International Journal of (2000) 24, 1011±1017 ß 2000 Macmillan Publishers Ltd All rights reserved 0307±0565/00 $15.00 www.nature.com/ijo The paradox of low and high body fat percentage among Chinese, Malays and Indians in Singapore

M Deurenberg-Yap1*, G Schmidt2, WA van Staveren3 and P Deurenberg3,4

1Department of Nutrition, Ministry of Health, Singapore; 2School of Physical Education, Nanyang Technological University, Singapore; 3Department of Nutrition and , Wageningen Agricultural University, Wageningen, The Netherlands; and 4Nutrition Consultant, Singapore

OBJECTIVE: To study the relationship between body fat percentage and body mass index (BMI) in three different ethnic groups in Singapore (Chinese, Malays and Indians) in order to evaluate the validity of the BMI cut-off points for obesity. DESIGN: Cross-sectional study. SUBJECTS: Two-hundred and ninety-one subjects, purposively selected to ensure adequate representation of range of age and BMI of the general adult population, with almost equal numbers from each ethnic and gender group. MEASUREMENTS: Body weight, body height, sitting height, wrist and femoral widths, skinfold thicknesses, total body water by deuterium oxide dilution, densitometry with Bodpod1 and bone mineral content with Hologic1 QDR- 4500. Body fat percentage was calculated using a four-compartment model. RESULTS: Compared with body fat percentage (BF%) obtained using the reference method, BF% for the Singaporean Chinese, Malays and Indians were under-predicted by BMI, sex and age when an equation developed in a Caucasian population was used. The mean prediction error ranged from 2.7% to 5.6% body fat. The BMI=BF% relationship was also different among the three Singaporean groups, with Indians having the highest BF% and Chinese the lowest for the same BMI. These differences could be ascribed to differences in body build. It was also found that for the same amount of body fat as Caucasians who have a body mass index (BMI) of 30 kg=m2 (cut-off for obesity as de®ned by WHO), the BMI cut-off points for obesity would have to be about 27 kg=m2 for Chinese and Malays and 26 kg=m2 for Indians. CONCLUSIONS: The results show that the relationship between BF% and BMI is different between Singaporeans and Caucasians and also among the three ethnic groups in Singapore. If obesity is regarded as an excess of body fat and not as an excess of weight (increased BMI), the cut-off points for obesity in Singapore based on the BMI would need to be lowered. This would have immense public health implications in terms of policy related to obesity prevention and management. International Journal of Obesity (2000) 24, 1011±1017

Keywords: body fat percentage; body mass index; body build; obesity; cut-off values; Singaporeans; Chinese; Malays; Indians; Caucasians; race; ethnicity; public health; four-compartment model

