European Journal of Clinical Nutrition (2013) 67, S79–S85 & 2013 Macmillan Publishers Limited All rights reserved 0954-3007/13 www.nature.com/ejcn

ORIGINAL ARTICLE The use of bioelectrical impedance analysis for in epidemiological studies

ABo¨ hm and BL Heitmann

BACKGROUND/OBJECTIVES: Bioelectrical impedance analysis (BIA) is a relatively simple, inexpensive and non-invasive technique to measure body composition and is therefore suitable in field studies and larger surveys. SUBJECTS/METHODS: We performed an overview of BIA-derived body fat percentages (BF%) from 55 published studies of healthy populations aged 6–80 years. In addition, the relationship between (BMI) and body composition is documented in the context of BIA as a good alternative to closely differentiate which composition of the body better relates to the risk of cardiovascular diseases (CVDs)and all-cause mortality. RESULTS AND CONCLUSIONS: BIA-estimated percentage of BF varies greatly with population and age. BIA-estimated BF% is directly and closely related to various health outcomes such as CVDs, which is in contrast to BMI where both high and low BMIs are associated with increased risk of developing chronic diseases. Studies, among others using BIA, suggest that low BMI may reflect low muscle and high BMI fat mass (FM). BIA-derived lean and FM is directly associated with morbidity and mortality. To the contrary, BMI is rather of limited use for measuring BF% in epidemiological studies.

European Journal of Clinical Nutrition (2013) 67, S79–S85; doi:10.1038/ejcn.2012.168 Keywords: bioelectrical impedance; percentage body fat; body mass index; mortality

INTRODUCTION , which are associated with hydration alterations of FFM, is a condition in which fat accumulates in the body1 and may not be sufficiently accurate in estimating BF, and that four- or the (BF%) is therefore the relevant measure three-compartment models are a better alternative under these 12 of obesity. In the 1980s, bioelectrical impedance analysis (BIA) was circumstances. However, in field studies, or surveys including introduced as a new method to be used for estimating body many subjects, clearly such advanced three- or four-compartment composition,2 and since then many studies have investigated its models are not useful and the simpler methods, such as validity—indeed, because then BIA has been widely applied for impedance, are needed. predicting body composition (for example, fat-free mass (FFM), A number of studies have measured BIA in random population total body water and BF) in healthy subjects with normal fluid samples to either derive reference values for BF and FFM or relate distribution,3–5 and the method is considered useful in relation to such body composition measures to subsequent development of estimating BF% in both epidemiological and clinical research.6 disease and/or mortality. Body mass index (BMI) measures the Moreover, Wells and Fewtrell7 described BIA as the ‘only predictive degree of relative and does not differentiate between 13,14 technique that estimates lean mass’. lean and fat mass. Previous studies suggest that this may be It is widely recognised that calculation of body composition part of the reason for the general finding of a U-formed measures from BIA requires population-specific equations, as also association between BMI and cardiovascular, as well as total, illustrated by the results by Deurenberg et al.8 examining validity mortality. The present review extracts estimations of average BF% of BIA among various European population groups where from published studies among healthy populations and discusses significant differences in biases for the prediction of BF% among the relationship between overweight and body composition in the participants from the European centres were reported. However, context of using BIA to measure body composition. relatively few studies have in fact developed their own specific equations, and suitable equations therefore often have to be looked for from other validation studies. In addition, BIA seems to APPLICATION OF BIA FOR BODY COMPOSITION be a good method to estimate BF% in healthy subjects with BIA has been used in large cohort studies, such as NHANES (USA), normal BF distribution;9 however, several studies9,10 have NUGENOB (EU) or MONICA (DK), to predict body composition (for suggested that this may not be the case in obese individuals, example, fat mass (FM) and BF%) in individuals, and has been where BIA tends to underestimate BF% (BF% 430%). Further, BIA found to be useful in large-scale epidemiological studies.15 measurements allow the determination of anatomic locations of Consequently, BIA has an important part in contributing to develop BF depositions (for example, central and peripheral), and thus are and compare body composition across populations, but a summary also applied to compare BF proportions.11 of results from such studies has rarely been presented. Table 1 gives It has also been argued that composition measurement an overview of such studies published in the period 1991–2009. instruments such as BIA that rely on constant body hydration However, comparison of BIA-derived body composition and thus do not regard health inequalities such as obesity and measures must be done with care, as BIA measures are dependent

