International Journal of Obesity (1998) 22, 915±922
ß 1998 Stockton Press All rights reserved 0307±0565/98 $12.00
http://www.stockton-press.co.uk/ijo
Body fatness: Longitudinal relationship of body
mass index and the sum of skinfolds with other
risk factors for coronary heart disease
JWR Twisk1, HCG Kemper1, W van Mechelen1,2, GB Post1 and FJ van Lenthe1
1
Institute for Research in Extramural Medicine and 2Department of Social Medicine, Faculty of Medicine, Vrije Universiteit,
Amsterdam, The Netherlands
OBJECTIVE: To analyse the longitudinal relationships between body mass index (BMI)/sum of skinfolds (SSF) and
biological and lifestyle risk factors for coronary heart disease (CHD).
DESIGN: An observational longitudinal study; that is, the Amsterdam Growth and Health Study.
SUBJECTS: 181 males and females, initially aged 13 y. Over a period of 15 y, six repeated measurements were carried
out.
MEASUREMENTS: BMI and SSF, biological CHD risk factors; that is, total cholesterol (TC), high density lipoprotein
_ 2-max) and lifestyle
(HDL), TC:HDL ratio, systolic/diastolic blood pressure (SBP/DBP) and cardiopulmonary ®tness (VO
CHD risk factors (that is, daily physical activity, dietary parameters, smoking, and alcohol consumption). The
longitudinal relationships were analysed by an autoregressive model, in which the value of the outcome variable
at time-point t is not only related to the value of the predictor variable at t, but also to the value of the outcome
variable at t 7 1.
RESULTS: Both BMI and SSF were positively related to TC and the TC:HDL ratio. Only BMI was positively related to
_ 2-max. Physical activity was negatively related to SSF. None of the
SBP and only SSF was negatively related to VO
other lifestyle parameters were related to SSF and/or BMI.
CONCLUSIONS: Both BMI and SSF were related to a high risk pro®le regarding CHD. Different relationships for SSF
and BMI are found, because BMI not only re¯ects body fatness, but also lean body mass. Analyses with BMI as an
indicator for body fatness should therefore be interpreted cautiously.
Keywords: coronary heart disease; body mass index; epidemiology; life style; longitudinal studies; skinfold thickness;
statistical methods
Introduction
It is well known that coronary heart disease (CHD) is
one of the most important chronic diseases in western
countries and that the origin of CHD lies in early
childhood.1 One of the risk factors for CHD is obesity
or more general body fatness. The importance of body
fatness as a risk factor for CHD lies ®rstly in the fact
that body fatness is directly related to mortality and
morbidity of CHD.2±4 Secondly, body fatness seems
to be related to other biological risk factors for CHD,
like lipoprotein levels and blood pressure.5±6 Furthermore, body fatness is regarded to be an intermediate
factor between lifestyle parameters, like dietary intake
and daily physical activity, and the earlier mentioned
biological risk factors.7±9
There have been many epidemiologically based
investigations of the relationships mentioned above.
Correspondence: Jos WR Twisk, EMGO-Institute, Faculty of
Medicine, Vrije Universiteit, vd Boechorststraat 7, 1081 BT
Amsterdam, The Netherlands.
Received 13 January 1998; revised 15 April 1998; accepted
12 May 1998
One of the major problems in this ®eld of research is
how to de®ne body fatness. Most commonly, body
fatness is estimated on the basis of body mass index
(BMI), which is easy to measure from body height and
body weight, and therefore widely used as an indicator
for body fatness (under the assumption that the variation in body weight is mainly caused by the variation
in body fat mass). Using the sum of two or more
skinfold thicknesses is another way to estimate body
fatness. Although both BMI and skinfolds are
assumed to be indicators of the same parameter (that
is, body fatness), analysis with both parameters can
lead to different results.10
The purpose of this study was ®rstly to clarify the
role of body fatness as a risk factor for CHD. Therefore, data were derived from the Amsterdam Growth
and Health Study (AGHS); an observational longitudinal study over a period of 15 years covering
adolescence and young adulthood.11 During this
period of time, six repeated measurements were carried out in which body fatness, lifestyle and biological
CHD risk factors were measured.11 In this paper the
following longitudinal relationships were analysed:
(1) between body fatness and biological CHD risk
factors (that is, lipoprotein levels, blood pressure and
Body mass index and sum of skinfolds
JWR Twisk et al
916
cardiopulmonary ®tness) and (2) between lifestyle
CHD risk factors (that is, physical activity, dietary
intake, smoking behaviour and alcohol consumption)
and body fatness. The second purpose of this study
was to make a comparison between two widely used
indicators for body fatness: BMI and the sum of
skinfold thickness (SSF) regarding the above mentioned longitudinal relationships.
Methods
Subjects
The AGHS is an observational longitudinal study
which started in 1977 with healthy subjects from the
®rst and second forms of a secondary school in
Amsterdam. Compared to the general Dutch population, the socio-economic background of the subjects
(initially based on profession, education and income
of the parents) was above average. At the start of the
study 307 subjects (148 male and 159 female) with a
mean age of 13 y were recruited. During the ®rst four
years of the study, annual measurements were carried
out. In total 233 subjects (102 male and 131 female)
completed these four annual measurements. Later on
two follow-up measurements were carried out. In
1985, at the age of 21 y, 200 subjects (93 male and
107 female) were measured for the ®fth time and in
1991 at the age of 27 y, a sixth measurement was
carried out on 181 subjects (83 male and 98 female).