Introduction impossible or too prohibitive in terms of cost and risk. Indirect methods that are commonly used include The WHO has de®ned obesity as a condition with densitometry, dilution techniques and dual energy X- excessive fat accumulation in the body to the extent ray absorptiometry (DEXA). Each of these methods that health and well-being are adversely affected.1 has its own advantages and limitations.2±4 These Implicit in this de®nition is the need to be able to techniques depend on the use of equations to calculate accurately measure the amount of body fat and the BF% from the measured body parameter, and such level at which disease risk is increased. equations were developed mainly in the `normal' Indirect and doubly indirect methods have been Caucasian population and based on certain assump- developed to measure body fat percentage (BF%) tions. Densitometry, using underwater weighing or for most studies, as direct methods (cadavar studies more recently air-displacement,5 is a classical and in vivo neutron activation analysis) are either method, long regarded as a method of reference. Dilution techniques, for example deuterium oxide dilution, depend on the assumption that hydration of 6 *Correspondence: M Deurenberg-Yap, Department of Nutrition, fat-free mass (FFM) is ®xed and constant. DXA Ministry of Health, Level 5, Institute of Health, 3, Second Hospital scanning has been found to be highly valid for bone Avenue, Singapore 168937. mineral density, but unfortunately not as good for E-mail: [email protected] Received 24 September 1999; revised 5 January 2000; accepted determination of BF%. Varying body sizes could 7 April 2000 affect the accuracy of DXA in the measurement of Body fatness in Singapore M Deurenberg-Yap et al 1012 BF%.7 There are also differences between machines8 by WHO may not adequately re¯ect the actual obesity and even between different models,9 rendering the status and thus the discase risk of Singaporeans.19,20 standardization of methodology dif®cult. However, no adequate information on the relationship For best and least biased information, more-com- between BMI and BF% was available for Singaporeans. partment models in which the variations in water For this reason a study was content and mineral content in the fat-free mass are performed to study the relationship between BMI accounted for,10 should be used. However, the com- and BF% in the three ethnic groups (Chinese, bination of techniques required makes this method Malays and Indians) in Singapore. To avoid systema- expensive and not suitable for use in large epidemio- tic biases in the determination of BF% due to violation logical studies. Doubly indirect methods (predictive of assumptions in one or more ethnic groups, BF% methods) are normally employed for large population was determined using a four-compartment model studies as they are generally affordable, easy to per- based on densitometry, deuterium dilution and dual form and require minimal equipment and specialized energy X-ray absorptiometry (DXA) measurements.21 manpower. The body mass index (BMI), de®ned as weight=height squared (kg=m2), is one such method that is commonly used as a surrogate measure for BF%. BMI is generally well correlated with BF% and is a good indicator of level of `risk'.11 Cut-off points Subjects and methods for obesity as de®ned by the WHO1 are based on BMI-values, but these cut-off points are based on During the 1998 National Health Survey, 300 volun- studies on the relationship between BMI and morbid- teers were invited to participate in a body composition ity and mortality in western populations 12,13 and it study. The inclusion criteria for selection were that the may be questionable whether they are valid in other subjects need to be distributed over the whole range of populations. In recent years several studies have shown age and BMI of the main study, with almost equal a different relationship between BMI and BF% among numbers in each ethnic (Chinese, Malay, Indian) and ethnic groups. For example Wang et al 14 in a study in sex group. Of the ®nal participants that underwent all New York, found that Asians had a lower mean BMI measurements, 108 were Chinese, 76 were Malays but a higher BF% than Caucasians of the same age and 107 were Indians. Their age ranged from 18 to and sex. Guricci et al 15 reported that Indonesians had 75 y and their body mass index from 16 to 40 kg=m2. for the same BF% a BMI about 3 units lower than Table 1 gives some characteristics of the subjects. All Dutch Caucasians. On the other hand, Gallagher measurements were performed at the study site situ- et al 16 did not ®nd differences in the relationship ated at the laboratory of the School of Physical between US Whites and Blacks. In a meta-analysis of Education, Nanyang Technological University, Singa- available data from the literature it was clearly shown pore. Subjects fasted from food and drinks at least 6 h that differences in the BMI=BF% relationship exists and voided preceding the measurements. All anthro- among ethnic groups.17 In Singapore, the mean BMI pometric measurements were performed by trained is low compared to most Western countries, but the observers. The Singapore National Medical Research mortality from cardiovascular diseases is similar to Council approved the study protocol and all subjects these countries.18 In an earlier study in Singaporean gave their informed consent before the measurements. Chinese it was found that the odds ratio of having Body weight was measured to the nearest 0.1 kg in cardiovascular risk factors was high at low BMI light indoor clothing without shoes, using a digital levels.19 The above mentioned studies raised the scale. A correction of 0.5 kg was made for the weight suspicion that the BMI cut-off for obesity as de®ned of the cloths. Body height was measured without

Table 1 Characteristics of the subjects (meanÆ s.d.)

Chinese Malays Indians

Females Males Females Males Females Males

Mean s.d. Mean s.d. Mean s.d. Mean s.d. Mean s.d. Mean s.d.

Age (y) 36.3 12.8 40.7 13.6 35.6 13.9 41.4 12.3 36.6 10.1 43.4 12.8 Weight (kg) 54.7 11.1 65.0 10.8 58.1 11.5 69.0 12.4 61.2 13.9 69.8 11.9 Height (cm) 157 5.9 169 5.2 154 6.1 166 6.4 157 6.0 170 6.9 BMI (kg=m2) 22.1 4.8 22.8 3.5 24.5 4.8 25.0 3.7 24.9 5.2 24.2 3.6 BF%4c 33.4 7.5 24.4 6.1 37.8 6.3 26.0 7.5 38.3 6.9 28.0 5.4 BF%BMI 29.5 7.4 20.5 5.7 32.2 7.9 23.3 6.0 32.9 7.6 22.9 5.5 BF%skfd 33.3 6.3 24.4 6.3 35.8 6.4 26.2 7.6 35.8 5.6 27.3 6.0 Rel SH 0.54 0.15 0.54 0.15 0.53 0.14 0.54 0.13 0.53 0.12 0.52 0.15 Slenderness 11.5 0.8 10.9 0.8 11.3 0.9 11.2 0.5 10.9 0.4 11.2 0.5

BMI, body mass index; BF%4C, body fat percentage from four-compartment model; BF%BMI, body fat percentage predicted from BMI; BF%skfd, body fat percentage predicted from skinfolds; Rel SH, relative sitting height (sitting height=standing height); slenderness, height=sum (wrist ‡ knee diameter).

