Overweight (BMI ≥ 25 kg/m2) and obesity (BMI ≥ 30 kg/m2) are major public health issues and significant contributors to the burden of disease worldwide(Reference Haslam, Sattar and Lean1). The prevalence of obesity is rising and projections suggest that 73 % of women in the UK will be overweight and 36 % obese by 2015(2).
During pregnancy, women who are overweight or obese are at greater risk of developing gestational diabetes, pre-eclampsia, hypertension and have increased complications during labour(Reference Heslehurst, Lang, Rankin, Wilkinson and Summerbell3–Reference Catalano and Ehrenberg7). Maternal obesity affects fetal growth and development, increasing the risk of neural tube defects and macrosomia(Reference King8, Reference Abenhaim, Kinch, Morin, Benjamin and Usher9). Recent evidence has shown that maternal adiposity is associated with increased adiposity in neonates and children at age 9 years(Reference Sewell, Huston-Presley, Super and Catalano10, Reference Gale, Javaid, Robinson, Law, Godfrey and Cooper11).
Various anthropometric indices have been proposed to assess adiposity, BMI being the most extensively used in clinical practice. A BMI ≥ 18·5 and < 25 kg/m2 is considered normal, ≥ 25 and < 30 kg/m2 is overweight and ≥ 30 kg/m2 is obese. Obesity is associated with increased mortality(Reference Flegal, Graubard, Williamson and Gail12). However, increasing evidence from studies of cardio-metabolic disease shows that waist circumference, an index of central adiposity, may be more closely associated with risk of abnormal metabolic function than BMI(Reference Snijder, van Dam, Visser and Seidell13). Given that BMI is not easily nor accurately estimated by the general public, waist circumference has been suggested as an alternative measure(Reference Lean, Han and Morrison14, Reference Han, Sattar and Lean15). Cut-off points have been suggested to identify individuals for whom weight management would be recommended. For women, it has been proposed that a waist circumference of ≥ 80 cm would identify almost all women with a BMI ≥ 25 kg/m2 and a waist circumference of ≥ 88 cm those with a BMI ≥ 30 kg/m2(Reference Lean, Han and Morrison14). An alternative index, waist circumference expressed relative to height, the waist:height ratio, has also been suggested as a screening tool(Reference Ashwell and Hsieh16). Individuals with values of ≥ 0·5 are advised to ‘Take care’ and ≥ 0·6 to take ‘Action’. A simple public health message, ‘Keep your waist circumference to less than half your height’ has been put forward. Waist:height ratio predicts intra-abdominal fat(Reference Ashwell, Cole and Dixon17) and two studies have shown that it is a better predictor of death than BMI(Reference Ashwell, Cole and Dixon17, Reference Cox and Whichelow18). Proponents suggest that the cut-off points for waist:height ratio would apply to both men and women, different ethnic groups and children, in contrast to those for waist circumference(Reference Ashwell and Hsieh16, Reference Hsieh, Yoshinaga and Muto19).
The extent to which proposed cut-off points for BMI waist circumference and waist:height ratio identify the same or different women is unclear and there are few large studies of anthropometry in young women of reproductive age. It is of considerable importance to identify individuals ‘at risk’, and to determine whether these indices mark the same or differing risk of ill health. Here, we present results from a large general population sample of young women living in Southampton (Hants, UK) on whom we have detailed anthropometric measurements. We sought to explore whether high BMI, high waist circumference and high waist:height ratio occur in the same or different women and how these indices relate to total and regional adiposity.
Methods
Data were collected as part of the Southampton Women's Survey, which is a study of a population sample of non-pregnant women aged 20–34 years living in the city of Southampton and registered with a general practitioner. The survey was carried out between 1998 and 2002. A profile of the cohort has been published(Reference Inskip, Godfrey, Robinson, Law, Barker and Cooper20). The Southampton Women's Survey was approved by the Southampton and South West Hampshire Local Research Ethics Committee.