For the variables of interest in this study, no drop-out
effects were observed.
The population used in the present study consists
of the 181 subjects who were present at the last
measurement, at the age of 27 y. The total number
of missing observations for this population during the
measurement period, was about 2%. The primary aim
of the AGHS was to describe the development of
growth, health and lifestyle of Dutch boys and girls,
although nowadays interest has moved towards the
analyses of longitudinal relationships between health
and lifestyle parameters. Extensive information about
the design of the AGHS is given by Twisk and
Kemper.12
Measurements
Body fatness. Body fatness was estimated in two
ways. Firstly as BMI, which was calculated as body
weight (expressed in kg) divided by body height2
(expressed in m2). Secondly, as the sum of the thickness of four skinfolds (triceps, biceps, subscapular and
suprailiac). SSF (expressed in mm) was measured
with a Harpenden skinfold caliper, according to the
method described by Weiner and Lourie.13 Because
BMI not only re¯ects body fatness, but also lean body
mass (LBM), LBM was estimated as [(100 7 %body
fat) body weight]; where %body fat was estimated
from SSF, by regression equations as given by Durnin
and Rahaman14 and Durnin and Womersley.15
Biological CHD risk factors. Blood cholesterol was
determined by taking approximately 10 ml of venous
blood from the antecubital vein with a vacutainer.
Besides the concentration of total serum cholesterol
(TC), which was analysed according to the methods
of Huang et al16 and Abell et al,17 also the concentration of high density lipoprotein cholesterol
(HDL) and the ratio TC:HDL ratio was determined.
HDL were analysed according to the method of
Burstein and Samaille.18 External quality control
took place with target samples from a World
Health Organisation (WHO) reference laboratory
(Lipid Standardization Laboratory, Atlanta, GA).
Blood pressure was measured with an indirect
method. A standard pressure cuff was placed
around the left upper arm. With a sphygmomanometer, diastolic blood pressure (DBP) and systolic
blood pressure (SBP) were measured twice and the
lowest value was recorded. Cardiopulmonary ®tness
was assessed by measuring maximal oxygen uptake
_ 2-max) and was expressed in ml.min71.kg72/3.
(VO
_ 2-max was measured with a maximal test, whilst
VO
running on a treadmill with the speed at 8 km/h and
increasing slope.19
Lifestyle CHD risk factors. Dietary intake was
measured by a modi®cation of the cross-check dietary history interview. In this dietary history interview, which was specially tailored for use in the
AGHS,20,21 all subjects were asked to recall their
usual dietary intake by reporting frequency, amounts
and methods of preparation of the foods consumed.
The method was use to assess the usual food intake
during the previous month. All consumed food items
were transformed into nutrients by the Dutch Food
and Nutrition Table.22 In relation to CHD risk
factors, the daily intake of the following nutrients
was assessed: 1) total energy intake, 2) the intake of
fat (absolute in g and expressed as percentage of
total energy intake) , and 3) the intake of carbohydrates (absolute in g and expressed as percentage of
total energy intake). With the dietary history interview, alcohol consumption (expressed in g per
week) was also measured. Smoking behaviour, that
is, the amount of tobacco (g) smoked per week, was
asked for by a separate questionnaire. Daily physical
activity was measured by a structured interview,
which was developed for the AGHS.23,24 With this
interview, the total time spent on physical activities
in relation to school and work, and on other activities (for example, organized sports activities, unorganized sports activities, other leisure time activities,
stair climbing, etc.) was measured. The measured
Body mass index and sum of skinfolds
JWR Twisk et al
times were combined with the intensity of the
different activities to calculate a total weighted
activity score (expressed in METs/week). The activity interview covered the period of the three months
prior to the interview. Both the cross-check dietary
history and the daily physical activity interview
were shown to be valid measurements of dietary
intake and daily physical activity.20,21,23,24
Other parameters. The development of CHD risk
factors during adolescence, depend on maturation
rate; that is, biological age.11,25 In the AGHS, biological age was determined by measuring skeletal age
from X-ray photographs of the let hand, according to
the Tanner-Whitehouse II method.26 Before full
maturity is reached, biological age can be different
from calendar age. Compared to their calendar ages,
rapidly maturing subjects have higher biological ages,
while slowly maturing subjects have lower biological
ages.
Extensive information about the methods used in
the AGHS is given by Kemper and van Mechelen.27
Analysis
To analyse the longitudinal relationships between
body fatness and biological CHD risk factors and
between lifestyle CHD risk factors and body fatness
a `®rst order' autoregressive model was used:
Yit ˆ b0 ‡
‡
J
X
jˆ1
M
X
mˆ1
b1j Xijt ‡ b2 Yitÿ1 ‡
K
X
kˆ1
b3k Zikt
b4m Gim ‡ eit
where: Yit ˆ observations of subject i at time t (where
2 t number of measurements); b0 ˆ intercept;
Xijt ˆ independent variable j of subject i at time t;
b1j ˆ regression coef®cient of independent variable
j; J ˆ number of independent variables; Yit71 ˆ
observation of subject i at time t 7 1; b2 ˆ
autoregression coef®cient; Zikt ˆ time dependent covariate k of subject i at time t; b2k ˆ regression coef®cient of time dependent covariate k; K ˆ number of
time dependent covariates; Gim ˆ time independent
covariate m of subject i; b4m ˆ regression coef®cient
of time independent covariate m; M ˆ number of time
independent covariates; and eit ˆ measurement error
of subject i at time t.
