Predicting overweight and obesity in young adulthood from childhood body-mass index: comparison of cutoffs derived from longitudinal and cross-sectional data

Abstract

Background

Historically, cutoff points for childhood and adolescent overweight and obesity have been based on population-specific percentiles derived from cross-sectional data. To obtain cutoff points that might better predict overweight and obesity in young adulthood, we examined the association between childhood body-mass index (BMI) and young adulthood BMI status in a longitudinal cohort.

Methods

In this study, we used data from the International Childhood Cardiovascular Cohort (i3C) Consortium (which included seven childhood cohorts from the USA, Australia, and Finland) to establish childhood overweight and obesity cutoff points that best predict BMI status at the age of 18 years. We included 3779 children who were followed up from 1970 onwards, and had at least one childhood BMI measurement between ages 6 years and 17 years and a BMI measurement specifically at age 18 years. We used logistic regression to assess the association between BMI in childhood and young adulthood obesity. We used the area under the receiver operating characteristic curve (AUROC) to assess the ability of fitted models to discriminate between different BMI status groups in young adulthood. The cutoff points were then compared with those defined by the International Obesity Task Force (IOTF), which used cross-sectional data, and tested for sensitivity and specificity in a separate, independent, longitudinal sample (from the Special Turku Coronary Risk Factor Intervention Project [STRIP] study) with BMI measurements available from both childhood and adulthood.

Findings

The cutoff points derived from the longitudinal i3C Consortium data were lower than the IOTF cutoff points. Consequently, a larger proportion of participants in the STRIP study was classified as overweight or obese when using the i3C cutoff points than when using the IOTF cutoff points. Especially for obesity, i3C cutoff points were significantly better at identifying those who would become obese later in life. In the independent sample, the AUROC values for overweight ranged from 0·75 (95% CI 0·70–0·80) to 0·88 (0·84–0·93) for the i3C cutoff points, and the corresponding values for the IOTF cutoff points ranged from 0·69 (0·62–0·75) to 0·87 (0·82–0·92). For obesity, the AUROC values ranged from 0·84 (0·75–0·93) to 0·90 (0·82–0·98) for the i3C cutoff points and 0·57 (0·49–0·66) to 0·76 (0·65–0·88) for IOTF cutoff points.

Interpretation

The childhood BMI cutoff points obtained from the i3C Consortium longitudinal data can better predict risk of overweight and obesity in young adulthood than can standards that are currently used based on cross-sectional data. Such cutoff points should help to more accurately identify children at risk of adult overweight or obesity.

Introduction

The obesity pandemic is a major threat to public health.¹ High body-mass index (BMI) in childhood has been shown to be associated with high BMI in adulthood, and childhood BMI is an important independent predictor of cardiovascular risk factors and cardiovascular morbidity in adulthood.^2–9 For these reasons, accuracy in defining the BMI in children that predicts the risk of obesity in adulthood is important, because children who could benefit from direct interventions would be identified more accurately.

The BMI cutoff points most commonly used for defining childhood overweight and obesity, such as those recommended by the US Centers for Disease Control and Prevention (CDC) and WHO, were developed using percentiles for BMI within the chosen child population to define the thresholds for obesity.^10,11 However, these defined childhood standards have not been examined in relation to adult obesity outcomes, and they might vary depending on the population selected to define them and the time period in which that population was measured.

The International Obesity Task Force (IOTF) addressed this issue by relating the distribution of BMI in children to BMIs known to be associated with risk of obesity in adults in the same population as the children.¹² In 2000, the data used to establish the childhood cutoff points produced by the IOTF were obtained from cross-sectional surveys of 192727 participants aged 0–25 years between 1963 and 1993 from six populations, including population in Brazil, the UK, Hong Kong, the Netherlands, Singapore, and the USA. In the IOTF approach, the proportion of the adult population that was overweight or obese at age 18 years was first estimated using conventional adult BMI thresholds of 25 kg/m² for overweight and 30 kg/m² for obese. The cutoff point for each population was calculated using the lambda-musigma (LMS) method so that each population-specific centile curve corresponded to the percentile of adulthood overweight or obesity in that population. To obtain the single cutoff points, the mean of the population-specific curves was taken.¹² A concern about this method of defining childhood BMI risk for adult obesity is that if it is used in a setting in which a secular change in obesity prevalence is occurring in the populations from which the sample was chosen, the inferences on cutoff points in childhood would not be valid. If the trend in overweight and obesity was increasing, the estimates from the cross-sectional approach would underestimate the proportion of children who would become overweight adults. Furthermore, the method used specifies that the same percentage of children will be at risk in all age groups, which is unlikely to be the case.

