American Journal of Epidemiology
Copyright O 1996 by The Johns Hopkins University School of Hygiene and Public Health
All rights reserved
Vol. 143, No 4
Printed In U S.A
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)
Norman Anderssen,1 David R. Jacobs, Jr.,2 Stephen Sidney,3 Diane E. Bild,4 Barbara Stemfeld,3
Martha L Slattery,5 and Peter Hannan2
exercise; longitudinal studies; racial stocks
Habitual physical activity (1-6) and physical fitness
(7, 8) are inversely related to the incidence of coronary
heart disease, cardiovascular disease, and all-cause
mortality. Data indicate that at the population level
there are potentially major health benefits from increasing fitness levels in the least fit. One way to
achieve this would be for these groups to increase their
levels of habitual leisure-time physical activity ( 8 10).
To develop programs to promote physical activity, it
is important to know when lasting habits of leisuretime physical activity are developed. Perry et al. (11)
strongly stress the need for examination and research
on the etiology of health behavior patterns, including
physical activity. They state, "Risk-related behaviors
can be viewed as social behaviors that are learned,
developed, and become prevalent in childhood and
adolescence" (11, p. 409). Some evidence indicates
that physical activity patterns learned during adolescence carry over into adulthood (12), while other studies report no association (13, 14). All in all, there is no
conclusive evidence supporting the idea of tracking of
physical activity (15-18) from adolescence into early
adulthood ("tracking" refers to people tending to
maintain their level of activity relative to that of others
(19)). Despite scanty evidence regarding the tracking
of activity from young to middle adulthood, there are
data to suggest that the strongest predictor of currentparticipation in an adult exercise program is past participation, at least for structured programs (20). During
this stage of life, increases in family and work responsibilities may significantly influence individual healthpromoting behaviors. Thus, it is important to know to
what degree physical activity patterns remain stable or
change through late adolescent and early adult years
into middle age and to identify factors that may influence the processes of change (21).
Received for publication January 9, 1995, and in final form November 28, 1995.
Abbreviation: CARDIA, Coronary Artery Risk Development in
Young Adults Study.
1
Center for Health Promotion, University of Bergen, Bergen,
Norway.
2
Division of Epidemiology, University of Minnesota School of
Public Health, Minneapolis, MN.
3
Division of Research, Kaiser-Permanente Medical Care Program, Oakland, CA.
* Division of Clinical Applications, National Heart, Lung, and
Blood Institute, Bethesda, MD.
8
University of Utah, School of Medicine, Salt Lake City, UT.
Reprint requests to Dr. David R. Jacobs, Jr., Division of Epidemiology, University of Minnesota School of Public Health, 1300
South Second Street, Suite 300, Minneapolis, MN 55454-1015.
351
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on April 24, 2014
Levels and changes in self-reported physical activity over a 7-year penod were examined to determine
tracking and to estimate the proportion of total cohort change attributable to secular trends. A populationbased sample of 2,328 men and 2,787 women aged 18-30 years at baseline (52% black and 48% white) from
Birmingham, Alabama, Chicago, Illinois, Minneapolis, Minnesota, and Oakland, California, were examined four
times between 1985-1986 and 1992-1993. The intraclass correlation for up to four measures was 0.57 for the
entire sample, varying between 0.57 for white men and 0.42 for black women, indicating a moderate tendency
for tracking. The energy expenditure in physical activity at each examination was greatest in black men and,
compared with black men, about 5% less in white men, 30% less in white women, and 50% less in black
women. The total cohort decrease in mean physical activity was approximately 30% in each race-sex group.
The secular trend accounted for 38% of the total cohort change in black men, 43% in black women, 52% in
white men, and 81 % in white women. Physical activity declined sharply during the early years of adulthood,
partly because of secular trend. Young adults are therefore an important target group for physical activity
promotion programs to reverse individual and populationwide declines prior to middle age. Am J Epidemiol
1996;143:351-62.
352
Anderssen et al.
Leisure-time physical activity seems to decline in
late adolescent and young adult years (22-26). However, most of these studies are cross-sectional and do
not permit separating total cohort change into period,
birth cohort, and aging effects. One study suggests that
the general population of men in the upper northwestern United States exhibited a secular trend for increasing mean leisure time physical activity from 1957 to
1987 (27). However, another recent analysis, focused
only on the 1980s, suggests little change and perhaps
a decline in vigorous intensity activity during that
decade (28).
Thus, the purpose of the present analysis is to
MATERIALS AND METHODS
The Coronary Artery Risk Development in Young
Adults Study (CARDIA) population
The CARDIA is a longitudinal observational study
of the development of coronary risk factors in relation
to physiologic and lifestyle variables in young black
and white men and women. The study population was
recruited from four geographic areas by communitybased sampling in Birmingham, Alabama, Chicago,
Illinois, and Minneapolis, Minnesota, and by sampling
from the membership of a large prepaid health care
program in Oakland, California. Details about recruitment procedures have been reported (29, 30). At baseline, the sample consisted of 2,328 men and 2,787
women aged 18-30 years. Fifty-two percent were
black and 48 percent were white. Participants were
examined four times using a variety of physiologic and
self-report measures, including a physical activity
questionnaire. Baseline (year 0) information was recorded in 1985-1986. The second (year 2) examination was in 1987-1988 (response rate = 90.4 percent),
the third (year 5) examination was in 1990-1991
(response rate = 85.7 percent), and the fourth (year 7)
was in 1992-1993 (response rate = 80.6 percent). The
dropout rate was greater for blacks than for whites; 61
percent (701/1,157) of black men and 66 percent (983/
1,479) of black women attended all four examinations,
compared with 79 percent (926/1,170) of white men
and 78 percent (1,022/1,307) of white women. Despite
this differential response rate, findings concerning
physical activity were similar in analyses including all
Measurement of physical activity
Assessments of physical activity levels were obtained using the interviewer-based CARDIA Physical
Activity History, which covers 13 different types of
activities. The eight vigorous intensity activities were
jogging, racket sports, bicycling, swimming, vigorous
exercise/dancing, weight lifting, vigorous job activity,
and other strenuous sports. The five moderate intensity
activities were nonstrenuous sports, walks/hikes,
bowling/golf, home exercise, and home maintenance.