In this autoregressive model the actual value of the
outcome variable Y at time-point t is not only related
to the actual value of the predictor variable X at time
point t, but also to the value of the outcome variable Y
at t 7 1.28±30 The presented model is called a `®rst
order' autoregressive model, because the outcome
variable Y at time point t is only related to the value
of the outcome variable Y at t 7 1 (in a `second or
third order' autoregressive model, the outcome at time
point t is also related to the value of the outcome
variable Y at t 7 2 or t 7 3). The idea behind this kind
of model is that the value of an outcome variable at
each time-point is primarily caused by the value of
this variable, one measurement earlier. To estimate
the `real' in¯uence of the predictor variables (Xijt) on
the outcome variable (Yit) the model should therefore
correct for the value of the outcome variable at time
point t 7 1 (Yit71). Besides this autoregression, in the
model there is also the possibility to correct for both
time-dependent (Zikt) and time independent covariates
(Gim). Because the repeated observations of a particular subject are not independent of each other, the
estimation of the parameters of the statistical model
should take into account the within-subject correlations. Therefore the parameters of the autoregressive
model were estimated by generalized estimating equations (GEE).10,31
In the present study the following analyses were
carried out: 1) The longitudinal relationships between
body fatness (either BMI or SSF) and other biological
CHD risk factors; because body fatness is supposed to
in¯uence other biological CHD risk factors, in these
analyses, BMI and SSF are predictor variables (Xijt),
while the other biological CHD risk factors are the
outcome variables (Yit). 2) The longitudinal relationship between lifestyle parameters and body fatness
(either BMI or SSF); because it is assumed that body
fatness is in¯uenced by lifestyle CHD risk factors, in
these analyses BMI and SSF are outcome variables
(Yit), and the lifestyle parameters are the predictor
variables (Xijt). The third purpose of this study was to
compare BMI and SSF as indicators for body fatness,
and because of the possible importance of LBM in this
comparison, all analyses were also carried out for
LBM.
In principle all biological and lifestyle CHD risk
factors were treated as continuous variables. However because of the skewness of the distribution of
smoking behaviour and alcohol consumption, these
lifestyle parameters were treated differently. At each
point of measurement, smoking behaviour was
dichotimized (smokers vs non smokers) and alcohol
consumption was trichotomized; non drinkers, low
drinkers and heavy drinkers. The distinction between
low and heavy drinkers was based on the median
of the amount of alcohol consumed among the
drinkers.
In order to facilitate the interpretation of the results
of the autoregressive analysis, standardized regression
coef®cients were calculated for each of the predictor
variables. After assessing the main effects for each
predictor variable, the interactions with gender and
time were added to the model. When a signi®cant
interaction with gender was found, different regression coef®cients were calculated for males and
females. Signi®cant interactions with time would
indicate that the observed relationship varied between
different parts of the longitudinal period.
917
Body mass index and sum of skinfolds
JWR Twisk et al
918
All longitudinal analyses were carried out by the
Statistical Package for Interactive Data Analysis
(SPIDA).32
Results
Table 1 shows descriptive information of both SSF
and BMI at each of the repeated measurements. SSF
was higher for females than for males, while BMI was
more or less equal between the genders. For males and
females the values for both BMI and SSF increased
during the longitudinal period.
The results of the autoregressive analysis in order to
assess the longitudinal relationships between SSF and
biological CHD risk factors are given in Table 2. SSF
was signi®cantly positively related to TC and the
TC:HDL ratio (standardized regression coef®cient
b ˆ 0.08). For TC, a signi®cant interaction between
SSF and gender was found, so different coef®cients
were calculated for males (b ˆ 0.12) and females
(b ˆ 0.07). Between SSF and HDL, SBP and
DBP, no signi®cant relationships were observed,
while a signi®cant inverse relationship was found
_ 2-max. Also for this relationship,
between SSF and VO
Table 1 The number of subjects and mean values (standard
deviation, s.d.) of the sum of four skinfolds (SSF) and body mass
index (BMI) at each of the repeated measurements
Calendar
age (y)
n
(M/F)
13
82/97
14
79/97
15
83/98
16
79/98
21
78/94
27
83/98
2
BMI (kg/m )
SSF (mm)
M
F
M
F
28.4
(10.9)
27.7
(10.7)
27.8
(10.3)
29.1
(9.7)
34.9
(11.9)
36.5
(13.5)
37.5
(12.8)
40.0
(13.8)
43.4
(14.2)
45.6
(14.6)
52.3
(16.9)
46.3
(16.5)
17.3
(1.6)
18.1
(1.8)
18.9
(1.9)
19.6
(1.7)
21.3
(1.8)
22.6
(2.2)
18.1
(2.1)
18.7
(2.3)
19.5
(2.3)
20.0
(2.3)
21.4
(2.6)
21.9
(2.5)
M ˆ male; F ˆ female.
