New age-adjusted measure of body fat distribution in children and adolescents: standardization of waist-hip ratio using multivariate analysis

Transcription

1 International Journal of Obesity (1997) 21, 594±599 ß 1997 Stockton Press All rights reserved 0307±0565/97 $12.00 New age-adjusted measure of body fat distribution in children and adolescents: standardization of waist-hip ratio using multivariate analysis K Asayama 1, K Hayashi 2,3, Y Kawada 1, T Nakane 1, N Uchida 1, H Hayashibe 1, K Kawasaki 2 and S Nakazawa 1 1 Department of Pediatrics, Yamanashi Medical University, 1110 Shimokato, Tamahocho, Nakakomagun, Yamanashi , Japan; 2 Biostatistics Department, Yamanouchi Pharmaceutical Co., Azusawa, Itabashiku, Tokyo 174, Japan; and 3 Present address: Department of Basic Allied Medicine, School of Health Sciences, Faculty of Medicine, Gunma University, Showa-machi, Maebashi, Gunma 371, Japan OBJECTIVES: To explore a new anthropometric index of body fat distribution adjusted for ages ranging from 6±15 y in both boys and girls. DESIGN: Sex, age, and 11 anthropometric variables were subjected to principal component analysis. Based on these results, we developed a new anthropometric index, namely an age-adjusted measure of body fat distribution. This index was evaluated statistically for suitability of use in epidemiological surveys. SUBJECTS: Japanese children, including obese and nonobese subjects, in one elementary and one junior high school in Yamanashi Prefecture, Japan: 508 boys and 549 girls whose ages ranged from 6 y 1 mon±15 y. MEASUREMENTS: Measurements included the height (Ht), body weight, circumference of the waist, hip and thigh. Body mass index, the ratios of the waist, hip or thigh to the Ht, waist-hip ratio (WHR) and waist-thigh ratio were calculated. RESULTS: The rst principal component (PC 1) accounted for 49.8% of the total variation, and was interpreted as an indicator of the general size of an individual. PC 2 accounted for 25.9%, and was interpreted as a shape measure that indicates body fat distribution. Calculation of WHR/Ht, a parameter that represented PC 2 adjusted by PC 1, gave an highly robust linear regression equation for age by gender. The residuals from the regression line for WHR/Ht deviated from normal distribution only in the boys, whereas the mean residual was nearly zero and distribution of the residuals was similar in three age subgroups by gender, supporting the use of the common standard deviation score in all age groups as an indicator of body fat distribution. CONCLUSION: The common standard deviation score of WHR/Ht can serve as an epidemiological index of body fat distribution adjusted for ages between 6 and 15 y. Keywords: anthropometric methods; body fat distribution; children and adolescents; principal component analysis; regression analysis Introduction Biochemical risk factors for atherosclerosis in children are more common among those who are obese. 1 Anthropometric standards that can predict such potential health problems should be established for use in the school or clinical setting. The amount of intraabdominal fat is more signi cantly related to metabolic derangement induced by obesity than is the amount of subcutaneous fat. Although in previous reports, anthropometric measurement did not always re ect the amount of intra-abdominal fat, 2,3 evidence suggests that waist-hip ratio (WHR) is useful in evaluating the risk factors for atherosclerosis in adults 4±6 and children. 7±10 At least in adults, WHR is generally accepted as a measure that re ects the Correspondence: Dr K Asayama. Received 29 November 1996; revised 18 March 1997; accepted 20 March 1997 distribution of fat between intra-abdominal and subcutaneous sites. Findings in several previous studies of children have discouraged the use of WHR as an epidemiological index of body fat distribution. However, several factors need to be considered in deciding whether WHR, and related anthropometric indices, are suitable for estimating the extent of abdominal obesity in children. Changes in body composition, adiposity and distribution of body fat during the growth and development of children are well established. 9±11 In general, biochemical complications as well as excess weight in obese children tend to worsen during growth, 1 whereas the normal values for WHR decrease with age. 10,12 Accordingly, an arithmetical adjustment for age using analysis of covariance is not very successful in obese children and adolescents. Standardization of the criteria for obesity should be based on the actual measures of body build observed in a large population of children.