Each activity was scored according to whether the
activity was performed for at least an hour during any
1 month during the past year (activity ever performed
in the past year), the number of months it was performed at that level (consistency of performance
throughout the year), and the number of months the
activity was performed frequently (based on a cutpoint
for frequent participation that varied according to the
activity, i.e., frequency/duration of activity in each
month). The physical activity score was expressed in
exercise units and consisted of a sum of moderate
intensity and vigorous intensity scores. Exercise units
are related to caloric expenditure (31, 32), but the
questionnaire did not ask about the duration of exercise sessions and therefore permits no direct estimate
of caloric expenditure. To assist in understanding the
score, we note that performing at least one of the
vigorous activities on a regular, frequent basis for at
least 6 of the previous 12 months roughly meets the
American College of Sports Medicine recommendation (33). At baseline, 23 percent of persons with
scores of 100-199 exercise units met this criterion.
Corresponding percentages reporting frequent activity
in at least 6 of the past 12 months, of persons with
scores of 300-399 exercise units, 500-599 exercise
units, and 800 or more exercise units, were 74 percent,
93 percent, and 100 percent. The questionnaire shows
a test-retest reliability in the range of 0.77- 0.84, and
scores on the questionnaire are associated with external validation criteria (treadmill test performance, caloric intake, skinfold thickness, high density lipoprotein cholesterol, other measures of physical activity)
(31, 32, 34). Details of the scoring system have been
published (32).
Am J Epidemiol
Vol. 143, No. 4, 1996
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on April 24, 2014
1. examine to what degree the level of physical
activity in late adolescence and young adulthood
predicts physical activity later (tracking) and
2. describe changing physical activity levels in
young adults during a 7-year period, with particular reference to what part of the change was
attributable to secular trends.
available participants or only those who participated in
all examinations.
Women who reported being pregnant or breastfeeding at the time of an examination may have lower
levels of physical activity. One-way analyses of variance of physical activity at the first three examinations, however, did not show statistically significant
differences between pregnant/lactating women and the
rest, and therefore all women were included.
Trends in Physical Activity
Demographic variables included the number of
years of education (queried at each examination), age,
clinical center, race, and sex.
Statistical methods
Am J Epidemiol
Vol. 143, No. 4, 1996
her personal average, i.e., nonzero for those participants who continued in school during CARDIA and
zero for those who did not continue).
Estimation of the age-related and time-related components of the physical activity level and change is
complicated by three facts: 1) the change in age and
change in time are synonymous; 2) the level of physical activity at a given age may depend on which
calendar years the person lived through (birth cohort
effect/historical time); and 3) the slope of physical
activity on age may vary by age (curvature in the age
trend). We defined age at baseline, a time-independent
variable, and age at examination, a time-dependent
variable. We computed the coefficients of time and
age in regressions of physical activity 1) on time and
age at baseline and 2) on time and age at examination.
The age coefficients are algebraically identical in the
two regressions, estimating between person difference
in physical activity per year of age, which we call the
cross-sectional age effect. The time coefficient in the
first regression with age at baseline estimates the total
within person time trend from both age- and timerelated sources, which we call the total cohort change
in physical activity. The time coefficient in the second
regression with age at examination estimates the agematched time trend, namely, the change in physical
activity after removing between person differences in
age; this coefficient includes purely time-related effects, such as the secular trend and methodological
artifacts. Both cross-sectional age effects and the agematched time trend are possibly confounded with historical and age curvature effects. To assess the consistency of age and time trends, we examined the
graphs of mean physical activity levels for each single
year of age at each examination, adjusted for education and clinical center.
RESULTS
There were small, but statistically significant differences between examination attendance categories in
age and educational attainment with a tendency for
nonresponders to be younger and less well educated
(table 1). The physical activity level varied by attendance category only in white women. Age and education did not differ between black men and women who
attended all four examinations (table 1). White women
attending all four examinations were 0.2 years older
than were white men at baseline, with no differences
in educational attainment. Black participants attending
all four examinations were 1.1 year younger than were
white participants and had 1.6 years less of educational attainment.
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on April 24, 2014
The data were analyzed as a repeated measures
design. Missing observations were primarily due to
nonresponse at the 2-, 5-, and 7-year examinations (see
table 1). The baseline physical activity score (mean)
was compared among persons who attended only baseline, persons who attended two or three examinations,
and persons who attended all four examinations. As
qualitative findings were similar for either arithmetic
or geometric means (using the natural logarithmic
scale), most analyses were done on the less complex
arithmetic scale, specific to race and sex.