Institute of Preventive Medicine, Research Unit for Dietary Studies, Frederiksberg Hospital, Frederiksberg, Denmark. Correspondence: Dr BL Heitmann, Institute of Preventive Medicine, Research Unit for Dietary Studies, Frederiksberg Hospital, Nordre Fasanvej 57, Hovedvejen, Entrance 5, Ground floor, DK-2000 Frederiksberg, Denmark. E-mail: [email protected] BIA and epidemiology ABo¨hm and BL Heitmann S80 Table 1. Percentage of BF% from BIA by gender and age from published studies across different populations

Study Population Country of Number of Gender Age Mean implementation subjects (years) BF% þ s.d.

Antal et al.28 Healthy, randomly selected, Hungary 1928 m 7 12.6±7.4 Caucasian schoolchildren 8 13.8±7.4 9 17.1±7.6 10 19.4±9.2 11 19.3±9.0 12 19.4±8.0 13 16.7±8.4 14 15.6±7.3 f 7 14.3±6.5 8 16.5±8.6 9 19.9±8.0 10 21.3±8.7 11 20.8±7.7 12 20.3±7.0 13 23.3±7.6 14 23.5±7.2

Sung et al.49 Healthy, randomly selected Hong Kong 14 842 m 6 17.3±4.9 Hong Kong schoolchildren 7 17.9±5.4 8 18.7±5.9 9 19.7±6.7 10 20.6±6.9 11 19.7±7.4 12 18.0±7.1 13 17.1±6.4 14 17.4±6.1 15 18.9±6.3 16 19.7±6.0 17 20.2±6.0 18 19.1±4.0 f 6 14.1±4.8 7 15.4±5.7 8 16.2±5.9 9 17.2±6.1 10 18.3±6.5 11 19.3±6.6 12 21.1±6.9 13 23.5±6.8 14 24.3±6.4 15 25.4±6.5 16 25.5±6.3 17 25.7±6.2 18 25.5±5.7

Chumlea et al.50 Healthy, randomly selected USA 2880 m 12–13.9 18.4±7.3 non-Hispanic white Americans 14–15.9 18.4±8.3 16–17.9 17.7±6.8 18–19.9 19.6±6.9 20–29.9 21.8±6.2 30–39.9 23.6±5.8 40–49.9 24.2±5.7 50–59.9 25.1±6.0 60–69.9 26.2±5.5 70–79.9 25.1±5.5 3277 f 12–13.9 24.8±9.7 14–15.9 29.1±6.5 16–17.9 30.7±6.9 18–19.9 30.8±7.9 20–29.9 31.1±7.5 30–39.9 33.0±8.5 40–49.9 35.4±6.9 50–59.9 37.3±7.1 60–69.9 36.9±6.9 70–79.9 35.9±6.9

USA 2348 m 12–13.9 19.5±8.9

European Journal of Clinical Nutrition (2013) S79 – S85 & 2013 Macmillan Publishers Limited BIA and epidemiology ABo¨hm and BL Heitmann S81

Table 1. (Continued )

Study Population Country of Number of Gender Age Mean implementation subjects (years) BF% þ s.d.

Healthy, randomly selected non-Hispanic black Americans 14–15.9 17.8±7.5 16–17.9 18.6±6.4 18–19.9 19.9±6.0 20–29.9 23.7±7.0 30–39.9 23.6±6.7 40–49.9 24.9±6.1 50–59.9 25.1±6.7 60–69.9 24.9±6.6 70–79.9 24.3±6.3 2606 f 12–13.9 26.9±8.8 14–15.9 30.9±8.0 16–17.9 32.6±8.5 18–19.9 33.3±8.7 20–29.9 35.5±7.5 30–39.9 38.0±7.7 40–49.9 39.4±7.0 50–59.9 40.0±7.5 60–69.9 39.8±6.9 70–79.9 38.5±6.7

Heitmann51 Healthy, randomly selected Denmark 1527 m 35 20.7 Caucasians 45 23.6 55 25.7 65 26.9 1467 f 35 26.2 45 30.2 55 33.8 65 35.7

Pichard et al.52 Healthy, randomly selected 1838 m 15–24 14.5±4.3 Caucasians 25–34 16.3±4.9 35–44 17.8±5.8 45–54 19.2±6.0 55–64 20.9±7.2 1555 f 15–24 24.5±4.2 25–34 24.6±4.7 35–44 25.2±5.1 45–54 25.9±5.5 55–64 30.1±5.8