2 The values and physiological signi cance of WHR differ between the races. 13 Japanese people have a different genetic background, as well as lifestyle, as compared with other cultures. Their body build also differs. In our previous study, after adjusting for age using linear regression equations for nonobese subjects, the standard deviation score (SDS) for WHR was found to be correlated with the serum levels of lipids and apolipoproteins in obese Japanese children attending elementary school (namely 6±12 y old). Indices for being overweight or for adiposity were unrelated to such serum biochemical ndings, especially in the girls. 14 The present study was designed to develop a new anthropometric index of body fat distribution in boys and girls that was adjusted for age ranging from 6± 15 y. The distribution pattern of the value of WHR in cohorts is skewed to the right, as with body mass index. It is well known that, in girls, the normal value of WHR declines rapidly during adolescence. 12 To solve these problems, we evaluated data from a large group of children by means of principal component analysis (PCA). A new anthropometric index that was statistically valid in this age group was constructed according to the results of PCA. Materials and methods Subjects A total of 1057 Japanese children who attended one elementary or one junior high school in Yamanashi Prefecture, Japan, were subjected to anthropometric measurements. There were 508 boys and 549 girls who ranged in age from 6 y 1 mon±15 y (Table 1). The subjects included both obese and nonobese children, and were subdivided into three groups by age and separated according to sex (Table 1). This survey was approved by the Ethics Committee of Yamanashi Medical University, Japan. Informed parental consent for participation in this survey was obtained. Anthropometric measurements Anthropometric measurements were performed, as described previously, 14 by the medical staff in the Department of Paediatrics, Yamanashi Medical University. In brief, height (Ht) was measured to the nearest 0.1 cm and body weight to the nearest 0.1 kg using a stadiometer. A plastic measuring tape was used to determine the circumference of the waist at the level of the umbilicus, and that of the hip at the level of maximum extension of the buttocks, to the nearest 0.1 cm, with the subject standing and following a normal expiration. Thigh circumference was measured 3 cm above the upper border of the patella on the left side if the subject was right-handed, and on the right side if the subject was left-handed. Body mass index (BMI) was calculated for each subject by dividing the body weight (in kilograms) by the Ht 2 (in meters). To express body shape, we calculated the ratio of the waist, hip or thigh to height, respectively, waist-to-hip circumference ratio (WHR) and waist-to-thigh circumference ratio (WTR). 595 Table 1 Characteristics of the study population: basic statistics of the anthropometric measurements Age group (year) 6^<9 9^<12 12^15 All subjects Numbers (boys/girls) 185 (96/89) 250 (116/134) 622 (296/326) 1057 (508/549) Age (y) M s.d range (6.17±8.96) (9.01±11.99) (12.00±15.01) (6.17±15.01) Ht (cm) M s.d range (104.0±140.8) (118.3±168.3) (134.4±182.5) (104.0±182.5) Weight (kg) M s.d range (14.2±41.2) (20.6±62.5) (28.4±82.0) (14.2±82.0) Waist (cm) M s.d range (44.8±85.0) (47.2±90.0) (50.7±99.0) (44.8±99.0) Hip (cm) M s.d range (50.0±82.6) (58.0±98.4) (67.5±111.0) (50.0±111.0) Thigh (cm) M s.d range (23.3±39.3) (17.0±47.1) (28.1±50.6) (17.0±50.6) BMI (kg/m 2 ) M s.d range (13.0±26.3) (12.7±27.6) (13.5±32.8) (12.7±32.8) Waist/Ht M s.d range (0.373±0.689) (0.351±0.606) (0.337±0.642) (0.337±0.689) Hip/Ht M s.d range (0.462±0.669) (0.454±0.651) (0.460±0.698) (0.454±0.698) Thigh/Ht M s.d range (0.200±0.318) (0.139±0.319) (0.190±0.347) (0.139±0.347) WHR M s.d range (0.754±1.029) (0.728±1.026) (0.632±0.978) (0.632±1.029) WTR M s.d range (1.48±2.40) (1.49±2.99) (1.34±2.27) (1.34±2.99) WHR/Ht M s.d range (0.576±0.942) (0.459±0.743) (0.389±0.733) (0.389±0.942) Abbreviations: M s.d. indicates mean standard deviation; Ht, height; BMI indicates body mass index (weight/ height 2 ); WHR, waist±hip ratio; WTR, waist±thigh ratio.

