# Identifying young children without overweight at high risk for adult overweight: the Terneuzen Birth Cohort.

**ABSTRACT** To develop a tool to identify children with high risk of adult overweight (AO), especially before developing overweight, based on body mass index (BMI) standard deviation score(s) (SDS) changes between 2-6 years (y) of age.

We fitted a linear spline model to BMI SDS of 762 young Caucasian adults from the Terneuzen Birth Cohort at fixed ages between birth and 18 y. By linear regression analysis, we assessed the increase in explained variance of the adult BMI SDS by adding the BMI SDS at 2 y to the models including the BMI SDS at 4 y, 6 y and both 4 y and 6 y. AO risk was modelled by logistic regression. The internal validity was estimated using bootstrap techniques. Risk models were represented as risk score diagrams by gender for the age intervals 2-4 y and 2-6 y.

In addition to the BMI SDS at certain ages, the previous BMI SDS during childhood is positively related to adult weight. Receiver Operating Curves analysis provides insight into sensible cut-offs (AUC varied from 0.76 to 0.83). The sensitivity and specificity for 2-6 y at the cut-off of 0.25 and 0.5 are respectively, 0.76 and 0.74, and 0.36 and 0.93, whereas the PPV is 0.52 and 0.67, respectively.

The risk score diagrams can serve as a tool for young children for primary prevention of adult overweight. To avoid wrongly designating children at risk for AO, we propose a cut-off with a high specificity at the risk of approximately 0.5. After external validation, wider adoption of this tool might enhance primary AO prevention.

**0**Bookmarks

**·**

**97**Views

- [Show abstract] [Hide abstract]

**ABSTRACT:**In het Terneuzen Onderzoek naar Preventie zijn overgewicht en cardiometabole risicofactoren op jongvolwassen leeftijd bestudeerd in relatie tot de veranderingen in body mass index (BMI) tussen de geboorte en 18-jarige leeftijd. Deze studie is gebaseerd op het Terneuzen Geboorte Cohort, waarbij prospectief verkregen gegevens zijn verkregen via de jeugdgezondheidszorg (JGZ) van GGD Zeeland. Bij de analyses is gebruik gemaakt van de brokenstickmethode en van lineaire en logistische regressieanalyses. Het leeftijdsinterval 2-6 jaar was het meest voorspellend voor overgewicht en de meeste cardiometabole uitkomsten op jongvolwassen leeftijd. Deze resultaten vragen speciale aandacht van de JGZ voor een stijgende BMI-standaarddeviatiescore (SDS) gedurende het leeftijdsinterval 2-6 jaar, ook als er nog geen sprake is van overgewicht. Hierbij kunnen predictie-instrumenten nuttig zijn. Het monitoren en stabiliseren van de BMI SDS gedurende dit leeftijdsinterval draagt waarschijnlijk niet alleen bij aan de preventie van overgewicht op jongvolwassen leeftijd, maar ook aan een goede cardiometabole gezondheidstoestand op latere leeftijd.JGZ Tijdschrift voor jeugdgezondheidszorg. 04/2013; 45(2). - SourceAvailable from: PubMed CentralLise Graversen, Thorkild I A Sørensen, Liselotte Petersen, Ulla Sovio, Marika Kaakinen, Annelli Sandbæk, Jaana Laitinen, Anja Taanila, Anneli Pouta, Marjo-Riitta Järvelin, Carsten Obel[Show abstract] [Hide abstract]

**ABSTRACT:**Pre- and perinatal factors and preschool body size may help identify children developing overweight, but these factors might have changed during the development of the obesity epidemic. We aimed to assess the associations between early life risk indicators and overweight at the age of 9 and 15 years at different stages of the obesity epidemic. We used two population-based Northern Finland Birth Cohorts including 4111 children born in 1966 (NFBC1966) and 5414 children born in 1985-1986 (NFBC1986). In both cohorts, we used the same a priori defined prenatal factors, maternal body mass index (BMI), birth weight, infant weight (age 5 months and 1 year), and preschool BMI (age 2-5 years). We used internal references in early childhood to define percentiles of body size (<50, 50-75, 75-90 and >90) and generalized linear models to study the association with overweight, according to the International Obesity Taskforce (IOTF) definitions, at the ages of 9 and 15 years. The prevalence of overweight at the age of 15 was 9% for children born in 1966 and 16% for children born in 1986. However, medians of infant weight and preschool BMI changed little between the cohorts, and we found similar associations between maternal BMI, infant weight, preschool BMI, and later overweight in the two cohorts. At 5 years, children above the 90th percentile had approximately a 12 times higher risk of being overweight at the age of 15 years compared to children below the 50th percentile in both cohorts. The associations between early body size and adolescent overweight showed remarkable stability, despite the increase in prevalence of overweight over the 20 years between the cohorts. Using consequently defined internal percentiles may be a valuable tool in clinical practice.PLoS ONE 01/2014; 9(4):e95314. · 3.53 Impact Factor - [Show abstract] [Hide abstract]