International Journal of Obesity Body fatness in Singapore M Deurenberg-Yap et al 1013 shoes with Frankfurt plane horizontal, to the nearest Lunar DPXL (software version 1.35) was determined. 0.1 cm using a wall-mounted stadiometer. Sitting The found correction factor of 1.167 for the phantoms height was measured with a wall-mounted stadiometer was con®rmed by three sets of measurements per- with the subject sitting on a small stool with a ¯at and formed on two subjects over a period of 1 y. For each hard sitting board with the lower legs hanging. The set these two subjects were measured twice within 1 height of the stool was subtracted from the reading. week with both systems. Total body mineral was Relative sitting height was calculated as sitting calculated as 1.235ÂBMC.21,25 Body fat percentage height=standing height. A low relative sitting height was calculated using the four-compartment model as indicates relatively long legs. From weight and height described by Baumgartner et al,21 the body mass index (BMI) was calculated as BF% ˆ 100  2:75  BV 0:714  TBW weight=height squared (kg=m2). Body fat percentage was predicted from BMI, age and sex with a formula ‡ 1:148  M 2:05  BW†=weight developed and validated in Caucasians:22 BF% ˆ 1:2  BMI ‡ 0:23  age 10:8  sex 5:4 where BV ˆ body volume, M ˆ total body mineral and BW ˆ body weight. Fat free mass (FFM kg) was Wrist width was measured with an anthropometric calculated as body weight minus fat mass. Data were calliper at the left and the right sides over the distal analysed using the SPSS for Windows program.26 ends of the radius and the ulna to the nearest 0.1 cm. Correlations are Pearson's correlation coef®cients or Knee width was measured to the nearest 0.1 cm, at the partial correlation with correction for possible con- left and right sides in the sitting position, lower legs founders. The relationship between BMI and BF% relaxed with the knee ¯exed at a 90 angle over the was analysed using stepwise multiple regression, with femur condyles. The mean values of left and right age, sex (females ˆ 0 and males ˆ 1) and ethnicity as widths were used in the statistical calculations. A independent variables. The dummy variables for eth- 20 parameter for slenderness was calculated as nicity were E1 and E2. For Chinese E1 ˆ 0 and E2 ˆ 1, height=(sum of wrist and knee widths, cm=cm). for Malays E1 and E2 ˆ 0, and for Indian E1 ˆ 1 and Obviously, a higher index indicates a more slender E2 ˆ 1. Parallelism of the regression equations for the body build or frame. sexes and the ethnic groups was tested using interac- Biceps, triceps, subscapular and supra-iliac skin- tion factors between the independent variables.27 folds were measured in triplicate to the nearest Differences between the sexes and differences 0.2 mm, on the left side of the body, according to between the ethnic groups were tested using analyses Durnin and Womersley.23 The mean value of each of (co)variance. Bland and Altman statistics28 were measurement was used in the prediction of BF%.23 used to test the bias of predicted BF%. Differences in For the determination of total body water (TBW) the variables within groups were tested with the paired t- subject drank a precisely weighed amount of deuter- test. Level of signi®cance is set at P < 0.05. Values ium oxide (amount given varied between 10 and 11 g). are presented as meanÆ s.d., unless otherwise stated. After 3 h, 10 ml venous blood was taken and plasma was obtained and stored in well-sealed tubes at 720 C until analyses. Deuterium was analysed after sublimation of the plasma using infrared spectro- Results scopy.24 From the given dose and the deuterium concentration in plasma, TBW was calculated, assum- Table 1 gives characteristics for the males and females ing a 5% non-aqueous dilution of the deuterium in the of each ethnic group. Normal differences between body.2 males and females were observed, males being taller Body density was determined using air plethysmo- and heavier, and having a lower body fat percentage. graphy (BODPOD1, Body composition System, Life Age did not differ signi®cantly between the groups. Measurements Instruments, Concord, CA) according Indian females had signi®cantly higher body weight to the instructions of the manufacturer. The method is than Chinese females. Body height of Malays males described in detail by Dempster and Aitkens.5 Body and females was shorter than that of their Chinese and volume was calculated as weight=density. Indian counterparts. The BMI of Chinese females was Bone mineral content (BMC) was measured using a lower than that of Malays and Indian females, but in Hologic DXA whole body X-ray densitometer (QDR- males only the BMI from Chinese was signi®cantly 4500, Hologic, Waltham, MA; software version lower than that of Malays. BF% was lower in Chinese V8.23a:5). As Hologic measurements generally result females compared to Malays and Indian females, but in systematic lower BMC measurements compared in males the difference in BF% was only between to Lunar measurements,8 the BMC data were cor- Chinese and Indians. Body fat predicted from BMI rected to Lunar values. This was found to be neces- using a Caucasian prediction equation was signi®- sary, as a Lunar machine was used for development cantly (P < 0.001) underestimated in all subgroups. of the equation of Baumgartner's four-compartment This shows that Singaporeans have higher body fat model.21 A correction factor based on phantom mea- at the same BMI compared with Caucasians. Body surements (Lunar aluminium `spine' phantom) using a fat predicted from skinfolds was only signi®cantly

