Relation of body mass index and waist-to-height ratio to
cardiovascular disease risk factors in children and adolescents:
the Bogalusa Heart Study1 4
David S Freedman, Henry S Kahn, Zuguo Mei, Laurence M Grummer-Strawn, William H Dietz, Sathanur R Srinivasan,
and Gerald S Berenson
ABSTRACT
Background: Several investigators have concluded that the waist-
to-height ratio is more strongly associated with cardiovascular dis-
ease risk factors than is the body mass index (BMI; in kg/m2).
Objectives: We examined the relation of the BMI-for-age z score
and waist-to-height ratio to risk factors (lipids, fasting insulin, and
blood pressures). We also compared the abilities of these 2 indexes
to identify children with adverse risk factors.
Design: Children aged 517 y (n 2498) in the Bogalusa Heart
Study were evaluated.
Results: As assessed by the ability of the 2 indexes to 1) account for
the variability in each risk factor and 2) correctly identify children
with adverse values, the predictive abilities of the BMI-for-age z
score and waist-to-height ratio were similar. Waist-to-height ratio
was slightly better (0.01 0.02 higher R2 values, P 0.05) in pre-
dicting concentrations of total-to-HDL cholesterol ratio and LDL
cholesterol, but BMI was slightly better in identifying children with
high systolic blood pressure (0.03 higher R2, P 0.05) in predicting
measures of fasting insulin and systolic and diastolic blood pres-
sures. On the basis of an overall index of the 6 risk factors, no
difference was observed in the predictive abilities of BMI-for-age
and waist-to-height ratio, with areas under the curves of 0.85 and
0.86 (P 0.30) and multiple R2 values of 0.320 and 0.318 (P
0.79). This similarity likely results from the high intercorrelation (R2
0.78) between the 2 indexes.
Conclusions: BMI-for-age and waist-to-height ratio do not differ in
their abilities to identify children with adverse risk factors. Although
waist-to-height ratio may be preferred because of its simplicity,
additional longitudinal data are needed to examine its relation to
disease. Am J Clin Nutr 2007;86:33 40.
KEY WORDS BMI, body mass index, waist, height, waist-to-
height ratio, children, lipids, blood pressure, insulin
INTRODUCTION
Vague (1) was the first to observe that women with android
obesity had a high prevalence of diabetes and atherosclerosis.
Subsequent studies have shown that abdominal obesity, as mea-
sured by the waist circumference or related indexes such as the
waist-to-hip ratio, is associated with the subsequent development
of type 2 diabetes (25) and ischemic heart disease (6 8), as well
as with risk factors for cardiovascular disease (CVD) (9). Fur-
thermore, despite the relatively low amount of intraabdominal fat
Am J Clin Nutr 2007;86:33 40. Printed in USA. 2007 American Society for Nutrition
among children (10), several indexes of abdominal obesity are
associated with CVD risk factors among children and adoles-
cents (1116).
The waist-to-height ratio was first used in the Framingham
Downloaded from ajcn.nutrition.org by guest on September 10, 2012
Study (17), and several studies of children (1315) and adults
(18, 19) have concluded that this ratio is more strongly associated
with CVD risk factors than is the body mass index (BMI; in
kg/m2). In addition, waist-to-height ratio may be simpler to use.
For example, because waist-to-height ratio is only weakly asso-
ciated with age, measures among children do not have to be
expressed relative to their sex and age peers [by using z scores
(20)] as do measures of BMI. In addition, the same cutoff (eg,
0.5) could possibly be used to identify adverse measures of waist-
to-height ratio among both children and adults (21, 22), which
would simplify the expression of obesity-related disease risk.
However, relatively few studies have examined the relation of
waist-to-height ratio to CVD risk factors, and it is important to
examine these associations in other data.
The current study compares the relation of BMI and waist-to-
height ratio to measures of lipids, fasting insulin, and blood
pressure among 517-y-olds (n 2498) in the Bogalusa Heart
Study. In addition, we examine the abilities of these 2 indexes to
correctly identify children with adverse risk factors.
SUBJECTS AND METHODS
Study population
The Bogalusa (Louisiana) Heart Study is a community-based
(Ward 4 of Washington Parish) study of CVD risk factors in early
life (23). Seven cross-sectional examinations of schoolchildren
were conducted since 1973, and the current analyses are based on
1
From the Divisions of Nutrition and Physical Activity (DSF, ZM,
LMG-S, and WHD) and Diabetes Translation (HSK), Centers for Disease
Control and Prevention, Atlanta, GA, and the Tulane Center for Cardiovas-
cular Health, Tulane University School of Public Health and Tropical Med-
icine, New Orleans, LA (SRS and GSB).
2
The findings and conclusions in this report are those of the authors and
not necessarily those of the CDC.
3
Supported by grant no. AG-16592 from the National Institutes of Aging.
4
Address reprint requests to DS Freedman, CDC K-26, 4770 Buford
Highway, Atlanta GA 30341. E-mail: dfreedman@cdc.gov.
Received September 26, 2006.
Accepted for publication March 16, 2007.
33
34 FREEDMAN ET AL
the 19931994 examination. Written informed consent was ob-
tained from all parents, and study protocols were approved by
human subjects review committees at the Tulane University
School of Public Health and Tropical Medicine.
Of the 3135 children and adolescents (aged 517 y) examined,
we excluded 9 girls who reported being pregnant, 7 children who
were not white or black, 30 children who reported taking insulin
(or were unsure), 13 children for whom we did not have a systolic
(SBP) or diastolic (DBP) blood pressure measurement, and 14
children for whom information on measurements of waist,
weight, or height was missing; these categories were not mutu-
ally exclusive. Of the remaining 3066 children, cholesterol (total,
LDL, and HDL) and triacylglycerol determinations were avail-
able for 2961. Nonfasting children were excluded from the anal-
yses of triacylglycerol and fasting insulin concentrations, and
another 130 children did not have an insulin determination. After
these exclusions, sample sizes for the various risk factors are
3066 (for SBP and DBP), 2961 (for LDL and HDL cholesterol),
2624 (for triacylglycerol), and 2494 (for insulin).
Because obesity is associated positively with LDL cholesterol