**ABSTRACT:**OBJECTIVE: To assess whether waist-to-height-ratio (WHtR) is a better estimate of body fat percentage (BF%) and a better indicator of cardiometabolic risk factors than BMI or waist circumference (WC) in young children. METHODS: WHtR, WC and BMI were measured by trained staff according to standardized procedures. (2)H2O and (2)H2(18)O isotope dilution were used to assess BF% in 61 children (3-7 years) from the general population, and bioelectrical impedance (Horlick equation) was used to assess BF% in 75 overweight/obese children (3-5 years). Cardiometabolic risk factors, including diastolic and systolic blood pressure, HOMA2-IR, leptin, adiponectin, triglycerides, total cholesterol, HDL- and LDL-cholesterol, TNFα and IL-6 were determined in the overweight/obese children. RESULTS: In the children from the general population, after adjustments for age and gender, BMI had the highest explained variance for BF% compared to WC and WHtR (R(2) = 0.32, 0.31 and 0.23, respectively). In the overweight/obese children, BMI and WC had a higher explained variance for BF% compared to WHtR (R(2) = 0.68, 0.70 and 0.50, respectively). In the overweight/obese children, WHtR, WC and BMI were all significantly positively correlated with systolic blood pressure (r = 0.23, 0.30, 0.36, respectively), HOMA2-IR (r = 0.53, 0.62, 0.63, respectively), leptin (r = 0.70, 0.77, 0.78, respectively) and triglycerides (r = 0.33, 0.36, 0.24, respectively), but not consistently with other parameters. CONCLUSION: In young children, WHtR is not superior to WC or BMI in estimating BF%, nor is WHtR better correlated with cardiometabolic risk factors than WC or BMI in overweight/obese children. These data do not support the use of WHtR in young children.Clinical nutrition (Edinburgh, Scotland) 05/2013; · 3.27 Impact Factor

Page 1

C orrespondence: Marlou L. A. de Kroon, Department of Public and Occupational Health, Institute for Research in Extramural Medicine (room no. C574),

VU University Medical Centre, Van der Boechorststraat 7, 1081 BT Amsterdam, the Netherlands. Fax 31 204 448 387. E-mail: top@fms.demon.nl

(Received 24 December 2009; Accepted 29 August 2010 )

ORIGINAL ARTICLE

Identifying young children without overweight at high risk for

adult overweight: The Terneuzen Birth Cohort

MARLOU L. A. DE KROON 1 , CARRY M. RENDERS 2 , JACOBUS P. VAN WOUWE 3 ,

REMY A. HIRASING 1 & STEF VAN BUUREN 3,4

1 Department of Public and Occupational Health, EMGO-Institute for Health and Care Research, VU University

Medical Centre, Amsterdam, the Netherlands, 2 Section of Prevention and Public Health, Department of Health

Sciences and EMGO Institute for Health and Care Research, VU University Amsterdam, Amsterdam, 3 Netherlands

Organisation for Applied Scientifi c Research, TNO Quality of Life, Prevention and Health Care, Leiden, the

Netherlands, 4 Dept of Methodology and Statistics, FSS, University of Utrecht, the Netherlands

Abstract

Objective. To develop a tool to identify children with high risk of adult overweight (AO), especially before developing

overweight, based on body mass index (BMI) standard deviation score(s) (SDS) changes between 2 – 6 years (y) of age.

Methods. We fi tted a linear spline model to BMI SDS of 762 young Caucasian adults from the Terneuzen Birth Cohort at

fi xed ages between birth and 18 y. By linear regression analysis, we assessed the increase in explained variance of the adult

BMI SDS by adding the BMI SDS at 2 y to the models including the BMI SDS at 4 y, 6 y and both 4 y and 6 y. AO risk

was modelled by logistic regression. The internal validity was estimated using bootstrap techniques. Risk models were

represented as risk score diagrams by gender for the age intervals 2 – 4 y and 2 – 6 y. Results. In addition to the BMI SDS at

certain ages, the previous BMI SDS during childhood is positively related to adult weight. Receiver Operating Curves

analysis provides insight into sensible cut-offs (AUC varied from 0.76 to 0.83). The sensitivity and specifi city for 2 – 6 y at

the cut-off of 0.25 and 0.5 are respectively, 0.76 and 0.74, and 0.36 and 0.93, whereas the PPV is 0.52 and 0.67, respec-

tively. Conclusions. The risk score diagrams can serve as a tool for young children for primary prevention of adult overweight.

To avoid wrongly designating children at risk for AO, we propose a cut-off with a high specifi city at the risk of approximately

0.5. After external validation, wider adoption of this tool might enhance primary AO prevention.

Key words: Adult overweight risk , birth cohort , body mass index standard deviation scores (BMI SDS) , childhood , prediction tool

Introduction

Overweight and obesity cause serious health

hazards (1,2), especially if obesity develops during

childhood and is sustained into adulthood (3 – 6). In

young adulthood, not only obesity (Body mass index

[BMI] ? 30), but also overweight (BMI ? 25) is

associated with a considerable increase in cardiovas-

cular risk (2). The increasing prevalences of over-

weight and the signifi cantly increased risk for adult

overweight in overweight children (7) underline the

need for effective prevention programmes. Therefore

much attention has been paid to identifying and

treating children with overweight. However, the

results of treatment for overweight and obesity are

disappointing, especially in the long term. Conse-

quently, today’s challenge for Youth Health Care

(YHC) is not only to reduce overweight and obesity

in childhood, but especially to identify non-over-

weight children at high risk for developing adult over-

weight (AO), including obesity, and to offer them

primary prevention. It makes sense to consider not

only the actual BMI status, but also the change in

BMI level, especially in non-overweight children, as

this change is an additional risk factor for later over-

weight (8 – 10). To enable YHC workers to offer tar-

geted primary prevention to normal-weight children

with a high AO risk, a tool to assess this risk is needed.

However, no such tool has been developed. Others

International Journal of Pediatric Obesity, 2011; 6: e187–e195

ISSN Print 1747-7166 ISSN Online 1747-7174 © 2011 Informa Healthcare

DOI: 10.3109/17477166.2010.526220

Int J Pediatr Obes Downloaded from informahealthcare.com by University Library Utrecht on 12/22/11

For personal use only.

Page 2

e188 M. L. A. de Kroon et al.

have shown that from the age of 2 years (y) onwards

abnormally high weight gain is associated with the

risk of later obesity, also in normal weight children

(11 – 16). Because overweight at the age of 6 y often

translates into overweight in adulthood (17), primary

prevention especially before this age seems worth-

while. Moreover, at a young age lifestyle and risk

factors of overweight and obesity are easier to mod-

ify (18). In a previous study we have shown that the

age interval 2 – 6 y is very sensitive in predicting

overweight (19). The aim of our current study is to

develop a tool enabling the identifi cation of young

children at high risk of adult overweight, based on

the BMI changes between 2 and 6 years of age.