Table 2 Standardized regression coef®cients (b), 95%
con®dence intervals (CI) and P-values regarding the
relationships between sum of four skinfolds (predictor variable)
and biological coronary heart disease risk factors (outcome
variables); estimated with an autoregressive model (n ˆ 181)
b
95% CI
P-value
Total serum cholesterol (TC) M
0.12
0.07 ± 0.18
< 0.01
F
0.07
0.03 ± 0.11
< 0.01
High density lipoprotein
70.02 70.05 ± 0.01
0.10
(HDL) cholesterol
TC:HDL ratio
0.08
0.05 ± 0.12
< 0.01
Systolic blood pressure
0.04 70.01 ± 0.08
0.09
Diastolic blood pressure
0.02 70.03 ± 0.08
0.37
Cardiopulmonary ®tness
M 70.20 70.28 ± 70.13 < 0.01
F 70.06 70.11 ± 70.02 < 0.01
M ˆ male; F ˆ female.
a signi®cant interaction between SSF and gender was
found (bmales ˆ 70.20 and bfemales ˆ 70.06). In
Table 3, the results of the same analysis with BMI
as predictor variable are shown. BMI was signi®cantly
positively related to TC (b ˆ 0.05), the TC:HDL ratio
(b ˆ 0.04) and SBP (b ˆ 0.06). No signi®cant relationships were observed between BMI and HDL, DBP
_ 2-max. In Table 4, the longitudinal relationand VO
ships between lifestyle CHD risk factors and SSF are
Table 3 Standardized regression coef®cients (b), 95%
con®dence intervals (CI) and P-values regarding the
relationships between body mass index (BMI) (predictor
variable) and biological coronary heart disease (CHD) risk
factors (outcome variables); estimated with an autoregressive
model (n ˆ 181)
Total serum cholesterol (TC)
High density lipoprotein
(HDL) cholesterol
TC:HDL ratio
Systolic blood pressure
Diastolic blood pressure
Cardiopulmonary ®tness
b
95% CI
P-value
0.05
70.01
0.02 ± 0.08
70.04 ± 0.01
< 0.01
0.29
0.04
0.06
0.03
0.00
0.01 ± 0.07
0.02 ± 0.10
70.01 ± 0.08
70.03 ± 0.03
< 0.01
< 0.01
0.16
0.96
Table 4 Standardized regression coef®cients (b), 95%
con®dence intervals (CI) and P-values regarding the
relationships between lifestyle coronary heart disease (CHD)
risk factors (predictor variables) and sum of four skinfolds
(SSF) (outcome variable); estimated with an autoregressive
model (n ˆ 181)
b
Daily physical activity
Smoking behaviour
Alcohol consumption
(`moderate')
Alcohol consumption
(`heavy')
Dietary intake
Energy intake
Absolute fat intake
Fat intake (% energy intake)
Absolute carbohydrate intake
Carbohydrate intake
(% energy intake)
95% CI
P-value
70.06 70.09 ± 70.03 < 0.01
70.02 70.11 ± 0.07
0.67
70.00 70.09 ± 0.08
0.98
0.06 70.03 ± 0.15
70.02
70.01
0.00
70.02
70.00
70.06 ± 0.01
70.05 ± 0.02
70.03 ± 0.03
70.06 ± 0.01
70.03 ± 0.03
0.18
0.23
0.41
0.92
0.24
0.96
Table 5 Standardized regression coef®cients (b), 95%
con®dence intervals (CI) and P-values regarding the
relationships between lifestyle coronary heart disease (CHD)
risk factors (predictor variables) and body mass index (BMI)
(outcome variable); estimated with an autoregressive model
(n ˆ 181)
Daily physical activity
Smoking behaviour
Alcohol consumption
(`moderate')
Alcohol consumption
(`heavy')
Dietary intake
Energy intake
Absolute fat intake
Fat intake (% energy intake)
Absolute carbohydrate intake
Carbohydrate intake
(% energy intake)
b
95% CI
P-value
0.01
70.05
70.05
70.03 ± 0.04
70.13 ± 0.03
70.13 ± 0.03
0.72
0.21
0.23
0.04
70.04 ± 0.12
0.32
0.04
0.02
0.01
0.02
70.01
70.01 ± 0.06
70.01 ± 0.06
70.03 ± 0.04
70.02 ± 0.05
70.04 ± 0.02
0.18
0.23
0.69
0.33
0.53
Body mass index and sum of skinfolds
JWR Twisk et al
shown. Besides an inverse relationship with daily
physical activity (b ˆ 70.06) no signi®cant relationships were observed. When the same analysis was
carried out for BMI as outcome variable (Table 5), the
relationship with daily physical activity was not signi®cant any more (b ˆ 0.01; P ˆ 0.72). Furthermore,
no signi®cant relationships between lifestyle parameters and BMI were found. In Table 6 and Table 7, the
results of the longitudinal analyses with LBM are
shown. Regarding the relationship with other biological CHD risk factors (Table 6), a signi®cant positive
relationship with SBP (b ˆ 0.08), DBP (b ˆ 0.08), and
_ 2-max (b ˆ 0.12) was found, while no signi®cant
VO
relationships were observed with lipoprotein levels.
