Diagnostic Performance of BMI Percentiles to Identify Adolescents With Metabolic Syndrome. abstract

Transcription

1 Diagnostic Performance of BMI Percentiles to Identify Adolescents With Metabolic Syndrome WHAT S KNOWN ON THIS SUBJECT: The Centers for Disease Control and FITNESSGRAM BMI percentile thresholds are commonly used for obesity screening in youth. It is assumed that these thresholds are predictive of metabolic health risk, but little diagnostic data are available. WHAT THIS STUDY ADDS: Both thresholds are predictive of metabolic syndrome, more so for boys than for girls, although with differing sensitivity and specificity. The diagnostic details of the thresholds can inform clinicians and practitioners about how these standards perform in practice. abstract OBJECTIVES: To compare the diagnostic performance of the Centers for Disease Control and Prevention (CDC) and FITNESSGRAM (FGram) BMI standards for quantifying metabolic risk in youth. METHODS: Adolescents in the NHANES (n = 3385) were measured for anthropometric variables and metabolic risk factors. BMI percentiles were calculated, and youth were categorized by weight status (using CDC and FGram thresholds). Participants were also categorized by presence or absence of metabolic syndrome. The CDC and FGram standards were compared by prevalence of metabolic abnormalities, various diagnostic criteria, and odds of metabolic syndrome. Receiver operating characteristic curves were also created to identify optimal BMI percentiles to detect metabolic syndrome. RESULTS: The prevalence of metabolic syndrome in obese youth was 19% to 35%, compared with,2% in the normal-weight groups. The odds of metabolic syndrome for obese boys and girls were 46 to 67 and 19 to 22 times greater, respectively, than for normal-weight youth. The receiver operating characteristic analyses identified optimal thresholds similar to the CDC standards for boys and the FGram standards for girls. Overall, BMI thresholds were more strongly associated with metabolic syndrome in boys than in girls. CONCLUSIONS: Both the CDC and FGram standards are predictive of metabolic syndrome. The diagnostic utility of the CDC thresholds outperformed the FGram values for boys, whereas FGram standards were slightly better thresholds for girls. The use of a common set of thresholds for school and clinical applications would provide advantages for public health and clinical research and practice. Pediatrics 2014;133:e330 e338 AUTHORS: Kelly R. Laurson, PhD, a Gregory J. Welk, PhD, b and Joey C. Eisenmann, PhD c a School of Kinesiology and Recreation, Illinois State University, Normal, Illinois; b Department of Kinesiology, Iowa State University, Ames, Iowa; and c Department of Radiology, Michigan State University, East Lansing, Michigan KEY WORDS weight status, youth, obesity, overweight, healthy fitness zone ABBREVIATIONS ATP III Adult Treatment Panel III CDC Centers for Disease Control and Prevention FGram FITNESSGRAM HDL high-density lipoprotein NHANES National Health and Nutrition Examination Survey NLR negative likelihood ratio NPV negative predictive value PLR positive likelihood ratio PPV positive predictive value ROC receiver operating characteristic %BF percentage body fat Dr Laurson contributed to the conception and design of the project, analytical design, and interpretation of the data and wrote the draft version of the manuscript; Dr Welk contributed to the conception and design of the project, analytical design, and interpretation of the data and revised the manuscript for intellectual content; Dr Eisenmann contributed to the analytical design and interpretation of the data and revised the manuscript for intellectual content; and all authors approved the final manuscript as submitted. doi: /peds Accepted for publication Nov 18, 2013 Address correspondence to Kelly R. Laurson, PhD, Illinois State University, McCormick Hall, School of Kinesiology and Recreation, Normal, IL klaurso@ilstu.edu PEDIATRICS (ISSN Numbers: Print, ; Online, ). Copyright 2014 by the American Academy of Pediatrics FINANCIAL DISCLOSURE: Dr Welk serves as the Scientific Director of the FITNESSGRAM program and oversees the activities of the Scientific Advisory Board; the other authors have indicated they have no financial relationships relevant to this article to disclose. FUNDING: Supported by The Cooper Institute, Dallas, TX. POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose. e330 LAURSON et al

2 ARTICLE The percentile-based growth charts for BMI from the Centers for Disease Control and Prevention (CDC) are widely used for obesity screening in children and adolescents. 1 The charts are used to categorize youth into weight status groups (ie, normal weight, overweight, and obesity). 2 Alternative standards linking BMI and percentage body fat (% BF) were recently developed in FIT- NESSGRAM (FGram), a commonly used school-based fitness testing program. These standards were empirically derived by using nationally representative data from the National Health and Examination Survey (NHANES). Initially, %BF percentiles were created. 3 Then receiver operating characteristic (ROC) analyses identified %BF percentile thresholds that corresponded with metabolic risk factors. 4 Finally, %BF thresholds were equated to BMI values by using another ROC procedure. 5 Although BMI is unable to discern fat mass from fat-free mass, 6 these standards were linked to health risks on the basis of corresponding risks for body fatness. Although there are advantages of direct assessment of body fatness for screening, use of BMI is very common in schools as well as in clinical practice. Both CDC and FGram thresholds were created by using the same CDC ageand gender-specific BMI percentiles. 