Research design and methods

Population and setting

We analyzed the data of weight and length of 762

Caucasians from the Terneuzen Birth Cohort from

birth until young adulthood. The original cohort con-

sists of all 2 604 Caucasian children born between

1977 and 1986 in the city of Terneuzen. Data for

weight and length as routinely registered by the

Municipal Health Services were available from birth

for 1 701 subjects. Of these subjects, 762 persons

(45%) were willing to participate in a follow-up study

in 2004 – 2005, when they were between 18 and 28

years of age. This follow-up study included measure-

ments of weight and height and a questionnaire to

collect socio-demographic characteristics, which is

described in more detail elsewhere (2). The partici-

pants in the follow-up study did not differ from the

original cohort regarding baseline characteristics, i.e.,

age, birth weight, BMI standard deviation score (SDS)

at birth, and parity and age of the mother, except for

gender (41% males vs. 51% in the original cohort p

? 0.05). We used BMI values (kg/m 2 ) as the measure

for (over)weight, converted to age-specifi c standard

deviation scores (BMI SDS) based on Dutch refer-

ence data (20), because these are most comparable to

our study population. The criterion for being over-

weight in young adulthood is defi ned as BMI ? 25.

The study protocol was approved by the Medical

Ethics Committee of the VU University Medical

Centre Amsterdam, and written informed consent

was obtained from all participants.

Statistical analyses

We fi tted the so-called ‘ broken stick ’ model (21)

to BMI SDS at fi xed ages between birth and 18 y

(n ? 762), which approximates the observed BMI

SDS trajectory of each individual by a series of

straight lines that connect to each other at fi xed ages

(21). Multiple linear regression analysis was applied

to assess the proportion of explained variance of the

BMI SDS at young adulthood by adding the BMI

SDS at 2 y to the models that include the BMI SDS

at 6 y, the BMI SDS at 4 y and the BMI SDS at both

ages 6 y and 4 y, respectively. Gender and age were

analyzed as possible explanatory variables. Gender

was analyzed as a potential confounder. Risk of AO

was modeled by logistic regression. To test for inter-

nal validity, model optimism on the proportion of

explained variance, R 2 , was estimated by the boot-

strap procedure, as given by Steyerberg (22), using

1 000 bootstrap samples. In Addendum 1 the statis-

tical methods are explained further . Risk models for

AO were graphically represented as risk score dia-

grams with contour lines, given BMI SDS at the start

and the end of the age intervals. For convenience, in

the risk score diagrams intended for clinical practice,

the axes are labeled by BMI values instead of BMI

SDS values. Using Receiver Operating Curves (ROC)

analysis we calculated the sensitivity and specifi city

at various cut-off values for the probability of AO.

We used S Plus 8.0 to fi t the ‘ broken stick model ’

and to perform the statistical analyses.

Results

The mean age of the participants was 23.1 years

(Standard deviation [SD] 2.9), 23.2 years for males

(SD 2.9) and 23.0 years (SD 2.9) for females.

The prevalence of overweight (BMI ? 25) in young

adults was 25.1% for males and 28.4% for females

(p ? 0.05). Pearson correlations of BMI SDS at the

ages of 2 y, 4 y and 6 y, with BMI SDS at adulthood

are 0.36, 0.52, and 0.62, respectively (p ? 0.001).

Linear regression analyses

Because gender appeared to be a confounder, but

not an effect-modifi er, males and females could be

analyzed as one group in the multiple regression

analyses (Table I). The proportion of explained vari-

ance in the multiple linear regression model of BMI

SDS at adulthood as a function of BMI SDS at 4 y

increased from 0.28 to 0.34 after extending the

model with BMI SDS at 2 y (p ? 0.001). Likewise,

this proportion increased from 0.39 to 0.47 and from

0.39 to 0.48 by extending the model as a function of

BMI SDS at 6 y with the BMI SDS at 4 y and the

BMI SDS at 2 y, respectively (p ? 0.001). Finally

the proportion of explained variance increased from

0.47 to 0.48 by extending the model as a function

of BMI SDS at 6 y and 4 y with the BMI SDS at

2 y (p ? 0.001), and this proportion remained

almost constant, i.e., 0.48, by extending the model

as a function of BMI SDS at 6 y and 2 y with the

Int J Pediatr Obes Downloaded from informahealthcare.com by University Library Utrecht on 12/22/11

For personal use only.

Page 3

Identifying children at risk for adult overweight e189

BMI SDS at 4 y (p ? 0.001). Therefore, augmenting

the model by a second observation obviously

improved the prediction of BMI SDS at adult age,

whereas the third observation had very little addi-

tional value. The positive value of the regression coef-

fi cient of the BMI SDS in the models including one

BMI SDS increased by adding the BMI SDS at an

earlier age, whereas the regression coeffi cient of the

added BMI SDS became negative. This implies, as

we showed previously (19), that an increase of BMI

SDS in the age intervals is correlated with a higher

BMI SDS at adulthood, and a decrease with a lower

BMI SDS at adulthood.

Logistic regression analyses

Four logistic regression models were fi tted. The

models incorporate respectively the BMI SDS at 4 y

and 2 y, 6 y and 4 y, 6 y and 2 y, and fi nally, 6 y,

4 y and 2 y. All models except the last one predict

signifi cantly better by adding the last mentioned

BMI SDS to the model (p ? 0.05). Because the

last model was of no surplus value in predicting

AO in comparison to the second and third model,

this model was not elaborated further. Based on the

prediction models, it is possible to calculate the AO

risk by hand, using the equations of Cole et al. (23),

the LMS parameters of the Dutch reference standard

of BMI (20) (Table II) and the results of the logistic

regression models (Table III). An example of such a

calculation is elaborated on in Addendum 2. As

shown in this example, it appears that, despite the

fact that this boy has a normal BMI at age 6 y, his

AO risk is substantial considering the prevalence

of overweight of young adult males in this cohort.