Regarding the relationship between lifestyle parameters and LBM (Table 7), daily physical activity
was signi®cantly positively related to LBM
(b ˆ 0.04). The results showed further positive relationships with energy intake, absolute fat intake and
absolute carbohydrate intake; however these relationships were only observed for males. When fat intake
and carbohydrate intake were expressed as percentage
Table 6 Standardized regression coef®cients (b), 95%
con®dence intervals (CI) and P-values regarding the
relationships between lean body mass (LBM) (predictor
variable) and biological coronary heart disease (CHD) risk
factors (outcome variables); estimated with an autoregressive
model (n ˆ 181)
Total serum cholesterol (TC)
High density lipoprotein (HDL)
cholesterol
TC:HDL ratio
Systolic blood pressure
Diastolic blood pressure
Cardiopulmonary ®tness
b
95% CI
P-value
70.01
70.03
70.04 ± 0.03
70.06 ± 0.01
0.96
0.10
0.02
0.08
0.08
0.12
70.01 ± 0.06
0.03 ± 0.13
0.02 ± 0.15
0.07 ± 0.17
0.23
< 0.01
0.01
< 0.01
Table 7 Standardized regression coef®cients (b), 95%
con®dence intervals (CI) and P-values regarding the
relationships between lifestyle coronary heart disease (CHD)
risk factors (predictor variables) and lean body mass (LBM)
(outcome variable); estimated with an autoregressive model
(n ˆ 181)
b
Daily physical activity
Smoking behaviour
Alcohol consumption
(`moderate')
Alcohol consumption
(`heavy')
Dietary intake
Energy intake
P-value
0.04
0.02 ± 0.06 < 0.01
0.00 70.04 ± 0.04
0.94
70.03 70.08 ± 0.02
0.19
0.01 70.04 ± 0.06
0.07
0.01
Absolute fat intake
0.06
0.01
Fat intake (% energy intake)
0.01
Absolute carbohydrate intake M
0.06
F
0.01
Carbohydrate intake
70.01
(% energy intake)
M ˆ male; F ˆ female.
95% CI
M
F
M
F
0.63
0.04 ± 0.11 < 0.01
70.01 ± 0.03
0.42
0.02 ± 0.09 < 0.01
70.02 ± 0.03
0.58
70.02 ± 0.02
0.73
0.03 ± 0.09 < 0.01
70.01 ± 0.03
0.46
70.03 ± 0.01
0.34
of energy intake, no signi®cant relationships were
observed.
In none of the analyses was a signi®cant interaction
between a particular predictor variable and time
observed.
Discussion
The purpose of this longitudinal analysis was to
analyse the longitudinal relationships between body
fatness and other biological CHD risk factors, and
between lifestyle CHD risk factors and body fatness,
in order to get a better view of the role of body fatness
in the etiology of CHD. The data used in the analyses
were derived from the AGHS.11 Although this is a
unique observational longitudinal study, the interpretation is limited by the fact that the study sample was
relatively small, healthy and with a high socioeconomic status.
The longitudinal relationships were analysed with a
`®rst order autoregressive' model. Autoregressive
models are also called conditional models, because at
each time point the value of an outcome parameter is
conditional on the value of the same parameter measured one time point earlier. Conditional models are to
be distinguished from marginal models in which the
values of the outcome parameter at each time point are
only related to the values of one or more predictor
variables. With these marginal models a combined
analysis of within subject (longitudinal) and between
subject (cross-sectional) relationships is carried out.30,33
With an autoregressive model only the within-subject
relationships are analysed; that is, an autoregressive
model will give a better estimation of the relationships
between the changes in an outcome variable and the
changes in one or more predictor variables over
time.30,34 Because of this, the magnitude of the parameters estimated with a marginal model, can be an
overestimation of the `real' longitudinal relationship.
In this study the same analyses were made for BMI
and for SSF; that is, the two most widely used
indicators for body fatness in large epidemiological
studies. Both BMI and SSF were found to be positively related to TC and the TC:HDL ratio. A small
difference between the two indicators was found for
the relationship with SBP. BMI was signi®cantly
positively related to SBP, while the positive relationship between SSF and SBP did not reach signi®cance.
However, when the corresponding 95% con®dence
intervals (CI) are considered, no `real' differences
were present. The most pronounced difference was
_ 2-max and regardfound for the relationship with VO
ing the relationships between lifestyle CHD risk
factors and body fatness for the relationship with
daily physical activity. For SSF a highly signi®cant
_ 2-max was found
negative relationship with VO
(P < 0.01), while for BMI, no relationship was
919
Body mass index and sum of skinfolds
JWR Twisk et al
920
found. In the same way the amount of daily physical
activity was found to be negatively related to SSF
(P < 0.01), but not to BMI. One of the reasons for
these different results, is the fact that BMI is not only
an indicator for fat mass, but also for LBM or muscle
mass.10 Subjects with high muscle mass and moderate
fat mass will have high values for BMI, but only
moderate values for SSF. To explore this important
issue further, the longitudinal relationships between
_ 2-max and between daily physical activLBM and VO
ity and LBM were investigated (Table 6 and Table 7).
_ 2-max
LBM was found to be positively related to VO
and daily physical activity was positively related to
LBM. Thus when BMI was used as indicator for body
fatness, the negative relationship between body fat_ 2-max, and between daily physical activness and VO
ity and body fatness were more or less
counterbalanced by the opposite relationships with
LBM. This indicates that results obtained with BMI
as indictor for body fatness should be interpreted
cautiously.