Changes in physical activity with time and within
person correlation and standard deviation were assessed using SAS version 6.07 PROC MIXED software (35) using restricted maximum likelihood, with
each person contributing baseline and up to three
subsequent observations, depending on which examinations were attended. Because the assumption of a
distinct within person correlation coefficient between
each pair of examinations (unstructured covariance
structure) yielded results similar to those assuming a
common correlation, in most analyses the correlation
between successive physical activity measurements
was assumed not to depend on the time elapsed between examinations (compound symmetry covariance
structure). The exception was in analyses of tracking.
Tracking was measured by 1) the within person
correlations for physical activity measured in pairs of
examinations in the same person up to four times,
under the unstructured covariance assumption, and 2)
the within person standard deviation of physical activity, pooled across examinations, under the compound
symmetry covariance assumption. In addition, the
intraclass correlation coefficient (the ratio of the
between person variance component to the total
variance) was reported as a summary within person
correlation.
The description of changes in physical activity over
7 years as linear was adequate. Race-sex-specific linear physical activity changes were adjusted for education and clinical center. Interactions between changes
in physical activity, time and age, educational attainment, and clinical center were also studied. Educational attainment was represented as two variables,
one representing the between person component (level
of education averaged over all examinations within
each person) and the other representing the within
person component (each person's deviations from his/
353
354
Anderssen et al.
Tracking of physical activity
3i
K
r- r- i-
p
Sf 8 o o o o
o
00
CM
eg
o
1(1 CO
in cvj
fl) S
1
o
8
o
o
a
is - s i s
i
8888
o o o d
T-
CVl
co co
o «
o.
o co
CM CO
'5 y» I
a CM
Mean changes in physical activity
If
if
«3 • * T - CJ
CM CM i r i CO
T-
op CM O
«
in in N
SS
o ^r o p
o o o d
1?
<t ^ i r i u i
CM CM CM OJ
O
is Si
O
CU CM CM CO
V Tt Tf <f
CM CM CM CM
(B D)
(^ O i- r^
m op r- o
S
O
|
O o
CO
E
s
»-
CO <O CM
I o
aO
a ^
ft
CO a >
If"
» - CD
co co
•- o
i
UJ
IIII
8
•6 -g 2 S
n a f f
55 m 5 s
Figure 1 gives arithmetic and geometric means of
the physical activity score by year of examination,
according to race and sex, adjusted for age, education,
and clinical center. The level of physical activity score
at each examination was greatest in black men and,
compared with black men, about 5 percent lower in
white men, about 30 percent lower in white women,
and about 50 percent lower in black women. A general
tendency toward decreasing physical activity was seen
in each race-sex group, with less change between
years 2 and 5 than between years 0 and 2 or years 5
and 7. Given the skewness of the physical activity
variable, the geometric mean, which in this case was
consistently about 100 exercise units lower than the
arithmetic mean (130 exercise units lower for black
men), gave the truer picture of the center of the physical activity distribution. Nevertheless, the time-,
race-, and sex-related patterns of geometric means
were similar to those of the arithmetic means. The
total cohort changes between years 0 and 2 and between years 5 and 7 were statistically significant,
while those between years 2 and 5 were only marginally statistically significant. The overall reduction in
the geometric mean physical activity score between
years 0 and 7 was 27 percent in black men, 32 percent
in black women, 22 percent in white men, and 33
percent in white women.
The total cohort change in physical activity score
per year, computed within individuals, is given in table
3, taken from models in which time was represented
linearly. The absolute decline was least in black
women (whose mean level of physical activity was
also lowest) and about twice as great in the other three
Am J Epidemiol
Vol. 143, No. 4, 1996
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on April 24, 2014
Q. * E
Table 2 shows that the intraclass correlation between physical activity measures at the different examinations was 0.57 for the sample as a whole and
slightly higher in men than in women and in whites
than in blacks. There was a minor violation of the
compound symmetry assumption, in that there was a
slight drop-off in the level of correlation over longer
intervening time periods. Further analyses (not shown)
showed that the level of tracking was similar for
participants aged 18-20, 21-25, and 26-30 years at
baseline. Though these correlation coefficients suggest
that there was a tendency for each person to maintain
his/her initial rank, there was also considerable within
person variation in this measure, ranging from a standard deviation of 167 exercise units in black women to
233 exercise units in black men.
Trends in Physical Activity
355
TABLE 2. Measure* of tracking of physical activity score determined at four CAROIA* examinations
between 1985-1986 and 1992-1993 adjusted for age, education, and clinical center
Within person corretationir
All participants
YearO
Year 2
Year 5
Black men
YearO
Year 2
Year 5
White men
YearO
Year 2
Year5
White women
YearO
Year 2
Year5
Between
person
standard
deviation*
Intraclass
correlation
for years
0-7§
Year 2
Yaar5
Year 7
0 61
0.52
0.63
0.49
0.57
0.66
189
217
0.57
0 55
0 47
0.59
0.42
0 52
063
233
244
053
0.47
0 36
0 49
0.34
0.42
0.50
167
142
0.42
0.60
0.53
0.60
0.49
0.55
0.69
187
217
0 57
0.52
044
058
0.41
0.50
0.60
171
169
0.50
* CARDIA, Coronary Artery Risk Development In Young Adults Study,
t Computed under the unstructured covariance structure
j Computed under the compound symmetry covariance structure.
§ Intraclass correlation is defined as between person varianca/(between person variance plus within person
variance)
race-sex groups. In general, the decline in physical
activity score per year was 2-3 percent of baseline
levels in each race-sex group. The total cohort change
did not vary by baseline age but did vary by the level
of educational attainment and by the clinical center, as
detailed below. The mean activity scores specific to
most moderate and vigorous intensity activities declined, including those for jogging, racket sports,
swimming, vigorous exercise/dancing, other strenuous
sports, nonstrenuous sports, walks/hikes, and home
exercise. The mean scores for bicycling, weight lifting, and bowling/golf did not change during the study.