and negatively with HDL cholesterol, we did not examine asso-
ciations with total cholesterol. However, the ratio of total cho-
lesterol to HDL cholesterol (total:HDL cholesterol) is included
in the analyses.
General examinations
Height was measured to the nearest 0.1 cm with the use of an
Iowa Height Board, and weight was measured to the nearest 0.1
kg with the use of a balance beam metric scale; BMI was calcu-
lated as a measure of relative weight. No adjustments were made
for the weight of the gown, underpants, or socks that were worn
during the examination.
BMI z scores were calculated from the 2000 Centers for Dis-
ease Control and Prevention (CDC) Growth Charts (20, 24) to
account for the differences in BMIs by sex and age. These growth
charts express the BMIs of children in the current study relative
to their sex and age peers in the United States between 1963 and
1980; the calculated z scores are termed BMI-for-age in the
current analyses. (BMIs among 5-y-olds in the CDC Growth
Charts also include data from 1988 1994.) Overweight is de-
fined as a BMI-for-age z score 1.645 (corresponding to the
95th percentile of normally distributed data) of these growth
charts (25, 26). BMI-for-age z scores were used in all analyses in
the current study. BMI-for-age percentiles are used only to clas-
sify children into 4 categories in one table that cross-classifies
BMI-for-age and waist-to-height ratio.
The waist circumference was measured midway between the
rib cage and the superior border of the iliac crest while the child
was standing. Three measurements were obtained with a non-
stretchable tape, and the mean value was used in the calculation
of the waist-to-height ratio. In analyses that compared the abil-
ities of BMI and waist-to-height ratio to correctly identify chil-
dren with adverse risk factors, we dichotomized waist-to-height
ratio at 0.512 (without considering the childs sex or age) so that
the same proportion (17%) of children would be overweight and
have a high waist-to-height ratio.
On each examination day, a 10% sample of the children was
randomly selected to be reexamined 23 h later by the same
observer. We use these data to compare the reproducibilities of
BMI and waist-to-height ratio.
Risk factors
Concentrations of serum total cholesterol and triacylglycerols
were measured by using enzymatic procedures in a centralized
laboratory that met the requirements of the CDCs Lipid Stan-
dardization Program. For LDL- and HDL-cholesterol measure-
ments, we used a combination of heparin-calcium precipitation
and agar-agarose gel electrophoresis (27). Plasma insulin mea-
surements were obtained with the use of a radioimmunoassay
procedure (Phadebas Insulin Kit; Pharmacia Diagnostics AB,
Uppsala, Sweden).
As previously described (23), sitting SBP and DBP in the right
arm were measured 6 times by trained observers with a mer-
cury sphygmomanometer (Baumanometer; WA Baum Co Inc,
Copiague, NY). The cuff size was based on the length and
circumference of the upper arm and was chosen to be as large
as possible without having the elbow skin crease obstruct the
stethoscope (28).
The distributions of lipid and lipoprotein concentrations in the
Bogalusa Study were similar to those in the third National Health
Downloaded from ajcn.nutrition.org by guest on September 10, 2012
and Nutrition Examination (NHANES III) conducted from 1988
to 1994 (29). For example, the 90th percentiles of LDL choles-
terol among 1215-y-old white children (data were not cross-
classified by race, sex, or age group) in NHANES III were 122
mg/dL (whites) and 133 mg/dL (blacks); corresponding values in
the Bogalusa Study were 127 mg/dL (whites) and 133 mg/dL
(blacks). Similarly, the 10th percentiles of HDL cholesterol were
35 mg/dL (boys) and 36 mg/dL (girls) among 1215-y-olds in
NHANES III and were 37 mg/dL among both boys and girls in
the Bogalusa Study. However, because of differences in methods
of measuring blood pressure (28), recorded measures of blood
pressure are 510 mm Hg lower in the Bogalusa Study than in
other studies.
Measures of adverse risk factors
Because measures of lipids, insulin, and blood pressures vary
substantially by sex and age, we defined adverse measures in
relation to a childs sex and age peers in the Bogalusa Study
sample. After log-transformation of measures of the risk factors
to improve normality, each risk factor was regressed on sex, race,
and age. Age was modeled with the use of restricted cubic splines
with 5 knots (see Statistical analyses) (30), and we allowed for
interactions with age (age BMI and age waist-to-height
ratio) in the prediction of each risk factor. Regression models for
SBP and DBP also included height (cubic splines) as a predictor.
The standardized residuals (adjusted risk factor measures) from
these models represent measures relative to children of the same
sex, race, and age. All adjusted risk factors had a mean SD
value of 0 1.0. With the exception of HDL cholesterol (10th
percentile), adverse risk factor measures were defined as a mea-
sure 90th percentile.
Although the identification of children with adverse risk fac-
tors in the current study is based solely on the distribution of risk
factors in the Bogalusa Study, the use of cutoffs from NHANES
III (29) identified similar children with adverse concentrations of
lipids and lipoproteins. For example, all of the 1215-y-olds
(whites and blacks combined; n 106) in the current study who
were classified as having a high LDL-cholesterol concentration
(according to the Bogalusa Study cutoffs) also had a concentra-
tion 90th percentile (119 mg/dL) in NHANES III. However, 45
(5%) of the 950 children aged 1215 y who we considered to have
 CHILDHOOD BMI, WAIST-HEIGHT RATIO, AND CARDIOVASCULAR RISK
a normal LDL-cholesterol concentration were in the 90th
percentile in NHANES III. [It should be noted that some esti-
mates of the 90th percentile in NHANES III were considered to
be unstable because of the relatively small sample size (29).]
Because of differences in methods of measuring blood pressure
(28), few children in the Bogalusa Study had a SBP or DBP
90th percentile of the National High Blood Pressure Education
Program (31).
The risk factor sum was used as a summary measure of the 6
risk factors and was derived by combining adjusted measures of
triacylglycerols, LDL cholesterol, HDL cholesterol, fasting in-
sulin, SBP, and DBP. Adjusted measures of most risk factors