Similar calculations apply to other pairs of BMI

values observed at ages 2 y, 4 y and 6 y. Model

optimism of the logistic regression models, as calcu-

lated by the procedure of Steyerberg (22), was small:

the estimates were all lower than 0.01, so the expected

R 2 in a similar, but new, sample will achieve almost

the same value as the reported R 2 .

The risk score diagram and the BMI for

age diagram

How are these models related to the conventional

BMI diagram? Figure 1a plots the trajectories of fi ve

hypothetical children A – E on the Dutch BMI for age

diagram. Child A is at low risk and child E at high

risk. However, it is not clear how we should distin-

guish between children B, C and D, who have exactly

the same BMI at the age of 6 years. Figure 1b graphs

the trajectories for the same children on our risk

score diagram. Because the mean age of the cohort

is 23.1 years, the risk score diagrams have been

developed for 23 years of age. The risk score diagram

in this example contains fi ve contour lines, which

correspond to 10%, 25%, 50%, 75% and 90% risk

values for AO at various combinations of BMI SDS

at 2 years and BMI SDS at 6 years. The line through

the origin (angle of 45 degrees) consists of all com-

binations for which the change between the BMI

SDS at these two ages equals zero. Children A, C

and E are located on this line as their BMI SDS at

2 y is identical to the BMI SDS at 6 y. As expected,

child A has the lowest risk of adult overweight and

child E the highest. Children located above the main

diagonal move upwards through the centiles. Child

B increases from -1.0 SD to 0.0 SD (which equals a

rise in BMI from 15.0 to 15.5) and has a much

higher risk of AO than children C or D, although

the BMI (SDS) at the age of 6 years are exactly the

same for children B, C and D. According to their risks,

the children should be ordered as A, D, C, B, and E.

Receiver Operating Curves (ROC) analysis, positive

predictive value (PPV), sensitivity and specifi city

Figure 2a graphs the histogram of AO risk under the

girls ’ model 2 y 6 y. About half of the girls have a

negligible AO risk ( P O ? 0.1). In YHC practice, it

is useful to set a cut-off value π on AO risk such that

Table I. Prediction of BMI SDS at young adulthood by BMI SDS

at one, two and three ages at childhood, adjusted for gender in

models by multiple regression analysis: regression coeffi cients and

adjusted R 2 (N ? 761).

Prediction

model

Independent

variables β (Standard error) Adj R 2

1

2

3

4

BMI SDS at 2 y

BMI SDS at 4 y

BMI SDS at 6 y

BMI SDS at 2 y

BMI SDS at 4 y

BMI SDS at 4 y

BMI SDS at 6 y

BMI SDS at 2 y

BMI SDS at 6 y

BMI SDS at 2 y

BMI SDS at 4 y

BMI SDS at 6 y

0.54 (0.05) ∗

0.91 (0.06)∗

1.07 (0.05)∗

−0.85 (0.10)∗

1.79 (0.12)∗

−1.75 (0.17)∗

2.72 (0.17)∗

−0.46 (0.08)∗

1.47 (0.07)∗

0.57 (0.14)∗

−3.05 (0.34)∗

3.45 (0.24)∗

0.14

0.28

0.39

0.34

50.47

6

0.48

7

0.48

All models are adjusted for gender and age.

∗ p ? 0.001.

Table II. The Dutch reference for body mass index at the ages of

2, 4 and 6 years (20).

Age

(years)

Boys Girls

m s l m s l

2

4

6

16.42

15.61

15.52

0.0790

0.0882

0.0967

−0.007

−0.375

−1.324

16.07

15.51

15.47

0.0785

0.0865

0.1024

−0.815

−1.416

−1.663

Int J Pediatr Obes Downloaded from informahealthcare.com by University Library Utrecht on 12/22/11

For personal use only.

Page 4

e190 M. L. A. de Kroon et al.

all children with P O ? π are eligible for intervention.

A nice property of such a rule is that the PPV of the

group of children P O ? π is equal to π . Thus if

we set π ? 0.5 and refer those with P O ? π , we

expect that at least half of this group will be over-

weight as an adult. Figure 2b shows how the actual

AO prevalence in the eligible group depends on the

cut off π . At π ? 0 the AO prevalence in the eligible

group is equal to the prevalence of overweight at

young adulthood. Increasing π leads to a progres-

sively higher AO proportion in this group, until the

remaining group becomes so extreme (at π ? 0.82)

that all members fall into the AO group. Occasional

drops in AO prevalence occur at π values where

many subjects with AO are placed. Changing π also

affects the sensitivity and specifi city of the rule. Fig-

ure 3 plots ROC under models 2 y 6 y and 2 y 4 y.

Model 2 y 6 y is more informative than model 2 y

4 y, i.e., at the same specifi city; model 2 y 4 y has a

lower sensitivity than model 2 y 6 y. The AUC for

the models 2 y 4 y and 2 y 6 y was 0.79 (95% CI:

0.73 – 0.85) and 0.83 (95% CI: 0.78 – 0.88), respec-

tively, for boys, and 0.76 (95% CI: 0.71 – 0.81) and

0.79 (95% CI: 0.75 – 0.84), respectively, for girls. On

the basis of the ROC analyses, the cut-off values for

AO risk should be chosen around 0.25. In clinical

practice this means that we single out those children

with a risk of AO of 0.25 and higher, and subse-

quently offer them targeted preventive interventions.

In Table IV, the PPV, the sensitivity and specifi city

of the models are given for different cut-offs on AO

risk. At a rising cut-off the PPV rises, the sensitivity

decreases and the specifi city rises. The percentage

(%) of false-positive children can be derived from

this Table by calculating ‘ 1-specifi city ’ , e.g., at a cut-

off of 0.25 the % false positive children varies from

26 to 29%, whereas at a cut-off of 0.50, these values

vary from 7 to 8%.

