10126486032902Obesity (Silver Spring)Obesity (Silver Spring)Obesity (Silver Spring, Md.)1930-73811930-739X28245098537398010.1002/oby.21782NIHMS841403ArticleBMI z-scores are a poor indicator of adiposity among 2- to 19-year-olds with very high BMIs, NHANES 1999-2000 to 2013-14FreedmanDavid SPhDaButteNancy FRD, PhDbTaverasElsie MMPH, MDcLundeenElizabeth APhDaBlanckHeidi MPhDaGoodmanAlyson BMDaOgdenCynthia LPhDdDivision of Nutrition, Physical Activity and Obesity, Centers for Disease Control and Prevention, Atlanta, GAChildren's Nutrition Research Center, Baylor College of Medicine, Houston TXDepartment of Pediatrics, MassGeneral Hospital for Children, Boston MANational Center for Health Statistics, Centers for Disease Control and Prevention, Hyattsville, MDAddress correspondence to: David S Freedman: CDC F-77, 4770 Buford Highway, Atlanta GA 30341 - 3724; phone: 770-488-6016, fax: 815-572-8152; dxf1@cdc.gov10120172822017420172882017254739746Users may view, print, copy, and download text and data-mine the content in such documents, for the purposes of academic research, subject always to the full Conditions of use:http://www.nature.com/authors/editorial_policies/license.html#termsObjective
Although the CDC growth charts are widely used, BMIz is known to be uninformative above the 97^{th} percentile. We compared the relations of BMIz and other BMI metrics (%BMI_{p95}, percent of 95^{th} percentile, and ΔBMI_{p95}, BMI minus 95^{th} percentile) to circumferences, skinfolds and fat mass. We were particularly interested in the differences among children with severe obesity (%BMI_{p95} ≥ 120).
Methods
We used data from 30,003 2- to 19-year-olds who were examined from 1999-2000 through 2013-14 in NHANES.
Results
The theoretical maximum BMIz based on the growth charts varied by more than 3-fold across ages. The BMI metrics were strongly intercorrelated, but BMIz was less strongly related to the adiposity measures than were ΔBMI_{p95} and %BMI_{p95}. Among children with severe obesity, circumferences and triceps skinfold showed almost no association with BMIz (r ≤ 0.10), whereas associations with %BMI_{p95} and ΔBMI_{p95} ranged from r=0.32 to 0.79. Corresponding associations with fat mass ÷ height^{2} ranged from r=0.40 (BMIz) to r=0.82 (%BMI_{p95}) among 8- to 19-year-olds.
Conclusions
Among children with severe obesity, BMIz is only weakly associated with other measures of body fatness. Very high BMIs should be expressed relative to the CDC 95^{th} percentile, particularly in studies that evaluate obesity interventions.
The 2000 Centers for Disease Control and Prevention (CDC) growth charts (1, 2) are widely used to classify obesity (BMI ≥ 95^{th} percentile for a child's sex and age) among 2- to 19-year-olds. In these growth charts, 10 percentiles of BMI between the 3^{rd} and 97^{th} were estimated using various smoothing methods (1, 3). These percentiles were then used to derive L (normality transformation for skewness), M (central tendency) and S (dispersion) parameters that allow for the estimation of z-scores and percentiles for any child (4).
The use of the LMS parameters in the CDC growth charts, however, is known to yield z-scores for very high BMIs that can differ substantially from the estimates based on the data (3, 5). This is, in part, because these parameters in the CDC growth charts were derived from already-smoothed percentiles ≤ 97^{th} (z-score of 1.88), rather than from the underlying data as originally proposed (4). In addition, it is difficult to accurately estimate extreme values. This has led to the use of 120% of the 95^{th} percentile of BMI rather than the LMS-extrapolated 99^{th} percentile, to classify severe obesity (5, 6).
Although the CDC growth charts were constructed as references for clinical care and in the estimation of prevalences, BMI-for-age z-scores (BMIz) have been widely used in all types of analyses, including obesity interventions (7–10). This has occurred despite investigators having emphasized the limitations of very high BMIz values (11–15). A wide range of very high BMIs can map to similar z-scores, and BMIz values in the growth charts have a theoretical maximum (14, 16). Further, when neither the baseline nor follow-up BMIz can be estimated accurately, an examination of ΔBMIz can lead to erroneous conclusions.
The objective of the current study is to describe the relationship of various BMI metrics to other anthropometric indices of obesity (arm and waist circumferences, and triceps skinfold thickness) and to fat mass assessed by dual X-ray absorptiometry (DXA). We were particularly interested in comparing the magnitudes of these associations among children with severe obesity.