were simply added together, but adjusted measures of HDL cho-
lesterol were subtracted from the total. In addition, because of the
high correlation (r 0.66) between SBP and DBP, these 2
characteristics were first divided by 2. The resulting risk factor
sum had a mean SD value of 0 2.9 (range: 111). Correla-
tions between the risk factor sum and the individual risk factors
ranged from r 0.37 (DBP) to r 0.73 (triacylglycerols); the
association with HDL cholesterol was r 0.59.
The risk factor sum was highly correlated (r 0.97) with the
first principal component (32) of the 6 risk factors. Furthermore,
with the exception of LDL-cholesterol concentrations (r 0.39),
the absolute value of the correlation coefficients with the first
principal component ranged from 0.50 (HDL cholesterol) to 0.70
(triacylglycerols). (The second principal component was diffi-
cult to interpret because it contrasted measures of DBP, SBP, and
HDL cholesterol with measures of triacylglycerols and fasting
insulin, and it was not considered further.) Although risk factor
summaries can be derived by adding together the number of
adverse risk factors (16, 33), our method allows the risk factor
sum to be used as a continuous variable.
Statistical analyses
The analyses, which were performed with the use of SAS
software (version 9.1; SAS Institute Inc, Cary, NC) and R [ver-
sion 2.4.1; R Foundation for Statistical Computing, Vienna, Aus-
tria (34)], first examined the ability of BMI-for-age and waist-
to-height ratio to identify children with adverse measures of each
risk factor. We calculated the positive predictive value (the pro-
portion of children with a high BMI or waist-to-height ratio who
actually have adverse risk factors) and the sensitivity (the pro-
portion of children with adverse risk factors who have a high BMI
or waist-to-height ratio) for each risk factor. Because these val-
ues depend on the cutoff used for BMI and waist-to-height ratio,
we also examined the receiver operating characteristic curve for
each risk factor. These curves are constructed by plotting the
sensitivity at each value of BMI-for-age or waist-to-height ratio
compared with the corresponding 1-specificity, and the area un-
der the curve (AUC) quantifies the screening performance over
all cutoffs. An AUC of 0.5 indicates that the screening test is no
better than chance, and 1.0 indicates perfect classification. The
statistical significance of the difference (35) in AUCs between
BMI and waist-to-height ratio was calculated by using MED-
CALC software (version 9.1.0.1; MedCalc Software, Mari-
akerke, Belgium).
Regression models were also used to quantify the prediction of
risk factor measures by both indexes. (The original, unadjusted
measures of the risk factors were used as the dependent variable
in these models.) These analyses compared the increases in the
multiple R2 values achieved by adding either BMI-for-age z score
35
or waist-to-height ratio to a model already containing age, sex,
and race. Continuous variables were modeled by using restricted
cubic splines with 5 knots (30) to allow for nonlinearity, and we
allowed for an interaction between each index and age. In con-
trast to the use of higher-order polynomials, models based on
splines do not have peaks and valleys, and the fit in one region
does not influence the fit in all other regions of the data.
To assess the differences between the R2 values of models that
contained either BMI-for-age or waist-to-height ratio, we first
calculated predicted risk factor measures from each model. We
then examined the statistical significance of the difference in the
correlation between the actual risk factor measures and the 2 sets
of predicted risk factor measures coefficients (36). We also ex-
amined whether the relation of BMI-for-age and waist-to-height
ratio to each risk factor was nonlinear.
We then cross-classified categories of BMI-for-age (50th
percentile, 50th 84th percentile, 85th94th percentile, and
95th percentile) and waist-to-height ratio. Cutoffs for the 4
categories of waist-to-height ratio were selected so that the num-
ber of children in each category would equal the number in the
Downloaded from ajcn.nutrition.org by guest on September 10, 2012
corresponding BMI-for-age category. We focused on measures
of the risk factor sum among children whose BMI-for-age stra-
tum was lower or higher (discordant) than the corresponding
waist-to-height ratio stratum. We also show the relation of waist-
to-height ratio to BMI-for-age by using lowess to smooth the data
(37).
In addition to determining whether differences between BMI-
for-age and waist-to-height ratio were statistically significant,
we also focused on the magnitudes of the differences in the AUCs
and R2 values for each risk factor. It should be realized that a
small difference between the 2 indexes, which indicates that their
predictive abilities are similar, could be statistically significant.
However, this small difference would have little practical im-
portance.
RESULTS
Mean measures of various characteristics are shown in Table
1. The mean BMI-for-age z score was 0.46, and 17% of the
children were overweight; BMIs did not differ significantly be-
tween boys and girls. The mean waist-to-height ratio was 0.458,
and measures of waist, height, waist-to-height ratio, and SBP
were slightly but significantly higher among boys than among
girls. In contrast, girls had slightly but significantly higher con-
centrations of total cholesterol:HDL cholesterol, triacylglycer-
ols, LDL cholesterol, and fasting insulin than did boys. Measures
of the risk factor sum, an overall summary of the 6 age- and
sex-adjusted risk factors, ranged from 9 to 11, with a mean
value of 0. Age was associated with BMI (r 0.48), waist
circumference (r 0.59), and height (r 0.90) but not with
BMI-for-age (r 0.02) or waist-to-height ratio (r 0.01).
Among the 276 children who were reexamined by the same
observer, the intraclass correlations between the repeated mea-
surements were 0.997 (BMI) and 0.949 (waist-to-height ratio),
and the coefficients of variation were 1% (BMI) and 3% (waist-
to-height ratio).
The abilities of BMI-for-age and waist-to-height ratio to cor-
rectly identify children with adverse risk factors are compared in
Table 2. The AUCs, positive predictive values, and sensitivities
varied substantially across risk factors, with the most accurate
classification seen for fasting insulin concentrations. For each
36 FREEDMAN ET AL