The risk score diagrams and general practice

Figures 4 and 5 contain the risk score diagrams for

males and females for the age intervals 2 – 6 y and 2 – 4

y, respectively, which make it easy to identify children

at high risk of AO. The risk score diagrams for the

age interval 4 – 6 y is not given as its practical value

seems less obvious. For practical purposes the four

risk score diagrams that can be used to estimate AO

risk are expressed as a function of BMI instead of

BMI SDS. The risk of AO can be read from the

contour lines of these diagrams, and is based on the

BMI at two ages, of which the BMI at the start of

the interval is given by the value on the X-axis, and

the end of the interval by the value on the Y-axis. If

the child has the BMI at the age that is given on the

X-axis, an indication of AO risk can be given for the

combination of the BMI on the X-axis and various

values of BMI at the age that will be reached as given

on the Y-axis .

Discussion and conclusion

We developed a tool to identify children with a high

risk of AO and in particular those who are not yet

overweight. The tools consist of several risk score

diagrams, which are all based on two measurements

of the BMI, because including a third did not improve

the performance of the tools. The explained variance

of adult BMI by the BMI development between

2 and 6 years of more than 40% is considerable,

especially taking into account that this age interval

concerns a very early growth period in human life

and the age interval 2 – 6 y only covers 22% of the

age range between 0 – 18 years. The BMI changes in

the age intervals 2 – 4 y and 4 – 6 y contribute equally

to AO risk. We have developed risk score diagrams

and illustrated the use of these diagrams.

Cut-off values

An indication of a normal growth of a child from 2

years onwards can be extracted from the risk score

diagrams. The diagrams show how the BMI should

develop to 4 and 6 years of age, respectively, to secure

a low AO risk. In addition, the diagram for 2 – 4 y

offers a mid-term estimate of AO risk that could be

used to evaluate weight change at the age of 4 y. After

an evaluation with the help of the diagram for 2 – 4 y,

the diagram for 2 – 6 y should be applied to determine

Table III. Parameters of three risk models logit(PO) = a + bageA +bxZa + byZß, where PO stands for probability of adult overweight, bage is

the regression coeffi cient of the variable A, A equals the variable age minus 23, bx and by are the regression coeffi cients, and Za and Zß

stand for body mass index standard deviation scores (BMI SDS) at ages 2 y and 4 y, 2 y and 6 y, and 4 y and 6 y, respectively.

Period

Boys Girls

a

b age b 2 b 4 b 6

a b age b 2 b 4

b 6

2 – 4 y

2 – 6 y

4 – 6 y

−1.26

−1.08

−0.97

0.33

0.34

0.33

−1.89

−1.03

–

3.93

–

−3.71

–

−0.85

−0.67

−0.75

0.07

0.08

0.08

−1.34

−3.02

–

2.90

–

−0.73

–

3.40

6.02

4.78

2.52

At the age of 23 years, A ? 0, so logit( P O ) ? a ? b x Z a ? b y Z ß .

Int J Pediatr Obes Downloaded from informahealthcare.com by University Library Utrecht on 12/22/11

For personal use only.

Page 5

Identifying children at risk for adult overweight e191

if the BMI development of the child is normal or

whether it should be adjusted.

The ROC plots of the risk score diagrams suggest

cut-off values for the risk at approximately 0.25. At

this cut-off about 30% of the children that did not

develop AO are wrongly designated as ‘ high risk ’ .

Therefore the choice of a cut-off at 50% seems more

sensible because this is associated with only 8% of

false positive results. At the cut-off around 0.5, we

fi nd that the PPV is 67% of the 2 – 6-year-old children

with an estimated overweight risk of ? 0.5. Another

important consideration in deciding to offer preven-

tive intervention is its cost-effectiveness.

Context of the study results

The prevalence of adult overweight (BMI ? 25) in

the Netherlands is still rising: in 2004 it was 51%

and 42% for adult males and females, respectively.

In addition the prevalences are higher in later birth

cohorts and tend to evolve into obesity at older ages

(24). Therefore primary prevention of AO is very

important in lowering these fi gures. In addition to

interventions targeting the total population of chil-

dren (universal prevention) it will be particularly

effi cient to identify children at high risk for develop-

ing overweight. Therefore tools are needed that can

1234567

14

16

18

20

Age (years)

BMI

ab

−2.5

−2

−1

0

1

2

2.5

●●●●

●●

●●●●

●●

● ● ●● ● ●

●●

A

B

C

D

E

BMI at 2 years

BMI at 6 years

10

25

50

75

90

1314 15161718 1920

13

14

15

16

17

18

19

20

21

−3 −2 −10123

−3

−2

−1

0

1

2

3

●●

A

●●

C

●●

E

●●

B

●●

D

Figure 1. Five body mass index (BMI) trajectories (A – E) plotted on the conventional diagram (a) and the risk score diagram (b).

Adult overweight risk

Frequency

0.0

ab

0.2 0.40.60.81.0

0

10

20

30

40

50

60

0.0 0.20.40.60.81.0

0.0

0.2

0.4

0.6

0.8

1.0

Cut−off value

AO prevalence in those above the cutoff

Figure 2. (a) Histogram of frequency of girls (Y-axis) as a function of the risk of adult overweight (AO) under the model 2 y 6 y (X-axis),

and (b) the prevalence of adult overweight (Y-axis) as a function of the cut-off value (X-axis).

Int J Pediatr Obes Downloaded from informahealthcare.com by University Library Utrecht on 12/22/11

For personal use only.

Page 6

e192 M. L. A. de Kroon et al.

be easily incorporated within preventive health care.

We developed this tool, which is aimed at the age

interval 2 – 6 y, just before the AR, which is known to

be crucial for developing overweight (15,25).

