Body mass index cut offs to define thinness in children and adolescents: international survey

Objective To determine cut offs to define thinness in children and adolescents, based on body mass index at age 18 years. Design International survey of six large nationally representative cross sectional studies on growth. Setting Brazil, Great Britain, Hong Kong, the Netherlands, Singapore, and the United States. Subjects 97876 males and 94851 females from birth to 25 years. Main outcome measure Body mass index (BMI, weight/ height). Results The World Health Organization defines grade 2 thinness in adults as BMI <17. This same cut off, applied to the six datasets at age 18 years, gave mean BMI close to a z score of −2 and 80% of the median. Thus it matches existing criteria forwasting in childrenbased onweight for height. For each dataset, centile curves were drawn to pass through the cut off of BMI 17 at 18 years. The resulting curves were averaged to provide age and sex specific cut-off points from 2-18 years. Similar cut offs were derived based on BMI 16 and 18.5 at 18 years, together providing definitions of thinness grades 1, 2, and 3 in children and adolescents consistent with the WHO adult definitions. Conclusions The proposed cut-off points should help to provide internationally comparable prevalence rates of thinness in children and adolescents. INTRODUCTION Much has been written about the epidemic of child obesity but malnutrition—meaning undernutrition— in infants, children, and adolescents poses a considerably larger public health problem internationally, and in the developed world anorexia nervosa is the third most common chronic condition of adolescence. Obesity and malnutrition represent opposite extremes on the spectrum of adiposity, and both are routinely quantified in terms of weight and height relative to the child’s age. Yet the classification of malnutrition in later childhood and adolescence is currently unsatisfactory because of the lack of suitable cut offs for international use. Fifty years ago Gomez introduced his malnutrition classification of weight below a specified percentage of median weight for the child’s age. This included three components: a measurement, a reference for age adjustment, and a set of cut offs. Later Seoane and Latham proposed splitting weight for age into weight for height and height for age, allowing underweight to be defined as wasting or stunting, or both. Subsequently Waterlow et al recommended the use of z scores for the definitions of underweight, wasting, and stunting, with the cut offs defined in terms of standard deviations (SDs) below the median rather than as percentages of themedian. This ensures that the false positive screening rate is constant across age as applied to the reference population. In 1983 theWorldHealthOrganization (WHO) formally recognised the US National Center for Health Statistics (NCHS) classification as the international reference and has used it since to classify children as underweight, wasted, or stunted, each based on a cut off of −2 z scores. Wasting in particular is assessed with the NCHS/WHO weight for height reference, which compares the child’s weight to the average weight of children of the same height. This ignores the child’s age, which allows nutritional status to be assessed when age is not known. It also assumes that, on average, children of a given height weigh the same whatever their age; in infancy and adolescence, however, the weight-height relation depends on age. 19 This can be seen by considering the index weight/ height, where the height power p is allowed to vary with age. The index is adjusted for age and sex by dividing it by the same ratio based on median weight and height for the child’s age and sex. For a weight for height index such asNCHS, the value of p is the ratio of the percentage growth rates in weight and height at each age, so it is largest when weight is growing fastest relative to height—that is, in infancy and adolescence when p is 3 or more as against 1.5 in mid-childhood. In later adolescence, as weight growth continues after height growth has stopped, p increases to infinity and height adjustment becomes impossible. This is an important general limitation of weight for height references in that they cannot be used in adolescence. 20 For this reason the NCHS weight for height reference was truncated at age 10 in girls and 11.5 in boys. The weight/height index can alternatively be adjusted for height for age, where p is chosen to make the index uncorrelated with height among children at Centre for Paediatric Epidemiology and Biostatistics, UCL Institute of Child Health, London WC1N 1EH National Center for Health Statistics, Centers for Disease Control and Prevention, Hyattsville MD 20782, USA Department of Child and Adolescent Mental Health, Great Ormond Street Hospital for Children, London Institute of Human Nutrition, University of Southampton, Southampton Correspondence to: T J Cole tim. [email protected] doi:10.1136/bmj.39238.399444.55 BMJ | ONLINE FIRST | bmj.com page 1 of 8 each age. This leads to a different pattern of p changing with age, with p=2 in infancy, rising to 3 in adolescence and then falling back to 2 in adulthood. 