3 596 Statistical analyses Data are presented as means and standard deviations (s.d.). At the rst stage of the present study, all samples (n ˆ 1057) were pooled, and gender 0 indicating boys and 1 the girls, age, and 11 anthropometric variables (Table 1) were subjected to PCA. Then we explored a new anthropometric index that would represent an age-adjusted measure of body fat distribution by utilizing various statistical methods. Other statistical methods used were linear regression analysis, Shapiro±Wilks test for normality, Kruskal± Wallis and Median tests for the difference in the central values, and the Kolmogorov±Smirnov twosample test for the comparison of the distribution of variables in different subgroups. All P-values are twosided. A level P < 0.05 was accepted as statistically signi cant. Statistical analyses were performed using Statistical Analysis System software (SAS Institute Inc., Cary, NC, USA). Results The means s.d. and ranges of anthropometric measures in the three age subgroups are summarized in Table 1. The mean values for Ht, body weight, waist, hip and thigh were greater in the older groups than in the younger groups, indicating that the body size increased with age. Similarly, the BMI tended to increase with age. On the other hand, both WHR and WTR showed a gradual decrease with age. Table 2 summarizes the results of PCA in the subjects studied. We retained factor loadings and respective scores for the rst four major axes, which jointly accounted for 96.2% of the total variation. Only the rst three principal components (PC) exhibited eigen-values greater than 1. PCA clearly separated the variables that gave different loadings to the respective axes. The rst PC accounted for 49.8% of the total variation, and could serve as a credible indicator of an individual's general size. PC 2 accounted for 25.9% of the total variation, and was considered to be a measure of shape that indicated body fat distribution, being mainly loaded by waist/ht and WHR in contrast with size variables. PC 3 and 4 were factors for combinations of gender and the several variables for shape. The factor patterns were plotted by assigning PC 1 to the ordinate and PC 2 to the abscissa (Figure 1). The gure clearly contrasts the different characteristic between BMI (index for being overweight) and WHR (index of body fat distribution). WHR was divided by the Ht (in meters) of each individual, in an attempt to obtain a new measure of body fat distribution adjusted for body size. Among the size variables studied here, Ht appeared to be the most suitable for standardizing the WHR because it is normally distributed in the general population and it is the standard measure for linear growth. Since our previous study revealed that WHR was signi cantly smaller in girls than in boys even in ages younger than 12 y old, 14 data for boys and girls were treated separately in the further analyses. We next evaluated the correlation of WHR/Ht with age in each gender, and compared that to the correlation of waist/ht and WHR with age. Table 3 summarizes the univariate linear regression analysis assigning age as an independent variable. In both sexes, the correlation with age was highly signi cant for waist/ht, WHR, as well as WHR/Ht. The r 2 values indicated that the linear association of WHR/Ht with age in both sexes was more stable than was the association of waist/ht and WHR with age: the correlation coef cient of WHR/Ht with age was for the boys and for the girls (Figures 2 and 3). This close linear association of the two parameters in both sexes clearly indicates that WHR/Ht can be adjusted for age by simply introducing a linear function, and that the residual (namely the deviation of the observed value of WHR/Ht from Table 2 Results of principal component analysis (n ˆ 1057) Factor1 Factor 2 Factor 3 Factor 4 Gender Age Ht Weight BMI Waist Hip Thigh Waist/Ht Hip/Ht Thigh/Ht WHR WTR Eigen-value Proportion Cumulative Abbreviations are the same as in Table 1. Figure 1 Factor patterns in principal component analysis. Factor loadings are plotted assigning the rst principal component to the ordinate and the second component to the abscissa. Abbreviations: Ht, height; BMI, body mass index; WHR, waist-hip ratio; WTR, waist-thigh ratio.