Several studies have assessed the relationship

between a relatively fast BMI increase (or upwards

centile crossing) between 2 to 5 or 6 years, and adult

overweight or obesity (11,15,25,26). One of these

studies also constructed risk charts based on serial

BMI SDS in a non-Caucasian cohort (26). More-

over, these charts are meant to identify children at

risk of the metabolic syndrome and diabetes.

Strengths and limitations

A methodological diffi culty of our study is that

we had to deal with missing values, which can

cause the individual broken stick models to shrink

further towards the overall mean. Therefore, any tests

of differences will be conservative, and possibly

underestimate the effects of BMI changes in age

intervals in which fewer measurements are recorded.

Another limitation was that as in most cohort studies

there was a substantial loss to follow-up (10). There-

fore sampling bias might be possible. However, there

is no reason to assume that the loss to follow-up is

related to the strength of the relationship between

BMI changes in childhood and adult BMI. More-

over, no signifi cant differences were found for the

baseline characteristics for males and females between

those that participated in the follow-up study and the

original cohort.

We should be aware that no data on the represen-

tativeness of well-known risk factors for overweight,

such as socio-economic status, parental weight status

and parenting, were available. It is not clear if and

how these risk factors infl uence the performance of

the tool. The study population of Terneuzen differs

slightly from the total Dutch population regarding

e.g., the prevalence of overweight, which was higher

in the Terneuzen cohort than in 15 – 25-year-olds in

the general Dutch population in 2006 (27.0 vs.

20.4%) (27), although this difference might be largely

due to the age distribution. Therefore cohort effects

cannot be excluded.

Because of the above mentioned limitations, the

tool should be validated in younger cohorts, before

implementing the tool in YHC. This will improve its

generalisibility. Beyond validation, adaptations of the

tool to other ethnicities or other possible risk factors

0.00.2 0.4

1 − Specificity

0.60.81.0

0.0

0.2

0.4

0.6

0.8

1.0

Boys

ab

Sensitivity

● ●

0.1

● ●

●●

●●

●●

●●

●●

●●

●●

0.25

0.5

0.75

0.9

● ●

● ●

2y6y

2y4y

● ●

● ●

2y6y

2y4y

0.0 0.20.4

1 − Specificity

0.6 0.8 1.0

0.0

0.2

0.4

0.6

0.8

1.0

Girls

Sensitivity

●●

0.1

●●

●●

●●

●●

●●

●●

●●

●●

●●

0.25

0.5

0.75

0.9

Figure 3. ROC plots of models 2y6y and 2y4y, including the risk of AO at several points. The AUC was respectively 0.83 (95%CI

0.78–0.88) and 0.79 (95%CI 0.73–0.85) for boys (fi gure a), and respectively 0.80 (95%CI 0.75–0.84) and 0.76 (95%CI 0.71–0.81) for

girls (fi gure b).

Table IV. The positive predictive value (PPV), sensitivity and

specifi city of the three risk models 2 y 6 y, 2 y 4 y and 4 y 6 y at

23 y of age at three different cut-offs.

Cut-offs 25% 50% 75%

PPV of model 2 y 4 y

4 y 6 y

2 y 6 y

2 y 4 y

4 y 6 y

2 y 6 y

2 y 4 y

4 y 6 y

2 y 6 y

0.49

0.54

0.52

0.75

0.76

0.76

0.71

0.76

0.74

0.58

0.66

0.67

0.28

0.38

0.36

0.92

0.93

0.93

0.94

0.80

0.86

0.08

0.15

0.15

1.00

0.97

0.99

Sensitivity of model

Specifi city of model

Int J Pediatr Obes Downloaded from informahealthcare.com by University Library Utrecht on 12/22/11

For personal use only.

Page 7

Identifying children at risk for adult overweight e193

might be necessary. It is to be expected that the PPV

of the tool will increase in younger birth cohorts as

the higher prevalences of AO in younger cohorts will

be in favor of the PPV of the tools. Also, we should

realize that BMI at young adulthood possibly under-

estimates ultimate adult obesity (24). However, by

developing a tool aimed at the risk estimation of

overweight (including obesity) at young adulthood,

this tool will probably also predict the more severe

cases of overweight at later adulthood.

A limitation of the risk score diagram as presented

is that it will only work if the children have been

BMI at 2 years

BMI at 4 years

10

50

90

13 1415 1617181920

13

14

15

16

17

18

19

20

ab

BMI at 2 years

BMI at 6 years

10

25

50

75

90

13 1415 1617181920

13

14

15

16

17

18

19

20

21

Figure 4. (a) Risk score diagram for boys measured at ages 2 y and 4 y; (b) Risk score diagram for boys measured at ages 2 y and 6 y.

The risk on adult overweight (AO) at 23 years of age (in %) can be read from the contour lines of these diagrams, and is based on the

body mass index (BMI) at two ages, of which the BMI at the start of the age interval is given by the value on the X-axis, and at the end

by the value on the Y-axis. If the child has approximately the age as given on the X axis, an indication of AO risk can be given for various

values of BMI at the age which will be reached as given on the Y axis.

BMI at 2 years

BMI at 4 years

10

25

50

75

90

13141516 1718 19 20

13

14

15

16

17

18

19

20

21

BMI at 2 years

BMI at 6 years

10

25

50

75

90

131415161718 1920

13

14

15

16

17

18

19

20

21

ab

Figure 5. (a) Risk score diagram for girls measured at ages 2 y and 4 y; (b) Risk score diagram for girls measured at ages 2 y and 6 y.

The risk on adult overweight (AO) at 23 years of age (in %) can be read from the contour lines of these diagrams, and is based on the

body mass index (BMI) at two ages, of which the BMI at the start of the age interval is given by the value on the X-axis, and at the end

by the value on the Y-axis. If the child has approximately the age as given on the X axis, an indication of AO risk can be given for various

values of BMI at the age which will be reached as given on the Y axis.