Kyle et al.53 Healthy, randomly selected Switzerland and USA 3714 m 20–29 17.3±4.7 Caucasians 30–39 19.0±4.9 40–49 20.1±5.1 50–49 20.7±5.6 60–69 22.5±5.4 70–79 24.6±5.1 3199 f 20–29 26.3±5.1 30–39 26.1±5.5 40–49 26.9±5.7 50–49 29.3±5.6 60–69 32.6±6.6 70–79 35.9±5.7

Lahmann et al.54 Healthy, randomly selected Sweden 5464 f 45–49 28.8±4.9 Caucasian women 50–54 29.4±4.8 55–59 31.1±5.0 60–64 31.9±4.9 65–69 32.5±4.7 70–73 31.5±4.8

Nagaya et al.55 Healthy Japanese adults 12 287 m 30–34 21.7±5.0 35–39 21.0±4.7 40–44 21.2±4.5

& 2013 Macmillan Publishers Limited European Journal of Clinical Nutrition (2013) S79 – S85 BIA and epidemiology ABo¨hm and BL Heitmann S82

Table 1. (Continued )

Study Population Country of Number of Gender Age Mean implementation subjects (years) BF% þ s.d.

45–49 20.6±4.4 50–54 20.3±4.5 55–59 20.1±4.3 60–64 19.4±4.4 65–69 18.4±4.4 6657 f 30–34 25.6±5.7 35–39 25.9±5.2 40–44 26.7±5.2 45–49 27.0±5.2 50–54 27.5±5.2 55–59 27.7±5.3 60–64 27.6±5.1 65–69 27.4±5.5 Abbreviations: BF, body fat; BIA, bioelectrical impedance analysis; f, female; m, male.

on several factors such as, for instance, age, gender, ethnicity and the presence of medical conditions.16 In this regard, differences in limb-to-trunk length contribute greatly to variations in BIA for more or less the same body composition. Indeed, one study17 compared four groups of different ethnic identities, including the Aboriginals from Australia, and found that, except for the Aboriginals, the associations between body weight and resistance (impedance) were generally constant in the different ethnic groups, once height and age differences had been considered. Thus, this study indicated that the relationship between body size and body composition (total body water or FFM) after all may involve a certain universality that is independent of the population specificity for impedance measurement. Table 1 gives BF% data from European, Asian and US studies published previously, and shows that there are marked differences in BF% between groups of the same age and gender. Such differences are also seen for BMI. Figures 1 and 2 show the distribution of BF% with age in men and women from some of the larger cohorts published between 1991 and 2009. The figure shows that BF% generally is higher the higher the age, especially Figure 1. BF% distribution with age in women from published within the European population, whereas the Japanese and US cohort studies: Denmark;51 Switzerland;53 USA;50 Japan;55 and Hong populations tend to level off above the age of 60. between the Kong.49 For illustration of BF% from data of Sung et al.,49 average of Asian and European populations—with higher BF% values for the BF% was calculated for subjects aged 14–15.9, 16–17.9 and 18–19.9 US population, particularly among black Americans and in the age years. group of 20–60 years. Thereafter, differences appear to be smaller between the Caucasian populations in Europe and the US, least before the age of 60–65 years.1,13,18,21,27 However, in reality, whereas only little difference in BF% is seen with age for the the association between BMI and morbidity or mortality is Asian population. In fact, it appears that among Asian men, BF% is U-shaped, and compared with a low BMI, a high BMI is lower for older men compared with the BF% of younger men. generally not related to excess morbidity or mortality. See below. Thus, BMI is a crude surrogate for obesity,13 and may introduce a bias in relation to understanding the importance of obesity for BF% AND BMI health outcomes.23 For instance, a cross-sectional study with 1928 BMI has been used as a key component for measuring risk related schoolchildren categorised 17.9% of boys and 12.8% of girls as to obesity in a variety of epidemiological studies, and is generally obese using BF% from BIA, whereas only 7.4% of boys and 6.3% of considered to be a good indicator of obesity. As BMI is calculated girls were considered obese based on BMI, and the authors by total body mass (FM and FFM),14 it allows no differentiation concluded that determination of BF% in addition to BMI seems to between and lean mass. BMI correlates with total be a necessity to correctly identify .28 These BF content18 and BF%.13,19 However, some studies6,20,21 have results are in accordance with those of Dugas et al.,29 who suggested that BMI is better correlated with BF (kg) than BF% suggested that BMI alone may not be an equivalent method to because BMI does not differentiate between BF and body lean determine adiposity in a group of multi-ethnic (non-Hispanic mass. Other studies6,22–25 have found that BMI was not very good white, non-Hispanic black and Mexican-American) adolescents. It was at quantifying either BF or BF%. argued that BMI standard cut-off points for a healthy weight range of Moreover, although BMI in reality is somewhat of limited use in 18.5–o25 kg/m2 that had been developed for international use,1 but relation to correctly assessing BF and obesity,26 BMI is widely used mostly refer to data from Caucasian populations, are not to be used as a marker for risk of disease and premature death because of its for populations other than Caucasians.30,31 simplicity, and because a high BMI generally associates with Variations in BF% vs BMI are large, and such variations are, as increased risk of disease compared with average levels of BMI, at indicated earlier, dependent on age, ethnicity and degree of