Trained research nurses visited the women at home, administered a questionnaire and took anthropometric measurements. The questionnaire included details of the women's current smoking and previous obstetric history. Height was measured with a stadiometer to the nearest 0·1 cm with the head in the Frankfort plane. Weight was measured after the women removed their shoes and any heavy items of clothing or jewellery, with calibrated electronic scales to the nearest 0·1 kg. Waist circumference was measured midway between the lower rib margin and the iliac crest (both palpated in the mid axillary line) at the end of expiration over bare skin(Reference Fidanza and Fidanza21). Hip circumference was measured as the maximum circumference over the buttocks over thin clothing(Reference Callaway, Chumlea, Bouchard, Himes, Lohman, Martin, Mitchell, Mueller, Roche, Seefeldt, Lohman, Roche and Martorell22). Both circumferences were measured to the nearest 0·1 cm using a fibreglass tape measure. Four skinfold thicknesses (triceps, biceps, subscapular and supra-iliac) were measured in triplicate on the non-dominant side using Harpenden skinfold calipers to the nearest 0·1 mm(Reference Harrison, Buskirk, Carter, Johnston, Lohman, Pollock, Roche, Wilmore, Lohman, Roche and Martorell23). Up to two further readings were taken if necessary. The mean of the three closest readings was used in the statistical analysis.
Statistical analysis
The data were analysed using STATA (version 8; StataCorp LP, College Station, TX, USA). Data were available for 12 551 women, but we restricted our analysis to 11 786 white Caucasian women (94 % of the population), because of the known differences in body composition between different ethnic groups(Reference Deurenberg, Deurenberg Yap, Wang, Lin and Schmidt24). We excluded twenty-two women from the analysis because we did not know their ethnic group. We used the maximum number of observations available for each analysis.
BMI was calculated as weight/height2 (kg/m2). Fat mass was estimated from skinfold thickness measurements using the method of Durnin & Womersley(Reference Durnin and Womersley25). The appropriate equation was used depending on how many and which skinfold thickness measurements were available. A total of 11 594 women had all four skinfold thickness measurements available, fifty had three skinfold thickness measurements available, fifteen had two skinfold thickness measurements available and three had only the triceps available. Fat mass index (FMI) was calculated as fat mass (kg)/heightn (m). To establish the most appropriate exponent needed to make fat mass completely independent of height, we regressed log fat mass on log height. The linear regression coefficient is the power (n) to which height should be raised to achieve no association between the derived variable and height(Reference Wells and Victora26, Reference Heymsfield, Gallagher, Mayer, Beetsch and Pietrobelli27). The value for n of 1·65 (95 % CI 1·48, 1·83) gave the least correlation between FMI and height.
We applied the same technique to derive indices of body weight and waist circumference adjusted for height. To assess fat distribution, we calculated subscapular:triceps skinfold ratio.
Variables that were not normally distributed (all variables except age, height and percentage body fat) were logged using loge. To present the findings graphically, kernel density plots were used to estimate the probability density functions for FMI. A kernel density plot can be considered a refinement of a histogram or frequency plot and is a graphical summary of the shape of the data. Separate plots were produced within the three categories defined by the cut-off points of each of the three indices. For BMI, the categories were: 1, < 25 kg/m2; 2, ≥ 25 and < 30 kg/m2; 3, ≥ 30 kg/m2. For waist circumference, the categories were: 1, < 80 cm; 2, ≥ 80 and < 88 cm; 3, ≥ 88 cm. For waist:height ratio, the categories were: 1, < 0·5; 2, ≥ 0·5 and < 0·6; 3, ≥ 0·6.
Results
Table 1 shows the general characteristics of the women. A total of 47 % of the women had had one or more children (including stillbirths) and 32 % were smokers at the time of the interview. A total of 42 % of the women had a BMI ≥ 25 kg/m2 and 16 % a BMI ≥ 30 kg/m2 (Table 2).
(Medians and interquartile ranges)
Numbers of women in categories of BMI according to World Health Organization definitions
Tables 3 and 4 show the number of women divided into four groups based on their BMI and waist circumference. Some 49 % of women had a BMI ≥ 25 kg/m2, a waist circumference ≥ 80 cm or both, and 14 % of women were classified differently according to the two indices, by either having the combination of a BMI < 25 kg/m2 and a waist circumference ≥ 80 cm, or having a BMI ≥ 25 kg/m2 but a waist circumference < 80 cm. Women with the combination of a BMI < 25 kg/m2 and a waist circumference ≥ 80 cm were taller than those with the combination of a BMI ≥ 25 kg/m2 and a waist circumference < 80 cm (167 v. 160 cm respectively). Using the higher cut-off points, 23 % of women had a BMI ≥ 30 kg/m2, a waist circumference ≥ 88 cm or both, and 9 % of women were differently classified by the cut-offs with 7 % of women having the combination of a BMI < 30 kg/m2 and a waist circumference ≥ 88 cm, and 2 % having a BMI ≥ 30 kg/m2 but a waist circumference < 88 cm.