Int J Pediatr Obes Downloaded from informahealthcare.com by University Library Utrecht on 12/22/11

For personal use only.

Page 8

e194 M. L. A. de Kroon et al.

measured at ages 2 y, 4 y and 6 y. As long as the age

of the measurement does not differ substantially from

the target by no more than 2 – 3 months, the risk score

diagrams will remain valid, especially if the length of

the age intervals remain close to two or four years.

Finally, because BMI SDS refl ects total body

mass and not body fatness, it might be possible that

a relatively high BMI increase during the age interval

2 – 6 years is also due to an increase in muscular

and bone tissue. Therefore future research should

take into account the predictive value of waist cir-

cumference or, less known, neck circumference at

childhood, both strongly related to the risk of car-

diometabolic diseases (11,28,29). However, the BMI

is still the most common measurement used to

estimate body fat. Moreover, several studies have

shown that an early AR, which is the result of upwards

centile crossing of the BMI just before the age of 6

years (30), is caused by a rapid elevation in the depo-

sition of body fat rather than lean tissue mass (25).

The strength of our study is that we have devel-

oped a tool suitable for primary prevention for chil-

dren who are not yet overweight. Two-dimensional

easy-to-use risk score diagrams could be developed,

because adding a third BMI SDS to the model did

not signifi cantly improve the performance of the

model. The accepted defi nition of overweight in

children is based on the cut-off values of the Inter-

national Obesity Task Force (IOTF), centile curves

with variable cut-off values for different ages (31).

However, the risk of AO at the IOTF cut-offs increases

with age. Therefore preventive interventions that are

offered to children with a BMI above the IOTF cut-

off point for overweight may have, depending on age,

quite different implications for future weight. The

advantage of the methodology proposed in this paper

is that it provides an alternative that is directly based

on risk of AO. Because the tools take both the actual

BMI SDS and BMI SDS change into account,

the new approach could lead to different interven-

tions for children of the same age and same BMI.

Relevance and usefulness within the setting

of the Youth Health Care (YHC)

In the Netherlands, the tool might be used within

YHC that reaches more than 90% of all Dutch

infants from birth onwards by a nationwide program

at set ages (32). During the YHC check-ups the

length and weight of each child are measured. Based

on the information in the risk score diagrams (Fig-

ures 4 and 5), parents can be given information and

an indication about the risk of AO, and thereby be

advised about the preferred growth and nutrition of

their child until the ages of 4 y and 6 y. This also

applies to parents of children who are already over-

weight at 2 y or 4 y, so they can be motivated to

modify the family ’ s and children ’ s lifestyle to prevent

AO. Within YHC it might also be considered to use

the tool selectively for those children with a high risk

of overweight, which can already be assessed before

the age of 2 years, e.g., by assessing risk factors, such

as the BMI of the parents, ethnicity, or SES (33 – 35).

Tailored primary prevention programs might be

offered to these high-risk children, aimed at e.g.,

stimulating breastfeeding, daily physical activity, and

eating breakfast, and preventing the watching of tele-

vision and drinking of sweetened beverages.

Conclusion

Our tool can support preventive healthcare profes-

sionals in the early detection of young children at

high AO risk with the aim of deciding as to whether

or not tailored preventive interventions should be

offered. Moreover, the tool can be used as an instru-

ment for primary prevention by informing parents

about the risks of upward centile crossing during the

age interval 2 – 6 y. The feasibility and effectiveness of

the tool in combination with offering tailored preven-

tive interventions should be studied, e.g., in ongoing

trials. After external validation and a positive evalua-

tion of related interventions, a wider adoption of this

tool might enhance primary prevention of overweight

during a very sensitive period in human growth.

Acknowledgements

This study received a grant from the Health Research

and Development Council of the Netherlands

(ZONMw Grants no. 2100.0092). The researchers

are not dependent on the funder. We gratefully thank

all participants for their time and efforts, the assis-

tants for their contribution to the research, the

Municipal Health Services of Terneuzen (GGD Zee-

land) for their support and cooperation, and Guus

A. de Jonge, PhD, professor emeritus, for laying the

foundations of this study in 1977 – 1986.

Declaration of interest: The authors report no

confl icts of interest. The authors alone are respon-

sible for the content and writing of the paper.

References

Deckelbaum RJ, Williams CJ. Childhood obesity: the health

issue. Obes Res. 2001;(Suppl 4):239 – 43S.

De Kroon ML, Renders CM, Kuipers EC et al. Identifying

metabolic syndrome without blood tests in young adults - the

Terneuzen Birth Cohort. Eur J Pub Health. 2008;6:656 – 60.

1.

2.

Int J Pediatr Obes Downloaded from informahealthcare.com by University Library Utrecht on 12/22/11

For personal use only.

Page 9

Identifying children at risk for adult overweight e195

Dietz WH. Overweight in childhood and adolescence. New

Engl J Med. 2004;350:855 – 7.

Singhal A, Lucas A. Early origins of cardiovascular dis-

ease: is there a unifying hypothesis? Lancet. 2004;363:

1642 – 5.

Raitakari OT, Juonala M, Viikari JS. Obesity in childhood

and vascular changes in adulthood: insights into the Cardio-

vascular Risk in Young Finns Study Int J Obes (Lond). 2005;

29 Suppl 2:S101 – 4.

Must A. Does overweight in childhood have an important

impact on adult health? Nutr Rev. 2003;61:139 – 42.

Freedman DS, Khan LK, Serdula MK et al. The relation of

childhood BMI to adult adiposity: the Bogalusa Heart Study.

Pediatrics. 2005;115(1):22 – 7.

Baird J, Fisher D, Lucas P et al. Being big or growing fast:

systematic review of size and growth in infancy and later

obesity. BMJ. 2005;331:929 – 31.

Lawlor DA, Chaturvedi N. Treatment and prevention of

obesity - are there critical periods for intervention? Int J

Epidemiol. 2006;35:3 – 9.