MethodsSample and Measurements
We used data from the National Health and Nutrition Examination Survey (NHANES) from 8 cycles conducted from 1999-2000 through 2013-2014 (17). NHANES employs a multi-stage, stratified, cluster sampling design to select a representative sample of the US civilian, non-institutionalized population. The surveys were approved by the ethics review board, and parental permission was obtained for subjects < 18 y of age.
We focus on 2- to 19-year-olds who had weight and height measurements, and after excluding pregnant girls, the resulting sample size was 30,003. Race and ethnicity were self-reported, and subjects were classified as white non-Hispanic, black non-Hispanic, Mexican-American, or other (which includes other Hispanics and multi-racial persons). Weight, height, waist circumference, and mid upper-arm circumference were measured in a standardized fashion (18). Triceps skinfold thickness was measured through the 2009-10 surveys, but not in more recent cycles.
DXA scans were acquired in NHANES 1999-2006 for boys and non-pregnant girls who were ≥ 8 y using a Hologic QDR 4500A fan-beam densitometer (Hologic Inc., Bedford MA) (19, 20). The current analyses focus on fat mass (kg) and fat mass index (fat mass/ht^{2}). The 1999-2000 DXA data for girls are not available in the publicly released data and are not included in the current analyses.
We used the NHANES DXA Multiple Imputation Data Files (19) in the analyses. These imputations, which were performed by the National Center for Health Statistics, used sequential regression to impute (estimate) missing DXA values for 5 complete datasets from non-missing DXA measurements and characteristics such as sex, race-ethnicity, age, BMI and waist circumference (21).
BMI Transformations
Body mass index (BMI) was calculated as kg/m^{2}. BMIz was calculated by expressing a child's BMI relative to children in the CDC growth charts (1). The L (power transformation for skewness) M (median), and S (dispersion) parameters in Cole's LMS method (4) allows for the estimation of a child's z-score with:
Z-score=[(BMIM)L−1]÷(L×S)
Figure 1 show the L, M, and S estimates in the CDC growth charts (http://www.cdc.gov/growthcharts/percentile_data_files.htm). Because L is large and negative, if a child's BMI is very large relative to the median BMI, (BMI ÷ M)^{L} approaches 0 and the maximum BMIz value that is possible (−1) ÷ (L × S). An 8-year-old boy with a BMI of 80 kg/m^{2}, for example, would have a (BMI ÷ M)^{L} of 0.006 and a BMIz of 3.1.
Obesity is defined as a BMI ≥ 95^{th} percentile of the CDC growth charts (1, 2). We refer to a BMI that is expressed as a percentage of the 95th percentile as %BMI_{p95}, and severe obesity as a %BMI_{p95} ≥ 120 (6). Several analyses also focus on children with %BMI_{p95} levels of 100 to <120; we refer to this category as moderate obesity.
We also examined other BMI metrics that account for sex and age. Because of the upper limit of BMIz, CDC proposed using ‘modified’ z-scores to identify extreme values; these were constructed by extrapolating ½ of the distance between 0 and 2 z-scores to more extreme values (16, 22). In addition to %BMI_{p95}, which expresses a child's BMI as a percentage of the (sex- and age-specific) 95^{th} percentile, we also examine ΔBMI_{p95} (BMI – 95^{th} percentile) which is the distance (in kg/m^{2}) from the 95^{th} percentile. For example, an 8-year-old boy (95^{th} percentile, 20.5 kg/m^{2}) with a BMI of 25.3 kg/m^{2}, would have a ΔBMI_{p95} of +4.8 kg/m^{2} and a BMI_{p95} of 124%, whereas a similarly aged boy with a BMI of 18 kg/m^{2} would have a ΔBMI_{p95} of -2.5 kg/m^{2} (2.5 units below the 95^{th} percentile).
The CDC SAS program for the Growth Charts (23) calculates the 95^{th} percentile of BMI, %BMI_{p95} and ΔBMI_{p95}, as well as BMIz and the modified z-score for each child.