Numbers of women in four groups according to BMI and waist circumference, with BMI cut-off 25 kg/m2 and waist circumference cut-off 80 cm
Numbers of women in four groups according to BMI and waist circumference, with BMI cut-off 30 kg/m2 and waist circumference cut-off 88 cm
Tables 5 and 6 show the number of women divided into four groups based on their BMI and waist:height ratio. We used cut-offs for waist:height ratio of 0·5 and 0·6(Reference Ashwell and Hsieh16). A total of 46 % of women had a BMI ≥ 25 kg/m2, a waist:height ratio ≥ 0·5 or both. A total of 13 % of women were classified differently according to the two indices. Using higher cut-off points, only 1 % of women had the combination of a BMI < 30 kg/m2 and a waist:height ratio of ≥ 0·6, while 8 % had the combination of a BMI ≥ 30 kg/m2 but a waist:height ratio of < 0·6.
Numbers of women in four groups according to BMI and waist:height ratio, with BMI cut-off 25 kg/m2 and waist:height ratio cut-off 0·5
Numbers of women in four groups according to BMI and waist circumference, with BMI cut-off 30 kg/m2 and waist:height ratio cut-off 0·6
Table 7 shows the proportion of women who would be classified as ‘at risk’ according to height for each of the three measures. More women in the lowest third of height had a BMI ≥ 25 kg/m2 compared with those in the highest (45 v. 39 %). Conversely, fewer women in the lowest third of height had a waist circumference ≥ 80 cm compared with those in the highest third of height (38 v. 48 %). Adjusting waist circumference for height by using the waist:height ratio overcompensated for the excess of women in the highest third as determined by waist circumference, resulting in a larger proportion of women exceeding the cut-off point in the lowest third: 44 % in the lowest third of height had a waist:height ratio ≥ 0·5 compared with only 32 % in the highest third of height. Differences in height accounted for 0·5, 1·4 and 2·7 % of the variation in BMI, waist circumference and waist:height ratio respectively.
Proportion (%) of women with BMI of 25 kg/m2 or more, waist circumference of 80 cm or more and waist:height ratio of 0·5 or more according to height
Proportion (%) of women with waist/height0·41 of 10 or more and waist/√height of 6·32 or more according to height
When we regressed log weight and log waist on log height, we found that the optimal power index was 1·68 (95 % CI 1·59, 1·76) for weight/height and 0·41 (95 % CI 0·35, 0·47) for waist/height, i.e. in this population there was no statistically significant association between height and weight/height1·68 and height and waist/height0·41. Rounding these powers indicates that, for this population, the standard BMI formula (i.e. using a power of 2) only approximates a height-independent measure of weight but is probably the most appropriate one, and that to adjust waist circumference for height, waist circumference should be divided by the square root of height rather than height itself.
The effect of the two height adjustments on waist circumference is shown in Table 8. We have derived cut-off points for the indices waist/height0·41 and waist/√height. Each cut-off point was chosen such that 40 % of the women would be at or above this level. This was to mirror the distribution of the other indices for which approximately 40 % of the women exceeded the cut-off points. For waist/height0·41 the cut-off point was 10 and for waist/√height was 6·32. Table 8 shows that, across the thirds of height, an equal proportion of women had a waist/height0·41 ratio of 10 or more, as expected. A slightly greater proportion of women in the lowest third of height had a waist/√height ratio of 6·32 or more, compared with women in the highest third of height (41 v. 39 %).
To explore how well the three indices of adiposity reflected fatness we looked at the association with the subscapular:triceps skinfold ratio and FMI. BMI, waist circumference and waist:height ratio explained 19, 24 and 25 % of the variation in the subscapular:triceps skinfold ratio but were more strongly associated with FMI, explaining 85, 76 and 75 % of the variation respectively.