An alternative approach to the cross-sectional comparisons used by the IOTF, WHO, and CDC would be to identify cutoff points by examining the association between childhood and adult BMI in the same individuals who had been followed in a cohort as they aged. We aimed to obtain cutoff points that might better predict overweight and obesity in adulthood by examining the association between childhood and adult BMI status in a longitudinal cohort, the International Childhood Cardiovascular Cohort (i3C) Consortium, that includes seven cohorts from three countries, the USA, Finland, and Australia.

Methods

Study design and participants

The i3C Consortium has been described previously.¹³ Briefly, the consortium includes seven large childhood cohorts, five from the USA (the Bogalusa Heart Study [BHS] cohort from Louisiana; the Minnesota cohort; the Iowa Muscatine Study cohort; the National Growth and Health Study [NGHS] Ohio cohort; and the Princeton Lipid Research Study [PLRS] Ohio cohort), one from Finland (the Cardiovascular Risk in Young Finns Study [YFS] cohort), and one from Australia (the Childhood Determinants of Adult Health Study [CDAH] cohort), which collectively recruited more than 40 000 children and adolescents from their respective communities in the 1970s and 1980s for assessment of a variety of cardiometabolic risk factors. The study participants were born between the early 1950s and early 1990s, with the median birth year being 1970. A subset of each of these cohorts has been re-evaluated at least once in adulthood.

For each study, ethical approval was obtained by the appropriate institutional review board. Informed consent was obtained from all parents and adult participants; assent was obtained from participants when they were children or adolescents.

Procedure and outcomes

There were 41086 participants who had childhood BMI data (aged 3–17 years). For the main analysis, to allow a direct comparison between i3C and IOTF cutoff points, the sample size was restricted to 3779 (9·2%) of 41086 participants (table 1) who had at least one BMI measurement between ages 6 years and 17 years and a BMI measurement specifically at age 18 years, the age defining adulthood used by the IOTF. Measurements in early childhood (ages 3–5 years) were excluded from the analyses because of insufficient sample size. We did parallel analyses using an extended young-adulthood age range of 18–20 years (5019 [12·2%] of 41086 participants) and a later young-adulthood age range of 21–29 years (9039 [22·0%] of 41086 participants). Data from all visits for these individuals were used in the analysis.

Table 1:

Number of i3C Consortium participants included in the main analysis, by sex

	BHS (n=878; year of birth 1958–82)		Muscatine (n=536; year of birth 1954–74)		NGHS (n=568; year of birth 1976–78; female participants*)	Minnesota cohorts (n=769; year of birth 1966–89)		PRLS (n=l6;yearof birth 1957–59)		YFS (n=1012; year of birth 1965–74)		Total (n=3779; year of birth 1954–89)
	Male	Female	Male	Female		Male	Female	Male	Female	Male	Female	Male	Female
Age 6 years	19 (13%)	21 (18%)	40 (27%)	24 (20%)	..	38 (26%)	27(23%)	..	..	49 (34%)	48 (40%)	146 (55%)	120 (45%)
Age 7 years	59 (25%)	40 (22%)	3 (1%)	..	..	172 (74%)	138 (78%)	..	..	..	..	234 (57%)	178 (43%)
Age 8 years	34 (9%)	33 (11%)	16 (4%)	9 (3%)	..	327 (87%)	257 (86%)	..	..	..	..	377 (56%)	299 (44%)
Age 9 years	106 (19%)	107 (13%)	52 (9%)	53 (7%)	320(40%)	324 (57%)	250 (31%)	..	..	84 (15%)	76 (9%)	566 (41%)	806 (59%)
Age 10 years	116 (22%)	81 (9%)	112 (21%)	74 (8%)	540 (59%)	297 (57%)	221(24%)	..	..	..	..	525 (36%)	916 (64%)
Age 11 years	66 (17%)	66 (8%)	38 (10%)	33 (4%)	528 (62%)	294 (74%)	223 (26%)	..	..	..	..	398 (32%)	850 (68%)
Age 12 years	157 (16%)	168 (12%)	166 (17%)	137 (10%)	529 (37%)	385(40%)	282 (20%)	..	..	264 (27%)	304 (21%)	972 (41%)	1420 (59%)
Age 13 years	156 (27%)	133 (14%)	17(3%)	14 (1%)	494 (52%)	396 (70%)	306 (32%)	..	..	..	..	569 (38%)	947 (62%)
Age 14 years	149 (21%)	146 (14%)	178 (25%)	169 (16%)	450 (42%)	391 (54%)	299 (28%)	2 (<1%)	1 (<1%)	..	..	720 (40%)	1065(60%)
Age 15 years	216 (21%)	209 (14%)	14 (1%)	28 (2%)	437 (30%)	359 (35%)	272 (19%)	2 (<1%)	1(<1%)	439 (43%)	514 (35%)	1030(41%)	1463 (59%)
Age 16 years	139 (21%)	108 (11%)	181 (28%)	171 (17%)	463 (47%)	330 (50%)	244 (25%)	4(1%)	..	..	..	654(40%)	986 (60%)
Age 17years	146 (33%)	161 (19%)	..	1 (<1%)	452 (54%)	291 (66%)	226 (27%)	3 (1%)	1(<1%)	..	..	440 (34%)	841 (66%)

