Performance of different adiposity measures for predicting cardiovascular risk in adolescents

This study aims to compare the performance of body mass index (BMI), waist circumference (WC), and waist-to-height-ratio (WHtR) to predict the presence of at least 3 main CV risk factors in US adolescents. A total of 3621 adolescents (boys: 49.9%) aged 12–17 years from the US National Health and Nutrition Examination Survey (1999–2012) were included in this study. Measured CV risk factors included systolic/diastolic blood pressure, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, and fasting plasma glucose. The AUC of BMI-z score, WC-z score and WHtR-z score to predict at least three CV risk factors were similar (~0.85), irrespective of criteria used to define abnormal levels of CV risk factors. A 1-SD increase in any of three indices to predict CV risk was also similar for the three adiposity scores. For instance, a 1-SD increase risk in BMI-z score, WC-z score and WHtR-z score was 3.32 (95%CI 2.53–4.36), 3.43 (95%CI 2.64–4.46), and 3.45 (95%CI 2.64–4.52), respectively, in the total population using the International Diabetes Federation definition. In addition, the most efficient WHtR cut-off for screening CV risk was ~0.50 in US adolescents. In summary, BMI, WC and WHtR performed similarly well to predict the presence of at least 3 main CV risk factors among US adolescents.

Centers for Disease Control and Prevention. NHANES is a program of studies designed to assess the health and nutritional status of adults and children in the United States, which includes household interviews and physical examinations in the mobile center. Data on demographics, questionnaire, examination, dietary and blood samples were collected. Our study was restricted to adolescents aged 12-17 years since fasting plasma glucose (FPG) was available only for subjects aged 12 years or older. A total of 3621 adolescents (boys: 49.9%) aged 12

Measurements.
Measurements included height, weight, WC, systolic BP (SBP), diastolic BP (DBP), total cholesterol, TG, HDL-C, LDL-C and FPG levels 27 . Weight and height were measured for each individual in light clothing without shoes. BMI was calculated as weight (kg)/height squared (m 2 ). WC was measured at the high point of the iliac crest 28 . WHtR was calculated as WC (cm)/height (cm).
SBP and DBP were measured using auscultatory mercury sphygmomanometers by trained examiners, following a standard protocol. BP was taken on the right arm with an appropriate cuff size (pediatric, normal or large cuffs) with the subject seating quietly for at least 5 minutes before BP measurement. The first phase (SBP) and the fifth phase (DBP) Korotkoff sounds were recorded. Data on DBP level as zero were excluded. Up to three BP measurements were taken for each individual and their average was used in the analyses.
Blood samples were obtained from the antecubital vein in at least 8-hour fasting subjects. TC, TG, HDL-C, and FPG were measured using enzymatic methods. LDL-C was estimated using the formula of Friedewald et al. 29 .

Definition of CV risk factors.
We defined abnormal levels of CV risk factors (BP, TG, HDL-C, and FPG) based on cut-offs used to define the metabolic syndrome (MetS) from the International Diabetes Federation (IDF) 30 or the revised National Cholesterol Education Program Adult Treatment Panel III (ATP III) modified for the pediatric population 31 . One, or combination of two or three more CV risk factors was used as the outcomes. The IDF criteria in adolescents aged 12-15 years are defined as triglycerides ≥ 150 mg/dL (1.7 mmol/L), HDL-cholesterol < 40 mgl/dL (1.03 mmol/L), SBP/DBP ≥ 130/85 mm Hg, and FPG ≥ 100 mg/dL (5.6 mmol/L) or known type 2 diabetes. Criteria for adults were applied for adolescents aged 16-17 years. The ATP III criteria are defined as triglycerides ≥ 110 mg/dL (1.24 mmol/L), HDL-C ≤ 40 mgl/dL (1.03 mmol/L), SBP/DBP ≥ 90 th percentile (age-, sex-and height specific), and FPG ≥ 110 mg/dL (6.1 mmol/L). High LDL-C was defined as LDL-C ≥ 100 mg/dL (2.6 mmol/L) 32 .
Statistical analysis. All data analyses were performed using SAS 9.2 (SAS Institute, Cary, NC). Sampling weights, primary sampling unit (PSU) and strata were taken into consideration to account for the complex survey design. Quantitative data are expressed as mean [standard error (SE)] and categorical data as percentage (SE). Correlation coefficients between BMI, WC and WHtR were calculated using partial correlation analyses by sex adjusting for age and race/ethnicity. As BMI, WC and WHtR have different units, we calculated sex and age specific z-scores within the study sample. The performance of BMI-z score, WC-z score and WHtR-z score to predict the presence of one, or combination of two or three more from the five CV risk factors (BP, TG, HDL-C, LDL-C, and FPG), defined by either IDF or ATP III criteria, was assessed by comparing the area under curve (AUC) derived from receiver operator characteristic (ROC) curve analysis. ROC curve was constructed to visually show the relationship between true-positive (sensitivity) and false positive (1-specificity) 33 . The AUC was used to evaluate the performance of three adiposity indices in detecting CVD risk factors. An AUC = 0.5 indicated the test performed no better than chance. An AUC of < 0.7 was considered as poor, 0.7-0.8 was rated as acceptable and of > 0.8 as good. We performed logistic regression analysis to assess a cumulative CV risk associated with a 1-SD increase of adiposity measures. ROC analysis was also performed to determine the optimal WHtR cut-off. The optimal WHtR cut-off was determined by the maximum of the Youden index (sensitivity + specificity-1) 33 . This optimal threshold corresponds to the point that is closest to the top left-handed corner in the ROC curve with optimal sensitivity and specificity.