Fig. 1 shows the kernel density estimation of the probability density function for FMI for the three anthropometric indices of adiposity. Each index has been divided into three categories according to their particular cut-off points. Regardless of the index used, mean FMI was higher for each progressively higher category. However, there was considerable overlap such that, for example, women whose waist circumference was ≥ 80 and < 88 cm could have a FMI in the same range as some women whose waist circumference was < 80 cm. For the lowest category (three left-most lines), the probability density functions using the three different indices were remarkably similar. For each progressively higher category, however, greater differences between the probability density functions for each index were observed, shown by the lines not overlapping so closely.
Kernel density estimation of the probability density function for fat mass index for three anthropometric indices of adiposity. Category 1 (far left lines):
, BMI < 25 kg/m2;
, waist circumference < 80 cm;
, waist:height ratio < 0·5. Category 2 (middle lines): 
, BMI 25·0–29·9 kg/m2; 
, waist circumference 80·0–87·9 cm; 
, waist:height ratio 0·5–0·59. Category 3 (far right lines):
, BMI ≥ 30 kg/m2;
, waist circumference ≥ 88 cm;
, waist:height ratio ≥ 0·6.
Table 9 shows the number of women who might be at risk of ill health according to which of the three indices or combination of indices is used. These data are for 11 611 women on whom complete data were available. Of the single indices, waist:height ratio identified the fewest women and waist circumference the most. Some 50 % of the women in the present study had a BMI ≥ 25 kg/m2 or a waist circumference ≥ 80 cm or a waist:height ratio ≥ 0·5.
Number of women who might be at risk of ill health according to which index or combination of indices were used*
* Complete data were available on 11 611 women.
Discussion
We determined BMI, waist circumference and waist:height ratio in young women aged 20–34 years living in Southampton, UK and found they were positively associated with adiposity. Women were differentially classified depending on which index of adiposity was used. Half the women in the study would be categorised as being ‘at risk’ using the proposed cut-off points for one or other index: a BMI ≥ 25 kg/m2, a waist circumference ≥ 80 cm or a waist:height ratio ≥ 0·5.
The demographic profile of the women in the present study was similar to women of the same age, although no study can claim to be wholly representative of the general population(Reference Inskip, Godfrey, Robinson, Law, Barker and Cooper20). However, the Southampton Women's Survey is one of the largest studies of its kind to date, making it a valuable resource in providing extensive information about the body size and shape of young women. Of our women, 42 % had a BMI ≥ 25 kg/m2, while in the UK National Diet and Nutrition Survey 2000–1, 44 % of women aged 25–34 years had a BMI >25 kg/m2 and in the Health Survey for England 2002, 48 % of women had a BMI >25 kg/m2(Reference Ruston, Hoare, Henderson, Gregory, Bates, Prentice, Birch, Swan and Farron28, 29). Wells et al. report the prevalence of overweight and obesity (BMI ≥ 25 kg/m2) as 24 % in women aged 21–30 years and 40 % in women aged 31–40 years who took part in the UK National Sizing Survey, carried out in 2000–1(Reference Wells, Treleaven and Cole30).