Data are n (%), where the % is the proportional contribution of each cohort in each of the age-sex specific analyses; on each row, the percentage of males and females equals 100%. The main analysis included participants who had at least one BMI measurement from age 6 years to 17 years and a BMI measurement at age 18 years. Empty cells indicate no data were collected for that age or sex, or both, for that cohort. BHS=Bogalusa Heart Study. NGHS=National Growth and Health Study. PRLS=Princeton Lipid Research Study. YFS=Cardiovascular Risk in Young Finns Study.

^*The NGHS cohort includes only female participants.

The outcome was defined as being overweight (BMI ≥25 kg/m²) or obese (BMI ≥30 kg/m²) at the age of 18 years in the main analyses and correspondingly being overweight or obese between ages 18–20 years and 21–29 years in the additional analyses. Because the participants in the CDAH cohort had their youngest adult measurement at age 26 years and did not have measurements between ages 18 years and 20 years, they were excluded from the main analysis. However, they were included in the analysis for the later young-adulthood age range (ages 21–29 years).

We validated the results in an independent sample using data from the Special Turku Coronary Risk Factor Intervention Project (STRIP) study, which is a longitudinal, prospective, randomised controlled study that investigated the prevention of atherosclerosis.¹⁴ The participants were enrolled in the study at their 5 month visit at Turku well baby clinics. At age 6 months, 1062 (56·5%) of 1880 infants from the eligible age cohort were randomly assigned to an intervention group (n=540), which was followed-up biannually, or a control group (n=522), which was followed-up biannually up to age 7 years and after that on a yearly basis. The intervention, which included dietary counselling and information on physical activity and smoking prevention, continued until the study participants were aged 20 years.¹⁴ For the analyses in this study, we included individuals with the same criteria as for the i3C Consortium sample: adult BMI at age 18 years and at least one BMI measurement in childhood between ages 6 years and 17 years (n=500). The participants in the STRIP study were aged 18 years during 2007–09.

In a longitudinal study that continues for as long as the STRIP study did, loss to follow-up is inevitable. However, the characteristics of the participants who remained in the study and those who discontinued have been compared on several occasions, with no systematic differences found between the groups.^14–16 Additionally, we compared participants who were in the study at the age of 18 years to those who were not and found no differences in sex, BMI, height, weight at baseline (age 0·7 months) or 10 year follow-up visit, or parental socioeconomic status (education and occupation) at the age of 13 months. Intervention was, however, modestly associated with loss to follow-up at age 18 years because only 44% of the participants in the sample at age 18 years were from the intervention group (appendix p 1). However, the intervention was not associated with BMI¹⁵ and thus the results presented here should not be biased by the difference in proportion of those who remained in the study.

Statistical analysis

All analyses were done separately by sex and integer age group for ages 6–17 years. The data from different i3C Consortium cohorts were pooled. For cohort members who had more than one BMI measurement in one age group, we used the first measurement.