Statistical Methods
All analyses used the examination sample weights and accounted for the sample design using the survey package in R (24). About 3% to 4% of the 30,003 2- to 19-year-olds in the analyses were missing information on triceps skinfold thickness and waist circumference, and we imputed these missing values using the ‘aregImpute’ function in the Hmisc package (25). Because children who were missing information on these measures tended to have high BMIs, an analysis of only non-missing data would be biased, and multiple imputation replaces these missing values with estimates based on information from correlated variables (26). Predictors in these imputations included sex, race-ethnicity, age, BMI, DXA-calculated fat mass and fat mass index (kg of fat/m^{2}), leg length, circumferences, and survey cycle. One imputed data set was created for each of the 5 DXA data sets, and the 5 sets of results were combined (27).
Descriptive characteristics of various characteristics are contrasted across three categories (non-obese, moderate obesity and severe obesity) of BMI_{p95}. Before examining the relation of the various BMI metrics to levels of the circumferences, skinfolds and fat mass, we adjusted these other characteristics for sex and age by regressing each characteristic on sex, age (modeled with splines) and the sex × age interaction. The residuals from these models, referred to as ‘adjusted levels’, were then used in the correlational analyses. We also examined the correlations between the various BMI metrics and levels of circumferences, skinfolds and DXA measures within categories of sex, age (<12 y vs. ≥ 12 y) and BMI. We used lowess, a non-parametric smoother, to illustrate the relation of several of the examined characteristics to age.
Results
Table 1 shows levels of various characteristics among non-obese children, children with moderate obesity (BMI_{p95}, 100% to 119%), and children with severe obesity. About 11% of the children had a BMI ≥ 97^{th} percentile, and 5% had a %BMI_{p95} ≥ 120%. In the current study, 15 children had a %BMI_{p95} > 200%, with a maximum of 230% (BMI, 63 kg/m^{2}).
Severe obesity tended to be more prevalent among boys than girls (p = 0.04) and among both black and Mexican-American children than among white children (p < 0.0001). Among children with severe obesity, the mean BMIz was 2.5 SDs and the mean ΔBMI_{p95} was 8.9 kg/m^{2}. As compared to children without obesity, those with severe obesity had higher mean levels of both weight-for age and height-for-age, as well as higher levels of the of circumferences, triceps skinfolds and DXA-calculated body fat.
The bottom, right panel of Figure 1 shows the maximum values of BMIz that are theoretically possible at each age, based on the LMS values in the CDC growth charts; these values have been previously published (14). Among boys, the theoretical maximum BMIz is greater than 5 among 2- to 4-year-olds, but decreases to about 3 between ages of 6 and 12 y, and then increases to about 4 at age 18 y. Among girls, the maximum possible BMIz value decreases rapidly from > 11 (age 2 y) to about 3.5 (ages 6 to 15 y); this maximum further decreases to 2.8 at age 19 y.
Figure 2 shows, for the CDC growth charts, BMIz values corresponding to levels of %BMI_{p95} between 100% (the CDC 95^{th} percentile) and 160%. As seen for the maximum possible BMIz (Figure 1), these very high BMIz values also vary by sex and age so that a constant %BMI_{p95} value is associated with substantially different z-scores. Among girls, z-scores become more compressed at older ages, so that z-score differences (between constant %BMI_{p95} levels) become smaller. For example, a 2-year-old girl with a %BMI_{p95} of 140 would have a BMIz of about 4.5, whereas a 17-year-old girl with the same %BMI_{p95} would have a BMIz of 2.3. Among boys, the BMIz differences across %BMI_{p95} levels was more complex, and were most compressed at about age 10 y. A very similar pattern (data not shown) was seen if BMIz values were plotted against values of ΔBMI_{p95} that ranged from +6 to +18 kg/m^{2}.
Table 2 shows correlations among the BMI metrics, circumferences, skinfolds and fat mass among all children (top rows), those with moderate obesity (middle), and those with severe obesity (bottom). Overall, correlations among the BMI metrics were r ≥ 0.90 except for BMIz. For example, %BMI_{p95} (3^{nd} column) was very strongly correlated with modified BMIz (r=0.93) and ΔBMI_{p95} (r=0.98). In contrast, BMIz was less strongly associated with %BMI_{p95} and ΔBMI_{p95} (r=0.81 and 0.87). In addition, BMIz consistently showed weaker (about 0.05 to 0.10 lower) correlations with the other adiposity measures than did both %BMI_{p95} and ΔBMI_{p95}. Correlations with adjusted fat mass, for example, ranged from r=0.81 (BMIz) to 0.93 (%BMI_{p95} and ΔBMI_{p95}.)