International Journal of Obesity Body fatness in Singapore M Deurenberg-Yap et al 1014 (P < 0.001) underestimated in Malay and Indian relationship between BF% and BMI) resulted in a females. Relative sitting height was signi®cantly lower (72% instead of 74%) explained variance. The lower in the Indian males and females compared to residuals of predicted BF% in the different population their Chinese and Malay counterparts. In males the groups and the correlation with the level of BF% are difference between Malays and Chinese was also given in Table 3. The bias was in neither group statistically signi®cant. Malay males and females signi®cant different from zero, but in all groups the had a signi®cant higher slenderness index than correlation with the level of BF% was high. For the Chinese and Indians. The correlation between BMI total population a Bland and Altman plot is given in and BF% was 0.75, 0.72 and 0.76 in Chinese, Malay Figure 1. and Indian females, respectively, and 0.76, 0.78 and The residuals of a regression model without E1 and 0.68 in Chinese, Malays and Indian males, respec- E2 showed a signi®cant partial correlation (after tively (all values P < 0.001). correction for the level of BF%) with slenderness Table 2 gives the regression coef®cients (s.e.) of the (part r ˆ 0.318, P < 0.001) and with relative sitting stepwise multiple regression in the order the indepen- height (part r ˆ 70.191, P < 0.001). dent variables entered the equation. The ®nal predic- From age and sex of the individual Singaporean tion equation, subjects it was calculated, using the Caucasian pre- diction equation, how high their BF% would be if BF% ˆ 1:04  BMI 10:9  sex ‡ 0:1  age their BMI would have been 30 kg=m2, ie the cut-off ‡ 2:0E1 ‡ 1:5  E2 ‡ 5:7 point for obesity as de®ned by WHO. This BF% level was used to recalculate their BMI using the Singapor- explains 74% of the variation in BF% when the ean prediction equation to obtain a BMI level that is independent factors are controlled for, and has a equivalent with the WHO cut-off point for obesity in standard error of estimate of 4.4% body fat. With Caucasians. Figure 2 shows that Singaporeans with a 2 BMI, age, sex and E1 and E2 in the model there was a BMI of about 27 kg=m have the same level of body non-signi®cant interaction between BMI and sex fatness as Caucasians with a BMI of 30 kg=m2. There (P ˆ 0.065), showing that the regression lines are are slight differences between Chinese and Malays but slightly but not signi®cantly different for the sexes. the difference with Indians is more pronounced. The No interaction between E1 and E2 with BMI or age BMI value in Singaporeans that corresponds to the was observed. However there was a signi®cant inter- same BF% in Caucasians with a BMI of 25 kg=m2 in 2 action between E2 and sex, showing that in the Malays about 21 kg=m . the regression coef®cient for sex was slightly higher compared with the Chinese and Indian females. Taking this interaction into account resulted in only a very minor improvement of the regression equation Discussion (change in s.e.e. 0.03%). Therefore it was decided to continue analyses without this interaction. Regression The subjects participating in this study were a selected analysis using BMI-squared (assuming curvilinear group of Singaporean Chinese, Malays and Indians,