2 FGram uses the 83rd and 92nd percentiles for boys and the 80th and 90th percentiles for girls. CDC thresholds are the 85th and 95th percentiles for both genders. Although routinely used, these CDC thresholds were not based on a health criterion but rather on a pre obesity epidemic population distribution. Essentially, the 85th and 95th percentiles equate to a BMI 1 and 2 SDs above the mean, respectively. Whereas population-norm thresholds serve as a preliminary strategy for defining screening classifications, empirical testing is required to determine if youth in the at risk groups are truly at risk. The relative utility compared with current FGram standards has also not been evaluated. Therefore, the purpose of this study was to detail the diagnostic performance of FGram and CDC BMI standards to quantify metabolic risk. Because neither was developed to predict metabolic syndrome, a secondary goal was to complete an independent analysis to determine optimal cut points for detecting the condition. METHODS Sample Participants were 3385 adolescents from NHANES, which is a series of surveys designed to assess the nutrition and health status of noninstitutionalized children and adults. Five waves of NHANES data were combined ( ). Certain metabolic variables (eg, fasting glucose) were available in a subset of participants $12 years old, so the sample was restricted to those aged 12.0 to 18.9 years. Pregnant female participants and participants missing analytic variables were excluded. Because BMI thresholds are applied without regard to race/ethnicity, all participants were represented. The weighted racial/ ethnic breakdown of the participants was as follows: 10.7% Mexican American (53.5% male), 63.4% non-hispanic white (50.8% male), 13.8% non- Hispanic black (51.2% male), 5.7% other Hispanic (51.0% male), and 6.4% other race/multiracial (54.6% male). The NHANES protocols were approved by the research ethics review board of the National Center for Health Statistics. Written informed consent was obtained from all participants. Anthropometric Variables Stature was measured to the nearest 0.1 cm via digital stadiometer (Seca Corporation, Hanover, MD). Body mass was measured to the nearest 0.1 kg via digital scale (Toledo; Columbus, OH). BMI was calculated by using stature and mass (kg/m 2 ). Waist circumference was measured to the nearest 0.1 cm with a steel measuring tape just above the uppermost lateral border of the ilium at the end of a normal expiration. Measurements were taken by trained technicians with regular calibration and quality-control checks. Protocol, procedures, calibration, and quality control are detailed in the NHANES manuals available at gov/nchs/data/nhanes/nhanes_03_04/ BM.pdf. BMI percentiles were calculated, and CDC standards were used to classify children as normal weight (BMI,85th percentile), overweight ($85th and,95th percentile), or obese ($95th percentile). FGram standards were used to classify children as Healthy Fitness Zone (boys:,83rd percentile; girls:,80th percentile), Needs Improvement: Some Risk (boys: $83rd and,92nd percentile; girls: $80th and,90th percentile), or Needs Improvement: High Risk (boys: $92nd percentile; girls: $90th percentile). From this point forward, the 3 groups created by both standards are referred to as normal weight, overweight, and obese to avoid confusion. Metabolic Variables Blood pressure was measured via mercury sphygmomanometer and recorded as the average of 3 or 4 consecutive measurements after sitting and resting quietly for 5 minutes. 7 Highdensity lipoprotein (HDL) cholesterol and fasting triglycerides were analyzed at the Johns Hopkins Lipoprotein Analytical Laboratory, and fasting glucose was analyzed at the University of Missouri-Columbia as detailed in the NHANES laboratory procedures manuals. 8 Waist circumference, blood pressure, triglycerides, HDL cholesterol, and PEDIATRICS Volume 133, Number 2, February 2014 e331

3 fasting glucose were used to determine metabolic syndrome status. The Joliffe and Janssen 9 metabolic syndrome thresholds were used to make all classifications. This definition of metabolic syndrome was generated by using growth curves that are linked to the National Cholesterol Education Program/ Adult Treatment Panel III (ATP III) adult values. 10 Children were considered as having an individual metabolic risk factor if they had a high waist circumference, systolic or diastolic blood pressure, triglycerides, or fasting glucose or low HDL cholesterol. Children with $3 metabolic risk factors were considered as having metabolic syndrome. Using this method, if youth with metabolic syndrome were to maintain their current growth trajectories above these thresholds into adulthood, they would be diagnosed with metabolic syndrome using the adult definition. Statistical Analysis Descriptive statistics were calculated. Multiple analyses were used to investigate the utility of CDC and FGram thresholds to detect metabolic syndrome. First, the prevalence of metabolic risk factors and metabolic syndrome by weight status was calculated, followed by the sensitivity (ability to detect adolescents with metabolic syndrome) and specificity (ability to detect adolescents without metabolic syndrome) of the thresholds. Then, positive predictive values (PPVs; proportion of high-bmi adolescents with metabolic syndrome) and negative