We used logistic regression to assess the relationship between childhood BMI and adult overweight and obesity status for each age and sex defined stratum. The linearity in the logit was assessed using fractional polynomials. If a non-linear transformation of BMI was found to best characterise the association between BMI and the logit, the validity of the original logistic regression model was ensured by using suitable transformations of BMI. The calibration of the models was assessed using the Hosmer-Lemeshow goodness-of-fit test.¹⁷

Receiver operating characteristic (ROC) curve analysis was used to estimate preliminary childhood cutoff points for overweight and obesity. The optimal cutoff points were defined from the ROC curve by calculating sensitivity and specificity and deriving Youden’s J index, calculated as follows: sensitivity + specificity − 1.¹⁸ Sensitivity describes the probability of correctly predicting a participant who will be overweight or obese; specificity describes the probability of correctly predicting a participant who will be of normal weight. Youden’s J index summarises the performance of a predictor in terms of a single statistic. The cutoff point was then calculated using the following equation:

$$\text{Cutoff} \text{point}=(\mathrm{logit}-\text{α})/\text{β}$$

in which α is the intercept and β is the slope from the logistic regression model, and logit describes log of the odds in favour of overweight or obesity.¹⁹

The final cutoff points at different ages in childhood and adolescence that best predicted being overweight or obese at age 18 years for men and women were obtained by fitting a locally estimated scatterplot smoothing (LOESS) curve through the series of preliminary cutoff points over age. The smoothing parameter on each loess was obtained by corrected-Akaike information criterion. Using the same methods, we did additional analyses by computing cutoff points for being overweight or obese in early adulthood during the ages 18–20 years, and later in adulthood during ages 21–29 years.

As an additional validation to our cutoff points, we investigated how the results would change if we further adjusted the models for region, with these being Finland (YFS cohort), midwestern USA (Minnesota cohort, NGHS cohort, PRLS cohort, and Muscatine study cohort), and southern USA (BHS cohort). To investigate the effects of excluding one of the cohorts on the cutoff points, we also reanalysed the data by excluding one cohort at a time. To take into account the possible effect of birth year on risk of overweight and obesity, we did additional sensitivity analyses adjusting for year of birth.

To establish which of the two estimated sets of age-specific and sex-specific cutoff points in childhood, IOTF or i3C, might be a more valid estimate risk of adult overweight or obesity, we applied the two sets of cutoff points to an independent cohort dataset, the STRIP study.¹⁴ We classified children in the STRIP study as overweight or obese on the basis of the cutoff points at each age point. To compare their predictive ability, we established the area-under-the-ROC curve (AUROC) for both sets of cutoff points and computed p values for comparing the AUROCs. Additionally, we compared i3C cutoff points to those defined by the CDC and WHO.

We did analyses using SAS 9.4 (SAS Institute Cary, NC, USA) and R version 3.6.1.

Role of the funding source

There was no funding for this study. The corresponding author had full access to all of the data and the final responsibility to submit for publication.

Results

We provide the cutoff points at different ages in childhood and adolescence that best predict being overweight or obese at age 18 years in the pooled i3C Consortium data (table 2). The cutoff points derived from the pooled i3C Consortium data are lower than the IOTF cutoff points for overweight and obesity in both sexes and across all ages considered (figure 1). For the early adulthood responses (ages 18–20 years), the cutoff points were similar to those that were derived from data at age 18 years (appendix pp 9,11,12). When BMI during ages 21–29 years was used, the cutoff points were even lower than when BMI at age 18 years or 18–20 years was used (appendix pp 10–12). Sensitivities, specificities, and Youden’s J indices of each of the cutoff points are shown in the appendix (pp 13,14).

Table 2:

Cutoff points for BMI that predict overweight and obesity at age 18, by sex and age

	Overweight at age 18 years		Obesity at age 18 years
	Male individuals	Female individuals	Male individuals	Female individuals
Age 6 years	16.13	15.68	16.81	16.78
Age 7 years	16.66	16.43	17.55	17.52
Age 8 years	17.22	17.20	18.29	18.29
Age 9 years	17.79	18.01	19.03	19.08
Age 10 years	18.36	18.79	19.78	19.86
Age 11 years	19.08	19.76	20.58	20.99
Age 12 years	19.84	20.76	21.38	22.25
Age 13 years	20.60	21.59	22.28	23.53
Age 14 years	21.43	22.12	23.48	24.60
Age 15 years	22.27	22.61	24.73	25.64
Age 16 years	23.10	23.08	26.00	26.65
Age 17 years	23.94	23.53	27.31	27.63

BMI=body-mass index.