4 Table 3 Univariate linear regression analysis of Waist/Ht, WHR and WHR/Ht Estimate SE P-value r 2 Waist/Ht Boys a < b < Girls a < b < WHR Boys a < b < Girls a < b < WHR/Ht Boys a < b < Girls a < b < Y ˆ a b 6 Age. the expected value of the linear regression equation) for each individual can be an age-adjusted estimate of body fat distribution. To further optimize the residual of WHR/Ht for clinical use, we calculated the SDS of WHR/Ht using common s.d. for all age subgroups. We examined the statistical characteristics of this parameter to determine whether it could be used as a clinical index of body fat distribution in children and adolescents. We statistically evaluated the residuals (or SDS) of WHR/ Ht to determine the normality of the distribution as a whole, and to compare the central values and distribution pattern in three different age subgroups by sex. The histograms of the SDS of WHR/Ht for the boys and girls are shown in the Figures 4 and 5. The normality of the distribution of the residuals was accepted at P ˆ for the girls, but was rejected at P ˆ for the boys as determined by the Shapiro±Wilk test. The central values did not differ signi cantly between the three age subgroups for each gender as determined by Kruskal±Wallis and Median tests. Each pair of data from the different age groups showed no Figure 3 Correlation of WHR/Ht with age in girls (n ˆ 549). The r indicates the Pearson's correlation coef cient. The relationship between the two variables is very close. signi cant difference in distribution from each other as determined in the boys and girls by Kolmogorov± Smirnov two-sample test. Thus, the uniformity of the central values and distribution in three different subgroups for each gender supported the use of the common SD for the calculation of SDS for WHR/Ht. Figure 4 Histogram of the common SDS of WHR/Ht in boys. The distribution slightly deviates from normal (see text), but does not look highly skewed. 597 Figure 2 Correlation of WHR/Ht with age in boys (n ˆ 508). WHR is divided by Ht (m). Pearson's correlation coef cient is indicated by r. The relationship between the two variables is very close. Figure 5 Histogram of the common SDS of WHR/Ht in girls. The distribution does not signi cantly deviate from normal (see text).

5 598 Discussion The present study explored an index of body fat distribution that was adjusted for ages ranging from 6±15 y for each sex. The PCA successfully contrasted the size and shape variables in the present group of children and adolescents. WHR divided by Ht was considered to represent PC 2 adjusted by PC 1. This estimate gave an extremely robust linear regression by age for each sex. Further statistical tests of normality revealed that the distribution of the residuals for WHR/Ht was deviated somewhat from normality only in the boys, whereas the central values and the distribution were similar in the three age groups of each sex. Thus, the common SDS of the WHR/Ht was veri ed to be a potential measure of body fat distribution over age in the population studied. The contribution of the androgyny of fat patterning to metabolic derangement seen in childhood obesity is supported by the previous observation that metabolic derangement is more common at older ages and in boys, than at younger ages and in girls. 15 In a study of obese women, metabolic derangement was linked to the WHR, but not to the indices for overweight or adiposity. 6 In adult men, sagittal diameter and WHR were considered almost equally good indices for predicting metabolic risk. 16 A lipoprotein abnormality was better re ected in the BMI in nonobese, but in the WHR in the obese young adults in Taiwan. 17 In children, the WHR appears to be a useful index when the age and sex of the subjects studied are relatively uniform. Waist measurement was reported to correlate with a potentially atherogenic lipoprotein pro le in obese 12 and 14 y old children. 