European Journal of Clinical Nutrition (2013) S79 – S85 & 2013 Macmillan Publishers Limited BIA and epidemiology ABo¨hm and BL Heitmann S83

Figure 2. BF% distribution with age in men from published cohort studies: Denmark;51 Switzerland;53 USA;50 Japan;55 and Hong Kong.49 For illustration of BF% from data of Sung et al.,49 average of BF% was calculated for subjects aged 14–15.9, 16–17.9 and 18–19.9 years. obesity. At higher levels of obesity, fatness is generally under- estimated when using BMI formulas, and because body composi- tion varies with age BMI also tends to underestimate BF% in Figure 3. Correlation between BMI and a four-compartment 6 younger subjects and overestimate it in older subjects.8 reference model in 72 men and 67 women. In addition, even when on average the same individuals are characterised as using BIA and BMI, variation is often greater for the estimates derived from BMI, as evidenced by a study by Heitmann,6 who compared different methods for evaluating BF based on population-specific equations derived from either BMI, skinfold or BIA, and showed that all three techniques provided reliable average BF estimates (as expected relating to the population- specific equations). However, fat estimated from BIA showed a lower variance than that from BMI and skinfolds, and both the s.e. of estimate and the s.d. were higher, and the R value lower when BF was estimated from BMI and skinfolds than from BIA. Figures 3 and 4 show the correlation between BMI and BF% estimated from BIA vs BF% estimated from a four-compartment reference model, which was derived from the measurement of total body water by dilutometry and whole-body potassium counting by scintigraphy. The data presented represent informa- tion from a subset of Danes from the Danish MONICA study conducted in 1987–1988 among 72 men and 67 women aged 35–65 years.6 The figures show, as expected, markedly greater variation in BF% by BMI than by BIA-estimated BF%. However, it also shows that men and women are largely separated into two groups, with either generally higher (women) or lower (men) BF% when BMI is expressed as the function of BF% measured by the reference method, whereas men and women are clustered in two distinct groups across the full range of BF% when BF% by BIA is expressed as a function of the reference measure. The large variation in BF% for the given BMI is further illustrated in Figure 5, which shows the Figure 4. Correlation between BF% estimated from BIA and a four- scatter of BF% vs BMI, here for the entire sample of 1528 men compartment reference model in 72 men and 67 women.6 aged 35–65 years from the Danish MONICA study conducted in 1987–1988.6 As can be seen, there were large variations in BF% for a given BMI—for instance, BF% varied between 7 and almost 40% for men with a BMI of 25 kg/m2. This observation is furthermore in agreement with several other studies,20,23,26,29 for instance, the RELATION BETWEEN BMI AND BF% TO CVD AND MORTALITY NHANES cohort, showing that for a BMI of 25 kg/m2, BF% in men A large number of studies14,19,22,33–36have demonstrated the J- or ranged widely from 13.8 to 35.3% and from 26.4 to 32.8% in U-shaped relation between BMI and mortality, where both a high women, respectively.13 On the contrary, a population-based and a low BMI is significantly associated with increased risk study32 including 26 942 subjects found that BF% measured of death.37 It has been hypothesised that the J- or U-shaped from BIA was strongly correlated with BMI and waist association was dependent on smoking status and pre-existing circumference more in women than in men. diseases in individuals.19,38 However, a recent study39 of almost

