the relation of age
with fasting insulin was complex because of an age by
time interaction. In the youngest members (<22
years), insulin declined with age (time), which likely
reflects waning growth between examinations 0 and 7.
A rise and fall in fasting insulin with puberty is well
documented (46). In the older CARDIA cohort members, there was little change in fasting insulin with age
over the brief 7 years studied. Thus, increasing body
mass index and decreasing physical activity, not aging
or secular trend, accounted for most of the overall
increase in insulin over time in CARDIA.
The response rate to baseline recruitment of the
CARDIA cohort was 50 percent (22), but about 80
percent of the cohort returned for the Year 7 examination. While selection factors may have operated to
produce biased associations, this seems unlikely because the glucose and insulin concentrations and most
other risk factors measured at the Year 0 examination
for CARDIA participants included in this analysis did
not differ from those of the original cohort. Furthermore, the longitudinal and cross-sectional associations
were for the most part quite consistent.
In conclusion, increasing weight and inactivity in
young adulthood contribute to increased insulin and
glucose concentrations. This finding is generally similar in men and women, blacks, and whites. Elevated
levels of insulin, and insulin resistance, are known to
be associated with adverse changes in several well
established cardiovascular risk factors. It remains to be
established whether higher insulin or glucose levels in
young adulthood are associated with greater risk of
future diabetes and atherosclerotic vascular disease.
Yet, assuming that they are (2, 3), our findings reinforce the need for young adults to be concerned with
energy balance and weight gain.
ACKNOWLEDGMENTS
The CARDIA Study is carried out as a collaborative
study supported by contract nos. N01-HC-48047, N01-HC48048, N01-HCM8049, N01-HC-48050 and N01-HC95095 from the National Heart, Lung and Blood Institute.
The authors thank Peter Hannan, Laura Kemmis, and
Heather McCreath for technical assistance.
REFERENCES
1. DeFronzo RA, Bonadonna RC, Ferrannini E. Pathogenesis of
NIDDM: a balanced overview. Diabetes Care 1992;15:
318-68.
2. Saad MF, Knowler WC, Pettitt DJ, et al. Sequential changes in
246
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
Folsom et al.
serum insulin concentration during development of noninsulin-dependent diabetes. Lancet 1989;l:1356-9.
Martin BC, Warram JH, Krolewski AS, et al. Role of glucose
and insulin resistance in development of type 2 diabetes
mellitus: results of a 25-year follow-up study. Lancet 1992;
340:925-9.
Laakso M. How good a marker is insulin level for insulin
resistance? Am J Epidemiol 1993;137:959-65.
Reaven GM. Role of insulin resistance in human disease.
Diabetes 1988;37:1595-1607.
DeFronzo RA, Ferrannini E. Insulin resistance: a multifaceted
syndrome responsible for NIDDM, obesity, hypertension, dyslipidemia, and atherosclerotic cardiovascular disease. Diabetes
Care 1991;14:173-94.
Juhan-Vague I, Vague P. Hypofibrinolysis and insulinresistance. Diabetes Metab 1991;17:96-100.
Folsom AR, Eckfeldt JH, Weitzman S, et al. Relation of
carotid artery wall thickness to diabetes mellitus, fasting glucose and insulin, body size, and physical activity. Stroke
1994;25:66-73.
Pybraia K. Relationship of glucose tolerance and plasma insulin to the incidence of coronary heart disease: results from
two population studies in Finland. Diabetes Care 1979;2:
131-41.
Welborn TA, Wearne K. Coronary heart disease incidence and
cardiovascular mortality in Busselton with reference to glucose and insulin concentrations. Diabetes Care 1979;2:
154-60.
Ducimetiere P, Eschwege E, Papoz L, et al. Relationship of
plasma insulin levels to the incidence of myocardial infarction
and coronary heart disease mortality in a middle-aged population. Diabetologia 1980; 19:205-10.
Jarrett RJ. Why is insulin not a risk factor for coronary heart
disease? Diabetologia 1994;37:945-7.
Reaven GM, Laws A. Insulin resistance, compensatory hyperinsulinaemia, and coronary heart disease. Diabetologia 1994;
37:948-52.
Stern MP. The insulin resistance syndrome: the controversy is
dead, long live the controversy! Diabetologia 1994;37:956-8.
Manolio TA, Savage PJ, Burke GL, et al. Correlates of fasting
insulin levels in young adults: the CARDIA Study. J Clin
Epidemiol 1991;44:571-8.
Nabulsi AA, Folsom AR, Heiss G, et al. Fasting hyperinsulinemia and cardiovascular disease risk factors in nondiabetic
adults: stronger associations in lean versus obese subjects.
Metabolism 1995;44:914-22.
Feskens FJM, Loeber JG, Kromhout D. Diet and physical
activity as determinants of hyperinsulinemia: the Zutphen
Elderly Study. Am J Epidemiol 1994;140:350-60.
Haffher SM, Stern MP, Hazuda HP, et al. Increased insulin
concentrations in nondiabetic offspring of diabetic parents.
N Engl J Med 1988;319:1297-1301.
Eriksson J, Franssila-Kallunki A, Ekstrand A, et al. Early
metabolic defects in persons at increased risk for non-insulindependent diabetes mellitus. N Engl J Med 1989:321:337-43.