Although the magnitudes of the observed correlations were weaker among children with moderate obesity (middle) and those with severe obesity (bottom), due at least in part to the restricted ranges of BMI values, the decreases were most evident for BMIz. Among children with severe obesity, for example, correlations between BMIz and levels of both ΔBMI_{p95} and %BMI_{p95} were of r < 0.5. Further, among children with severe obesity, BMIz showed almost no association with adjusted levels of the two circumferences and triceps skinfold (r ≤ 0.10), and only a moderate (r=0.40) correlation with fat mass index, which was not available among 2- to 7-year-olds. In general, ΔBMI_{p95} showed the strongest associations with the other body size measures among children with either moderate or severe obesity. For example, correlations with waist to height among children with severe obesity were r=0.18 (BMIz), r=0.56 (%BMI_{p95}) and r=0.62 (ΔBMI_{p95}), while the comparable associations among children with moderate obesity ranged from r=0.26 (BMIz) to 0.52 (ΔBMI_{p95}).
We then examined the relation of the various BMI metrics to the other adiposity measures within categories of BMI status and sex (Figure 3). For both boys (upper panels) and girls, associations with the other body size measures were generally stronger for %BMI_{p95} and ΔBMI_{p95} than for BMIz, and the magnitudes of the differences among the correlations were largest for children with severe obesity (right panels). Of the 36 sets of correlations (2 sexes × 3 obesity groups × 6 body size measures), 24 were strongest for ΔBMI_{p95} and 11 were strongest for %BMI_{p95}. Only for levels of fat mass index among non-obese girls, was the correlation with BMIz slightly higher than that with %BMI_{p95} (r = 0.77 vs 0.75). Among children with severe obesity, several of the associations between BMIz and adjusted levels of circumferences and skinfolds were < 0.30. The largest difference in the magnitudes of the 4 BMI metrics among children with severe obesity was seen for adjusted levels of arm circumference among girls, with correlations ranging from r=0.08 (BMIz) to r=0.79 (ΔBMI_{p95}).
The relations of BMIz, %BMI_{p95} and ΔBMI_{p95} to adjusted levels of waist circumference are shown in Figure 4 among children with severe obesity. BMIz (left panels) showed a strong association with adjusted waist circumference among 12- to 19-year-olds, but almost no association with BMIz values above 2.5 among younger children (solid line). In contrast, both %BMI_{p95} and ΔBMI_{p95} showed strong associations with waist circumference over their entire ranges in both age groups.
Discussion
About 6% of 2- to 19-year-olds currently have a BMI that is greater than or equal to 120% of the CDC 95^{th} percentile (i.e., %BMI_{p95} ≥ 120) (28), and there is much interest in the evaluation and treatment of these children (6, 29). Although the report accompanying the CDC growth charts (1) noted that extrapolation outside the 3^{rd} through 97^{th} percentiles (z-scores of ± 1.88) should be interpreted cautiously, BMIz values remain widely used in cross-sectional and longitudinal analyses of children with severe obesity (7, 8, 10, 30, 31). Our results emphasize that BMIz functions poorly as an indicator of adiposity among children with obesity, particularly among those with severe obesity. Among children with severe obesity, BMIz is, in general, much less strongly associated with circumferences, skinfolds and fat mass than are ΔBMI_{p95} and %BMI_{p95}. Because a wide range of very high BMI values can map to essentially the same z-score, which varies by sex and age, differences between children who have different levels of adiposity, as well as longitudinal BMIz changes (32), can be obscured. Many of the limitations of BMIz, particularly when based on the CDC growth charts, have been emphasized in previous studies (11–14), and our results highlight the weak relationship of BMIz to levels of other measures of fatness among children who are severely obese.
Because of the LMS transformation, BMIz values can differ substantially among children who have similar levels of %BMI_{p95}, ΔBMI_{p95} and other measures of adiposity. For example, there were 2 boys (ages 3 and 9 y) in the current study who had a %BMI_{p95} of about 140, but the older boy had a markedly lower BMIz (2.5 vs 4.7). This BMIz difference resulted from the sex and age changes in the L and S parameters in the CDC growth charts which affect z-scores > 1.88 (97^{th} percentile). Previous studies (11–15) have concluded that LMS-based z-scores should not be used to assess BMI changes. Our results indicate that among children with severe obesity, even very large (e.g., >1 SD) BMIz differences may simply reflect differences in sex or age rather than body size.