predictive values (NPVs; proportion of normal-bmi adolescents without metabolic syndrome) were calculated. Higher values for sensitivity, specificity, PPVs, and PPVs indicate more effective thresholds. Positive likelihood ratios (PLRs; = sensitivity/[1-specificity]) and negative likelihood ratios (NLRs; = [1-sensitivity]/specificity) were then calculated. A higher PLR and lower NLR are preferred. Logistic regression was used to estimate the odds of metabolic syndrome by weight status using CDC and FGram thresholds. A contrast was used to estimate the odds ratio between the overweight and obese groups. Sensitivity and specificity were also calculated for the CDC and FGram thresholds while excluding a portion of the sample on the basis of weight status. First, sensitivity and specificity were calculated for the low-bmi threshold when excluding the obese (CDC) or high-risk (FGram) group. The overweight (CDC) or some-risk (FGram) groups were then excluded to calculate sensitivity and specificity for the high- BMI threshold. In a separate analysis, ROC curves were used to determine optimal thresholds to detect metabolic syndrome. The analysis used presence of metabolic syndrome as the criterion and BMI percentile as the continuous test variable. The analysis evaluated the sensitivity and specificity of each possible BMI percentile to detect metabolic syndrome. The resulting data were used to select thresholds with a desired sensitivity and specificity. An ideal threshold would have a perfect 100% sensitivity and specificity (all cases would be either true positives or true negatives); however, this threshold is rarely possible. Sensitivity and specificity are inversely related and one must usually be sacrificed foragaininthe other. For more detailed descriptions of ROC procedures and interpretation, readers are referred to previous reviews. 11,12 All analyses accounted for the complex sample design of NHANES, using the weighting, stratification, and clustering for estimates in SAS version 9.2 (SAS Institute, Cary, NC) and IBM SPSS version 20 (IBM SPSS Statistics, IBM Corporation, Armonk, NY). RESULTS Descriptive statistics are provided in Table 1. The prevalence of metabolic syndrome was 7%. The prevalence of metabolic risk factors and metabolic syndrome by weight status are shown in Tables 2 and 3. FGram thresholds resulted in a greater prevalence of obesity, with 24.9% of boys and 27.5% of girls above the high threshold compared with 18.0% and 17.1% of boys and girls, respectively, by CDC thresholds. The prevalence of metabolic syndrome in obese children was between 19% and 35% compared with #2% in the normal-weight groups. The prevalence of metabolic risk factors varied by gender and weight status, and usually the lowest prevalence was found in normal-weight children and the highest in obese children. However, this was not consistent for all risk factors. In most instances, the 95% confidence intervals overlapped between CDC and FGram thresholds. Table 4 includes the diagnostic performance data for detecting children with and without metabolic syndrome. Most overweight thresholds tended to have sensitivities approximating 90%. The exception was the girls CDC threshold for which the sensitivity was 83%. Specificities for the overweight thresholds TABLE 1 Descriptive Statistics by Gender Variable Boys (n = 1785) Girls (n = 1600) Age, y 15.5 (0.1) 15.5 (0.1) Height, cm (0.4) (0.3) Weight, kg 67.2 (0.7) 60.8 (0.6) BMI, kg/m (0.2) 23.2 (0.2) BMI percentile 60.7 (1.2) 63.3 (1.0) Prevalence of metabolic syndrome, % (95% confidence interval) 7.9 ( ) 6.7 ( ) Data are presented as means (SE), except for the prevalence of metabolic syndrome. e332 LAURSON et al

4 ARTICLE TABLE 2 Prevalence of Metabolic Syndrome and Metabolic Abnormalities in Boys by Weight Status Boys CDC Standards FGram Standards Normal Weight (n = 1160) Overweight (n = 280) Obese (n = 345) Normal Weight (n = 1071) Overweight (n = 227) Obese (n = 487) MetS 0.8 ( ) 6.8 ( ) 35.4 ( ) 0.8 ( ) 2.7 ( ) 28.2 ( ) High WC ( ) 65.4 ( ) ( ) 48.7 ( ) High BP 6.6 ( ) 11.5 ( ) 22.7 ( ) 5.6 ( ) 11.9 ( ) 20.9 ( ) Low HDL 17.2 ( ) 30.7 ( ) 56.1 ( ) 17.7 ( ) 20.9 ( ) 50.2 ( ) High TGs 7.6 ( ) 17.9 ( ) 31.4 ( ) 7.5 ( ) 15.6 ( ) 27.1 ( ) High FG 16.4 ( ) 19.5 ( ) 24.1 ( ) 16.3 ( ) 21.1 ( ) 21.7 ( ) Data are presented as prevalence within weight status (95% confidence interval), where prevalence is the percentage of youth in that column with the condition. BP, blood pressure (either systolic or diastolic); FG, fasting glucose; HDL, HDL cholesterol; MetS, metabolic syndrome; TGs, triglycerides; WC, waist circumference. ranged between 67% and 73%. CDC obesity standards tended to have a higher specificity than FGram standards (CDC = 86% 87%, FGram = 76% 81%). The sums of the sensitivities and specificities were greater for boys than for girls by using either set of thresholds. The CDC standards typically had higher PPVs and +PRs, whereas the FGram standards generally had higher NPVs and lower (better) NLRs. The results of the logistic regression using weight status to predict metabolic syndrome are shown in Table 5. A higher weight status was predictive of a greater odds of metabolic syndrome, except for the FGram low-risk threshold in boys. The contrasts indicate that youth classified as obese maintained a higher odds of metabolic syndrome than the overweight groups. The odds of metabolic syndrome for obese boys were 14 times higher than for overweight boys using the FGram standards, and 7.5 times higher using CDC standards. These odds ratios were