18 Improvement of the atherogenic risk-factor pro le during weight reduction was more marked in obese adolescent girls with a higher than, in those with lower, WHR. 19 In cross-sectional studies that dealt with both children and adolescents in one group, 20 WHR per se did not appear to be as good an index as observed in adults. 6,16,17 WHR was reported to be higher in boys than in girls throughout pediatric ages, and to decrease gradually in boys with age, while there were two sharp reductions at ages 10±13 y and 13±16 y in girls because of the marked increase in hip circumference at 13 y but a decrease in waist circumference from 13± 17 y. 12 These results indicate the need for establishing an age-adjusted estimate of body fat distribution in the pediatric age group. There is some debate whether WHR measures the amount of visceral adipose tissue. 2 Ross et al 21 reported that in adult men and women the amount of visceral adipose tissue, as measured by magnetic resonance image (MRI), analysis was greater at 10± 15 cm above than at the level of lumbar spine 4 and 5 and that this level corresponded to the level of the last rib. They thought that waist circumference measured at this level was a better predictor than that measured at the level of umbilicus. Likewise, the amount of intra-abdominal fat measured by MRI did not correlate with WHR in obese and nonobese adolescents. 3 Unlike the ndings in adults, the amount of intraabdominal fat in children was much less than that of subcutaneous fat. 3 Thus, the measurement of visceral adipose tissue by MRI has its limitations, and it may be relatively imprecise for studying children. 19 Even a study in adults failed to demonstrate a relationship between the amount of visceral adipose tissue and serum lipid levels. 22 Armellini et al 5 reported that the WHR correlated with resting metabolic rate in obese women, but that the amount of visceral adipose tissue did not. These results suggest that the WHR is a unique index of body fat distribution that re ects individual metabolic activity, and is not just an alternative to measurement of visceral fat by an imaging technique. In the present study subjects, the ratio of WHR to Ht was expressed as an extremely robust linear function of age for each sex. This implies that Ht and age are the two major determinants of normal value of WHR in Japanese children with ages ranging from 6± 15 y. Thus, change in the normal value of WHR by age in each sex appears to largely depend on the factor of linear growth. The SDS of the WHR/Ht would be useful in performing cross-sectional studies dealing with body fat distribution in children and adolescents. Since the population used here is comprised of children with various ages in one group, this estimate should be equally good for individuals with any given age in this range. Thus, the SDS of the WHR/Ht is also suitable for longitudinal study. This new anthropometric index can help to resolve a problem of limited applicability of WHR to the study of children and adolescents. The usefulness of this new ratio has to be veri ed with other groups of children originating from different races. Further study will elucidate the physiological signi cance of the SDS of the WHR/Ht, and the relevance of abdominal obesity in children to the risk of atherosclerosis. References 1 Wattigney WA, Harsha DW, Srinivasan SR, Webber LS, Berenson GS. Increasing impact of obesity on serum lipids and lipoproteins in young adults: The Bogalusa Heart Study. Arch Intern Med 1991; 151: 2017± Tornaghi G, Raiteri R, Pozzato C, Rispoli A, Bramani M, Cipolat M, Craveri A. Anthropometric or ultrasonic measurement in assessment of visceral fat? A comparative study. Int