The attenuation and confounding of very high BMIz values could influence the results of longitudinal studies that include a large proportion of children who have severe obesity. For example, a 17-year-old girl with a BMI of 50 kg/m^{2} would have a BMIz of 2.6 based on the CDC LMS parameters (33). (The theoretical maximum BMIz for this sex/age is 3.1.) If this girl were to gain an additional 29 kg over 2 years, her BMI would increase by about 10 kg/m^{2}, but her BMIz would remain constant. In contrast, there would be large increases in %BMI_{p95} (from 169% to 193%) and ΔBMI_{p95} (from +29 to +38 kg/m^{2}) reflecting her large weight increase.
The influence of the sex/age differences in the L and S parameters on very high BMIz levels may account for some of the conflicting results in the literature. For example, an intervention study (34) of children with severe obesity (mean BMI, 37 kg/m^{2}) reported a small, statistically significant decrease in BMIz (-0.03 SD) along with a statistically significant 1 kg/m^{2} increase in BMI. These problems would be particularly relevant for intervention studies that contain a large proportion of children with severe obesity, and if analyses focused only on ΔBMIz (7, 30, 31, 35), it is possible that the conclusions could be incorrect. Both cross-sectional and longitudinal studies that include a large proportion of children with severe obesity should express BMI levels relative to the CDC 95^{th} percentile.
In the current study, BMIz showed a moderate to strong association with both ΔBMI_{p95} and %BMI_{p95} (r=0.81 to 0.87) among all children, but much weaker associations (r < 0.50) among children with severe obesity. Although the relationships of %BMI_{p95} and ΔBMI_{p95} to the other measures of adiposity were generally similar, there are situations when 1 might be preferred. A 1-unit increase in ΔBMI_{p95} for example, would indicate that a child's BMI increased by 1 kg/m^{2} more than expected based on sex and change in age, and this might be easier to interpret than a change in %BMI_{p95}. However, for comparisons of children across a wide range of ages over which BMI levels vary substantially, %BMI_{p95} might be preferred as a 1 unit BMI difference should likely be interpreted differently among 2-year-olds than among 18-year-olds. It is possible that neither ΔBMI_{p95} nor %BMI_{p95} would be best in all situations, and that both metrics could be investigated. It should also be noted that %BMI_{p95} and ΔBMI_{p95} resemble other BMI metrics that have been proposed, such as sympercents (11, 12, 36) and percent over BMI (12). Additional analyses (data not shown) of sympercents, defined in the current study as the difference, on the natural log scale, between a child's BMI and the 95th percentile of BMI, indicated that this metric was strongly associated (r=0.99) with %BMI_{p95}.
There are several limitations of the current study that should be considered when interpreting our results. Our results concerning the extrapolation of the LMS calculations apply to children with obesity, and particularly those with severe obesity. However, even in the entire sample of children, BMIz was less strongly correlated with the other adiposity measures than were ΔBMI_{p95} and %BMI_{p95} (Table 2). It should also be realized that attenuation of the LMS-estimated z-scores in the CDC growth charts was recognized by CDC, and a set of modified z-score were developed to address this limitation (16). These modified z-scores are based on extrapolating the distance between 0 and 2 SDs to more extreme BMI values and therefore do not have an upper limit. However, the appropriateness of applying this fixed distance to very high BMI values is uncertain, and a somewhat different approach was used in the WHO growth standards (37). It should also be realized that all of the adiposity measures in the current study, including circumferences, triceps skinfold and DXA measurements, become increasingly difficult to accurately assess among children and adults with severe obesity (38, 39). For example, about 27% of all children with severe obesity in the current study had a DXA measurement that was imputed from other characteristics (21). It should be also realized it would have been optimal to have examined differences between the BMI metrics using a gold-standard measure of adiposity rather than with indirect measures (e.g., circumferences) or with DXA measurements that were available for only a subset of the children.
Although LMS-extrapolated z-scores for children who have very high BMI levels are widely used (7, 30, 31, 35), with some investigators specifically recommending the use of ΔBMIz among children with severe obesity (http://stokes.chop.edu/web/zscore), this approach is not optimal. BMIz levels among children with severe obesity (1) can differ substantially from the empirical estimates, (2) have an effective upper limit resulting in the mapping of very high BMIs to similar z-scores, and (3) can differ solely due to differences in sex and age. Investigators using the CDC growth charts to calculate BMIz values for analyses should be aware of these limitations. Although the drawbacks of BMIz would be unlikely to markedly influence the analysis of population-based studies, the results of studies that include a large proportion of children who have severe obesity should be verified using %BMI_{p95} or ΔBMI_{p95}. For studies that include many children with extreme BMI values, including those focused on obesity interventions, the analyses should likely emphasize %BMI_{p95}, ΔBMI_{p95} or another metric based on the 95^{th} percentile rather than BMIz.