Table 2 Regression coef®cients (s.e.) of the stepwise multiple regression of body fat percentage as dependent variable

2 BMI (kg=m ) Sex Age (y) E1 E2 Intercept b s.e. b s.e. b s.e. b s.e. b s.e. b s.e. r 2 s.e.e. (%)

1.11 0.09 Ð Ð Ð Ð Ð Ð Ð Ð 4.7 2.2 0.34 6.99 1.16 0.06 7 10.3 0.5 Ð Ð Ð Ð Ð Ð 8.8 1.5 0.71 4.63 1.07 0.06 7 10.8 0.5 0.09 0.02 Ð Ð Ð Ð 7.2 1.5 0.73 4.48 1.03 0.06 7 11.0 0.5 0.10 0.02 1.4 0.6 Ð Ð 7.2 1.5 0.73 4.44 1.04 0.06 7 10.9 0.5 0.10 0.02 2.0 0.6 1.5 0.7 5.7 1.6 0.74 4.42

BMI, body mass index; sex, females ˆ 0, males ˆ 1; age, in years; E1 and E2 ethnicity Ð Chinese E1 ˆ 0 and E2 ˆ 0; Malays E1 ˆ 1 and 2 E2 ˆ 0; Indian E1 ˆ 1 and E2 ˆ 1; R explained variance; s.e.e. standard error of estimate.

Table 3 Bias of predicted BF% from BMI, age, sex and ethnicity in the different subgroups and the correlation of the bias with the level of body fat percentage

Females Males

Chinese Malays Indian Chinese Malays Indian

Bias (%)a 70.3 4.7 1.1 4.2 70.3 4.9 0.4 3.7 70.9 4.8 0.3 3.7 Correlation of bias with BF%b 0.66 0.43 0.54 0.75 0.83 0.66

BF%, body fat percentage, BMI, body mass index (kg=m2); bias, measured BF% minus predicted BF%. aBias in all subgroups not different from zero. bAll correlations signi®cant (P < 0.001).

International Journal of Obesity Body fatness in Singapore M Deurenberg-Yap et al 1015 population. This sample selection process was impor- tant for the purpose of this study, which was to study the relationship between BMI and BF%. There are apparent differences in height, weight, BMI and body fat percentage between the three ethnic groups, the Malays being the shortest, and the Chinese being the lightest. The BMI of the Chinese is 2 ± 3 units lower and their BF% is about 3% lower com- pared to those of their Malays and Indian counterparts. Compared to other population-based body compo- sition studies the BMI is relatively low in comparison with BF%. For example Lean et al 29 reported in Scottish males and females of the same age a BMI of about 25 kg=m2, whereas BF% was only 22% in males and 34% in females. In a study by Gallagher et al,16 BF% in 50-y old Caucasians females was 30% whereas their BMI was as low as 23.3 kg=m2. Males in that study had a higher BMI (25 kg=m2) but a rela- tively low BF% of 21%. Also the under-prediction of BF% from BMI using a Caucasian (Dutch) prediction formula shows that Singaporeans have a high BF% at a relatively low BMI.22 Applying the Caucasian regression equation from the meta-analysis17 even resulted in slightly higher differences between mea- sured and predicted BF% (results not shown). This con®rms earlier ®ndings in study of Wang et al,14 showing that Asians have lower BMI but higher BF% than Caucasians. A different relationship between BMI and BF% in Asians compared to Caucasians has been recently reported in several studies among Indonesian population groups,15,30 in Thai popula- tions,31,32 in Japanese (Gallagher et al, personal com- Figure 1 Bias of predicted body fat percent from body mass index, age, sex and ethnicity in Singaporean males and females. munication) as well as in young Singaporean Chinese.20 It has to be noted that such a different relationship was not reported in Beijing Chinese compared to Dutch Caucasians.20,33 Differences have also been observed between different Black popula- tions34 and between Caucasians and Polynesians.35 The present study con®rms that the relationship between BMI and BF% in Singaporeans is not only different compared to Caucasians, but is also different among the three main ethnic groups in Singapore: Chinese, Malays and Indians. For the same BMI, age and sex, Chinese have the lowest BF% while Indians have the highest. One possible drawback of earlier studies reported in the literature, when comparing different (ethnic) groups, is the possible limitation of the reference methods used to determine BF%. For example in the Figure 2 Calculated BMI cut-off point for obesity for Singapore 15,30 Chinese, Malays and Indians based on the WHO obesity cut oof studies of Guricci et al, deuterium oxide was point for Caucasians. used as the reference method to determine BF%. It was assumed that all the ethnic groups had the same constant hydration factor of fat-free mass aged 18 ± 75 y. They were purposely selected from a (hydration ˆ 0.73).2,6 It can not be excluded that larger representative population sample of the there might be differences in the hydration of the National Health Survey 1998 (NHS), to ensure ade- fat-free mass (FFM) among different ethnic groups, quate representation from the entire range of age and which could be attributed to, for example, climatic BMI, with almost equal numbers from all the ethnic conditions, which could cause a directional bias in the and gender groups, rather than having similar age, results and thus lead to the wrong conclusions. Also BMI, ethnic and gender distribution to the general the differences as shown in a meta-analysis17 of