The mean score for vigorous job activity increased in
blacks but did not change in whites, while the mean
score for home maintenance increased in whites but
not in blacks. Details of scores for specific activities
are too voluminous to present here.
Age and time trends in physical activity
Figure 2 presents the mean physical activity scores
at each single year of age at each examination for each
Am J Epidemiol
Vol. 143, No. 4, 1996
race-sex group, adjusted for education and center. The
left side of the figure plots the four examinations'
specific sequences of physical activity according to
baseline age; in this case, the vertical distance between
physical activity means represents the total cohort
change between examinations. In accord with table 3,
there is a decrease in physical activity at each baseline
age, particularly evident between the examinations at
years 0 and 7. The slopes of the four sequences are the
time-specific cross-sectional age effects and are seen
to be relatively consistent between examinations
within each race-sex group, implying no birth cohort
effects. The right side of the figure plots the four
examination-specific sequences of physical activity
according to age at examination; in this case, the
vertical distance between physical activity means is
the age-matched time trend. The similarity of the
cross-sectional age effects over the extended age range
of 18-37 years is emphasized in the right side of
figure 2.
The time-specific slopes of physical activity on age
are pooled as a single cross-sectional age effect in
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on April 24, 2014
Black women
YearO
Year 2
Year 5
Within
person
standard
deviattont
356
Anderssen et al.
arithmetic mean
geometric mean
o
M
0>
X
o
race-sex
0 2 5 7
0 2 5 7
black m a l e
black female
0 2 5 7
white male
0 2 5 7
white female
FIGURE 1. Mean physical activity at the four Coronary Artery Risk Development in Young Adults Study (CARDIA) examinations, 1985-1986
to 1992-1993, according to race and sex, adjusted for age, clinical center, within person average educational attainment, and time-specific
deviation from this average.
TABLE 3. Physical activity score changes In the total cohort, secular trend, and their difference, a
linear model for time, adjusted for age, education,* and center, under the compound symmetry
assumption: Coronary Artery Risk Development in Young Adults Study (CARDIA), 1985-1986 to
1992-1993
Total cohort
change
Black men
Black women
White men
White women
Age-matched
time trend
Cross-sectional
age effect
Slopet
ft
Slope§
f
Slope 1
f
-77 7
-42 0
-77 0
-85.4
-7.6
-6 5
-9.7
-12 9
^9.0
-29.4
-406
-20 3
-3 1
-3 6
-2 7
-1.8
-28.7
-13 3
-36 4
-65.1
-1.5
-1.3
-2 2
-4 9
1
Education is represented as two variables: each person's average over all the examinations and the deviaton
from this average at each examination.
t The slope Is the regression coefficient for time holding age at baseline constant, I e , the total cohort change
(exercise units per 7 years of time and age change)
X t is the regression coefficient divided by its standard error.
§ The slope is the regression coefficient for age (exercise units per 7 years of age drfference)
I The slope is the regression coefficient for time holding age at examination constant (exercise units per 7
years of time change).
table 3; the weighted average for black males, for
example, is —49 exercise units per 7 years of age
difference between two individuals. The physical activity score was estimated to decline with age by 5-10
percent of baseline levels in the four race-sex groups.
Age-matched time trends per 7 years (i.e., pooled
age-matched time trends from the right side of figure
2) are given in the rightmost columns of table 3;
exercise unit declines for 7 elapsed years were 29 in
black men, 13 in black women, 36 in white men, and
65 in white women. The age-matched time trend was
statistically significant only in whites. Most of the
total cohort change in physical activity among white
women was attributed to the age-matched time trend.
Education effects
The physical activity score (averaged within each
person over all the examinations attended) was greater
in those with higher average educational attainment,
Am J Epidemiol
Vol. 143, No. 4, 1996
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on April 24, 2014
e x a m i n a t i o n year
Trends in Physical Activity
357
Cross-Sectional Analysis: Graph Emphasizes Age-Matched
Time Differences (Differences Between Person)
Longitudinal Analysis: Graph Emphasizes
Total Cohort Change (Change Within Person)
650
black
men
350
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on April 24, 2014
400
black
women
600
white
men
500
500
450
white
women
350
300
250
IB
20
22
24
26
28
30
32
34
36
18
20
22
24
26
28
30
32
34
36
current age
FIGURE 2. Mean physical activity by age and year of examination in the Coronary Artery Risk Development in Young Adults Study
(CARDIA), 1985-1988 to 1992-1993. Gray curve, year 0; dark solid curve, year 2; broken curve, year 5; light solid curve, year 7.
Am J Epidemiol
Vol. 143, No. 4, 1996
358
Anderssen et al.
Clinical center effects
Physical activity levels (averaged over all examinations) were lower for participants in the Birmingham
TABLE 4. Repeated measures regression of physical activity
score on educational attainment, computed from a linear
model for each parson's average education and deviation
from average education, adjusted for time, baseline age, and
center, under the compound symmetry assumption: Coronary
Artery Risk Development In Young Adults Study (CARDIA),
1985-1986 to 1992-1993
Cross-sectional
Black men
Black women
White men
White women
Longitudinal
Slope'.t
f
Slope*
t
6.0
16.3
2.8
13.7
1.5
6.3
1.0
58
8.6
-2.6
-11.5
-2 2
2.7
-0.8
-2.8
-1 2
* Increase In physical activity score per year of education
attained, both averaged for each person over all examinations.
t Slope Is the regression coefficient (exercise units per yea/ of
education); f Is the regression coefficient divided by its standard
error.