Wing RR, Matthews KA, Kuller LH, et al. Environmental and
familial contributions to insulin levels and change in insulin
levels in middle-aged women. JAMA 1992;268:1890-5.
Friedman GD, Cutter GR, Donahue RP, et al. CARDIA: study
design, recruitment, and some characteristics of the examined
subjects. J Clin Epidemiol 1988;41:1105-16.
Jacobs DR Jr, Hahn LP, Haskell WL, et al. Validity and
reliability of short physical activity history: CARDIA and the
Minnesota Heart Health Program. J Cardiopulmonary Rehabil
1989;9:448-59.
McDonald A, Van Horn L, Slattery M, et al. The CARDIA
dietary history: development and implementation. J Am Diet
Assoc 1991;91:1104-12.
Hilner JE, McDonald A, Van Horn L, et al. Quality control of
dietary data collection in the CARDIA Study. Control Clin
Trials 1992;13:156-69.
25. Liu K, Slattery M, Jacobs D Jr, et al. A study of the reliability
and comparative validity of the CARDIA dietary history.
Ethnicity Dis 1994;4:15-27.
26. Herbert V, Lau KS, Gottlieb CW. Coated charcoal immunoassay of insulin. J Clin Endocrinol 1965;25:1375-84.
27. Haffner SM, Bowsher RR, Mykkanen L, et al. Proinsulin and
specific insulin concentration in high- and low-risk populations for NIDDM. Diabetes 1994;43:1490-3.
28. Harris MI, Hadden WC, Knowler WC, et al. Prevalence of
diabetes and impaired glucose tolerance and plasma glucose
levels in US population aged 20-74 years. Diabetes 1987;36:
523-34.
29. Anderssen N, Jacobs DR Jr, Sidney S, et al. Change and
secular trends in physical activity patterns in young adults: a
seven-year longitudinal follow-up in the Coronary Artery Risk
Development in Young Adults Study (CARDIA). Am J Epidemiol 1996;143:351-62.
30. Burke GL, Bild DE, Hilner JE, et al. Differences in weight
gain in relation to race, gender, age and education in young
adults: the CARDIA Study. Ethn Health (in press).
31. Kuczmarski RJ, Flegal KM, Campbell SM, et al. Increasing
prevalence of overweight among US adults: the National
Health and Nutrition Examination Surveys, 1960 to 1991.
JAMA 1994;272:205-ll.
32. Golay A, Felber JP, Dusmet M, et al. Effect of weight loss on
glucose disposal in obese and obese diabetic patients. Int J
Obes 1985;9:181-90.
33. Goldstein DJ. Beneficial health effects of modest weight loss.
Int J Obesity 1992;16:397-415.
34. Devlin JT. Effects of exercise on insulin sensitivity in humans.
Diabetes Care 1992;15:1690-3.
35. King H, Kriska AM. Prevention of Type II diabetes by physical training: epidemiological considerations and study methods. Diabetes Care 1992;15:1794-9.
36. Saad MF, Anderson RL, Laws A, et al. A comparison between
the minimal model and the glucose clamp in the assessment of
insulin sensitivity across the spectrum of glucose tolerance.
Insulin Resistance Atherosclerosis Study. Diabetes 1994;43:
1114-21.
37. Kubasik NP, Ricotta M, Hunter T, et al. Effect of duration and
temperature of storage on serum analyte stability: examination
of 14 selected radioimmunoassay procedures. Clin Chem
1982;l:164-5.
38. Kaye SA, Folsom AR, Sprafka JM, et al. Increased incidence
of diabetes mellitus in relation to abdominal adiposity in older
women. J Clin Epidemiol 1991;44:329-34.
39. Ohlson L-O, Larsson B, Svardsudd K, et al. The influence of
body fat distribution on the incidence of diabetes mellitus:
13.5 years of follow-up of the participants in the Study of Men
Born in 1913. Diabetes 1985;34:1055-8.
40. van der Kooy K, Leenen R, Seidell JC, et al. Waist-hip ratio
is a poor predictor of changes in visceral fat. Am J Clin Nutr
1993:57:327-33.
41. Grey N, Kipnis DM. Effect of diet composition on the hyperinsulinemia of obesity. N Engl J Med 1971 ;285:827-31.
42. Sims EAH, Horton ES, Salans LB. Inducible metabolic abnormalities during development of obesity. Annu Rev Med
1971;22:235-50.
43. Clandinin MT, Cheema S, Field CJ, et al. Dietary lipids
influence insulin action. Ann N Y Acad Sci 1993;683:151-63.
44. Eliasson B, Attvall S, Taskinen M-R, et al. The insulin resistance syndrome in smokers is related to smoking habits.
Arterioscler Thromb 1994;14:1946-50.
45. Albanes D, Jones Y, Micozzi MA, et al. Associations between
smoking and body weight in the US population: analysis of
NHANES n. Am J Public Health 1987;77:439-44.
46. Smith CP, Dunger DB, Williams AJK, et al. Relationship
between insulin, insulin-like growth factor I, and dehydroepiandrosterone sulfate concentrations during childhood, puberty, and adult life. J Clin Endocrinol Metab 1989:68:932-7.
Am J Epidemiol
Vol. 144, No. 3, 1996