& 2013 Macmillan Publishers Limited European Journal of Clinical Nutrition (2013) S79 – S85 BIA and epidemiology ABo¨hm and BL Heitmann S84 women aged 65 years and older BIA-measured BF% compared with BMI or waist circumference, and Simpson et al.46 concluded that waist circumference and WHR were better predictors of risk of all-cause mortality than BIA in the Melbourne Collaborative Cohort Study, carried out with 41 313 men and women (aged 27–75 years, with most (99.3%) aged 40–69 years). The same cohort found WHR, body height and waist circumference to be better predictors of colon cancer risk than general adiposity, expressed as BMI or BF% among 24 072 women.47 Further, a prospective study with 10 564 men of the Malmo¨ and Cancer cohort found no association between general adiposity and risk of prostate cancer, but did find a relationship with WHR and body height.48 There is also discussion about rather focusing on lean mass parameters (lean mass or lean mass index) than on BMI, as lean mass parameters, but not other body composition parameters (for Figure 5. Distribution of BF% for any given BMI value.51 example, FM), showed a significant association with risk factors for all-cause mortality in the elderly Asian population examined by Han et al.27

900 000 participants (aged 35–89 years) based on 57 prospective studies and four continents found that the U-shaped relationship CONCLUSION persisted when excluding smokers and even all subjects with a The present review shows that BIA is a valid and precise method presence of chronic pre-existing disease. Similar findings for predicting body composition under controlled conditions in were noted in a large Chinese40 and a large American cohort.41 22 33 healthy subjects. BIA can be used to estimate fat and lean mass, Heitmann et al. and also Allison et al. explained this both of which associate linearly with morbidity and mortality in phenomenon of the U-shaped association by the fact that BMI is contrast to BMI where a U-formed association is found. BMI, a compound of mainly FM and FFM which have opposite effects however, is a simple but inaccurate method to estimate both BF% on mortality, and where the relationship between FM and and risk of diseases/mortality, because BMI neither allows mortality increases monotonically, whereas that between FFM distinctions between FM and FFM nor considers changes that and mortality monotonically decreases. Until this date, remarkably occur with age. BF% measured by BIA increases with age, with little research has been carried out on how body composition, marked differences for a given age and gender between different rather than BMI, influences mortality risk. However, a few studies populations. should be mentioned. For instance, the long-term population- 32 However, further research may be necessary to conclude firmly based study of Calling et al. that showed that especially a high whether BIA measurements are more effective than BMI, waist BF% was associated with increased cardiovascular risk as BF% circumference and WHR for predicting all-cause mortality. emerged as an independent risk factor for coronary events and (CVD) death in men, respectively, and coronary event and ischemic stroke in women. Another study42 found significant positive correlations between BF% and coronary CONFLICT OF INTEREST heart disease risk factors, in white non-Hispanic and Hispanic male The authors declare no conflict of interest. and Hispanic female college students. The group of Marques-Vidal et al.,43 in their paper, discussed whether BF%, BMI or waist:hip ratio (WHR) was the better ACKNOWLEDGEMENTS predictor of high (45%) 10-year risk of fatal CVD, and found Publication of this article was supported by a grant from seca Gmbh & Co. KG, stronger associations between BF% and risk than between BMI , Germany. and risk. In fact, in this study, development of CVD over 10 years was three times higher for those with the higher BF% than for those with high BMI. These results are also in accordance with REFERENCES 44 those of Singh et al., who found a strong association between 1 Eveleth PB. Physical status: the use and interpretation of anthropometry. BF% and coronary artery disease and the coronary risk factors Report of a WHO Expert Committee. Am J Hum Biol 1996; 86: 786–787. , hypercholesterolemia, diabetes mellitus and 2 Baumgartner RN, Chumlea WC, Roche AF. Estimation of body composition from in Indian men. bioelectric impedance of body segments. Am J Clin Nutr 1989; 50: 221–226. In addition, Heitmann et al.45 reported significant associations 3 Kyle UG, Bosaeus I, De Lorenzo AD, Deurenberg P, Elia M, Gomez JM et al. between FM or FFM and total mortality, which were furthermore Bioelectrical impedance analysis--part I: review of principles and methods. found to vary with the level of physical activity. Finally, Lahmann Clin Nutr 2004; 23: 1226–1243. 19 4 Heitmann BL. Impedance: a valid method in assessment of body composition? et al. assessed the effect of body composition on all-cause Eur J Clin Nutr 1994; 48: 228–240. mortality in a Swedish cohort of 28 098 men and women, aged 5 Segal KR, Van LM, Fitzgerald PI, Hodgdon JA, Van Itallie TB. Lean body mass 45–73 years, and found a larger hazard associated with a high BF% estimation by bioelectrical impedance analysis: a four-site cross-validation study. than a high BMI in both sexes. BF% of 435% among middle-aged Am J Clin Nutr 1988; 47: 7–14. individuals (45- to 59-year-olds) was found to entail the highest 6 Heitmann BL. Evaluation of body fat estimated from body mass index, skinfolds estimated hazard ratio. Further, men with a BF% ranging from 17 and impedance. A comparative study. Eur J Clin Nutr 1990; 44: 831–837. to 20% were found to have the lowest mortality risk, whereas a 7 Wells JC, Fewtrell MS. Measuring body composition. Arch Dis Child 2006; 91: BF% of 425% was associated with the highest mortality. 612–617. Comparison of mortality with lean body mass showed that the 8 Deurenberg P, Andreoli A, Borg P, Kukkonen-Harjula K, de LA, van Marken Lichtenbelt WD et al. The validity of predicted body fat percentage from body higher the lean body mass, the lower the mortality risk in men. mass index and from impedance in samples of five European populations. However, not all studies find an advantage of using body Eur J Clin Nutr 2001; 55: 973–979. 41 composition measures over BMI. For instance, Dolan et al. 9 Sun G, French CR, Martin GR, Younghusband B, Green RC, Xie YG et al. demonstrated no obvious advantage of predicting mortality in Comparison of multifrequency bioelectrical impedance analysis with dual-energy