Figure 1: Cutoff points for overweight and obesity in male and female individuals.

BMI=body-mass index. IOTF=International Obesity Task Force. i3C=International Childhood Cardiovascular Cohort.

In additional validation analyses, we adjusted the models for region and found that the cutoff points changed very little. We reanalysed the data by excluding one cohort at a time; removing any individual cohort did not seem to cause major changes to the cutoff points (appendix pp 4–8). In additional sensitivity analyses, adjustment for birth year made little difference to the estimates (appendix pp 23,24).

We present the proportion of the participants of the STRIP study who were classified as overweight or obese on the basis of both sets of cutoff points (figure 2). Given that the cutoff points derived from the longitudinal i3C data are lower than the IOTF cutoff points, a larger percentage of the participants were classified as overweight or obese on the basis of i3C cutoff points.

Figure 2: Proportion of the STRIP study population classified as overweight or obese on the basis of i3C and IOTF cutoff points.

IOTF=International Obesity Task Force. i3C=International Childhood Cardiovascular Cohort. *Since the IOTF cutoff points were higher than the i3C cutoff points, those who were overweight or obese according to the IOTF cutoff points were also considered overweight or obese using the i3C cutoff points.

Based on the AUROCs and p values, the i3C childhood cutoff points have better overall ability to identify individuals who will be overweight or obese in adulthood than the IOTF cutoff points (table 3). For all age points for obesity, and for some age points for overweight, i3C cutoff points are significantly better at predicting weight status in young adulthood than IOTF cutoff points. The i3C cutoff points were also lower and had better predictive ability than the cutoff points recommended by the CDC and WHO (appendix pp 15–22). The AUROC for the i3C cutoff points ranged from 0·75 to 0·88 for overweight and 0·84 to 0·90 for obesity, whereas the corresponding values were 0·69–0·83 and 0·66–0·88 for the CDC cutoff points and 0·72–0·88 and 0·67–0·88 for WHO cutoff points.

Table 3:

The performance of i3C and IOTF BMI cutoff points in the prediction of overweight and obesity in adulthood in the STRIP study population

	Overweight			Obesity
	AUROC of the i3C cutoff points	AUROC of the IOTF cutoff points	p value	AUROC of the i3C cutoff points	AUROC of the IOTF cutoff points	p value
Age 6 years	075 (0.70–0.80)	0.69 (0.62–0.75)	0.048	0.88 (0.83–0.94)	0.60 (0.51–0.70)	<0.0001
Age 7 years	0.73 (0.67–0.79)	0.70 (0.64–0.76)	0.22	0.87 (0.80–0.95)	0.57 (0.49–0.66)	<0.0001
Age 8 years	0.78 (0.72–0.84)	0.73 (0.67–0.79)	0.065	0.84 (0.75–0.93)	0.57 (0.49–0.66)	<0.0001
Age 9 years	0.78 (0.72–0.84)	0.70 (0.64–0.76)	0.0050	0.85 (0.76–0.94)	0.61 (0.51–0.70)	<0.0001
Age 10 years	0.79 (0.73–0.84)	0.76 (0.70–0.82)	0.26	0.86 (0.78–0.93)	0.63 (0.53–0.74)	0.0002
Age 11 years	0.80 (0.75–0.86)	0.76 (0.70–0.82)	0.041	0.85 (0.77–0.94)	0.65 (0.55–0.76)	0.0010
Age 12 years	0.78 (0.73–0.84)	0.76 (0.70–0.82)	0.24	0.86 (0.78–0.95)	0.71 (0.59–0.82)	0.007
Age 13 years	0.81 (0.75–0.86)	0.75 (0.69–0.81)	0.024	0.85 (0.76–0.95)	0.71 (0.59–0.82)	0.012
Age 14 years	0.81 (0.75–0.86)	0.79 (0.73–0.85)	0.41	0.90 (0.82–0.98)	0.74 (0.63–0.86)	0.009
Age 15 years	0.84 (0.79–0.90)	0.80 (0.74–0.86)	0.050	0.89 (0.80–0.98)	0.75 (0.63–0.87)	0.011
Age 16 years	0.87 (0.83–0.92)	0.83 (0.77–0.88)	0.033	0.88 (0.79–0.98)	0.76 (0.65–0.88)	0.022
Age 17 years	0.88 (0.84–0.93)	0.87 (0.82–0.92)	0.41	0.86 (0.76–0.96)	0.73 (0.62–0.85)	0.018

95% CIs are given in parentheses and were calculated using the Wald method. AUROC=area under the receiver operating characteristic curve. i3C=International Childhood Cardiovascular Cohort. IOTF=International Obesity Task Force. BMI=body-mass index. STRIP=Special Turku Coronary Risk Factor Intervention Project.