NB and ET were supported by funds from the Centers for Disease Control and Prevention (RFA-DP-11-007). ET was supported by funds from the Centers for Disease Control and Prevention (U18DP003370) and by grant K24 DK10589 from the National Institute of Diabetes and Digestive and Kidney Diseases.
Disclosure: The authors declare no conflict of interest
The findings and conclusions in this report do not represent the official position of the Centers for Disease Control.
KuczmarskiRJOgdenCLGuoSS2000 CDC Growth Charts for the United States: methods and developmentOgdenCLFlegalKMChanges in terminology for childhood overweight and obesityFlegalKMColeTJConstruction of LMS parameters for the Centers for Disease Control and Prevention 2000 growth chartsColeTJGreenPJSmoothing reference centile curves: the LMS method and penalized likelihoodFlegalKMWeiROgdenCLFreedmanDSJohnsonCLCurtinLRCharacterizing extreme values of body mass index-for-age by using the 2000 Centers for Disease Control and Prevention growth chartsKellyASBarlowSERaoGSevere obesity in children and adolescents: identification, associated health risks, and treatment approaches: A Scientific Statement from the American Heart AssociationWangYCaiLWuYWhat childhood obesity prevention programmes work? A systematic review and meta-analysisSiwikVKutobRRitenbaughCIntervention in overweight children improves body mass index (BMI) and physical activityKelleyGAKelleyKSPateRREffects of exercise on BMI z-score in overweight and obese children and adolescents: a systematic review with meta-analysisMcCormickEVDickinsonLMHaemerMAKnierimSDHambidgeSJDavidsonAJWhat can providers learn from childhood body mass index trajectories: a study of a large, safety-net clinical populationColeTJFaithMSPietrobelliAHeoMWhat is the best measure of adiposity change in growing children: BMI, BMI%, BMI z-score or BMI centile?PaluchRAEpsteinLHRoemmichJNComparison of methods to evaluate changes in relative body mass index in pediatric weight controlBerkeyCSColditzGAAdiposity in adolescents: change in actual BMI works better than change in BMI z score for longitudinal studiesWooJGUsing body mass index Z-score among severely obese adolescents: a cautionary noteKakinamiLHendersonMChioleroAColeTJParadisGIdentifying the best body mass index metric to assess adiposity change in childrenCenters for Disease Control and PreventionModified z-scores in the CDC growth charts[WWW document]. URL http://www.cdc.gov/nccdphp/dnpa/growthcharts/resources/BIV-cutoffs.pdfCenters for Disease Control and Prevention (CDC)2016National Center for Health Statistics: National Health and Nutrition Examination Survey Data. Questionnaires, Datasets, and Related Documentation[WWW document]. URL http://www.cdc.gov/nchs/nhanes/nhanes_questionnaires.htmNational Center for Health Statistics (NCHS), Center for Health Statistics N, National Center for Health Statistics (NCHS)2014Anthropometry Procedures Manual. National Health and Nutrition Examination Survey (NHANES)[WWW document]. URL http://www.cdc.gov/nchs/data/nhanes/nhanes_11_12/Anthropometry_Procedures_Manual.pdfNational Center for Health Statistics2016The 1999-2006 Dual Energy X-ray Absorptiometry (DXA) Multiple Imputation Data Files and Technical Documentation[WWW document]. URL https://wwwn.cdc.gov/Nchs/Nhanes/Dxx/dxa.aspxNational Center for Health StatisticsNational Health and Nutrition Examination Survey: Technical Documentation for the 1999-2004 Dual Energy X-Ray Absorptiometry (DXA) Multiple Imputation Data files[WWW document]. URL http://www.cdc.gov/nchs/data/nhanes/dxa/dxa_techdoc.pdfSchenkerNBorrudLGBurtVLMultiple imputation of missing dual-energy X-ray absorptiometry data in the National Health and Nutrition Examination SurveyFreedmanDSLawmanHGSkinnerACMcGuireLCAllisonDBOgdenCLValidity of the WHO cutoffs for biologically implausible values of weight, height, and BMI in children and adolescents in NHANES from 1999 through 2012Centers for Disease Control and Prevention (CDC)A SAS Program for the 2000 CDC Growth Charts[WWW document]. URL https://www.cdc.gov/nccdphp/dnpao/growthcharts/resources/sas.htmLumleyTSurvey: Analysis of complex survey samplesHarrellFE2016Hmisc: Harrell Miscellaneous[WWW document]. URL