European Journal of Clinical Nutrition (2013) S79 – S85 & 2013 Macmillan Publishers Limited BIA and epidemiology ABo¨hm and BL Heitmann S85 X-ray absorptiometry for assessment of percentage body fat in a large, healthy 32 Calling S, Hedblad B, Engstrom G, Berglund G, Janzon L. Effects of body fatness population. Am J Clin Nutr 2005; 81: 74–78. and physical activity on cardiovascular risk: risk prediction using the bioelectrical 10 Pimentel GD, Bernhard AB, Frezza MR, Rinaldi AE, Burini RC. Bioelectric impedance impedance method. Scand J Public Health 2006; 34: 568–575. overestimates the body fat in overweight and underestimates in Brazilian obese 33 Allison DB, Zhu SK, Plankey M, Faith MS, Heo M. Differential associations of women: a comparison with Segal equation 1. Nutr Hosp 2010; 25: 741–745. body mass index and adiposity with all-cause mortality among men in the first 11 Risica PM, Ebbesson SO, Schraer CD, Nobmann ED, Caballero BH. Body fat dis- and second National Health and Nutrition Examination Surveys (NHANES I and tribution in Alaskan Eskimos of the Bering Straits region: the Alaskan Siberia NHANES II) follow-up studies. Int J Obes Relat Metab Disord 2002; 26: 410–416. Project. Int J Obes Relat Metab Disord 2000; 24: 171–179. 34 Troiano RP, Frongillo Jr EA, Sobal J, Levitsky DA. The relationship between body 12 Ritz P, Salle A, Audran M, Rohmer V. Comparison of different methods to assess weight and mortality: a quantitative analysis of combined information from body composition of in obese and diabetic patients. Diabetes Res Clin existing studies. Int J Obes Relat Metab Disord 1996; 20: 63–75. Pract 2007; 77: 405–411. 35 Chen Z, Yang G, Offer A, Zhou M, Smith M, Peto R et al. Body mass index and 13 Romero-Corral A, Somers VK, Sierra-Johnson J, Thomas RJ, Collazo-Clavell ML, mortality in : a 15-year prospective study of 220 000 men. Int J Epidemiol Korinek J et al. Accuracy of body mass index in diagnosing obesity in the adult 2012; 41: 472–481. general population. Int J Obes (Lond) 2008; 32: 959–966. 36 Song X, Pitkaniemi J, Gao W, Heine RJ, Pyorala K, Soderberg S et al. Relationship 14 Allison DB, Faith MS, Heo M, Kotler DP. Hypothesis concerning the U-shaped between body mass index and mortality among Europeans. Eur J Clin Nutr 2012; relation between body mass index and mortality. Am J Epidemiol 1997; 146: 66: 156–165. 339–349. 37 Calle EE, Thun MJ, Petrelli JM, Rodriguez C, Heath Jr CW. Body-mass index and 15 Sun SS, Chumlea WC, Heymsfield SB, Lukaski HC, Schoeller D, Friedl K et al. mortality in a prospective cohort of U.S. adults. NEnglJMed1999; 341: 1097–1105. Development of bioelectrical impedance analysis prediction equations for body 38 Manson JE, Stampfer MJ, Hennekens CH, Willett WC. Body weight and longevity. a composition with the use of a multicomponent model for use in epidemiologic reassessment. JAMA 1987; 257: 353–358. surveys. Am J Clin Nutr 2003; 77: 331–340. 39 Whitlock G, Lewington S, Sherliker P, Clarke R, Emberson J, Halsey J et al. 16 Dehghan M, Merchant AT. Is bioelectrical impedance accurate for use in large Body-mass index and cause-specific mortality in 900 000 adults: collaborative epidemiological studies? Nutr J 2008; 7: 26. analyses of 57 prospective studies. Lancet 2009; 373: 1083–1096. 