Discussion

We observed that the overweight and obesity cutoff points derived using pooled longitudinal data from the i3C Consortium are lower than those estimated by the IOTF and identify a larger proportion of the childhood population at risk for adult overweight or obesity. On the basis of the AUROC, the set of childhood age-specific and sex-specific cutoff points derived in the present analyses are able to better predict the risk of adult overweight or obesity than the commonly used IOTF standards, especially for obesity.

The reason for the i3C cutoff points being lower than those estimated from the cross-sectional survey approach used to derive the IOTF standards is important to consider. Secular change in childhood obesity from the time when the 18-year-old individuals in the IOTF surveys were themselves children could account for the difference observed. Using the percentage of participants found to be obese at age 18 years to define the percentage of overweight or obese individuals in concurrent child hood samples gives estimates that are higher than if the data for these people as children had been used. This is because the children in the cross-sectional samples have a higher BMI at the same age than did their predecessors when obesity prevalence is increasing across the population. The authors of the IOTF paper have argued that the period in which their survey samples were collected (1963–93) was before the upward shift in BMI that has occurred in recent decades.²⁰ In fact, a modest upturn in prevalence of overweight and obesity during the 1980s and 1990s in the populations from which their samples came was observed.²¹ Moreover, an increase in the prevalence of factors determining development of obesity from childhood to age 18 years in the i3C Consortium would mean that more children of lower BMI would become obese in adulthood than if there had been no secular change. In both scenarios, the i3C longitudinal data would generally estimate lower BMI cutoff points in childhood than would the IOTF cross-sectional data. Adjustment for birth year in our sensitivity analysis made little difference to the estimates.

It might be suggested that the greater screening validity observed when i3C Consortium cutoff points were applied to data from the STRIP study than when the IOTF cutoff points were used was because both the STRIP study and i3C included participants from Finland. However, this explanation seems unlikely, given that YFS contributed only 9% of the participants in the i3C dataset used for this analysis, with the majority being from the USA.

Based on our cutoff points, the share of children at risk of obesity was higher at younger ages. The models used for defining cutoff points are more precise and have less uncertainty the closer the response and the explanatory variable are temporally. Thus, the most precise of our cutoff points are the ones defined for adolescents in the models with response being overweight or obesity at age 18 years. When the temporal distance between the response and explanatory variable is bigger there is greater uncertainty. In our study, such temporal distance appears to have led to lower cutoff points and higher prevalence of people who are predicted to be overweight or obese. The phenomenon can clearly be seen by comparing the sensitivities, specificities, and Youden’s J indices. Within each BMI outcome in both sexes, the Youden’s J indices increase with the age of measurement of the explanatory variable (ie, the smaller the temporal distance between the exposure and response). Accordingly, the cutoff points defined with age 18 years as a response appear to have overall higher Youden’s J indexes than the cutoff points that were defined with age 18–20 years or 21–29 years as a response. It is inarguable that the cutoff points for childhood overweight and obesity vary depending on the adult age range chosen. This variation is partly explained by the increasing un certainty; however, the prevalence of overweight and obesity in the chosen age range also affects it. Data from the CARDIA study show that the prevalence of obesity increases into the 30s and early 40s.²² Which adult age group might be the best for comparison is not an issue we have addressed here, because our main focus was examining what would be obtained with the cohort approach if it was compared to another major internationally used set of standards that were derived from cross-sectional data. Our choice of age 18 years was to ensure we were able to make comparisons with the IOTF. When we used an older age for our cohort participants for comparison in adulthood (as shown in our analyses using ages 18–20 years or 21–29 years), we found that the estimated cutoff points were lower than the cutoff points established using data from people aged 18 years to identify adult obesity.