Results
A total of 3621 adolescents (boys: 49.9%) aged 12-17 years were included in this study. Table 1 shows the characteristics and CV risk factors in boys and girls, respectively. The prevalence of clustering of CV risk factors differed between sexes and analyses were therefore also performed separately in boys and girls. In addition, BMI, WC and WHtR were highly correlated with each other (r ≥ 0.93 between either two indices, p < 0.001).
The AUC of BMI-z score, WC-z score, and WHtR-z score to predict at least three CV risk factors were virtually identical using IDF or ATP III criteria (Table 2). In the total population, AUC ranged from 0.84 to 0.85 for three obesity indices using either IDF or ATP III criteria, indicating good performance of all three indices. Similar results were obtained in sex and race/ethnicity subgroups ( Table 2). The odds ratios associated with a 1-SD increase and the presence of at least three risk factors were largely similar for the three obesity indices ( Table 2). In the total population, the odds ratios ranged from 3.32 to 3.45 using the IDF criteria and from 3.71 to 3.99 using the ATP III criteria. The results were largely independent of sex or race/ethnicity ( Table 2). In addition, Scientific RepoRts | 7:43686 | DOI: 10.1038/srep43686 BMI-z score, WC-z score, and WHtR-z score also performed equally in predicting one or two CV risk factors based on ROC analysis and Logistic regression analysis (Supplemental Table 1).
The optimal cut-off to predict at least three CV risk factors was about 0.50 using either the IDF or ATP III criteria, with values ranging from 0.48 to 0.52 by sex and race/ethnicity (Table 3). In addition, ~0.50 as the cut-off of WHtR also performed well to predict one or two CV risk factors (Supplemental Table 2).

Discussion
In the present study, we found that BMI, WHtR and WC performed similarly well to predict the presence of at least three CV risk factors in US adolescents. In addition, our findings confirm that the message "keep your waist less than half of your height" is valid in US adolescents. Indeed, the optimal WHtR cut-off was ~0.50 and was largely independent of sex, race/ethnicity, and criteria used to define CV risk factors. Many studies have compared the performance of BMI, WC and WHtR for screening CV risk factors in children [8][9][10][11][12][13][14][15][16][17][18][19][20] . However, results have been inconsistent with some studies reporting that WHtR performed better than WC and BMI [8][9][10][11]14 , several studies reporting similar performance [16][17][18][19] , and others reporting that BMI performed best 20,21 . These conflicting results may be due to differences in study design, population selection, race/ethnicity, geographical location or methods used for assessing obesity-related indices. Our study is based on nationally representative data so that our findings can be generalized to all US adolescents. As there was no significant difference in using WHtR, WC and BMI for predicting CV risk, either obesity index are similarly valid for use in practice. However, because prediction based on WHtR only needs one value (0.50), this indicator may be more convenient in practice.
The optimal cut-off of WHtR for predicting CV risk has been debated and values of < 0.50, 0.50 or > 0.50 have been suggested 16,17,25,26 . An UK cohort study of 2710 children showed that 0.50 as WHtR cut-off had perfect specificity in predicting CV risk, but were poorly sensitive, suggesting that WHtR optimal cut-off should be less than 0.50 17 . In addition, a cross-sectional study conducted in South Africa reported that the cut-offs of 0.46 in boys and 0.45 in girls were optimal for screening CV risk 26 . In our study, a WHtR of 0.50 was most efficient in all US adolescents, independently of sex and tested race/ethnicity (Hispanic, White and Black). However, we have no WHtR data in other race/ethnicity such as Arabian and Asian. Thus, the optimal WHtR cut-off may differ in other populations and our conclusion should not be directly applied in Arabian or Asian. For instance, Asian children tend to be shorter than White or Black children for the same sex and age, and a lower WHtR cut-off has been suggested 34 .  Our study has several strengths. First, the sample of participants was nationally representative and our findings can be generalized to all US adolescents. Second, we assessed CV risk factors using different criteria and found that the results were independent of the criteria used to define CV risk factors. Third, measurements of CV risk factors followed strict quality control, including adequate calibration of instruments and training of the examiners. However, findings in our study should be validated in other countries and further studies should extend to children aged < 12 years. In addition, the positive predictive value of WHtR (using 0.50 as optimal cut-off) for predicting 3 + CV risk factors was about 20% using IDF criteria and 25% using ATPIII because of low prevalence of clustering of 3 + CV risk factors, suggesting that WHtR alone has limited value to predict CV risk in adolescents.

Conclusions
Our study suggests that BMI, WC. and WHtR performed well and almost identically for identifying the presence of at least 3 major CV risk factors in US adolescents and that a WHtR cut-off of ~0.50 is suitable for US adolescents. Our results were largely independent of sex and race/ethnicity. As the uses of BMI, WC or WHtR have each potential advantages and limitations in practice, translational research is needed to identify which index is most suitable in different screening and clinical conditions.