$ Total cohort change in physical activity score for each
additional year of education attained during CARDIA.
clinic than in the other three clinics by 115 exercise
units in black men (t = -5.1), 76 exercise units in
black women (f = —6.4), 82 exercise units in white
men {t — —4.0), and 115 exercise units in white
women (t - -7.2). Table 5 shows the effect modification of the total cohort change of physical activity
score by race and sex, according to center. There was
little decline in the physical activity score in Birmingham and Oakland blacks. In each race-sex group, the
decline was steepest in Chicago. Except for black men
in Minneapolis, rates of decline in the physical activity
score were not statistically significantly different in
Birmingham or Minneapolis from those in Oakland.
DISCUSSION
The purpose of these analyses was to examine to
what degree physical activity tracked through early
adult years and to describe change and the secular
trend during a 7-year period in overall physical activity levels in the CARDIA population according to the
demographic factors sex, race, age, education, and
clinical center.
Tracking does not mean perfect correlation. Thus,
our analyses supported a tendency for physical activity
to track, as others also have found (36-38). Tracking
was lowest in black women and highest in white men,
with intraclass correlations of 0.42 and 0.57, respectively. Tracking was equal in those initially aged
18-20 years as in those initially aged in their late
twenties. It weakened as the time interval increased.
However, the within person standard deviations were
substantial in all four race-sex groups. The present
study is in agreement with recent well-designed studies (37, 38) that demonstrated similar patterns of tracking of physical activity in adolescents followed for
3-6 years. In Finns initially aged 18 years, tracking
correlation coefficients are remarkably similar to those
reported here (37), while tracking coefficients for
younger participants in their study were lower. The
Finnish study and ours give evidence that tracking of
habitual physical activity in young adults is stronger
than tracking in adolescents.
The overall lack of conclusive evidence of tracking
in earlier studies (17, 18) might be due to use of
different measures of type, frequency, and intensity of
physical activity and to the fact that prospective longitudinal studies are the exception (12—14). Most earlier studies did not use the same activity measure in
younger as in older participants. Possible further
weaknesses of earlier studies are recall bias and undocumented validity and reliability of measures (13,
14). Tracking of physical activity might also vary for
different populations and might be different in recent
years from in earlier years.
Am J Epidemiol
Vol. 143, No. 4, 1996
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on April 24, 2014
but much more so (and statistically significant only) in
women than in men. Regression estimates (adjusted
for time, age, and clinical center in table 4) were 6.0
exercise units higher per year of attained education in
black men (t = 1.5), 16.3 units higher (t = 6.3) in
black women, 2.8 units higher (t = 1.0) in white men,
and 13.7 units higher {t = 5.8) in white women. The
total cohort change in physical activity was modified
by the average attained education (time by average
education interaction) only among white women, with
an estimated decline of 51.1 exercise units during 7
years in those with 12 years of education compared
with a decline of 96.3 exercise units in those with 16
years of education (t = —4.1). The time by average
education interactions were small and not statistically
significant in the other race-sex groups (data not
tabulated).
There also was effect modification in the total cohort change in physical activity score per year of
additional education attained during the 7 years of
follow-up (table 4). In black men, the loss of physical
activity was partially mitigated by attainment of additional years of education during CARDIA; the decline
in physical activity score was 8.6 exercise units per
year less for each additional year of education attained
in black men (t = 2.7), while in white men the decline
in physical activity score was 11.5 exercise units more
(t = —2.8) for each additional year of education
attained. However, in women, change in the physical
activity score showed little association with years of
additional attained education.
Trends in Physical Activity
359
TABLE 5. Estimated total cohort change in physical activity score during 7 years by clinical center, a
linear model for time, center, and linear interaction with center, adjusted for baseline age and education,
under the compound symmetry assumption: Coronary Artery Risk Development in Young Adults Study
(CARDIA), 1985-1986 to 1992-1993
Birmingham,
Alabama
Slopat
Black men
Black women
White men
White women
-39.2
-9 1
-86.8
-73 5
tt
-0 9
05
-1.5
-0.7
Chicago,
Illinois
Minneapolis,
Minnesota
Slope
t
Slope
t
Oakland,
California,
slope
-160 3
-111.3
-116.2
-139.3
-5 0
-55
-2 8
-4.2
-123 9
-476
-58.1
-78.4
-4 1
-17
-0.2
-1 1
-140
-17 5
-53 9
-60 2
Omnibus
p value*
00001
0 0001
0.0132
0.0002
Tracking of these risk factors through childhood and
adolescence is receiving growing attention, because
behavioral and physiologic cardiovascular disease risk
factors (e.g., smoking and food choice, blood lipid
level) are evident among children and adolescents
(39). Tracking in young people appears to be strongest
for blood cholesterol. A recent review (21) reports
beneficial effects of physical activity on physiologic
risk factors (high density lipoprotein cholesterol and
body composition) in children and adolescents. Further, the Finnish study (37) found that changes in
serum insulin and triglycerides were inversely associated with changes in physical activity in young boys.
There is a need for additional prospective, longitudinal
study of tracking of physical activity and its relation to
other cardiovascular disease risk factors in children,
adolescents, and young adults.