A major strength of the present study was its ability to use longitudinal BMI data from childhood and the inclusion of several international cohorts. However, because the present analysis was not planned when these cohorts were established, there are some limitations that need to be noted. Most importantly, not all participants in the i3C Consortium cohorts had BMI data at the age of 18 years and therefore data from a subset of the cohorts were used. Our additional analysis found that the eligible population for the analyses was slightly older and had a higher proportion of females than the original population (appendix p 26). However, given that the analyses were stratified by sex and age, these differences should not have affected the cutoff points. Although we found that the 3779 participants who had a BMI measurement at age 18 years and the 5019 who had a BMI measurement between ages 18–20 years had lower age-adjusted and sex-adjusted baseline BMI than those who were excluded from these samples, the participants who had BMI measured between ages 21–29 years did not differ from those who were excluded in this respect (appendix p 26). An additional limitation of this study is that there were few participants in the i3C Consortium who had been measured specifically at ages 3–5 years and then later in life at age 18 years, and therefore they could not be included in the analyses.

Furthermore, it is important to note that IOTF cutoff points were based on populations from countries of different socioeconomic settings and obesity prevalence (Brazil, Hong Kong, the Netherlands, Singapore, the UK, and the USA), whereas the i3C population consists of cohorts only from developed and industrialised countries (Australia, Finland, and the USA). It will be particularly important in the future to establish whether the assessment of standards derived from predominantly non-Hispanic white populations are generalisable to other racial and ethnic groups. Previous international standards have been developed without an attempt to adjust analyses for socioeconomic status, possibly because of the difficulty in specifying what might be an ideal population base from which to derive an expected socioeconomic status distribution. The i3C database does provide the opportunity for adjusting for socioeconomic status measurement on individuals. However, there were substantial missing data for this variable, and the issue about which base population we might use for com parison remained. Therefore, we felt that we would not advance our understanding about the generalisability of our findings beyond what we had achieved by adjusting for region if we attempted to adjust for socioeconomic status. In future, particularly when standards have been derived from a sample sourced from a single population, investigators might wish to adjust their results for socioeconomic status in the particular reference population.

The majority of the i3C participants were children during 1970–88, and adults approximately 20 years later. IOTF participants were children approximately 10 years earlier on average. Neither cohorts have measurements that make the samples contemporary. If the i3C estimates had been obtained from a more recent sample, they might be even lower than those reported here, given that adult overweight and obesity have increased since the i3C participants were young adults. The cutoff points will need to be revised with each opportunity that presents as time passes. We suggest that the approach using cohort data has advantages over using cross-sectional data and that it results in lower estimated cutoff points, identifying a higher proportion of the child population as being at risk of future overweight or obesity. As more recent cohorts are able to produce estimates in the same way, these cutoff points should be revised, and because of secular trends in adiposity they are likely to be lower than those we have produced.

Our analysis of data from multiple pooled longitudinal cohort studies provides childhood BMI cutoff points for adult overweight and obesity prediction, using data at age 18 years for the adult comparison. Compared with existing IOTF data, the present cutoff points are lower at each childhood age. Using adult data at an older age lowers the childhood cutoff points and would appear to improve prediction of subsequent overweight or obesity in adulthood.

Research in context.

Evidence before this study

Thus far the cutoff points for defining childhood overweight and obesity have been based on cross-sectional data in childhood, from which population-specific percentiles have been derived for children or, alternatively, on inference from concurrent adult data on the proportion of the population that is overweight and the application of this percentile to define the cutoff point for children from that population. The US Centers for Disease Control and Prevention, using population data from five nationally representative health examination surveys in the USA, and WHO, using three population datasets from cross-sectional surveys in the USA, developed cutoff points using percentiles for body-mass index (BMI) in representative child populations to define thresholds for obesity. The International Obesity Task Force obtained cross-sectional data from six populations: Brazil, Hong Kong, the Netherlands, Singapore, the UK, and the USA, and the cutoff point for each population was calculated in such a way that each population-specific percentile curve corresponds to the percentile of adulthood overweight or obesity in that population.

Added value of this study

This study provides new cutoff points using methodologically preferable longitudinal cohort data. We showed that these cutoff points provide better identification of children who will become overweight or obese adults than do the cutoff points previously derived from cross-sectional data.

Implications of all the available evidence

Using the more precisely defined cutoff points from this study will allow intervention efforts to more efficiently identify children and adolescents at risk of becoming overweight or obese in adulthood, compared with the estimates obtained with current standards.