TABLE 1 (sensitivities of 0.426 and 0.463 for total cholesterol:HDL cho-

Characteristics of the subjects1 lesterol). On the basis of statistically significant (P 0.05) dif-
Boys Girls ferences in the AUCs, BMI-for-age was a slightly better predic-
(n 1501) (n 1565) tor of adverse SBP measures, whereas waist-to-height ratio was
a slightly better predictor of adverse concentrations of total cho-
Age (y) 11 32 11 3
Black (%) 43 44
lesterol:HDL cholesterol. (All differences, however, were
BMI (kg/m2) 20.0 4.8 20.2 5.2 0.03.) The abilities of BMI-for-age and waist-to-height ratio to
BMI-for-age z score 0.5 1.1 0.4 1.1 correctly identify children with adverse measures of the risk
Overweight (%) 17 16 factor sum were similar (AUCs of 0.85 and 0.86). We also ex-
Waist circumference (cm) 68 143 66 123 amined possible differences in the AUCs for the risk factor sum
Height (cm) 148 203 145 173 by race, sex, and age group (11 y or 11 y old). Although some
Waist-to-height ratio 0.46 0.063 0.45 0.063 differences were observed across strata (eg, the AUC for waist-
Total-to-HDL cholesterol ratio 3.2 0.83 3.4 0.93
to-height ratio was 0.87 among white children and 0.82 among
Triacylglycerols (mg/dL)4 67 (52, 92)5 73 (57, 98)5
LDL cholesterol (mg/dL) 101 253 104 263
black children), within each strata, the AUCs for BMI-for-age
HDL cholesterol (mg/dL) 54 12 53 12 and waist-to-height ratio were almost identical.
Fasting insulin (mU/L)4 8.6 (6.4, 12.0)5 10.0 (7.7, 14.0)5 We then compared the additional information, quantified by
SBP (mm Hg) 103 103 102 103 the multiple R2, provided by BMI-for-age and waist-to-height
DBP (mm Hg) 62 9 62 9 ratio in predicting measures of the various risk factors (Table 3).
Risk factor sum 03 03 Race, sex, and age (first column) could account for 4% (LDL

Downloaded from ajcn.nutrition.org by guest on September 10, 2012

1
The numbers of children with missing data for the individual risk
factors ranged from 0 (SBP and DBP) to 572 (fasting insulin). There were 573
children with missing information on the risk factor sum. SBP, systolic blood
pressure; DBP, diastolic blood pressure.
2
x SD (all such values).
3
Significant difference for sex, P 0.001 (t test).
4 of the variability in risk factor measures. Although some of the
Values are median; 25th and 75th percentiles in parentheses.
5
Significant difference for sex, P 0.001 (t test, or Wilcoxon test for
triacylglycerols or fasting insulin).
risk factor, however, only small differences were observed be-
tween the accuracies of the BMI-for-age and waist-to-height
ratio. The largest differences between any of the AUCs, positive
predictive values, and sensitivities of the 2 indexes were 0.037
cholesterol) to 34% (DBP) of the variability in the individual risk
factors. BMI-for-age and waist-to-height ratio (second and third
columns) could account for an additional 3% (BMI-for-age com-
pared with LDL cholesterol and waist-to-height ratio compared
with DBP) to 27% (BMI-for-age compared with fasting insulin)
differences between the 2 indexes were statistically significant,
the magnitudes of all differences were small. For example, al-
though waist-to-height ratio was a better (P 0.001) predictor of
total cholesterol:HDL cholesterol, the difference in R2 values
was only 0.024. Similarly, BMI-for-age was a better predictor of
measures of fasting insulin, SBP, and DBP, but the largest R2
difference was 0.034. The prediction of the risk factor sum by the
2 indexes was almost identical (R2 0.32 for each index) As
indicated in the final column, the use of the 2 indexes together

TABLE 2
Classification of adverse risk factor by BMI-for-age z score and waist-to-height ratio1

Area under the curve Positive predictive value Sensitivity

BMI-for-age Waist-to-height BMI-for-age Waist-to-height BMI-for-age Waist-to-height

Risk factor2 Subjects z score ratio z score ratio3 z score ratio3
n
Total-to-HDL cholesterol ratio 2961 0.734 0.764 0.26 0.28 0.43 0.46
Triacylglycerols 2624 0.74 0.75 0.28 0.28 0.47 0.48
LDL cholesterol 2961 0.61 0.62 0.16 0.18 0.27 0.30
HDL cholesterol 2961 0.67 0.68 0.21 0.22 0.35 0.36
Fasting insulin 2494 0.81 0.79 0.34 0.33 0.57 0.56
SBP 3066 0.684 0.654 0.19 0.20 0.33 0.33
DBP 3066 0.59 0.60 0.13 0.14 0.22 0.24
Risk factor sum 2493 0.85 0.86 0.38 0.38 0.64 0.65
1
The area under the curve summarizes the sensitivity and specificity of BMI-for-age and waist-to-height ratio over all values of the index (see Subjects
and Methods). Positive predictive value is the proportion of children with a high BMI-for-age (or waist-to-height ratio) who had an adverse risk factor.
Sensitivity is the proportion of children with an adverse risk factor who had a high BMI-for-age (or waist-to-height ratio). SBP, systolic blood pressure; DBP,
diastolic blood pressure.
2
The race-, sex-, and age-adjusted risk factors were used in the analyses.
3
The cutoff (0.512) for waist-to-height ratio was selected so that equal (17%) proportions of the children would be overweight and would have a high
waist-to-height ratio.
4
Significant difference in area under the curve, P 0.05 [as assessed by formulas in Hanely et al (35)]. The SE of each area under the curve difference
was 0.01.
 CHILDHOOD BMI, WAIST-HEIGHT RATIO, AND CARDIOVASCULAR RISK 37
TABLE 3
Variability in risk factors that can be accounted for by BMI-for-age z scores or waist-to-height ratios in various regression models1
Increase in multiple R2 values

Multiple R2 values BMI-for-age z

based on race, sex, BMI-for-age Waist-to- score and waist-
Risk factor2 Subjects and age z score height ratio to-height ratio
n
Total-to-HDL cholesterol ratio 2961 0.05 0.143 0.163 0.16
Triacylglycerols 2624 0.10 0.12 0.13 0.14
LDL cholesterol 2961 0.04 0.033 0.043 0.05
HDL cholesterol 2961 0.13 0.09 0.09 0.10
Fasting insulin 2494 0.21 0.273 0.243 0.28
SBP 3066 0.29 0.083 0.053 0.09
DBP 3066 0.34 0.063 0.033 0.06
Risk factor sum 2493 0 0.32 0.32 0.34
1
Age, BMI-for-age, and waist-to-height ratio were modeled by using restricted cubic splines with 5 knots. Regression models also contained the interaction
between BMI-for-age z score or waist-to-height ratio with age. SBP, systolic blood pressure; DBP, diastolic blood pressure.
2
The original, unadjusted risk factors were used as the dependent variable. The multiple R2 for the model predicting the risk factor sum from race, sex,
and age is 0 because the risk factor sum is based on standardized measures of the risk factors.