The most striking finding in these data was an
approximate 30 percent decrease in the geometric
mean level of physical activity across all race-sex
groups. This striking decline occurred during only 7
years of early adulthood, with individuals generally
maintaining their rank in the population. The data
suggested that this 30 percent decrease occurred
equally at the beginning of the CARDIA age range
(i.e., 18-year-olds aging to 25 years old) as at the end
(i.e., 30-year-olds aging to 37 years old). One may
therefore expect that the level of the comprehensive
set of activities queried in this study may be reduced
by 50 percent between the ages of 18 and 37 years.
Furthermore, the data suggest that a substantial part
of the decline in physical activity is populationwide
secular trends. Declines in leisure activity occurring
through the latter half of the 1980s may indicate a
reversal in the trend toward increasing leisure time
physical activity reportedly occurring since 1957 (27,
40). The prevalence of participation in regular physical activity during leisure time, measured by a single
Am J Epidemiol
Vol. 143, No. 4, 1996
question, was reported to increase through the 1980s,
but total leisure time caloric expenditure was reported
to decrease in men aged 25-44 years and to be stable
in women aged 25-44 years (28). Declines in physical
activity are consistent with populationwide increasing
secular trends in body weight observed in this (41) and
other (28, 42-45) studies. Such a trend toward decreased physical activity would likely have adverse
health consequences, including increased weight gain,
obesity, and, ultimately, coronary heart disease (1-7).
The physical activity score generally declined with
age. The cross-sectional age effect was less for greater
educational attainment in women but was not related
to educational attainment in men. Longitudinally, the
physical activity decline was less in black men who
continued going to school during CARDIA, but the
decline was greater in white men who continued their
education and was unrelated to changes in education
during CARDIA in women. Further exploration is
needed of the observation that physical activity declined more in better educated than in less educated
whites, but that the opposite was true for black men;
physical activity declined least in black men with the
most education. In addition, we observed some differences in the change in physical activity between clinical centers, which may reflect regional differences in
leisure time physical activity habits. A smaller proportion of exercisers has been observed in the southern
than in the western, eastern, and northeastern United
States (22).
At all four examinations, men consistently reported
higher physical activity scores than did women, with
black men having the highest scores and black women
having the lowest. Depending on what physical activities are included (23), men are generally as active as
or more active than women (22). Blacks, up to age 74
years (42, 46) and including a subgroup aged 18-29
years (46), have previously been reported to engage in
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on April 24, 2014
» p value for the test that there is any difference In total cohort change in physical activity among the four clinical
centers.
t Slope is the regression coefficient for time holding age at baseline constant, i.e, the total cohort change
(exercise units per 7 years of time and age change)
t Ms the difference between regression coefficients for the given clinical center and Oakland, divided by the
standard error of this difference
360
Anderssen et al.
The lack of a birth cohort effect in the physical
activity score was expected given the narrow, 12-year,
range of age in this study and should not be taken as
evidence that there are no birth cohort effects over a
greater range of ages.
One reason to be cautious in interpreting the present
findings is that self-report measures of physical activity may be less valid than objective measures. Selfreported physical activity has, however, been shown to
be associated with a variety of validation criteria in
both adults (31-33, 49) and children (50). The accuracy of self-report measures in adults has been shown
to vary with sex (men tend to overestimate their activity), obesity status (obese persons tend to underestimate), and type of activity (sedentary activities are
underestimated, aerobic activities are overestimated)
(51). Although one might argue that the striking difference between black women and the rest of the
sample could result from a questionnaire that was
culturally and sex biased (i.e., not sensitive to aspects
of physical activity common among black women),
the validity of the questionnaire has been shown to be
similar in black as in white women (31), the long-term
test-retest correlation coefficients are only slightly
smaller in black than in white women (table 2), and
black women in this study were heavier than were
white women.
It should be noted that physical activity in these
analyses included predominantly leisure activities and
strenuous activity at work. Besides less strenuous activity in paid employment, these analyses omit child
care and household chores, which are important
sources of activity for women. Information about these
physical activities was obtained at only one of the first
four CARDIA examinations; changes will be studied
after further data collection.
These findings point strongly to a need for physical
activity promotion programs aimed toward young
adults. Programs should be especially tailored for and
targeted at those at risk for a sedentary lifestyle, for, as
Haskell notes, "The greatest difference in risk is between people who do almost nothing and those who do
a moderate amount of exercise on a regular basis"
(1, P- 414).
ACKNOWLEDGMENTS
This work was supported by contracts N01-HC-48047,
N01-HCM8048, N01-HCM8049, and N01-HC-48050 from
the National Heart, Lung, and Blood Institute.
REFERENCES
1. Haskell WL. Overview: health benefits of exercise. In:
Matarazzo JD, Weiss SM, Herd JA, et aL, eds. A handbook of
health enhancement and disease prevention. New York: John
Wiley & Sons, Inc, 1984:409-23.
2. Leon AS, Cortnett J, Jacobs DR, et al Leisure-time physical
activity levels and risk of coronary heart disease and death.
Am J Epidemiol Vol. 143, No. 4, 1996
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on April 24, 2014
less leisure time physical activity than do whites (42,
46). Lower exercise levels were found in young adolescents in a sample of blacks than in whites of the
same age (47). Black women in the United States are
less physically active than are white women (24, 48).
Black men are, however, more active than white men
are in CARDIA; part of this difference at the year 7
examination, but not at baseline, is increased vigorous
job activity in black men. The relative ranking between race-sex groups was stable over the 7-year
period. More evidence is required to determine which
group of men is the more active (22, 48).