smaller for girls, ranging from 2 to 3, although the odds ratio using the FGram standards was not significantly different from 1. The odds ratios using CDC thresholds were greater for boys, whereas odds ratios in girls were greater using FGram thresholds. This finding was reinforced by the model fit, where Nagelkerke pseudo r 2 values were larger for CDC thresholds than for FGram thresholds in boys. The opposite was true for girls, although the model fit was more similar between the 2 sets of thresholds. The odds of metabolic syndrome above the thresholds were larger for boys than for girls, regardless of the thresholds. For boys, when excluding the CDC obese group, the low-bmi threshold had a sensitivity of 64.2% and a specificity of 83.2%. In contrast, when excluding the FGram high-risk group, the low-bmi threshold had a sensitivity of 37.8% and a specificity of 84.4%. For the high- BMI threshold, when excluding the CDC overweight group, the sensitivity was 92.1% and specificity was 85.2%. Using the high-bmi threshold for FGram, excluding the some-risk group resulted in a sensitivity of 92.9% and a specificity of 77.7%. In girls, for the low-bmi threshold excluding the CDC obese group, the sensitivity was 54.0% and specificity was 83.4%. Using the low-bmi threshold for FGram, excluding the high-risk group resulted in a sensitivity of 37.8% and a specificity of 84.4%. When excluding the CDC overweight group, the high-bmi threshold had a sensitivity of 78.4% and a specificity of 83.9%. In contrast, when excluding the FGram some-risk group, the high-bmi threshold had a sensitivity of 88.4% and a specificity of 73.8%. To complete the secondary aim of the study, ROC analyses were used with varying objectives (Table 6, Fig 1). The area under the ROC curve, a global indicator of the accuracy of BMI percentile to detect metabolic syndrome, was and for boys and girls, respectively. There is no single method to determine an optimal threshold, and because multiple levels of risk were desired, the selection of the resulting cutoff values was based on combinations of sensitivity and specificity. First, the single threshold with the highest sum of sensitivity and specificity was TABLE 3 Prevalence of Metabolic Syndrome and Metabolic Abnormalities in Girls by Weight Status Girls CDC Standards FGram Standards Normal Weight (n = 1006) Overweight (n = 274) Obese (n = 320) Normal Weight (n = 918) Overweight (n = 182) Obese (n = 500) MetS 1.7 ( ) 9.2 ( ) 24.6 ( ) 1.1 ( ) 9.2 ( ) 19.0 ( ) High WC 4.5 ( ) 63.6 ( ) 98.9 ( ) 2.6 ( ) 34.8 ( ) 89.5 ( ) High BP 2.9 ( ) 2.2 ( ) 9.0 ( ) 1.6 ( ) 10.8 ( ) 6.5 ( ) Low HDL 32.9 ( ) 48.2 ( ) 58.6 ( ) 31.2 ( ) 55.7 ( ) 53.5 ( ) High TGs 8.4 ( ) 10.7 ( ) 18.3 ( ) 8.5 ( ) 5.7 ( ) 16.3 ( ) High FG 6.5 ( ) 8.3 ( ) 12.1 ( ) 7.0 ( ) 3.1 ( ) 11.0 ( ) Data are presented as prevalence within weight status (95% confidence interval), where prevalence is the percentage of youth in that column with the condition. BP, blood pressure (either systolic or diastolic); FG, fasting glucose; HDL, HDL cholesterol; MetS, metabolic syndrome; TGs, triglycerides; WC, waist circumference. PEDIATRICS Volume 133, Number 2, February 2014 e333

5 TABLE 4 Sensitivity and Specificity for CDC and FGram Thresholds by Gender Sensitivity Specificity PPV NPV PLR NLR Boys threshold CDC: $85th percentile 93.1 ( ) 72.7 ( ) FGram: $low risk 93.2 ( ) 68.0 ( ) CDC: $ 95th percentile 80.6 ( ) 87.4 ( ) FGram: $high risk 89.1 ( ) 80.5 ( ) Girls threshold CDC: $85th percentile 82.8 ( ) 71.8 ( ) FGram: $low risk 89.9 ( ) 67.3 ( ) CDC: $95th percentile 62.6 ( ) 86.2 ( ) FGram: $high risk 77.4 ( ) 76.1 ( ) Data in parentheses for sensitivity and specificity are 95% confidence intervals. Dichotomies were created for each threshold to maintain constant n and prevalence of metabolic syndrome. For example, CDC: $85th percentile compares normalweight with overweight 1 obese groups and FGram: $low risk threshold compares Healthy Fitness Zone with low-risk 1 high-risk groups, etc. TABLE 5 Odds Ratios and Fit for Logistic Models Predicting Metabolic Syndrome by Gender CDC FGram Boys Normal weight 1.0 (ref) 1.0 (ref) Overweight 8.9 ( ) 3.3 ( ) a Obese 66.9 ( ) 46.0 (19.5, 108.4) Contrast Obese versus overweight 7.5 ( ) 14.0 ( ) Model r Girls Normal weight 1.0 (ref) 1.0 (ref) Overweight 5.9 ( ) 9.4 ( ) Obese 18.9 ( ) 21.6 ( ) Contrast Obese versus overweight 3.2 ( ) 2.3 ( ) a Model r Model r 2 = Nagelkerke pseudo r 2. ref, referent. Data in parentheses for odds ratio are 95% confidence intervals. a Not statistically different from the referent group. TABLE 6 Results of the ROC Analyses to Differentiate Between Youth With and Without Metabolic Syndrome BMI Percentile Sensitivity Specificity PPV NPV PLR NLR Odds Ratio Boys Single ( ) 81.9 ( ) ( ) Low risk ( ) 74.9 ( ) ( ) High risk ( ) 89.8 ( ) ( ) Girls Single ( ) 70.6 ( ) ( ) Low risk ( ) 65.4 ( ) ( ) High risk ( ) 80.1 ( ) ( ) Data in parentheses for sensitivity, specificity, and odds ratios are 95% confidence intervals. The referent category is children in the normal-weight group based on the proposed thresholds. selected (Youden index). 