http://cran.r-project.org/web/packages/Hmisc/index.htmlDondersARTvan der HeijdenGJMGStijnenTMoonsKGMDondersARTReview: a gentle introduction to imputation of missing valuesLumleyTmitools: Tools for multiple imputation of missing data[WWW document]. URL https://cran.r-project.org/web/packages/mitools/OgdenCLCarrollMDLawmanHGTrends in obesity prevalence among children and adolescents in the United States, 1988-1994 through 2013-2014GulatiAKKaplanDWDanielsSRClinical tracking of severely obese children: a new growth chartKreierFGencoŞMBoreelMAn individual, community-based treatment for obese children and their families: the solution-focused approachHamplSOdar StoughCPoppert CordtsKBestCBlackburnKDreyer GilletteMLEffectiveness of a hospital-based multidisciplinary pediatric weight management program: Two-year outcomes of PHIT KidsButteNFCaiGColeSAMetabolic and behavioral predictors of weight gain in Hispanic children: the Viva la Familia StudyCenters for Disease Control and Prevention (CDC)Percentile data files with LMS values[WWW document]. URL http://www.cdc.gov/growthcharts/percentile_data_files.htmSkeltonJADeMattiaLGFloresGA pediatric weight management program for high-risk populations: a preliminary analysisBaughcumAEGramlingKEneliISeverely obese preschoolers in a tertiary care obesity program: characteristics and managementColeTJKryakinYVSympercents: symmetric percentage differences on the 100 log(e) scale simplify the presentation of log transformed dataWHO Multicentre Growth Reference Study GroupHarounDWellsJCWilliamsJEFullerNJFewtrellMSLawsonMSComposition of the fat-free mass in obese and nonobese children: matched case-control analysesBrayGAGreenwayFLMolitchMEDahmsWTAtkinsonRLHamiltonKUse of anthropometric measures to assess weight lossAbbreviationsBMI
body mass index
CDC
Centers for Disease Control and Prevention
NHANES
National Health and Nutrition Examination Survey
DXA
dual-energy x-ray absorptiometry
Sex- and age-specific values of L, M, and S in the CDC growth charts. The lower, right panel shows the maximum values of BMIz that are theoretically possible based on the CDC growth charts.
Calculated BMIz values from the CDC growth charts associated with various levels of %BMI_{p95} and with the CDC 95^{th} percentile of BMI.
Relation of BMIz, modified BMIz, %BMI_{p95} and ΔBMI_{p95} to other body size measures, by sex and BMI status. Values of arm circumference, waist circumference, triceps skinfold and fat mass were adjusted for sex and age.
Relation of various BMI metrics to the sex- and age-adjusted levels of waist circumference among children with severe obesity
Descriptive Characteristics among 2- to 19-year-olds in NHANES, 1999-2000 through 2013-14<xref rid="TFN1" ref-type="table-fn">a</xref>
Non-Obese (%BMI_{p95} < 100)
Moderate Obesity (%BMI_{p95} of 100% to 119
Severe obesity (%BMI_{p95} ≥ 120)
N (unweighted)
24,564
3,600
1,839
Prevalence
84%
11%
5%
Boys (%)
51 ± 1%b
53 ± 1%
54 ± 2%
White non-Hispanics (%)
59 ± 1%
53 ± 2%
44 ± 3%
Black non-Hispanics (%)
14 ± 1%
16 ± 1%
23 ± 2%
Mexican-Americans (%)
13 ± 1 %
17 ±1%
19 ± 2%
Age (y)
10.9 ± 0.1
11.4 ± 0.1
12.9 ± 0.1
BMI (kg/m^{2})
18.7 ± 0.04
26.0 ± 0.1
34.1 ± 0.2
BMIz (SD)
0.13 ± 0.01
1.95 ± 0.01
2.52 ± 0.01
Modified BMIz (SD)
-0.03 ± 0.01
1.92 ± 0.01
3.46 ± 0.03
ΔBMI_{p95} (kg/m^{2})
-4.9 ± 0.03
2.0 ± 0.04
8.9 ± 0.1
%BMI_{p95} (%)
80 ± 1
108 ± 1
135 ± 4
Weight-for-age z (SD)
0.13 ± 0.01
1.88 ± 0.01
2.69 ± 0.01
Height-for-age z (SD)
0.07 ± 0.01
0.50 ± 0.03
0.70 ± 0.04
Waist circumference (cm)
65.3 ± 0.1
84.2 ± 0.5
103.4 ± 0.5
Waist/height
0.47 ± 0.001
0.57 ± 0.001
0.67 ± 0.001
Triceps skinfold (mm)
12.1 ± 0.1
21.7 ± 0.2
28.3 ± 0.2
Subscapular skinfold (mm)
9.3 ± 0.1
19.2 ± 0.2
27.0 ± 0.3
% Body fatc
26.4 ± 0.1
37.8 ± 0.2
42.5 ± 0.3
Fat Mass Index (kg/m^{2})c
5.5 ± 0.04
10.8 ± 0.1
15.2 ± 0.2
Fat mass (kg)c
13.8 ± 0.1
28.0 ± 0.4
40.1 ± 0.6
Chi-square tests indicated that the difference in sex across the 3 BMI categories was statistically significant at the 0.05 level and that difference in race-ethnicity were statistically significant at the 0.00001 level. An examination of continuous variables (beginning with age) across the 3 BMI categories indicated that all trends were statistically significant (p < 0.001).