21-23 Cole suggested fixing p at 2—that is, the body mass index (BMI). This is now used throughout infancy, childhood, adolescence, and adulthood. BMI has been used since the 1960s to assess obesity in adults 25 and more recently in children. 27 Many countries now have their own national reference centile charts for BMI for age. International BMI cut offs for child overweight and obesity, based on data from six countries, have been developed. The WHO 1995 expert committee endorsed the use of BMI for assessing thinness in adolescence, based on the BMI reference data from Must et al, and the recently published 2006 WHO growth standard also includes BMI for children aged 0-5 years. However, this is insufficient for international use because the BMI cut offs from Must et al were based on US data from the early 1970s and the WHO standard is restricted in age. Thus there are no valid BMI cut offs for assessing underweight or wasting in adolescents or children over 5 years. The international BMI cut offs for child overweight and obesity cover the age range 2-18 years and are based on the adult cut offs of 25 and 30 at 18 years. They have been widely used, with over 1100 citations in the seven years since publication. It would be logical to produce BMI cut offs for underweight using the same principle. However, underweight does not have the same meaning in adults and children. In adults, underweight or thinness indicates low BMI, whereas in children underweight is lowweight for age andwasting is low weight for height. We have extended the adult term of thinness to children, meaning low BMI for age. METHODS Subjects and data We used the same methods as those used by the International Obesity TaskForce (IOTF) for the international overweight and obesity cut offs. We obtained BMI data from nationally representative surveys of children in six high and middle income countries: Brazil, Great Britain, Hong Kong, the Netherlands, Singapore, and the United States (table 1). Each survey had over 20 000 subjects aged 6-18 years, and height and weight were measured with standard methods and quality control measures to minimise measurement error. Four of the datasets came from one-off surveys, while the British and US data were pooled from surveys collected over a period of time. The US data came from the national health examination surveys II and III, and the national health and nutrition examination surveys (NHANES) I and II, while for comparison Must et al used NHANES I data for their BMI reference. The Brazilian and US surveys used multi-stage sampling designs, and their data were analysed with survey weights. A total of 192 727 subjects were involved, 97 876 males and 94 851 females from birth to 25 years (table 1). LMS method We analysed each dataset using the LMS method, which summarises the distribution of BMI by age and sex in terms of three curves called L (lambda), M (mu) and S (sigma). The M curve is median BMI by age, the S curve is the coefficient of variation of BMI, and the L curve expresses the skewness of the BMI distribution in terms of the Box-Cox power needed to transform the data to near normality. Any required BMI centile curve is defined in terms of the L, M, and S curves as follows: M(1+L×S×z) L where z is the z score corresponding to the required centile (for example, z=0 gives the median M or z=0.67 gives the 75th centile) and the values of L, M, and S vary with age and sex. The reverse process, of converting a child’s BMI to a z score, involves the equation: z=((BMI/M) −1)/(L×S) where the values of L,M, and S are for the child’s age and sex.Note that the ratio BMI/M in the second equation,multiplied by 100, corresponds to BMI expressed as a percentage of themedian (BMI%). So BMI% and z are linked in a way that depends on the variability S and skewness L, which in turn depend on age. Conventionally a BMI centile chart is based on a prespecified set of centiles (for example, 3rd, 10th, 25th, 50th, 75th, 90th, 97th) or z scores (−2 to +2 in increments of two thirds of a z score). Here by contrast, quasi-centile curves are constructed to pass through a given BMI cut off at a given age (we chose 18 as it was the oldest age with data available in all six datasets). To do this the required BMI is substituted into the second equation and the corresponding z Table 1 | Six nationally representative datasets of BMI in childhood (n=192 727) Country Year Description Boys Girls Age range Sample size Age range Sample size Brazil 1989 Second national anthropometric survey 2-25 15 947 2-25 15 859 Great Britain 1978-93 Data pooled from five national surveys 0-23 16 491 0-23 15 731 Hong Kong 1993 National growth survey 0-18 11 797 0-18 12 168 Netherlands 1980 Third nationwide growth survey 0-20 21 521 0-20 20 245 Singapore 1993 School health service survey 6-19 17 356 6-20 16 616 US 1963-80 Data pooled from four national surveys 2-20 14 764 2-20 14 232 Total 1963-93 — 0-25 97 876 0-25 94 851 RESEARCH