We found that BMI had the least dependence on height of the three indices. For the population in the present study, the exponent that was most suitable for expressing weight independent of height was 1·68. By convention BMI is expressed as weight relative to height squared and clearly this is an approximation which can have important implications when used uncritically(Reference Buchan, Bundred, Kitchiner and Cole31). In five groups of women, Han et al. found that raising height to the power 0·87–1·74 corrected weight for height(Reference Han, Seidell, Currall, Morrison, Deurenberg and Lean32). We also found a positive statistically significant association between height and waist circumference (r 0·11). This is in contrast to findings from Han et al. who found a non-significant association between height and waist circumference (r − 0·036). When we divided waist circumference by height the association remained but became negative, so that fewer women in the highest third of height had a waist:height ratio ≥ 0·5 compared with the lowest third. In exploring the optimal index power of height in the relation waist/height, Han et al. reported powers of between 0·02 and 0·58. The optimal index power in the present study was 0·41, within the range reported by Han et al. and which for practical purposes would be approximated by taking the square root of height as the denominator. Although expressing waist circumference divided by √height provides a measure of waist that is approximately independent of height (see Table 8), this ratio is not straightforward to calculate. With the rising prevalence of obesity in populations, it is important to identify simple markers of the risk of ill health, and anthropometric indices have considerable utility in this regard. Clearly, excessive adiposity is disadvantageous and abnormal fat patterning carries additional risk. We considered how comparable each index was in reflecting adiposity by examining the probability distributions of FMI. We chose FMI over percentage body fat as a measure of adiposity as it better reflects the metabolic load imposed by fat mass(Reference Wells and Victora26, Reference Van Itallie, Yang, Heymsfield, Funk and Boileau33). It is possible to achieve a high percentage body fat by having a low lean mass. Furthermore, the relationship between BMI and percentage body fat is a curvilinear one, such that at high BMI, percentage body fat fails to reflect increases in adiposity. We divided the data into categories according to the particular cut-off points for each index. In the lowest category, the probability distributions of FMI were almost identical for each of the three indices. However, the distributions were more variable when comparing the three indices in the middle and upper categories. This suggests that each index captures different aspects of increased size in terms of adiposity. The extensive overlap of the probability distributions of FMI for the different categories for each index emphasises the limited specificity of anthropometry for identifying those women with greatest adiposity. Using skinfold thickness measurements, an indirect method, to estimate body fat mass will have misclassified some individuals. However, this method is widely used and the most appropriate in large-scale studies such as this. If the degree of adiposity itself carries risk of ill health, it will be important to determine which of the three indices best reflects functional state, and hence best identifies disease risk.
Currently BMI is widely used to predict ill health, but waist circumference is becoming increasingly popular. Given the argument that BMI is difficult to calculate, waist:height ratio seems unlikely to find widespread use in practice. We showed that waist circumference explained more of the variation in the distribution of fat, reflected by the subscapular:triceps skinfold ratio, whereas BMI better explained variation in total adiposity (FMI). In a study by Wells et al. women with a BMI of 24–25 kg/m2 had waist circumferences ranging from 73 to 114 cm(Reference Wells, Treleaven and Cole30). There is a trend for increasing waist circumference over time, which is not matched by a similar increase in BMI(Reference Lissner, Bjorkelund, Heitmann, Lapidus, Bjorntorp and Bengtsson34, Reference McCarthy, Ellis and Cole35). This would support the use of waist circumference over BMI as a predictor of risk and studies show that waist circumference is useful in predicting risk of disease associated with central adiposity such as CVD, type 2 diabetes, levels of blood lipids and blood pressure(Reference Yusuf, Hawken and Ounpuu36–Reference Canoy, Luben, Welch, Bingham, Wareham, Day and Khaw40). However, BMI might better predict the risk of morbidities associated with overall adiposity, such as musculo-skeletal disorders(Reference Woolf, Breedveld and Kvien41). Furthermore, while individuals do not accurately estimate their BMI, measuring waist circumference with reliability and reproducibility may not be straightforward either. Less is known about the usefulness of anthropometric indices in predicting reproductive outcomes and thus the sole use of one or other index is questionable. However, in the present study, we have collected information from a large population of young women, which, in time, will enable us to determine associations between these indices and aspects of reproductive health.
We have shown that different women are identified depending on which index of adiposity is used. Currently, BMI and waist circumference are commonly used in defining risk. Waist:height ratio only identifies a small additional number of women. Each index captures different aspects of size in terms of adiposity and we need to determine which one or combination best identifies those at specific risk. Of note is that 50 % of the young women in the present study had excess body fat according to one or other index. This serves to emphasise the magnitude of the public health problem. It is clearly inadequate simply to target individuals at the extreme upper end of the distribution and there is the need to shift the adiposity distribution of this half of the population downwards.
Acknowledgements
We are grateful to all the general practitioners in Southampton who made the present study possible, to Southampton Women's Survey staff, and particularly to the women of Southampton for taking part. The Southampton Women's Survey is grateful for financial support from the UK Medical Research Council, the University of Southampton and the Dunhill Medical Trust. S. L. D. is supported by the Biotechnology and Biological Sciences Research Council. H. M. I. and K. M. G. initiated the study. H. M. I. and the Southampton Women's Survey Study Group were responsible for data collection. All authors contributed to analysis and preparation of the manuscript. We declare that we have no conflicts of interest.