Values are mean ± standard error
DXA measurements were available only for 8 – to 19-year-olds in the 1999-2000 through 2005-2006 cycles. There were 8345, 1340, and 780 children with DXA measurements in the 3 groups based on %BMI_{p95}.
Weighted correlations among the BMI metrics and other adiposity measures, NHANES 1999-2000 through 2013-14.
BMI Metrics
Circumferences
DXA Measuresa
BMIz
Modified BMIz
%BMI_{p95}
ΔBMI_{p95}
Armb
Waistb
Waist / Height
Triceps Skinfoldb
Fat massb
Fat mass index
All subjects (N=30,003)
BMIz
1.00
0.84
0.78
0.76
0.73
0.81
0.78
Modified BMIz
0.97
1.00
0.88
0.83
0.80
0.76
0.88
0.84
%BMI_{p95}
0.87
0.93
1.00
0.93
0.90
0.86
0.81
0.93
0.87
ΔBMI_{p95}
0.81
0.88
0.98
1.00
0.92
0.88
0.83
0.79
0.93
0.84
Moderate Obesity (%BMI_{p95}: 100 – 119)
BMIz
1.00
0.20
0.21
0.26
0.14
0.43
0.26
Modified BMIz
0.99
1.00
0.21
0.22
0.26
0.15
0.42
0.22
%BMI_{p95}
0.79
0.82
1.00
0.52
0.50
0.45
0.36
0.46
0.39
ΔBMI_{p95}
0.66
0.69
0.96
1.00
0.60
0.58
0.52
0.41
0.54
0.48
Severe Obesity (%BMI_{p95} ≥ 120
BMIz
1.00
0.10
0.10
0.18
0.07
0.65
0.40
Modified BMIz
0.87
1.00
0.36
0.30
0.35
0.19
0.66
0.55
%BMI_{p95}
0.48
0.80
1.00
0.70
0.55
0.56
0.32
0.71
0.71
ΔBMI_{p95}
0.30
0.61
0.93
1.00
0.79
0.63
0.62
0.34
0.83
0.82
DXA measurements were available only for 8 – to 19-year-olds in the 1999-2000 through 2005-2006 cycle (N=10,465). Of these subjects, 1340 were moderately obese, and 780 were severely obese.
Levels of arm circumference, waist circumference, triceps skinfold, and DXA-calculated fat mass were sex- and age-adjusted.
Of the 30,000 subjects, 3600 had moderate obesity and 1839 had severe obesity. All children with severe obesity had a LMS-extrapolated BMI percentile > 98^{th}.
What is already known on this subject
BMI-for-age z-scores based on the CDC growth charts are widely used
BMI percentiles and z-scores for very high BMIs based on the LMS parameters in these growth charts can differ substantially from values that are directly estimated from the data. Therefore, 120% of the 95^{th} percentile, rather than the LMS 99^{th} percentile, is used to classify severe obesity
Previous investigators have emphasized the difficulties in interpreting z-scores and changes in these z-scores for children with very high BMIs
What does this study add?
BMI-for-age z-scores among children who have obesity, and particularly those who have severe obesity, show weaker correlations with other measures of adiposity, such as triceps skinfold, circumference and DXA-calculated fat mass than do other BMI metrics that adjust for sex and age.
Rather than using BMI-for-age z-scores for very high BMIs, these BMIs should be expressed relative to the 95^{th} percentile either as the distance (kg/m^{2}) from the sex- and age-specific 95^{th} percentile or as a percentage of the 95^{th} percentile.