Fall CH, Sachdev HS, Osmond C et al. Adult metabolic

syndrome and impaired glucose tolerance are associated with

different patterns of body mass index gain during infancy;

Data from the New Delhi birth cohort. Diabetes Care .

2008;31:2349 – 56.

McCarthy A, Hughes R, Tilling K et al. Birth weight; post-

natal, infant, and childhood growth; and obesity in young

adulthood: evidence from the Barry Caerphilly Growth

Study. Am J Clin Nutr. 2007;86:907 – 13.

Botton J, Heude B, Maccario J et al. Postnatal weight and

height growth velocities at different ages between birth and

5y and body composition in adolescent boys and girls. Am J

Clin Nutr. 2008;87:1760 – 8.

Sachdev HS, Fall CH, Osmond C et al. Anthropometric

indicators of body composition in young adults: relation to

size at birth and serial measurements of body mass index in

childhood in the New Delhi birth cohort. Am J Clin Nutr.

2005;82:456 – 66.

Wells JC, Hallal PC, Wright A et al. Fetal, infant and child-

hood growth: relationships with body composition in Brazil-

ian boys aged 9 years. Int J Obes (Lond). 2005;29(10):

1192 – 8.

Ekelund U, Ong K, Linn é Y et al. Upward weight percentile

crossing in infancy and early childhood independently

predicts fat mass in young adults: the Stockholm Weight

Development Study (SWEDES). Am J Clin Nutr. 2006;

83:324 – 30.

Lagstr ö m H, Hakanen M, Niinikoski H et al. Growth patterns

and obesity development in overweight or normal-weight

13-year-old adolescents: The STRIP study. Pediatrics. 2008;

122:e876 – 83.

Magarey AM, Daniels LA, Boulton TJ et al. Predicting

obesity in early adulthood from childhood and parental obes-

ity. Int J Obes Relat Metab Disord. 2003;27:505 – 13.

Cheung YB, Machin D, Karlberg J et al. A longitudinal study

of pediatric body mass index values predicted health in

middle age. J Clin Epidemiol. 2004;57:1316 – 22.

3.

4.

5.

6.

7.

8.

9.

10.

11.

12.

13.

14.

15.

16.

17.

18.

De Kroon MLA, Renders CM, Van Wouwe JP et al. The

Terneuzen Birth Cohort: BMI Changes between 2 and 6

Years Correlate Strongest with Adult Overweight. PLos

ONE. 2010;5(2):e9155.

Fredriks AM, van Buuren S, Wit JM et al. Body index meas-

urements in 1996-7 compared with 1980. Arch Dis Child.

2000;82(2):107 – 12.

Ruppert D, Wand MP, Carroll RJ. Semiparametric regres-

sion. Cambridge: Cambridge University Press; 2003.

Steyerberg EW. Clinical Prediction Models. A practical

approach to Development, Validation, and Updating.

Springer, New York; 2009.

Cole TJ, Freeman JV, Preece MA. British 1990 growth refer-

ence centiles for weight, height, body mass index and head

circumference by maximum penalized likelihood . Stat Med.

1998;17:407 – 29.

Schokker DF, Visscher TL, Nooyens AC et al. National

Prevalence of Obesity. Prevalence of overweight and obesity

in the Netherlands. Obes Rev. 2007;8(2):101 – 8.

Taylor RW, Grant AM, Goulding A et al. Early adiposity

rebound: review of papers linking to this to subsequent obes-

ity in children and adults. Curr Opin Clin Nutr Metab Care.

2005;8:607 – 12.

Sachdev HPS, Osmond C, Fall CH et al. Predicting adult

metabolic syndrome from childhood body mass index:

follow-up of the New Delhi Birth Cohort. Arch Dis Child.

2009;94:768 – 74.

http://www.cbs.nl/nl-NL/menu/themas/gezondheid-welzijn/

cijfers/default.htm .

Preis SR, Massaro JM, Hoffman U et al. Neck circumference

as a novel measure of cardiometabolic risk: the framingham

heart study. J Clin Endocrinol Metab. 2010;95:3701 – 10.

Watts K, Bell LM, Byrne SM et al. Waist circumference

predicts cardiovascular risk in young Australian children.

J Paediatr Child Health. 2008;44:709 – 15.

Cole TJ. Children grow and horses race: is the adiposity rebound

a critical period for later obesity? BMC Pediatrics. 2004;4:6.

Cole TJ, Bellizi MC, Flegal KM et al. Establishing a standard

defi nition for child overweight and obesity worldwide:

international survey. BMJ. 2000;320:1240 – 3.

van Wouwe JP, Hirasing RA. Prevention of sudden unex-

pected infant death. Lancet. 2006;367(9507): 277 – 8.

De Wilde JA, Van Dommelen P, Middelkoop BJ et al. Trends

in overweight and obesity prevalence in Dutch, Turkish,

Moroccan and Surinamese South Asian children in the

Netherlands. Arch Dis Child. 2009;94(10):795 – 800.

Fairclough SJ, Boddy LM, Hackett AF et al. Associations

between children ’ s socioeconomic status, weight status, and

sex, with screen-based sedentary behaviours and sport

participation. Int J Pediatr Obes. 2009;4(4):299 – 305.

O ’ Dea JA, Dibley MJ. Obesity increase among low SES Aus-

tralian schoolchildren between 2000 and 2006: time for pre-

ventive interventions to target children from low income

schools? Int J Public Health. 2010;55(3):185 – 92.

Stasinopoulos DM, Rigby RA. Generalized additive models

for location scale and shape (GAMLSS) in R. J Stat Software.

2007;23:1 – 46.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

32.

33.

34.

35.

36.

Supplementary material available online

Addendum 1

Addendum 2

Int J Pediatr Obes Downloaded from informahealthcare.com by University Library Utrecht on 12/22/11

For personal use only.