Body fatness and biological CHD risk factors
Regarding lipoprotein levels, in the literature, a positive relationship between body fatness and TC and the
TC:HDL ratio was found in both cross-sectional and
longitudinal studies. In most studies, a negative relation between body fatness and HDL was also
found.35±38 The relationships between body fatness
and TC and the TC:HDL ratio, are equivalent to the
results of the present study, but in the data of the
AGHS, no relationship between body fatness and
HDL was observed. These contradictory results are
probably caused by the fact that in the present study,
an autoregressive model was used in which only the
within-subject (that is, longitudinal) relationships
were assessed, rather than a combined analysis of
within-subject and between-subject (that is, crosssectional) relationships. In fact, in an earlier study in
which a marginal longitudinal model was used to
assess the relationships between lipoprotein levels
and body fatness, it was shown that both BMI and
SSF were negatively related to HDL.10 In the present
study, a positive signi®cant relationship was found
between BMI and SBP. For SSF, a positive relationship was also found with SBP, although this relationship did not reach signi®cance. The differences in
observed relationships were however negligible. With
DBP, for both BMI and SSF, no signi®cant relationships were found. In the literature, in general a
positive relationship is found between body weight
or body fatness and both SBP and DBP in adolescents
and young adults.6,39,40 It is hypothesized that blood
pressure is more related to LBM than to fat body
mass.5 So the present results, showing that the relationship between BMI and SBP is slightly stronger
than the relationship between SSF and SBP, are
in agreement with this ®nding. The results of the
longitudinal analysis with LBM showed a positive
relationship between SBP and LBM, with a higher
coef®cient than that found for the relationship with
BMI. Also for DBP, a positive and signi®cant relationship was found with LBM. So the (marginal)
differences found between BMI and SSF in the
relationship with blood pressure are also caused by
the fact that BMI not only re¯ects body fatness, but
also LBM.
As mentioned before, a highly signi®cant inverse
_ 2-max
relationship was found between SSF and VO
and no signi®cant relationship was found between
_ 2-max. These results differ somewhat
BMI and VO
from the literature, because in the literature both BMI
and SSF were found to be inversely related to cardio_ 2-max,41±43 while in the
pulmonary ®tness; that is, VO
present study, only SSF was found to be inversely
_ 2-max. One of the problems in the
related to VO
_ 2-max
assessment of the relationship between VO
_
and body fatness is that VO2-max is mostly expressed
_ 2-max is
per kg body weight. It is known that when VO
expressed per kg body weight, it still correlates well
with body weight.44,45 So relationships found between
_ 2-max expressed per kg body
body fatness and VO
_ 2-max
weight, can partly be caused by the fact that VO
_
is expressed in a `wrong way'. VO2-max divided by
kg body weight to the 2/3 power is a better indicator
for `real' cardiopulmonary ®tness.42,43 When the data
_ 2-max
of the AGHS were reanalysed with VO
expressed per kg body weight and not per kg body
weight to the 2/3 power, a negative relationship with
both BMI (bmales ˆ 70.13; 95% CI: 70.20 to 70.06;
P 0.01 and bfemales ˆ 70.10; 95% CI: 70.14 to
70.05; P 0.01) and SSF (bmales ˆ 70.39; 95% CI:
70.47 to 70.31; P 0.01 and bfemales ˆ 70.18; 95%
CI: 70.23 to 70.13; P 0.01) was found.
Lifestyle CHD risk factors and body fatness
The only observed signi®cant relationship between
lifestyle CHD risk factors and body fatness, was the
negative relationship between daily physical activity
and SSF, which is in agreement with the literature.9,46
The reason for not ®nding a relationship between
physical activity and BMI has already been discussed.
Furthermore, no relationships were observed
between any of the lifestyle parameters and both
BMI and SSF. This is surprising because in the
literature there are indications of (1) an inverse
relationship between smoking behaviour and body
fatness,7,47 (2) positive relationships between dietary
intake (that is, energy intake, the intake of fat and the
intake of carbohydrates) and body fatness.8,48,49 However most of these relationships ware observed in
studies where obese subjects were compared with
non-obese subjects. Although no signi®cant relationships were found between the dietary parameters and
both BMI and SSF, the direction of the relationships
with energy intake and the absolute intakes of fat and
carbohydrates was different for SSF and BMI. Again
this is probably caused by LBM, which was found to
Body mass index and sum of skinfolds
JWR Twisk et al
be positively related to all three absolute dietary
parameters. Besides the fact that in the present
study, an autoregressive model was used to assess
the longitudinal relationships, a possible explanation
for not ®nding any relationship between most lifestyle
parameters and body fatness in our study is, that the
population of the AGHS is relatively healthy regarding obesity; that is, the mean values of SSF and BMI
(Table 1) indicated that the number of subjects with
relative high values for body fatness is quite low. For
instance de®ning subjects as moderate obese, when
BMI 25 kg/m2, during the ®rst four measurements
none of the subjects, at the age of 21 y, 4% of the
subjects and at the age of 27 y, 14% of the subjects
could be classi®ed as moderate obese. This lack of
obesity may explain the absence of any relationship
between dietary intake, smoking behaviour and alcohol consumption and body fatness.
Conclusions
Both BMI and SSF were related to a high biological
CDH risk pro®le. The relationships between lifestyle
parameters and both SSF and BMI were marginal. The
differences found in the relationships between SSF
and BMI are caused by the fact that BMI not only
re¯ects body fatness, but also LBM. Analysis with
BMI as indicator for body fatness should therefore be
interpreted with caution.
Acknowledgements
This study was granted by the Dutch Heart Foundation (grant no. 76051-79051 and 90-312), the Dutch
Prevention Fund (grant no. 28-189a, 28-1106 and 281106-1), the Dutch Ministry of Well Being and Public
Health (grant no. 90-170), the Dairy Foundation on
Nutrition and Health, and the Netherlands Olympic
Committee and the Netherlands Sports Federation.