Downloaded from ajcn.nutrition.org by guest on September 10, 2012

3
Significant P 0.05 difference in multiple R2 values of a model containing BMI-for-age z score and a model containing waist-to-height ratio in the
prediction of risk factor measures, P 0.05. To calculate statistical significance, predicted measures of each risk factor were first calculated from the 2 regression
models. The difference in the correlations between these predicted and actual measures was assessed as suggested by Meng et al (36).

yielded multiple R2 values that were only slightly higher than
those obtained with the use of either index. For example, the
multiple R2 for the risk factor sum based on both BMI-for-age
and waist-to-height ratio was 0.34, whereas the R2 for each index
alone was 0.32.
Despite the similarity of the multiple R2 values for the 2 in-
dexes, additional analyses indicated that associations with BMI-
for-age were more curvilinear than those with waist-to-height
ratio. Predicted measures of several risk factors based on regres-
had a high BMI-for-age relative to waist-to-height ratio (6 upper
right cells; x 0.21) was almost identical to the mean measure
among children who had a high waist-to-height ratio relative to
BMI-for-age (6 lower left cells; x 0.19; P 0.88). Compa-
rable analyses for the individual risk factors indicated that chil-
dren with a relatively high waist-to-height ratio had slightly
120 120

sion models containing either BMI-for-age (left panels) or waist- SBP (mm Hg)
100 100
to-height ratio (right panels) for an 11-y-old white girl are shown
in Figure 1. (Predicted measures for boys and black children TG
80 80
would be shifted vertically, but they would parallel the curves in (mg/dL)
Figure 1.) Nonlinearity was most evident in the relation of BMI- 60 60
for-age to concentrations of triacylglycerol (upper left panel), HDL-C
fasting insulin, and the risk factor sum (bottom left panel) but was 40 (mg/dL) 40
also observed for SBP and HDL cholesterol. Furthermore, for -2 -1 0 1 2 3 0.4 0.5 0.6 0.7

each risk factor, the strength of the association increased (steeper 25 25

slope) at higher measures of BMI-for-age. Associations with
Fasting
waist-to-height ratio (right panels), in contrast, were more linear, 20 insulin 20
and the difference between the 2 indexes was particularly evident (U/L)
for triacylglycerol concentrations. Although waist-to-height ra- 15 15
tio showed a nonlinear association (P 0.001) with fasting
insulin concentrations, the change in slope was less marked than 10 10
RF Sum
with BMI-for-age.
We then examined measures of the risk factor sum among 5 5
children after a cross-classification of categories of BMI-for-age -2 -1 0 1 2 3 0.4 0.5 0.6 0.7
and waist-to-height ratio (Table 4). Waist-to-height ratio mea- BMI-for-age z score Waist-to-height ratio
sures were categorized so that equal numbers of children would
FIGURE 1. The relation of BMI-for-age z score (left) and waist-to-height
be in each waist-to-height ratio and BMI group. Despite the small
ratio (right) to risk factor (RF) measures. Predicted measures (for an 11-y-old
number of children in some of the discordant categories (cells white girl) were calculated from regression models that included BMI-for-
above and below the shaded diagonal), the mean risk factor sum age or waist-to-height ratio, in addition to age, sex, race, and the interaction
tended to increase with measures of both BMI-for-age and waist- of BMI-for-age (or waist-to-height ratio) with age. Continuous variables
to-height ratio. Because of residual confounding, however, these were modeled by using splines with 5 knots, and the nonlinear effect of
BMI-for-age was statistically significant (P 0.001) for each of the 5 RFs.
apparent independent effects should be interpreted cautiously. So that the RF sum and fasting insulin could be plotted in the same figure, 9.0
A comparison of measures of the risk factor sum in the 2 discor- was added to the former. The units for each RF are shown in parentheses.
dant groups indicated that the mean measure among those who SBP, systolic blood pressure; TG, triacylglycerols.
38 FREEDMAN ET AL

TABLE 4
Measures of the risk factor sum cross-classified by BMI-for-age and waist-to-height ratio1

BMI-for-age percentile

50th 84th 85th94th 95th

50th Percentile Percentile Percentile Percentile
Waist-to-height ratio (n 869) (n 825) (n 382) (n 417)

0.426 (n 869) 1.5 2.3 (623) 1.0 2.4 (240) 1.6 2.9 (4) 0.9 3.3 (2)
0.4260.466 (n 825) 1.0 2.5 (233) 0.4 2.4 (468) 0.2 2.5 (115) 1.1 1.7 (9)
0.4670.512 (n 382) 1.4 2.0 (13) 0.4 2.4 (107) 1.1 2.6 (187) 1.6 2.4 (75)
0.513 (n 417) (0) 0.7 2.2 (10) 1.6 2.8 (76) 3.3 2.5 (331)
1
All values are x SD; n in parentheses. Four hundred forty-five children had a high BMI-for-age relative to waist-to-height ratio (6 upper right cells);
439 children had a high waist-to-height ratio relative to BMI-for-age (6 lower left cells), and 1609 children were concordant for measures of BMI-or-age and
waist-to-height ratio (shaded cells along diagonal).

higher concentrations of total cholesterol:HDL cholesterol and DISCUSSION

LDL cholesterol, whereas those with a high BMI-for-age had Our results show that there is little difference in the abilities of
slightly higher concentrations of fasting insulin (P 0.05 for BMI-for-age and waist-to-height ratio to identify children with