The longitudinal design of CARDIA brings into
focus the complex interrelations between age and time
as predictors of physical activity, including the possible curvature in the age trend, birth cohort or other
historical effects, secular trend, methodological artifact, and differences in estimates of aging within person versus between persons. The underlying effects
may be present in cross-sectional designs but are confounded. They are partially separable in longitudinal
designs. An example of age curvature would be a peak
in physical activity in late adolescence, followed by a
decline and another peak; there was no evidence for
such a pattern in CARDIA where the total cohort
change does not vary by baseline age. Age-matched
time trends might reflect methodological artifacts,
such as different administration of the physical activity
instrument at different examinations, but there were
tight quality control and no specific evidence of such
methodological problems. The uniform relations of
physical activity across age and time, as depicted in
figure 2, suggest that cross-sectional age effects are
linear through the age range of 18-37 years. One
might take total cohort change as the longitudinal
(within person) estimate of the effect of age on physical activity. Yet, the total cohort change in physical
activity exceeds what would be expected due to age
alone, as estimated cross-sectionally. If the between
person estimate of the effect of age on physical activity is correct (longitudinal estimate overestimates the
age effect), then the excess is secular trend occurring
during the 7 years of the study. On the other hand, if
total cohort change is the correct estimate of the age
effect (cross-sectional estimate underestimates the age
effect), then the excess is historical, for example, secular trend that occurred before the study began. It is
reasonable to believe that a combination of these two
types of secular trend is occurring.
Trends in Physical Activity
Am J Epidemiol
Vol. 143, No. 4, 1996
26. EngstrOm LM. Exercise adherence in sport for all from youth
to adulthood. Presented at the World Congress on Sport for
All, Tampere, Finland, June 3-7, 1990.
27. Jacobs DR, Hahn LP, Folsom AR, et al Time trends in
leisure-time physical activity in the upper Midwest
1957-1987: University of Minnesota Studies. Epidemiology
1991;2 8-15.
28. Jacobs DR, Sprafka JM, Hannan PJ, et al. Mortality and risk
factor trends in Minnesota: Minnesota Heart Studies. In: Keys
A, Toshima H, Blackburn H, eds. Lessons for science from
the Seven Countries Study. Tokyo- Springer Verlag, 1994145-61
29 Hughes GH, Cutter G, Donahue R, et al. Recruitment in the
Coronary Artery Disease Risk Development in Young Adults
(CARDIA) Study. Control Clin Trials 1987;8:68S-73S.
30. Friedman GD, Cutter GR, Donahue RP, et al. CARDIA: study
design, recruitment, and some characteristics of the examined
subjects. J Chn Epidemiol 1988;41:1105-16.
31 Sidney S, Jacobs DR Jr, Haskell WL, et al. Comparison of two
methods of assessing physical activity in the Coronary Artery
Risk Development in Young Adults (CARDIA) Study. Am J
Epidemiol 1991;133.1231-45.
32. Jacobs DR, Hahn LP, Haskell WL, et al Validity and reliability of a short physical activity history: CARDIA and the
Minnesota Heart Health Program. J Cardiopulmon Rehabil
1989;9-448-59.
33 American College of Sports Medicine. The recommended
quantity and quality of exercise for developmg and maintaining cardiorespiratory and muscular fitness in healthy adults.
Med Sci Sports Exerc 1990;22-265-74.
34. Jacobs DR, Ainsworth BE, Hartman TJ, et al A simultaneous
evaluation of ten commonly used physical activity questionnaires. Med Sci Sports Exerc 1993,25.81-91
35. SAS Institute, Inc SAS/STAT software: changes and enhancements. Release 6.07. Cary, NC: SAS Institute, Inc, 1992
(SAS technical report P-229).
36 Butcher J. Longitudinal analysis of adolescent girls' participation in physical activity. See Sport J 1985;2-130-43.
37 Kelder SH, Perry CL, Klepp KI, et al. Longitudinal tracking of
adolescent smoking, physical activity, and food choice behaviors. Am J Public Health 1994,84-1121-6.
38. Raitakari OT, Porkka KVK, Taimela S, et al. Effects of
persistent physical activity and inactivity on coronary risk
factors in children and young adults. The Cardiovascular Risk
in Young Finns Study Am J Epidemiol 1994,140.195-205.
39 Perry CL, Kelder SH, Klepp KI The rationale behind early
prevention of cardiovascular disease in young people. Eur J
Public Health 1994;4:156-62
40. Luepker RV, Rosamond WD, Murphy R, et al. Socioeconomic
status and coronary heart disease nsk factor trends. The Minnesota Heart Survey. Circulation 1993;88(5 Pt 1)2172-9.
41. Lewis CE, Smith DE, Williams OD, et al Seven year trends
in weight and weight gain in black and white young adults: the
CARDIA Study. (Abstract). Circulation 1994;89-8.
42. Folsom AR, Cook TC, Sprafka JM, et al. Differences in
leisure time physical activity levels between blacks and whites
in population-based samples: the Minnesota Heart Survey. J
BehavMed 1991,14.1-9.
43. Sprafka JM, Burke GL, Folsom AR, et al. Continued decline
in cardiovascular disease risk factors: results of the Minnesota
Heart Survey, 1980-1982 and 1985-1987 Am J Epidemiol
1990,132.489-500.
44. Flegal KM, Harlan WR, Landis JR. Secular trends in body
mass index and skinfold thickness with socioeconomic factors
in young adult men. Am J Chn Nutr 1988;48:544-51.