References
1 Newman WP III, Freeman DS, Voors AW, Gard PD, Srinivasan
SR, Cresanta JL, Berenson GS. Relation of serum lipoprotein
levels and systolic blood pressure to early atherosclerosis: The
Bogalusa Heart Study. N Engl J Med 1986; 314: 138 ± 144.
2 Hubert HB, Feinleib M, McNamara PM, Castelli WP. Obesity
as an independent risk factor for cardiovascular disease: A
26-year follow-up of participants in the Framingham Heart
Study. Circulation 1983; 67: 968 ± 977.
3 Seidell JC, Bakx KC, Deurenberg P, Hoogen HJM, Hautvast
JGAJ, Stijnen T. Overweight and chronic illness Ð a retrospective cohort study, with a follow-up of 6 ± 17 years, in men
and women of initially 20 ± 50 years of age. J Chron Dis 1986;
39: 585 ± 593.
4 Manson JE, Stampfer MJ, Hennekens CH, Willett WC. Body
weight and longevity. JAMA 1987; 257: 353 ± 358.
5 Voors AW, Webber LS, Frerichs RR, Berenson GS. Body
weight and body mass as determinants of basal blood pressure
in children Ð The Bogalusa Heart Study. Am J Epidemiol
1977; 106: 101 ± 108.
6 Lauer RM, Clarke WR. Childhood risk factors for high adult
blood pressure: The Muscatine Study. Pediatrics 1989; 84:
633 ± 641.
7 Jacobs DR, Gottenborg S. Smoking and weight. The Minnesota lipid research clinic. Am J Public Health 1981; 71:
391 ± 396.
8 Prewitt TE, Schmeisser D, Bowen PE, Aye P, Dolecet TA,
Langenberg P. Changes in body weight, body composition and
energy intake in women fed high- and low-fat diets. Am J Clin
Nutr 1991; 54: 304 ± 310.
9 Baranowski T, Bouchard C, Bar-Or O, Bricker T, Heath G,
Kimm SY, Malina R, Obarzanek E, Pate R, Strong WB.
Assessment, prevalence, and cardiovascular bene®ts of
physical activity and ®tness in youth. Med Sci Sports Exerc
1992; 24: S237 ± S247.
10 Twisk JWR, Kemper HCG, Mellenberg GJ, van Mechelen W.
Relation between the longitudinal development of lipoprotein
levels and biological parameters during adolescence and
young adulthood in Amsterdam. J Epidemiol Comm Health
1996; 50: 505 ± 511.
11 Kemper HCG (ed). The Amsterdam Growth Study: A longitudinal analysis of health, ®tness and lifestyle. HK Sports
Science Monograph Series, Vol 6. Human Kinetics Publishers
Inc: Champaign, 1995.
12 Twisk JWR, Kemper HCG. Design of the Amsterdam
Growth Study. In: Kemper HCG (ed). The Amsterdam
Growth Study: A longitudinal analysis of health, ®tness, and
lifestyle. Human Kinetics Publishers Inc: Champaign, 1995,
pp 6 ± 16.
13 Weiner JS, Lourie JA (eds) Human Biology. A guide to ®eld
methods. IBP Handbook No. 9. Blackwell: Oxford, 1968.
14 Durnin JVGA, Rahaman MM. The assessment of the amount
of fat in human body measurements of skinfold thickness. Br J
Nutr 1967; 21: 681 ± 689.
15 Durnin JVGA, Womersley J. Body fat assessed from total
body density and its estimation from skinfold thickness:
Measurements on 481 men and women aged from 16 ± 72
years. Br J Nutr 1974; 32: 77 ± 97.
16 Huang TC, Chen CP, We¯er V, Raftery A. A stable reagent
for the Lieberman Buchard reaction: application to rapid
serum cholesterol determination. Anal Chem 1961; 33:
1405 ± 1407.
17 Abell LL, Levy BB, Brody BB, Kendall FE. Simpli®ed
method for the estimation of total cholesterol in serum and
demonstration of its speci®city. J Biol Chem 1952; 195:
357 ± 366.
18 Burstein M, Samaille J. Sur une dosage rapide du cholesterol
lie aux-alpha-et aux beÂtalipoproteÂines du serum. Clin Chim
Acta 1960; 5: 609 ± 611.
19 Kemper HCG, Verschuur R. Maximal aerobic power in
13- and 14-year old teenagers in relation to biological age.
Int J Sports Med 1981; 2: 97 ± 100.
20 Kemper Han CG, Post G Bertheke, Twisk Jos WR. Rate of
maturation during the teenage years: Nutrient intake and
physical activity between ages 12 and 22. Int J Sports Nutr
1997; 7: 229 ± 240.
21 Post GB, Kemper HCG, Twisk J, van Mechelen W. The
association between dietary patterns and cardiovascular
disease risk indicators in healthy youngsters: Results covering
®fteen years of longitudinal development. Eur J Clin Nutr
1997; 51: 387 ± 393.
22 Stichting Nevo. Dutch food and nutrition table (in Dutch).
Voorlichtingsburo voor de voeding, Den Haag, The Netherlands, 1985.
23 Verschuur R, Kemper HCG. Habitual physical activity. Med
Sports Sci 1985; 20, 56 ± 65.
24 Verschuur R, Kemper HCG. The pattern of daily physical
activity. Med Sports Sci 1985; 20: 169 ± 186.
25 Twisk JWR, Kemper HCG, Mellenbergh GJ, van Mechelen
W. Factors in¯uencing tracking of cholesterol and high density
lipoprotein: The Amsterdam Growth and Health Study. Prev
Med 1996; 25: 355 ± 364.