12 This threshold maximizes the true-positive and true-negative rates but separates the sample into 2 groups, which could be thought of as normal weight versus overweight/obese. To create 3 risk groups, like the CDC and FGram thresholds, low- and high-risk thresholds were established. Sensitivity was prioritized for the low-risk threshold, and the BMI percentile that had a sensitivity of 90% with a specificity thatwasashighaspossiblewasselected, which equates to 90% of children with metabolic syndrome having a BMI above the cutoff value. These thresholds approximated the 87th and 80th percentiles for boys and girls, respectively. Specificity was prioritized for the highrisk threshold. The goal was a BMI percentile with a specificity close to 90% while maintaining the highest sensitivity possible. This threshold results in 90% of children without metabolic syndrome testing below the threshold. These thresholds were near the 97th percentiles for both genders. However, whereas the 97th percentile in girls had a specificity of 90%, the resulting sensitivity of the threshold was 50%, indicating that half of the girls with metabolic syndrome would test into the low-risk or normal-weight categories. Decreasing the desired specificity by 10% (to 80%) resulted in an increase of sensitivity of 20%, so the 91st percentile was selected. The resulting thresholds identified optimal thresholds similar to the CDC standards for boys and the FGram standards for girls. DISCUSSION This study highlights the utility of CDC and FGram thresholds to predict the presence of metabolic syndrome in a nationally representative sample of adolescents. Regardless of how such diagnostic thresholds are created, it is critical for practitioners in all settings (ie, clinical, public health, schools) to understand whether the thresholds can be used to differentiate between risk groups when applied in practice and to quantify such differences. This information also helps practitioners convey the consequences of overweight and obese BMI values to parents and children. We have shown that both the CDC and FGram thresholds are predictive of metabolic syndrome. Adolescents with a BMI percentile in the obese or high- e334 LAURSON et al

6 ARTICLE FIGURE 1 ROC plots and selected percentiles for the detection of metabolic syndrome by BMI percentile. AUC, area under the curve; P, percentile. risk groups have significantly higher odds of metabolic syndrome than their normal-weight or Healthy Fitness Zone counterparts. The diagnostic performance of these BMI thresholds have been provided so that clinicians and practitioners can be aware of how the standards perform in practice and the preferred thresholds can be selected for the desired result. Although there is no single measure of how well a particular BMI threshold performs, from a diagnostic standpoint adolescents with a higher BMI percentile should have a higher prevalence of metabolic disease than those with a lower BMI. Ideally, the prevalence of metabolic syndrome would be 0% in the normal-weight group and 100% in the highest-risk group. This situation is rare in practice. Instead, one would expect the lowest prevalence of metabolic syndrome in normal-weight children and the highest prevalence in obese children, with a graded relationship across normal, overweight, and obesity. For example, using CDC standards in boys, the prevalence of metabolic syndrome among normalweight children was 0.8%, but the prevalence among obese children was 35.4%. For both CDC and FGram standards, such a relationship was generally the case. However, for some risk factors, such as high blood pressure and high fasting glucose in girls, differences between thresholds were smaller or did not increase consistently by weight status. Similar to prevalence, odds ratios between groups of weight status provide insight into the utility of thresholds. The odds of the condition should be low for normal-weight adolescents compared with upper-weight-status groups, providing for relatively large odds ratios. In the current study, odds of metabolic syndrome for obese boys were 46 to 67 times greater than those of normalweight boys, whereas obese girls were 19 to 22 times more likely to have metabolic syndrome than normalweight girls. Odds of metabolic syndrome in the overweight group were generally 6 to 9 times those in the normal-weight group. However, the FGram overweight threshold for boys was not statistically different from normal weight, indicating no additional likelihood of metabolic syndrome. Similarly, Agirbasli et al 13 found that BMI thresholds predicted risk of metabolic syndrome in 9-year-olds, with odds ratios ranging from 3.5 to 4.7. In an adult sample from NHANES, Ervin 14 found odds ratios of 6.2 and 31.9 for metabolic syndrome in overweight and obese males and odds ratios of 5.5 and 17.1 in females, respectively, when compared with normal weight (using adult BMI thresholds of 25 and 30). This finding suggests that FGram and CDC thresholds have similar utility compared with adult thresholds to predict odds of metabolic syndrome. Such a relationship between sets of thresholds regardless of scale (raw BMI versus percentile) is beneficial, especially in longitudinal studies spanning adolescence into young adulthood. Freedman et al 15 previously compared the CDC and FGram BMI standards by using data from the Bogalusa Heart Study and found that the reliability of the FGram thresholds across the age span was low. FITNESSGRAM has since updated the BMI standards and now uses the same BMI growth percentiles as the CDC, partially in an effort to correct this age-related limitation. 16 To our knowledge, this is the first examination of the diagnostic performance of current FGram thresholds using PEDIATRICS Volume 133, Number 2, February 2014 e335