17 Heitmann BL, Swinburn BA, Carmichael H, Rowley K, Plank L, McDermott R et al. 40 Lin WY, Tsai SL, Albu JB, Lin CC, Li TC, Pi-Sunyer FX et al. Body mass index and Are there ethnic differences in the association between body weight and resis- all-cause mortality in a large Chinese cohort. CMAJ 2011; 183: E329–E336. tance, measured by bioelectrical impedance? Int J Obes Relat Metab Disord 1997; 41 Dolan CM, Kraemer H, Browner W, Ensrud K, Kelsey JL. Associations between body 21: 1085–1092. composition, anthropometry, and mortality in women aged 65 years and older. 18 Pi-Sunyer FX, Becker DM, Bouchard C, Carleton RA, Colditz GA, Dietz WH et al. Am J Public Health 2007; 97: 913–918. Clinical guidelines on the identification, evaluation, and treatment of 42 Koutoubi S, Huffman FG. Body composition assessment and coronary heart overweight and obesity in adults: executive summary. Am J Clin Nutr 1998; 68: disease risk factors among college students of three ethnic groups. J Natl Med 899–917. Assoc 2005; 97: 784–791. 19 Lahmann PH, Lissner L, Gullberg B, Berglund G. A prospective study of adiposity 43 Marques-Vidal P, Bochud M, Mooser V, Paccaud F, Waeber G, Vollenweider P. and all-cause mortality: the Malmo Diet and Cancer study. Obes Res 2002; 10: Obesity markers and estimated 10-year fatal cardiovascular risk in Switzerland. 361–369. Nutr Metab Cardiovasc Dis 2009; 19: 462–468. 20 Pasco JA, Nicholson GC, Brennan SL, Kotowicz MA. Prevalence of obesity and the 44 Singh RB, Niaz MA, Beegom R, Wander GS, Thakur AS, Rissam HS. Body fat percent relationship between the body mass index and body fat: cross-sectional, by bioelectrical impedance analysis and risk of coronary artery disease among population-based data. PLoS One 2012; 7: e29580. urban men with low rates of obesity: the Indian paradox. J Am Coll Nutr 1999; 18: 21 Garrow JS, Webster J. Quetelet’s index (W/H2) as a measure of fatness. Int J Obes 268–273. 1985; 9: 147–153. 45 Heitmann BL, Hills AP, Frederiksen P, Ward LC. Obesity, leanness, and mortality: 22 Heitmann BL, Erikson H, Ellsinger BM, Mikkelsen KL, Larsson B. Mortality asso- effect modification by physical activity in men and women. Obesity (Silver Spring) ciated with body fat, fat-free mass and body mass index among 60-year-old 2009; 17: 136–142. swedish men-a 22-year follow-up. The study of men born in 1913. Int J Obes Relat 46 Simpson JA, MacInnis RJ, Peeters A, Hopper JL, Giles GG, English DR. A compar- Metab Disord 2000; 24: 33–37. ison of adiposity measures as predictors of all-cause mortality: the Melbourne 23 Rothman KJ. BMI-related errors in the measurement of obesity. Int J Obes (Lond) Collaborative Cohort Study. Obesity (Silver Spring) 2007; 15: 994–1003. 2008; 32(Suppl 3): S56–S59. 47 MacInnis RJ, English DR, Hopper JL, Gertig DM, Haydon AM, Giles GG. Body 24 Gallagher D, Visser M, Sepulveda D, Pierson RN, Harris T, Heymsfield SB. How size and composition and colon cancer risk in women. Int J Cancer 2006; 118: useful is body mass index for comparison of body fatness across age, sex, and 1496–1500. ethnic groups? Am J Epidemiol 1996; 143: 228–239. 