Downloaded from ajcn.nutrition.org by guest on September 10, 2012

each difference).
Measures of BMI-for-age and waist-to-height ratio for each
child, with the triangles representing children who had a high risk
factor sum, are shown in Figure 2. The strong association be-
tween the 2 indexes is evident, and regression models indicated
that measures of BMI-for-age could account for 78% of the
variability in measures of waist-to-height ratio. Furthermore, the
identification of high measures of the risk factor sum by waist-
to-height ratio (horizontal line) and BMI-for-age (vertical line)
did not differ. Of the 250 children with a high risk factor sum, 145
(58%) had high measures of both indexes (upper right), 14 (6%)
had high measures of BMI-for-age only (lower right), 15 (6%)
had high measures of waist-to-height ratio only (upper left), and
76 (30%) did not have high measures of either (lower left).
0.8
0.7
Waist-to-height ratio
adverse CVD risk factors. In general, waist-to-height ratio
showed slightly stronger associations with lipid and lipoprotein
concentrations, whereas BMI-for-age showed slightly stronger
associations with measures of fasting insulin and blood pres-
sures. Although some of the differences between the 2 indexes
were statistically significant, the AUCs, positive predictive val-
ues, sensitivities, and multiple R2 values were similar for each
risk factor. Furthermore, the use of both BMI-for-age and waist-
to-height ratio resulted in only slightly better prediction of risk
factors than that achieved with only one index. The strong asso-
ciation between BMI-for-age and waist-to-height ratio (R2
0.78) probably accounts for their similar predictive abilities, as
well as for the small amount of additional information obtained
by using the 2 indexes together.
Various indexes of abdominal obesity (such as waist circum-
ference, waist-to-hip ratio, and waist-to-height ratio) are associ-
ated with adverse risk factors among children (1315) and are
predictive of type 2 diabetes and CVD in adulthood (2 8). The
limitations of these indexes, however, should be considered. For
example, although waist circumference is correlated with the
amount of intraabdominal visceral fat, which may be the most