International Journal of Obesity Body fatness in Singapore M Deurenberg-Yap et al 1016 reported literature data could be due to such metho- equals the currently used cut-off point in Indonesia dological differences. For this reason, in the present (with a predominantly Malay population), the validity study, a multi-compartment model (body weight ˆ fat of which was con®rmed by studies of Guricci et al.15 mass ‡ water ‡ mineral ‡ protein) was used for the A lowering of the cut-off point for obesity would also determination of BF%, to avoid any bias due to be justi®ed if elevated cardiovascular risk were to be violation of assumptions that form the base for present at low BMI levels, as was recently found in single methods such as densitometry or deuterium Chinese Singaporeans.19 The presence of cardio- oxide dilution. Such a model was also used by vascular risk factors at different BMI levels will Gallagher et al 16 in their study comparing US be further studied among the three ethnic groups Blacks and Whites. Because of the use of this four- in Singapore to test this hypothesis (paper in compartment model it can be assumed that the differ- preparation). ences found among the four ethnic groups (namely Generally, if the cut-off point for obesity in Singa- Caucasians and the three ethnic groups in Singapore) pore were lowered to 27 kg=m2, this would have discussed in this paper are not due to violation of immense impact on the prevalence of obesity among assumptions in the body composition methodology. the adult Singapore population. Compared to a BMI The differences in the relationship between BMI cut-off point of 30 kg=m2 the prevalence would and BF% that were found between the three Singa- increase in females from 6.5% to 15.4% and in porean ethnic groups can be ascribed to differences in males from 5.2% to 17.3%. These higher prevalences body build. The residuals of a regression equation are more in line with the relatively high level of without E1 and E2 show a signi®cant partial cor- chronic degenerative diseases in Singaporeans and relation (after correction for the level of BF%) with their increased relative risk for cardiovascular with slenderness and relative sitting height. Also, risk factors at lower BMI levels.19 It is unnecessary to in a stepwise multiple regression model with BMI, say that such an increase in prevalence ®gures would age, sex, relative sitting height and slenderness as have serious implications in terms of public health independent variables, ethnicity did not contribute policy. signi®cantly anymore. This con®rms the ®ndings in In summary, Singaporeans have higher body fat earlier studies20,30 that body build is at least partly percentage at a lower BMI compared to Caucasians, responsible for the different relationship between BMI but differences in the BMI=BF% relationship also exist and BF%. A stocky person (low slenderness index) is among Singaporean Chinese, Malays and Indians. likely to have more bone, connective tissue and The differences can be explained by differences in muscle mass, thus less body fat for a given body body build. If obesity is de®ned as excess body fat height and weight than a more slender person. In other rather than excess weight, the obesity cut-off point words, BF% will be relatively low compared to the for Singaporeans should be 27 kg=m2 instead of BMI. Subjects with relatively long legs (low relative 30 kg=m2. The lowering of the cut-off point for sitting height) have less mass per unit length, so their obesity would more than double the prevalence ®g- BMI will be relatively low, compared to their BF%. ures in Singapore. The effect of relative sitting height has been discussed earlier by Norgan.36,37 Other authors34 discussed as possible explanation for differences in the BMI=BF% Acknowledgements relationship differences in physical activity level. No The participation of the subjects in this study is highly valid information was available about the physical appreciated. Special thanks are due to Frans JM activity level of the subjects in this study. Data from Schouten (Wageningen Agricultural University) who the 1998 National Health Survey38 indicated that most analysed the plasma samples for deuterium and to Ms Singaporeans have a sedentary life style, engaging in Violette Lin and Mr Eddy Chong for performing the less than 1.5 h of physical activity per week. This ®eld measurements in Singapore. The authors also could be due to a combination of climate (warm and wish to document their appreciation to the Epidemio- humid), living and working conditions and well-orga- logy and Disease Control Department (Ministry of nized and highly effective public transport, all factors Health) for providing data from the National Health which are disincentives for any sort of physical Survey 1998 and to the Department of Nutrition activity or during leisure time other than in (Ministry of Health), for administrative and logistic air-conditioned rooms. support. The study was supported by the National If obesity is de®ned as an excess body fat, it seems Medical Research Council in Singapore and also in logical that a different relationship between BMI and part by DS MediGroup, Spa, Milan. BF% among populations would result in population- speci®c BMI cut-off points for obesity, rather than a uniform cut-off point as is currently recommended by WHO.1 For Singaporeans this would mean that the References 1 WHO. Obesity: Preventing and managing the global epi- BMI cut-off points for obesity should be about demic. Report of a WHO Consultation on Obesity, Geneva, 26 kg=m2 for Indians and about 27 kg=m2 for Chinese 3 ± 5 June, 1997, WHO=NUT=NCD=98.1. WHO: Geneva, and Malays. This cut-off point of 27 kg=m2 for Malays 1998.