J Obes 1994; 18: 771± Brambilla P, Manzoni P, Sironi S, Del Maschio A, di Natale B, Chiumello G. Peripheral and abdominal adiposity in childhood obesity. Int J Obes 1994; 18: 795± Krotkiewski M, BoÈrntorp P, SjoÈstroÈm L, Smith U. Impact of obesity on metabolism in men and women: Importance of regional adipose tissue distribution. J Clin Invest 1983; 72: 1150± Armellini F, Robbi R, Zamboni M, Tedesco T, Castelli S, Bosello O. Resting metabolic rate, body fat distribution, and visceral fat in obese women. Am J Clin Nutr 1992; 56: 981± Pedersen SB, Bùrglum JD, Schmitz O, Bak JF, Sùrensen NS, Richelsen B. Abdominal obesity is associated with insulin

6 resistance and reduced glycogen synthase activity in skeletal muscle. Metabolism 1993; 42: 998± Zwiauer KFM, Widhalm K, Kerbl B. Relationship between body fat distributions and blood lipids in obese adolescents. Int J Obes 1990; 14: 271± Zwiauer KFM, Pakosta R, Mueller T, Widhalm K. Cardiovascular risk factors in obese children in relation to weight and body fat distribution. J Am Coll Nutr 1992; 11: 41S±51S. 9 Zonderland ML, Erich BM, Erkelens DW, Kortland W, Wit JM, Huisveld IA, de Ridder CM. Plasma lipids and apoproteins, body fat distribution and body fatness in early pubertal children. Int J Obes 1990; 14: 1039± Gillum RF. The association of the ratio of waist and hip girth with blood pressure, serum cholesterol and serum uric acid in children and youth aged 6±17 years. J Chron Dis 1987; 40: 413± Deurenberg P, Pieters JJL, Hautvast JGAJ. The assessment of the body fat percentage by skinfold thickness measurements in children and young adolescents. Br J Nutr 1990; 63: 293± Martinez E, Devesa M, Bacallo J, Amador M. Percentiles of waist-hip ratio in Cuban scholars aged 4.5 to 20.5 years. Int J Obes 1994; 18: 557± Croft JB, Keenan NL, Sheridan DP, Wheeler FC, Speers MA. Waist-to-hip ratio in a biracial population: Measurement, implications, and cautions for using guidelines to de ne high risk for cardiovascular disease. J Am Diet Assoc 1995; 95: 60± Asayama K, Hayashibe H, Dobashi K, Uchida N, Kawada K, Nakazawa S. Relationships between biochemical abnormalities and anthropometric indices of overweight, adiposity and body fat distribution in Japanese elementary school children. Int J Obes 1995; 19: 253± Deutsch MI, Mueller WH, Malina RM. Androgyny in fat patterning is associated with obesity in adolescents and young adults. Ann Hum Biol 1985; 12: 275± Richelsen B, Pedersen SB. Associations between different anthropometric measurements of fatness and metabolic risk parameters in non-obese, healthy, middle-aged men. Int J Obes 1995; 19: 169± Ding YA, Chu NF, Wang TW, Lin CC. Anthropometry and lipoproteins-related characteristics of young adult males in Taiwan. Int J Obes 1995; 19: 392± Flodmark CE, Sveger T, Nilsson-Ehle P. Waist measurement correlates to a potentially atherogenic lipoprotein pro le in obese 12±14-year-old children. Acta Paediatr 1994; 83: 941± Wabitsch M, Hauner H, Heinze E, Muche R, Boeckmann A, Parthon W, Mayer H, Teller W. Body-fat distribution and changes in the atherogenic risk-factor pro le in obese adolescent girls during weight reduction. Am J Clin Nutr 1994; 60: 54± Moussa MAA, Skaik MB, Selwanees SB, Yaghy OY, Bin- Othman SA. Contribution of body fat and fat pattern to blood pressure level in school children. Eur J Clin Nutr 1994; 48: 587± Ross R, Shaw KD, Rissanen J, Martel Y, de Guise J, Avruch L. Sex differences in lean and adipose tissue distribution by magnetic resonance imaging: Anthropometric relationships. Am J Clin Nutr 1994; 59: 1277± Leenen R, van der Kooy K, Droop A, Seidell JC, Deurenberg P, Weststrate JA, Hautvast JG. Visceral fat loss measured by magnetic resonance imaging in relation to changes in serum lipid levels of obese men and women. Arterioscler Thromb 1993; 13: 487±