7 metabolic markers of disease. Both sets of thresholds were predictive of metabolic syndrome and performed appropriately well. Compared directly with CDC standards, a greater prevalence of obesity was found using the lower FGram thresholds (eg, CDC 85th percentile versus FGram 80th percentile). Because metabolic risk factors are positively associated with BMI percentile (ie, higher BMI values equate to more risk factors), higher thresholds will generally produce higher prevalence estimates. 17 Examples of this phenomenon are pronounced when comparing CDC and FGram obesity thresholds. In general, the CDC thresholds resulted in greater prevalence estimates of metabolic syndrome and metabolic abnormalities (Tables 2 and 3) and higher specificities, PPVs, and PLRs, with fewer false negatives (Table 4). In contrast, the lower FGram thresholds generally have higher sensitivities, NPVs, and lower (better) NLRs while identifying a greater portion of truepositive cases. The logistic models perhaps indicate that the CDC standards for boys and FGram standards for girls provide more predictive utility. However, it should be emphasized that there is no one measure of efficient standards and the thresholds can be selected depending on the desired result (falsenegative rate, feasibility, etc). The ROC analysis identified percentiles close to CDC thresholds for boys and close to FGram thresholds for girls, supporting the previous analyses contrasting the 2 standards. The odds ratios using the 86.5th and 96.5th thresholds for boys are greater than those when using the current CDC standards, but similar for the girls FGram thresholds. It should be pointed out that the low-risk ROC thresholds were designed to prioritize sensitivity over specificity, and the opposite was true for the high-risk ROC thresholds. This design ensures that the majority of metabolic syndrome cases ($90%) fall above the low-risk threshold, minimizing false negatives. Very few cases of metabolic syndrome occur in youth with a BMI below this threshold. In contrast, the high-risk ROC threshold prioritizes specificity, resulting in higher PPVs, where 40% of boys and 20% of girls who have a BMI in the highrisk zone have metabolic syndrome. A single threshold at the 83.8th percentile for girls might be equivalent to the girls multithreshold standards (and easier to use) because the odds ratio when using 1 threshold was nearly equal to the high-risk threshold. The ROC results also provide support for a severe obesity threshold near the 97th percentile for boys and girls, where the odds of metabolic syndrome and specificity are high, 18 although future investigation focusing on an additional level of risk stratification is needed. The areas under the ROC curve were and for boys and girls, respectively, indicating moderate accuracy. 19 BMI percentiles can be used with better diagnostic accuracy in boys than in girls. This situation was not only true for the ROC analysis but for all others in the current study. CDC and FGram standards were more predictive of metabolic syndrome, the prevalence of metabolic syndrome by weight status was higher, odds ratios by weight status were larger, and sensitivities and specificities were better when using the thresholds for boys compared with girls. In general, the prevalence of metabolic syndrome and most components of metabolic syndrome have been found to be higher in boys than girls, regardless of the definition used Whereas the specific mechanisms for these differences are difficult to elucidate within the context of the current study, it is clear that the association between BMI and metabolic syndrome is weaker in girls than in boys. From a diagnostic perspective, this finding is due to the higher prevalence of metabolic syndrome in the normal-weight and overweight categories of weight status for girls, whereas the boys have a greater prevalence of metabolic syndrome concentrated within the obese category. The same is true for most of the components, excluding high waist circumference and low HDL cholesterol, which equates to a wider distribution of metabolic abnormalities across the upper BMI percentiles in girls, leading to more misclassification errors overall. Whereas girls with BMI values above the thresholds still have a higher odds of metabolic syndrome than girls who are normal weight (with odds ratios ranging from 6 to 22.5), clinicians and schools should be aware that such values may pose less of a hazard in terms of metabolic abnormality than a positive result in boys. It could also be that differences in the definition of metabolic syndrome between boys and girls contribute to these gender differences. Similarly, Ervin 14 noted that, using the adult version of the ATP III definition used here, women had a greater prevalence of high waist circumference and low HDL cholesterol, with a weaker association between BMI status and metabolic syndrome than men. Regardless, additional research into this disparity is warranted. With regard to the CDC standards and general use of BMI percentiles, some previous research in youth exists, although direct comparisons are difficult depending on the methodology. Our findings are in agreement with those of Camhi and Katzmarzyk, 22 who reported an increasing prevalence of metabolic risk factors by CDC weight status in adolescents, with 9% of normalweight adolescents having $2 metabolic risk factors compared with 21% and 35% of overweight and obese adolescents, respectively. Although not e336 LAURSON et al