48 Wallstrom P, Bjartell A, Gullberg B, Olsson H, Wirfalt E. A prospective Swedish 25 Roubenoff R, Dallal GE, Wilson PW. Predicting body fatness: the body mass study on body size, body composition, diabetes, and prostate cancer risk. index vs estimation by bioelectrical impedance. Am J Public Health 1995; 85: Br J Cancer 2009; 100: 1799–1805. 726–728. 49 Sung RY, So HK, Choi KC, Li AM, Yin J, Nelson EA. Body fat measured by bioe- 26 Luke A, Durazo-Arvizu R, Rotimi C, Prewitt TE, Forrester T, Wilks R et al. Relation lectrical impedance in Hong Kong Chinese children. Hong Kong Med J 2009; 15: between body mass index and body fat in black population samples from Nigeria, 110–117. Jamaica, and the United States. Am J Epidemiol 1997; 145: 620–628. 50 Chumlea WC, Guo SS, Kuczmarski RJ, Flegal KM, Johnson CL, Heymsfield SB et al. 27 Han SS, Kim KW, Kim KI, Na KY, Chae DW, Kim S et al. Lean mass index: a better Body composition estimates from NHANES III bioelectrical impedance data. predictor of mortality than body mass index in elderly Asians. J Am Geriatr Soc Int J Obes Relat Metab Disord 2002; 26: 1596–1609. 2010; 58: 312–317. 51 Heitmann BL. Body fat in the adult Danish population aged 35-65 years: an 28 Antal M, Peter S, Biro L, Nagy K, Regoly-Merei A, Arato G et al. Prevalence of epidemiological study. Int J Obes 1991; 15: 535–545. , overweight and obesity on the basis of body mass index and body 52 Pichard C, Kyle UG, Bracco D, Slosman DO, Morabia A, Schutz Y. Reference values fat percentage in Hungarian schoolchildren: representative survey in metropoli- of fat-free and fat masses by bioelectrical impedance analysis in 3393 healthy tan elementary schools. Ann Nutr Metab 2009; 54: 171–176. subjects. Nutrition 2000; 16: 245–254. 29 Dugas LR, Cao G, Luke AH, Durazo-Arvizu RA. Adiposity is not equal in a multi- 53 Kyle UG, Genton L, Lukaski HC, Dupertuis YM, Slosman DO, Hans D et al. race/ethnic adolescent population: NHANES 1999-2004. Obesity (Silver Spring) Comparison of fat-free mass and body fat in Swiss and American adults. Nutrition 2011; 19: 2099–2101. 2005; 21: 161–169. 30 Craig P, Halavatau V, Comino E, Caterson I. Differences in body composition 54 Lahmann PH, Lissner L, Gullberg B, Berglund G. Sociodemographic factors asso- between Tongans and Australians: time to rethink the healthy weight ranges? Int ciated with long-term , current body fatness and central adiposity in J Obes Relat Metab Disord 2001; 25: 1806–1814. Swedish women. Int J Obes Relat Metab Disord 2000; 24: 685–694. 31 Gurrici S, Hartriyanti Y, Hautvast JG, Deurenberg P. Relationship between body fat 55 Nagaya T, Yoshida H, Takahashi H, Matsuda Y, Kawai M. Body mass index (weight/ and body mass index: differences between Indonesians and Dutch Caucasians. height2) or percentage body fat by bioelectrical impedance analysis: which variable Eur J Clin Nutr 1998; 52: 779–783. better reflects serum lipid profile? IntJObesRelatMetabDisord1999; 23: 771–774.

& 2013 Macmillan Publishers Limited European Journal of Clinical Nutrition (2013) S79 – S85