International Journal of Obesity Body fatness in Singapore M Deurenberg-Yap et al 1017 2 Forbes GB. Human body composition. Springer: New York, 22 Deurenberg P, Weststrate JA, Seidell JC. Body mass index as 1987. a measure of body fatness: age and sex speci®c prediction 3 Lukaski HC. Methods for the assessment of human body formulas. Br J Nutr 1991; 65: 105 ± 114. composition: traditional and new. Am J Clin Nutr 1987; 46: 23 Durnin JVGA, Womersley J. Body fat assessed from total 537 ± 556. body density and its estimation from skinfold thickness: 4 Deurenberg P. The assessment of body composition: use and measurements on 481 men and women aged from 17 to misuse. Annual Report Nestle Foundation 1992; 35 ± 72. 72 years. Br J Nutr 1974; 32: 77 ± 97. 5 Dempster P, Aitkens S. A new air displacement method for the 24 Lukaski HC, Johnson PE. A simple, inexpensive method of determination of human body composition Med Sci Sports determining total body water using a tracer dose of D2O and Exerc 1995; 27: 419 ± 425. infrared absorption of biological ¯uids. Am J Clin Nutr 1985; 6 Wang ZM, Deurenberg P, Wang W, Pietrobelli A, Baumgart- 41: 363 ± 370. ner RM, Heyms®eld SB. Hydration of fat-free body mass: 25 Wang Z-M, Deurenberg P, Guo, SS, Pietrobelli A, Wang J, review and critique of a classic body composition constant. Am Pierson RN, Heyms®eld SB. Six-compartment body composi- J Clin Nutr 1999; 69: 833 ± 841. tion model: Inter-method comparison of total body fat mea- 7 Jebb SA. Measurement of soft tissue composition by dual surement. Int J Obes 1998; 22: 329 ± 337. energy X-ray absorptiometry. Br J Nutr 1997; 77: 151 ± 163. 26 SPSS=Windows, Version 8.01. SPSS: Chicago, 1998. 8 Tothill P, Avenell A, Reid DM. Precision and accuracy of 27 Kleinbaum DG, Kupper LL, Muller KE, Nizam A. Applied measurements of whole-body bone mineral: comparison regression analysis and other multivariate methods, 3rd edn. between Hologic, Lunar and Norland dual energy X-ray Duxbury Press: Paci®c Grove, 1998. absorptiometers. Br J Radiol 1994; 67: 1210 ± 1217. 28 Bland JM, Altman DG. Statistical methods for assessing 9 Paton N, Macallan D, Jebb S, Pazianas M, Grif®n G. Dual agreement between two methods of clinical measurements. energy X-rays absorptiometry results differ between machines. Lancet 1986; I: 307 ± 310. Lancet 1995; 346: 899 ± 900. 29 Lean MEJ, Han TS, Deurenberg P. Predicting body composi- 10 Heyms®eld SB, Wang Z, Baumgartner RN, Ross R. Human tion by densitometry from simple anthropometric measure- body composition: advances in models and methods. A Rev ments. Am J Clin Nutr 1996; 63: 4 ± 14. Nutr 1997; 17: 527 ± 558. 30 Guricci S, Hartriyanti Y, Hautvast JGAJ, Deurenberg 11 Bray GA. Clinical evaluation of the obese patient. Clin Endocr P. Differences in the relationship between body fat and & Metab 1999; 13: 71 ± 92. body mass index between two different Indonesian ethnic 12 Lew EA, Gar®nkel L. Variations in mortality by weight groups: the effect of body build. Eur J Clin Nutr 1999; 53: among 750 000 men and women. J Chron Dis 1979; 32: 468 ± 472. 563 ± 576. 