detrimental fat depot (9), it is also associated with subcutaneous

0.6
abdominal fat and with total body fat (38, 39). In addition, a
recent study of adults found that waist-to-height ratio and BMI
were more strongly associated with each other (r 0.85 0.91)
0.512 than with percentage of body fat (r 0.69 0.76), as determined
0.5
by air-displacement plethysmography (19). These associations
emphasize the potential problems in using waist-to-height ratio
and BMI as indexes of abdominal and generalized adiposity,
0.4 respectively. The interpretation of associations with BMI and
CDC 95th percentile waist-to-height ratio is further complicated by the possible rela-
tion of disease risk to height (40), which is in the denominator of
0.3 both indexes.
-2 -1 0 1 2 3
Some investigators have concluded that, compared with BMI,
waist-to-height ratio is more strongly associated with CVD risk
BMI-for-age z score factors among children (1315) and adults (18, 19). It has been
FIGURE 2. Each point represents the BMI-for-age z score and waist-to- emphasized, however, that many of the differences between
height ratio of an individual child, and the overlaid curve was constructed by waist-to-height ratio and BMI are relatively small (19). For ex-
using lowess. , children with an adverse risk factor (RF) sum. The vertical
line represents the cutoff for overweight (a z score of 1.654 corresponding to
ample, Hara et al (14) reported that the logarithm of a risk factor
the 95th percentile), and the horizontal line is the cutoff (0.512) for waist- score showed correlations of r 0.50 (waist-to-height ratio) and
to-height ratio. CDC, Centers for Disease Control and Prevention. r 0.45 (BMI), and Hsieh et al (33) reported correlations of r
 CHILDHOOD BMI, WAIST-HEIGHT RATIO, AND CARDIOVASCULAR RISK
0.37 (waist-to-height ratio) and r 0.33 (BMI) with a morbidity
index among men. The slightly stronger relation of BMI-for-
age (compared with waist-to-height ratio) to measures of SBP
and DBP that we observed was also noted by others (15, 19).
Several explanations are possible for the contrasting findings
about the relative importance of BMI-for-age and waist-to-
height ratio. Various subsets of risk factors have been included in
each study, and only one previous study included an index of
insulin resistance (19). (Of the risk factors we examined, fasting
insulin concentrations showed the strongest association with
BMI.) Furthermore, a study of 36 obese children found that
insulin resistance was more strongly associated with total fat
mass than with visceral abdominal fat (41). The weaker associ-
ations with BMI that were found in previous studies of children
may be due to the investigators use of BMI rather than BMI-
for-age (13, 14) or due to the fact that associations with BMI-
for-age were constrained to be linear (14 16, 19). We found that
forcing the association with concentrations of fasting insulin to
be linear reduced the R2 for BMI-for-age from 0.48 (nonlinear) to
0.43 (linear), whereas the R2 for waist-to-height ratio decreased
from 0.45 only to 0.44. These nonlinear associations may arise
because BMI-for-age is a good indicator of adiposity among
relatively fat children, but it is an index of both fat and fat-free
mass among thinner children (42). If BMI-for-age differences
among some relatively thin (eg, BMI-for-age z score 1.0)
children largely reflect differences in fat-free mass, it would be
expected that the relation of BMI-for-age to risk factor measures
would be flatter (Figure 1) among these children.
The current study has several potential limitations that should
be considered. Although the sample was not randomly selected,
measures of BMI, lipids, and lipoproteins were fairly comparable
to those reported in national studies (29). However, because of
differences in methods of measuring blood pressure (28), few
children in the Bogalusa Study had an SBP or a DBP 90th
percentile of the National High Blood Pressure Education Pro-
gram (31). Furthermore, although it can be difficult to compare
the magnitudes of the observed associations across studies be-
cause of differences in statistical modeling techniques, age
ranges, and the specific anthropometric index examined, the
magnitudes of the associations that we observed between BMI-
for-age and the examined risk factors agree well with those of
other studies (43).
Although several prospective studies found the indexes of
abdominal obesity to be stronger predictors of CVD and type 2
diabetes than is BMI (4 7), there are conflicting findings. For
example, the predictive abilities of BMI and waist-to-height ratio
for type 2 diabetes among Pima Indians were almost identical (3),
and several studies found that various indexes of abdominal
obesity predict disease no better than does BMI (2, 8, 17). For
example, the relative risk for coronary heart disease among men
in the upper quintile of waist circumference in the Physicians
Health Study was 1.60, whereas the corresponding relative risk
for BMI was 1.73 (8).
Some investigators have suggested that, even if the predictive
abilities of waist-to-height ratio and BMI-for-age are similar,
waist-to-height ratio may be preferred as an indicator of obesity-
related risk (15, 21, 22). The concept of a large waist relative to
height may be easier to explain than is the division of weight by
the square of height, particularly for people accustomed to using
pounds and inches. Furthermore, because waist-to-height ratios
vary only slightly by age and by sex among children, it is not
39
necessary to express measures as percentiles or z scores, relative
to a reference population, as is the case for BMI. (The correlation
between unadjusted and sex- and age-adjusted measures of
waist-to-height ratio was r 0.99, and the use of either adjusted
or unadjusted measures yielded virtually identical results.) The
calculation of waist-to-height ratio is also simpler, requiring only
the division of numbers in the same units. Furthermore, the
possible use of a single cutoff (0.5) to identify adverse measures
among both children and adults (21) would result in a simple
public health message: Keep your waist circumference to less
than half your height. In the current study, 85% of overweight
children had a waist-to-height ratio 0.5.
However, the disease risks associated with BMI have been
studied much more extensively than have those for waist-to-
height ratio, and additional longitudinal data on waist-to-height
ratio are needed. Furthermore, although the relation of childhood
BMIs to those in adulthood was examined in numerous studies
(reviewed in reference 44), we know of no study that has exam-
ined the tracking of waist-to-height ratios. In addition, although
the reproducibility of waist circumference measurements is high
Downloaded from ajcn.nutrition.org by guest on September 10, 2012
(45), we and others have found that it is lower than that of BMI
(19). This difference may limit the ability of waist-to-height ratio
to detect small changes in obesity-related risk. Furthermore,
waist circumference has been measured at numerous sites be-
tween the lowest rib and iliac crest, and there are differences
between the recommendations of the Anthropometric Standard-
ization Reference Manual (46), the World Health Organization,
and the National Institutes of Health (reviewed in reference 45).
Small changes in the location of the waist measurement can alter
associations with risk factor measures (47 49) and possibly with
disease risk.
In summary, we found that waist-to-height ratio and BMI-for-
age showed similar associations with CVD risk factors. Al-
though the use of waist-to-height ratio among children has the
potential to simplify the assessment of obesity-related risk, ad-
ditional information is needed on the tracking of waist-to-height
ratio from childhood to adulthood, as are data relating waist-to-
height ratio to morbidity and mortality.
The authors responsibilities were as followsDSF: analyzed data, in-
terpreted results, and wrote the manuscript; HSK: interpreted results; ZM and
LMG-S: analyzed data, interpreted results, and formulated study objectives;
WHD: formulated study objectives and drafted the manuscript; SRS and
GSB (principal investigator): data collection; and all authors: revised the
manuscript. None of the authors had a personal or financial conflict of
interest.
REFERENCES
1. Vague J. The degree of masculine differentiation of obesities: a factor
determining predisposition to diabetes, atherosclerosis, gout, and uric
calculous disease. Am J Clin Nutr 1956;4:20 34.
2. Stevens J, Couper D, Pankow J, et al. Sensitivity and specificity of
anthropometrics for the prediction of diabetes in a biracial cohort. Obes
Res 2001;9:696 705.
3. Tulloch-Reid MK, Williams DE, Looker HC, Hanson RL, Knowler WC.