921
Body mass index and sum of skinfolds
JWR Twisk et al
922
26 Tanner JM, Whitehouse RH, Marshall WA, Healy MJR,
Goldstein H. Assessment of skeletal maturity and prediction
of adult height (TW2 method). Academic Press: London, 1975.
27 Kemper HCG, van Mechelen W. Methods of measurements
used in the longitudinal study. In: Kemper HCG (ed). The
Amsterdam Growth Study: A longitudinal analysis of health,
®tness and lifestyle. HK Sports Science Monograph Series,
Vol 6. Human Kinetics Publishers Inc: Champaign, 1995, pp
28 ± 51.
28 Rosner B, Munoz A, Tager I, Speizer F, Weiss S. The use of
an autoregressive model for the analysis of longitudinal data in
epidemiologic studies. Stat Med 1985; 4: 457 ± 467.
29 Rosner B, Munoz A. Autoregresssive modelling for the
analysis of longitudinal data with unequally spaced examinations. Stat Med 1988; 7: 59 ± 71.
30 Twisk JWR. Different statistical models to analyse epidemiological observational longitudinal data. Int J Sports Med 1997;
18: S216 ± S224.
31 Zeger SL, Liang K-Y. Longitudinal data analysis for discrete
and continuous outcomes. Biometrics 1986; 42: 121 ± 130.
32 Gebski V, Leung O, McNeil D, Lunn D. SPIDA User manual.
Version 6. Macquarie University: Eastwood, NSW, Australia,
1992.
33 Liang K-Y, Zeger SL. Regression analysis for correlated data.
Ann Rev Publ Health 1993; 14: 43 ± 68.
34 Rosner B, Munoz A. Letter to the editor: A cautionary note on
the use of autoregressive modes in analysis of longitudinal
data. Stat Med 1990; 9: 492 ± 495.
35 Garrison RJ, Wilson PW, Castelli WP, Feinleib M, Kannel
WB, McNamara PM. Obesity and lipoprotein cholesterol in
the Framingham Offspring Study. Metabolism 1980; 29:
1053 ± 1060.
36 Lauer RM, Lee J, Clarke WR. Factors affecting the relationship between childhood and adult cholesterol levels: The
Framingham Study. Pediatrics 1988; 82: 309 ± 318.
37 Van Horn LV, Ballew C, Liu K, Ruth K, McDonald A, Hilner
JE, Burke GL, Savage PJ, Bragg C, Caan B, Jacobs Jr D,
Slattery M, Sidney S. Diet, body size and plasma lipidslipoproteins in young adults: Differences by race and sex.
The Coronary Artery Risk Development in Young Adults
(CARDIA) Study. Am J Epidemiol 1991; 133: 9 ± 23.
38 Marti B, Tuomilehto J, Salomaa V, Kartovaara L, Korhonen
HJ, Pietinen P. Body fat distribution in the Finnish population:
environmental determinants and predictive power for cardiovascular risk factor levels. J Epidemiol Commun Health 1991;
45: 131 ± 137.
39 Seidman DS, Laor A, Gale R, Stevenson DK, Mashiach
S, Danon YL. Birth weight, current body weight, and
blood pressure in late adolescence. BMJ 1991; 302:
1235 ± 1237.
40 Holland FJ, Stark O, Ades AE, Peckham CS. Birth weight and
body mass index in childhood, adolescence, and adulthood as
predictors of blood pressure at age 36. J Epidemiol Commun
Health 1993; 47: 432 ± 435.
41 Fripp RR, Hodgson JL, Kwierovitch PO, Werner JC, Schuler
HG, Whitman V. Aerobic capacity, obesity and atherosclerotic
risk factors in male adolescents. Pediatrics 1985; 75:
813 ± 819.
42 Suter E, Hawes MR. Relationship of physical activity, body
fat, diet, and blood lipid pro®le in youths 10 ± 15 yr. Med Sci
Sports Exerc 1993; 25: 748 ± 754.
43 Malina RM, Beunen GP, Claessens AL, Lefevre J, Vanden
Eynde BV, Renson R, Vanreusel B, Simons J. Fatness and
physical ®tness of girls 7 to 17 years. Obes Res 1995; 3:
221 ± 231.
44 Welsman J, Armstrong N, Nevill A, Winter E, Kirby B.
_ 2 for differences in body size. Med Sci
Scaling peak VO
Sports Exerc 1996; 28: 259 ± 265.
45 Bergh U, SjoÈdin B, Forsberg A, Svedenhag J. The relationship
between body mass and oxygen uptake during running in
humans. Med Sci Sports Exerc 1991; 23: 205 ± 211.
46 Moussa MAA, Skaik MB, Selwanes SB, Yaghy OY,
Bin-Othman SA. Factors associated with obesity in school
children. Int J Obes 1994; 18: 513 ± 515.
47 Shimokata H, Muller DC, Andres R. Studies in the distribution
of body fat. III Effects of cigarette smoking. JAMA 1989; 261:
1169 ± 1173.
48 Sheppard L, Kristal AR, Kushi LH. Weight loss in women
participating in a randomized trial of low-fat diets. Am J Clin
Nutr 1991; 54: 821 ± 828.
49 Rolls BJ, Shide DJ. The in¯uence of dietary fat on food intake
and body weight. Nutr Rev 1992; 50: 283 ± 290.