page 2 of 8 BMJ | ONLINE FIRST | bmj.com score obtained, by using L, M, and S values by sex for age 18 specific for the dataset. This z score is then substituted into the first equation and defines the required curve by age. We constructed centile curves of this form for each of the six datasets separately and then averaged the curves by age. The result is a single curve, based on all six datasets, that passes through the specified cut off at age 18. This exercise was repeated for each sex and for each of several distinct BMI cut offs at age 18. Choice of cut offs at age 18 The international cut offs for overweight and obesity were based on thewidely accepted adult BMI values of 25 and 30. These values are related to health, indicating points on the BMI spectrum where risk increases appreciably, and are widely used. Health related cut offs for thinness in adults also exist, but there is less consensus in their use. WHO defines thinness grades 1, 2, and 3 as BMI below 18.5, 17, and 16; the malnutrition universal screening tool of the British Association for Parental and Enteral Nutrition (BAPEN) scores 1 and 2 for BMI below 20 and 18.5, respectively; and theWHO ICD-10 criteria for anorexia nervosa include BMI below 17.5 or weight below 85% of expected weight for height. 45 In children, the diagnostic criteria for anorexia nervosa use BMI below the 5th or 10th centile, corresponding to −1.6 or −1.3 SD (z scores), to define underweight, 47 while the criteria for malnutrition, based on weight for height rather than BMI, use the graded WHO cut offs of −1, −2, and −3 SD, corresponding roughly to 90%, 80%, and 70% of expected weight for height. 48 Anomalously the WHO Expert Committee defined thinness in adolescence as BMI below the 5th centile rather than below −2 SD, probably because theNHANES I reference did not provide −2 SD cut offs. At age 18 the 5th centiles in Must et al were 17.5 for males and 16.7 for females, reflecting US youth in the early 1970s. An important question here is which cut off is the more appropriate, the 5th centile or −2 SD. WHO recommended the −2 SD criterion back in 1977, while the 5th centile was a pragmatic alternative at a time when a −2 SD BMI cut off was not available. For this reason we feel that −2 SD is the more appropriate cut off to use. On this basis, the simplest way to transfer the child cut offs fromweight for height to BMI is to treat the two −2 SD cut offs as equivalent. Weight for height is weight adjusted for height while BMI for age is weight adjusted for height and age. So ifweight for heightwere independent of age, as it is at certain ages, then the two cut offs would coincide. At other ages the variability in BMI is theoretically slightly less than for weight for height, as variability caused by age is adjusted for. Against that, the height adjustment for BMI is imperfect later in childhood, so on balance the variability is likely to be similar for the two indices. Thus the optimal cut off for our purposes would be a value of BMI at age 18 that coincidedwith a previously published adult cut off and which was also close to a child BMI cut off of −2 SD. But this introduces ambiguity as the z score corresponding to a given cut off will depend on the growth reference used. Here we use the six datasets as internal references to test the alternative cut offs. We also investigate the relation between z score and BMI%. RESULTS Table 2 gives BMI z scores and centiles corresponding to various published BMI cut offs at age 18, averaged across the six datasets,where the centiles correspond to the sex averaged z scores. In general the results are similar for boys and girls, and the cut offs range from the 0.6th to the 16th centile. BMI 18.5 is on the 16th centile and approximates to a z score of −1, while BMI 17 is on the 3rd centile and close to z score −2, and hence is near optimal for our purposes. Table 3 looks at the BMI cut off of 17 in z score terms by dataset. The four Western countries are close to z score −2.0 in females and −2.1 in males, while the data from Hong Kong and Singapore are near to −1.4. The centiles indicate the prevalence of thinness in each country at age 18 when the survey was done, at which time the East Asian children were appreciably thinner. Figure 1 shows the separate thinness curves for BMI 17 at age 18 by country and sex.Within each graph the country curves are largely superimposed and more so for girls than boys. Looking at the individual countries, Brazil is relatively low in both sexes while Hong Kong is high in boys, and for Singapore the boys’ curve stands out at age 6. The BMI cut off of 17 is not only near to z score −2, it is also the WHO definition of thinness grade 2 in adults. Thus theWHO classification provides a bridge between child and adult, in that a young person with BMI 17 at age 18 is both a borderline thin adult (grade 2) and a borderline thin child (z score −2). For this reason we propose to use the cut off of 17 plus the other Table 2 | z scores corresponding to different BMI cut offs at age 18, averagedby sex across six datasets