Do measures of body fat distribution provide information on the risk of
type 2 diabetes in addition to measures of general obesity? Comparison
of anthropometric predictors of type 2 diabetes in Pima Indians. Diabetes
Care 2003;26:2556 61.
4. Wang Y, Rimm EB, Stampfer MJ, Willett WC, Hu FB. Comparison of
abdominal adiposity and overall obesity in predicting risk of type 2
diabetes among men. Am J Clin Nutr 2005;81:555 63.
5. Schulze MB, Heidemann C, Schienkiewitz A, Bergmann MM,
Hoffmann K, Boeing H. Comparison of anthropometric characteristics
40 FREEDMAN ET AL
in predicting the incidence of type 2 diabetes in the EPIC-Potsdam
Study. Diabetes Care 2006;29:19213.
6. Folsom AR, Kushi LH, Anderson KE, et al. Associations of general and
abdominal obesity with multiple health outcomes in older women: the
Iowa Womens Health Study. Arch Intern Med 2000;160:211728.
7. Yarnell JW, Patterson CC, Thomas HF, Sweetnam PM. Central obesity:
predictive value of skinfold measurements for subsequent ischaemic
heart disease at 14 years follow-up in the Caerphilly Study. Int J Obes
Relat Metab Disord 2001;25:1546 9.
8. Rimm EB, Stampfer MJ, Giovannucci E, et al. Body size and fat distri-
bution as predictors of coronary heart disease among middle-aged and
older US men. Am J Epidemiol 1995;141:111727.
9. Despres JP. Is visceral obesity the cause of the metabolic syndrome? Ann
Med 2006;38:52 63.
10. Goran MI, Kaskoun M, Shuman WP. Intra-abdominal adipose tissue in
young children. Int J Obes Relat Metab Disord 1995;19:279 83.
11. Freedman DS, Srinivasan SR, Harsha DW, Webber LS, Berenson GS.
Relation of body fat patterning to lipid and lipoprotein concentrations in
children and adolescents: the Bogalusa Heart Study. Am J Clin Nutr
1989;50:930 9.
12. Maffeis C, Pietrobelli A, Grezzani A, Provera S, Tato L. Waist circum-
ference and cardiovascular risk factors in prepubertal children. Obes Res
2001;9:179 87.
13. Savva SC, Tornaritis M, Savva ME, et al. Waist circumference and
waist-to-height ratio are better predictors of cardiovascular disease risk
factors in children than body mass index. Int J Obes Relat Metab Disord
2000;24:1453 8.
14. Hara M, Saitou E, Iwata F, Okada T, Harada K. Waist-to-height ratio is
the best predictor of cardiovascular disease risk factors in Japanese
schoolchildren. J Atheroscler Thromb 2002;9:12732.
15. Kahn HS, Imperatore G, Cheng YJ. A population-based comparison of
BMI percentiles and waist-to-height ratio for identifying cardiovascular
risk in youth. J Pediatr 2005;146:482 8.
16. Katzmarzyk PT, Srinivasan SR, Chen W, Malina RM, Bouchard C,
Berenson GS. Body mass index, waist circumference, and clustering of
cardiovascular disease risk factors in a biracial sample of children and
adolescents. Pediatrics 2004;114:e198 205.
17. Higgins M, Kannel W, Garrison R, Pinsky J, Stokes J III. Hazards of
obesitythe Framingham experience. Acta Med Scand Suppl 1988;723:
2336.
18. Hsieh SD, Muto T. The superiority of waist-to-height ratio as an anthro-
pometric index to evaluate clustering of coronary risk factors among
non-obese men and women. Prev Med 2005;40:216 20.
19. Bosy-Westphal A, Geisler C, Onur S, et al. Value of body fat mass vs
anthropometric obesity indices in the assessment of metabolic risk fac-
tors. Int J Obes (Lond) 2006;30:475 83.
20. Kuczmarski RJ, Ogden CL, Guo SS, et al. CDC Growth Charts for the
United States: methods and development. Vital Health Stat 11 2000;
2002:1190.
21. Ashwell M, Hsieh SD. Six reasons why the waist-to-height ratio is a
rapid and effective global indicator for health risks of obesity and how its
use could simplify the international public health message on obesity. Int
J Food Sci Nutr 2005;56:3037.
22. McCarthy HD, Ashwell M. A study of central fatness using waist-to-
height ratios in UK children and adolescents over two decades supports
the simple messagekeep your waist circumference to less than half
your height. Int J Obes (Lond) 2006;30:988 92.
23. Berenson GS. Cardiovascular risk factors in children. New York, NY:
Oxford University Press, 1980:240 257.
24. Ogden CL, Kuczmarski RJ, Flegal KM, et al. Centers for Disease Con-
trol and Prevention 2000 growth charts for the United States: improve-
ments to the 1977 National Center for Health Statistics version. Pediat-
rics 2002;109:45 60.
25. Himes JH, Dietz WH. Guidelines for overweight in adolescent preven-
tive services: recommendations from an expert committee. Am J Clin
Nutr 1994;59:30716.
26. Kuczmarski RJ, Flegal KM. Criteria for definition of overweight in
transition: background and recommendations for the United States. Am J
Clin Nutr 2000;72:1074 81.
27. Srinivasan SR, Frerichs RR, Webber LS, Berenson GS. Serum lipopro-
tein profile in children from a biracial community. The Bogalusa Heart
Study. Circulation 1976;54:309 18.
28. Berenson GS, Cresanta JL, Webber LS. High blood pressure in the
young. Annu Rev Med 1984;35:535 60.
29. Hickman TB, Briefel RR, Carroll MD, et al. Distributions and trends of
serum lipid levels among United States children and adolescents ages
4 19 years: data from the third National Health and Nutrition Exami-
nation Survey. Prev Med 1998;27:879 90.
30. Harrell FR Jr. Regression modeling strategies with applications to linear
models, logistic regression, and survival analysis. New York, NY:
Springer, 2001:16 26.
31. Update on the 1987 Task Force Report on High Blood Pressure in
Children and Adolescents: a working group report from the National
High Blood Pressure Education Program. National High Blood Pressure
Education Program Working Group on Hypertension Control in Chil-
dren and Adolescents. Pediatrics 1996;98:649 58.
32. Everitt B. An R and S-Plus companion to multivariate analyses. London,
United Kingdom: Springer, 2005.
33. Hsieh SD, Yoshinaga H, Muto T. Waist-to-height ratio, a simple and
practical index for assessing central fat distribution and metabolic risk in
Japanese men and women. Int J Obes Relat Metab Disord 2003;27:
610 6.
34. R Development Core Team. R: a language and environment for statis-
Downloaded from ajcn.nutrition.org by guest on September 10, 2012
tical computing. Vienna, Austria: R Foundation for Statistical Comput-
ing, 2005. Internet: http://www.R-project.org (accessed 8 May 2007).
35. Hanley JA, McNeil BJ. A method of comparing the areas under receiver
operating characteristic curves derived from the same cases. Radiology
1983;148:839 43.
36. Meng XL, Rosenthal R, Rubin DB. Comparing correlated correlation
coefficients. Psychol Bull 1992;111:172 4.
37. Cleveland WS. Visualizing data. Summit, NJ: Hobart Press, 1993:93
130.
38. Lean ME, Han TS, Deurenberg P. Predicting body composition by den-
sitometry from simple anthropometric measurements. Am J Clin Nutr
1996;63:4 14.
39. Molarius A, Seidell JC. Selection of anthropometric indicators for clas-
sification of abdominal fatness: a critical review. Int J Obes Relat Metab
Disord 1998;22:719 27.
40. Hebert PR, Rich-Edwards JW, Manson JE, et al. Height and incidence of
cardiovascular disease in male physicians. Circulation 1993;88:1437
43.
41. Tershakovec AM, Kuppler KM, Zemel BS, et al. Body composition and
metabolic factors in obese children and adolescents. Int J Obes Relat
Metab Disord 2003;27:19 24.
42. Freedman DS, Wang J, Maynard LM, et al. Relation of BMI to fat and
fat-free mass among children and adolescents. Int J Obes Relat Metab
Disord 2005;29:1 8.
43. Freedman DS, Srinivasan SR, Berenson GS. The relation of obesity to
coronary heart disease risk factors among children. In: Burniat W, Cole
TJ, Lissau I, Poskitt EME, eds. Child and adolescent obesity: causes and
consequences, prevention and management. London, United Kingdom:
Cambridge University Press, 2002:22139.
44. Power C, Lake JK, Cole TJ. Measurement and long-term health risks of
child and adolescent fatness. Int J Obes Relat Metab Disord 1997;21:
50726.
45. Wang J, Thornton JC, Bari S, et al. Comparisons of waist circumferences
measured at 4 sites. Am J Clin Nutr 2003;77:379 84.
46. Lohman TG, Roche AF, Martorell R. Anthropometric standardization
reference manual. Champaign, IL: Human Kinetics Books, 1988.
47. Seidell JC, Cigolini M, Charzewska J, Ellsinger BM, Deslypere JP, Cruz
A. Fat distribution in European men: a comparison of anthropometric
measurements in relation to cardiovascular risk factors. Int J Obes Relat
Metab Disord 1992;16:1722.
48. Jakicic JM, Donnelly JE, Jawad AF, Jacobsen DJ, Gunderson SC,
Pascale R. Association between blood lipids and different measures of
body fat distribution: effects of BMI and age. Int J Obes Relat Metab
Disord 1993;17:1317.
49. Houmard JA, Wheeler WS, McCammon MR, et al. An evaluation of
waist to hip ratio measurement methods in relation to lipid and carbo-
hydrate metabolism in men. Int J Obes 1991;15:181 8.