8 ARTICLE the primary purpose of the study, our analysis adds to the literature showing that weight status is a powerful predictor of metabolic syndrome in youth. Others who used BMI percentiles have also reported increasing metabolic risk with increasing weight status or BMI z score cross-sectionally and prospectively, despite the wellknown limitations of BMI. 6 This study is not without limitations. Although the definition of metabolic syndrome was based on ATP III criteria and growth curves from NHANES, 9 there is no universal definition. 32 Theprevalenceof metabolic syndrome may vary depending on the definition chosen. Additionally, differences may exist in the prevalence of metabolic syndrome between race/ ethnicity groups. 33,34 However, neither the CDC or FGram BMI thresholds are race/ethnicity-specific. We recalculated the odds of metabolic syndrome by weight status (as in Table 5), controlling for race/ethnicity in an alternative analysis. We found no meaningful differences in the results (data not shown), indicating that the impact of race/ethnicity on the association between BMI and metabolic syndrome likely plays a small role in this analysis comparing the CDC and FGram standards. The ideal thresholds and diagnostic performance of any classification system will vary depending on the characteristics of the target population (race/ethnicity, socioeconomic status, the outcome selected, prevalence of the outcome, etc). Although NHANES provides a large, nationally representative sample of adolescents, data are cross-sectional. Our investigation details how weight status is associated with having metabolic syndrome, not prospectively being diagnosed. Additional research investigating the relationship of childhood weight status with future metabolic syndrome and providing prospective diagnostic information about these thresholds is necessary. CONCLUSIONS Regardless of how BMI standards and clinical thresholds are developed, they should be investigated to ensure that they impart meaningful health-related information to clinicians, teachers, parents, and children. Even though many practitioners already use BMI, there are still concerns over the lack of specific cutoffs relative to health risk. 35 We believe that our findings help fill this void by providing information on the 2 most widely used BMI standards in schools and clinics across the United States, although more research is still needed. Both CDC and FGram weight status thresholds were predictive of metabolic syndrome and metabolic abnormalities in this nationally representative sample of US adolescents. The independent ROC analyses identified ideal thresholds to detect metabolic syndrome that were more similar to the CDC standards for boys but aligned more closely with the FGram standards for girls. Although similar, the CDC, FGram, and ROC standards each present a different set of strengths and shortcomings. Although the FGram standards are popular, especially for fitness testing, the CDC thresholds are already more widely used by pediatricians and those working in public health care. Therefore, we recommend using the CDC standards so that children and parents receive consistent messages about weight status throughout growth and maturation. Additionally, we hope that these results inform practitioners and aid them in conveying the implications of having a BMI above (or below) these thresholds to parents and children. REFERENCES 1. Barton M; US Preventive Services Task Force. Screening for obesity in children and adolescents: US Preventive Services Task Force recommendation statement. Pediatrics. 2010;125(2): Kuczmarski RJ, Ogden CL, Guo SS, et al CDC growth charts for the United States: methods and development. Vital Health Stat ;(246): Laurson KR, Eisenmann JC, Welk GJ. Body fat percentile curves for U.S. children and adolescents. Am J Prev Med. 2011;41(4 suppl 2): S87 S92 4. Laurson KR, Eisenmann JC, Welk GJ. Development of youth percent body fat standards using receiver operating characteristic curves. Am J Prev Med. 2011;41(4 suppl 2):S93 S99 5. Laurson KR, Eisenmann JC, Welk GJ. Body mass index standards based on agreement with health-related body fat. Am J Prev Med. 2011;41(4 suppl 2):S100 S Prentice AM, Jebb SA. Beyond body mass index. Obes Rev. 2001;2(3): Centers for Disease Control and Prevention; National Center for Health Statistics. National Health and Nutrition Examination Survey Examination Procedures Manual. Hyattsville, MD: US Department of Health and Human Services, Centers for Disease Control and Prevention; Available at: nchs/data/nhanes/nhanes_03_04/pe.pdf. Accessed May 1, Centers for Disease Control and Prevention; National Center for Health Statistics. National Health and Nutrition Examination Survey Laboratory Methods. Hyattsville, MD: US Department of Health and Human Services, Centers for Disease Control and Prevention; Available at: www. cdc.gov/nchs/nhanes/nhanes / lab_methods_03_04.htm. Accessed May 1, Jolliffe CJ, Janssen I. Development of agespecific adolescent metabolic syndrome criteria that are linked to the Adult Treatment Panel III and International Diabetes Federation criteria. J Am Coll Cardiol. 2007;49(8): PEDIATRICS Volume 133, Number 2, February 2014 e337