45. Flegal KM, Harlan WR, Landis JR. Secular trends in body
mass index and skinfold thickness with socioeconomic factors
in young adult women. Am J Clin Nutr 1988,48:535-43.
46. Washburn RA, Kline G, Lackland DT, et al. Leisure time
physical activity: Are there black/white differences9 Prev Med
1992;21:127-35.
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on April 24, 2014
JAMA 1987;258:2388-95.
3. Obeiman A. Exercise and the primary prevention of cardiovascular disease. Am J Cardiol 1985;55:10D-20D.
4. Paffenbarger RS Jr, Wing AL, Hyde RT. Physical activity as
an mdex of heart attack nsk in college alumni. Am J Epidemiol 1978;108:161-75.
5 Berlin JA, Colditz GA A meta-analysis of physical activity in
the prevention of coronary heart disease Am J Epidemiol
199O;132-612-28.
6. Powell KE, Thompson PD, Caspersen CJ, et al. Physical
activity and the incidence of coronary heart disease. Annu Rev
Public Health 1987,8:253-87.
7. Slattery ML, Jacobs DR Jr. Physical fitness and cardiovascular
disease mortality. The US Railroad Study. Am J Epidemiol
1988;127:571-80.
8 Blair SN, Kohl HW, Paffenbarger RS, et al. Physical fitness
and all-cause mortality. JAMA 1989;2622395-4Ol.
9. Haskell WL. Physical activity and health: need to define the
required stimulus Am J Cardiol 1985.55-4D-9D.
10. Koplan JP, Caspersen CJ, Powell KE. Physical activity, physical fitness, and health: time to act (Editorial) JAMA 1989,
262-2437.
11 Perry C, Stone EJ, Parcel GS, et al. School-based cardiovascular health promotion: the Child and Adolescent Trial for
Cardiovascular Health (CATCH). J Sch Health 1990,60:
406-13.
12. Dennison BA, Straus JH, Melhts ED, et al Childhood physical fitness tests: predictors of adult physical activity levels?
Pediatrics 1988;82:324-30.
13. Dishman RK. Supervised and free-living physical activity, no
differences in former athletes and nonathletes. Am J Prev Med
1988;4:153-60
14. Salhs JF, Hovell MF, Hofstetter CR, et al. A multivariate
study of determinants of vigorous exercise in a community
sample. Prev Med 1989;1820-34.
15. Blair SN, Clark DG, Cureton KJ, et al. Exercise and fitness in
childhood, implications for a lifetime of health In' Gisolfi
CV, Lamb DR, eds. Perspectives in exercise science and
sports medicine. Vol 2. Indianapolis: Benchmark Press, 1989:
401-30
16. Dishman RK, Dunn AL Exercise adherence in children and
youth: implications for adulthood. In: Dishman RK, ed. Exercise adherence. Its impact on public health. Champaign, IL:
Human Kinetics Books, 1988:155-200.
17. Powell KE, Dysinger W. Childhood participation in organized
school sports and physical education as precursors of adult
physical activity. Am J Prev Med 1987;3:276-81
18. Salhs JF, McKenzie TL. Physical education's role in public
health. Res Q Exerc Sport 1991;62:124-37.
19. Berenson GS, Snnivasan SR, Hunter SM, et al Risk factors in
early life as predictors of adult heart disease- the Bogalusa
Heart Study. Am J Med Sci 1989;298:141-51.
20. Dishman RK, Salhs JF, Orenstein DR. The determinants of
physical activity and exercise. Public Health Rep 1985;100:
158-71.
21. Baranowski T, Bouchard C, Bar-Or O, et al. Assessment,
prevalence, and cardiovascular benefits of physical activity
and fitness in youth. Med Sci Sports Exerc 1992,24(6 suppl):
S237-47.
22. Stephens T, Jacobs DR, White CC A descriptive epidemiology of leisure-time physical activity. Public Health Rep 1985;
100.147-58.
23. Ainsworth BE, Richardson MT, Jacobs DR, et al. Gender
differences in self-reported physical activity Women Sports
Phys Activ 1993;2:11-16.
24. Bild D, Jacobs DR, Sidney S, et al. Physical activity in young
black and white women: the CARDIA Study Ann Epidemiol
1993;3:636-44.
25. D0lvik JE, Danielsen 0, Hernes G. Kluss I vekslinga. Fritid,
idrett og organisering. (Li Norwegian). 2nd ed. Oslo: Fagbevegelsens Senter for Forskning, Utredning og Dokumentasjon,
1988.
361
362
Anderssen et al.
47. Gottlieb NH, Chen MS. Sociocultural correlates of childhood
sporting activities, their implications for heart health Soc Sci
Med 1985,21:533-9.
48. Schoenbom CA. Health-habits of US adults, 1985. the "Alameda 7" revisited. Public Health Rep 1986;101.571-80.
49. Blair SN, Haskell WL, Ho P, et al. Assessment of habitual
physical activity by a seven-day recall in a community survey
and controlled experiments Am J Epidemiol 1985; 122794-804
50. Sallis J. Self-report measures of children's physical activity. J
Sch Health 1991 ;61.215-19.
51. Klesges RC, Eck LH, Mellon MW, et al. The accuracy of
self-reports of physical activity. Med Sci Sports Exerc 1990;
22:690-7.
Downloaded from http://aje.oxfordjournals.org/ at Pennsylvania State University on April 24, 2014
Am J Epidemiol
Vol. 143, No. 4, 1996