31 Stevens NHC. The correlation and prediction of body 13 WHO. , nutrition and the prevention of chronic diseases. composition from dual energy X-ray absorptiometry and Report of a WHO study group. TRS 797. WHO: Geneva, anthropometric measurements in Thai women aged 30 1990. to 39 years. Thesis Mahidol University, Bangkok, Thailand, 14 Wang J, Thornton JC, Russell M, Burastero S, Heyms®eld SB, 1997. Pierson RN. Asians have lower BMI (BMI) but higher percent 32 Tanphainchitr V, Kulapongse S, Pakpeankitvatana R, Leela- body fat than do Whites: comparisons of anthropometric hagul P, Tamwiwat C, Lochya S. Prevalence of obesity and its measurements. Am J Clin Nutr 1994; 60: 23 ± 28. associated risks in urban Thais. In: Oomura Y et al (eds). 15 Guricci S, Hartriyanti Y, Hautvast JGAJ, Deurenberg P. Progress in Obesity Research John Libbey: London, 1990, pp Relationship between body fat and body mass index: differ- 649 ± 652. ences between Indonesians and Dutch Caucasians. Eur J Clin 33 Deurenberg P, Ge K, Hautvast JGAJ, Wang J. Body mass Nutr 1998; 52: 779 ± 783. index as predictor for body fat: comparison between Chinese 16 Gallagher D, Visser M, Sepulveda D, Pierson RN, Harris T, and Dutch adult subjects. Asia Paci®c J Clin Nutr 1997; 6: Heyms®eld SB. How useful is BMI for comparison of body 102 ± 105. fatness across age, sex and ethnic groups. Am J Epidemiol 34 Luke A, Durazo-Arvizzu R, Rotimi C, Prewitt E, Forrester 1996; 143: 228 ± 239. T, Wilks R, Ogunbiyi OL, Schoeller DA, McGee D, 17 Deurenberg P, Yap M, Van Staveren WA. Body mass index Cooper RS. Relation between BMI and body fat in and percent body fat: a meta analysis among different ethnic black population samples from Nigeria, Jamaica and groups. Int J Obes 1998; 22: 1164 ± 1171. the United States. Am J Epidemiol 1997; 145: 620 ± 18 WHO. World Health Statistics 1996. WHO, Geneva, 1998. 628. 19 Deurenberg-Yap M, Tan BY, Chew SK, Deurenberg P, 35 Swinburn BA, Craig PL, Daniel R, Dent DPD, Strauss BJG. Staveren van WA. Manifestation of cardiovascular risk factors Body composition differences between Polynesians and Cau- at low levels of body mass index and waist±hip ratio in casians assessed by bioelectrical impedance. Int J Obes 1996; Singaporean Chinese. Asia Paci®c J Clin Nutr 1999; 8: 20: 889 ± 894. 177 ± 183. 36 Norgan NG. Population differences in body composition 20 Deurenberg P, Deurenberg-Yap M, Wang J, Lin Fu Po, in relation to BMI. Eur J Clin Nutr 1994; 48(Suppl 3), Schmidt G. The impact of body build on the relationship S10 ± S27. between body mass index and body fat percent. Int J Obes 37 Norgan NG. Interpretation of low body mass indices: 1999; 23: 537 ± 542. Australian aborigines. Am J Physical Anthropol 1994; 94: 21 Baumgartner RN, Heyms®eld SB, Lichtman SM, Wang J, 229 ± 237. Pierson RN Jr. Body composition in elderly people: effect of 38 Ministry of Health. Report of the National Health Survey criterion estimates on predictive equations. Am J Clin Nutr 1998. Epidemiology and Disease Control Department: Singa- 1991; 53: 1345 ± 1353. pore, 1999.

International Journal of Obesity