9 10. Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults. Executive Summary of the Third Report of The National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III). JAMA. 2001;285(19): Greiner M, Pfeiffer D, Smith RD. Principles and practical application of the receiver-operating characteristic analysis for diagnostic tests. Prev Vet Med. 2000;45(1 2): Akobeng AK. Understanding diagnostic tests 3: receiver operating characteristic curves. Acta Paediatr. 2007;96(5): Agirbasli M, Agaoglu NB, Ergonul O, et al. Comparison of anthropometric indices in predicting metabolic syndrome components in children. Metab Syndr Relat Disord. 2011;9(6): Ervin RB. Prevalence of metabolic syndrome among adults 20 years of age and over, by sex, age, race and ethnicity, and body mass index: United States, Natl Health Stat Report. 2009;(13): Freedman DS, Fulton JE, Dietz WH, et al. The identification of children with adverse risk factor levels by body mass index cutoffs from 2 classification systems: the Bogalusa Heart Study. Am J Clin Nutr. 2010;92(6): WelkGJ,GoingSB,MorrowJRJr,Meredith MD. Development of new criterion-referenced fitness standards in the FITNESSGRAM program: rationale and conceptual overview. Am JPrevMed. 2011;41(4 suppl 2):S63 S Okosun IS, Boltri JM, Lyn R, Davis-Smith M. Continuous metabolic syndrome risk score, body mass index percentile, and leisure time physical activity in American children. J Clin Hypertens (Greenwich). 2010;12(8): Harrington DM, Staiano AE, Broyles ST, Gupta AK, Katzmarzyk PT. BMI percentiles for the identification of abdominal obesity and metabolic risk in children and adolescents: evidence in support of the CDC 95th percentile. Eur J Clin Nutr. 2013;67(2): Swets JA. Measuring the accuracy of diagnostic systems. Science. 1988;240(4857): Daratha KB, Bindler RC. Effects of individual components, time, and sex on prevalence of metabolic syndrome in adolescents. Arch Pediatr Adolesc Med. 2009;163(4): Ford ES, Li C, Zhao G, Pearson WS, Mokdad AH. Prevalence of the metabolic syndrome among U.S. adolescents using the definition from the International Diabetes Federation. Diabetes Care. 2008;31(3): Camhi SM, Katzmarzyk PT. Prevalence of cardiometabolic risk factor clustering and body mass index in adolescents. J Pediatr. 2011;159(2): Lambert M, Delvin EE, Levy E, et al. Prevalence of cardiometabolic risk factors by weight status in a population-based sample of Quebec children and adolescents. Can J Cardiol. 2008;24(7): Cook S, Weitzman M, Auinger P, Nguyen M, Dietz WH. Prevalence of a metabolic syndrome phenotype in adolescents: findings from the third National Health and Nutrition Examination Survey, Arch Pediatr Adolesc Med. 2003;157(8): Jago R, Drews KL, McMurray RG, et al. Fatness, fitness, and cardiometabolic risk factors among sixth-grade youth. Med Sci Sports Exerc. 2010;42(8): Messiah SE, Arheart KL, Luke B, Lipshultz SE, Miller TL. Relationship between body mass index and metabolic syndrome risk factors among US 8- to 14-year-olds, 1999 to J Pediatr. 2008;153(2): de Ferranti SD, Gauvreau K, Ludwig DS, Neufeld EJ, Newburger JW, Rifai N. Prevalence of the metabolic syndrome in American adolescents: findings from the Third National Health and Nutrition Examination Survey. Circulation. 2004;110(16): Garnett SP, Baur LA, Srinivasan S, Lee JW, Cowell CT. Body mass index and waist circumference in midchildhood and adverse cardiovascular disease risk clustering in adolescence. Am J Clin Nutr. 2007;86(3): Weiss R, Dziura J, Burgert TS, et al. Obesity and the metabolic syndrome in children and adolescents. N Engl J Med. 2004;350 (23): Burns TL, Letuchy EM, Paulos R, Witt J. Childhood predictors of the metabolic syndrome in middle-aged adults: the Muscatine study. J Pediatr. 2009;155(3):S5.e Morrison JA, Friedman LA, Wang P, Glueck CJ. Metabolic syndrome in childhood predicts adult metabolic syndrome and type 2 diabetes mellitus 25 to 30 years later. J Pediatr. 2008;152(2): Ford ES, Li C. Defining the metabolic syndrome in children and adolescents: will the real definition please stand up? J Pediatr. 2008;152(2): Walker SE, Gurka MJ, Oliver MN, Johns DW, DeBoer MD. Racial/ethnic discrepancies in the metabolic syndrome begin in childhood and persist after adjustment for environmental factors. Nutr Metab Cardiovasc Dis. 2012;22(2): Gurka MJ, Ice CL, Sun SS, Deboer MD. A confirmatory factor analysis of the metabolic syndrome in adolescents: an examination of sex and racial/ethnic differences. Cardiovasc Diabetol. 2012;11(1): Flegal KM, Ogden CL. Childhood obesity: are we all speaking the same language? Adv Nutr. 2011;2(2):159S 166S e338 LAURSON et al

10 Diagnostic Performance of BMI Percentiles to Identify Adolescents With Metabolic Syndrome Kelly R. Laurson, Gregory J. Welk and Joey C. Eisenmann Pediatrics 2014;133;e330 DOI: /peds originally published online January 27, 2014; Updated Information & Services References Subspecialty Collections Permissions & Licensing Reprints including high resolution figures, can be found at: This article cites 31 articles, 7 of which you can access for free at: This article, along with others on similar topics, appears in the following collection(s): Endocrinology Metabolic Disorders ub Obesity Information about reproducing this article in parts (figures, tables) or in its entirety can be found online at: Information about ordering reprints can be found online:

11 Diagnostic Performance of BMI Percentiles to Identify Adolescents With Metabolic Syndrome Kelly R. Laurson, Gregory J. Welk and Joey C. Eisenmann Pediatrics 2014;133;e330 DOI: /peds originally published online January 27, 2014; The online version of this article, along with updated information and services, is located on the World Wide Web at: Pediatrics is the official journal of the American Academy of Pediatrics. A monthly publication, it has been published continuously since Pediatrics is owned, published, and trademarked by the American Academy of Pediatrics, 141 Northwest Point Boulevard, Elk Grove Village, Illinois, Copyright 2014 by the American Academy of Pediatrics. All rights reserved. Print ISSN: