Adrian Cameron, International Diabetes Institute, 250 Kooyong Road, Caulfield 3162, Victoria, Australia, (fax: 61 3 9258 5090; e-mail: firstname.lastname@example.org).
Objectives. To compare body mass index (BMI), waist circumference and waist–hip ratio (WHR) as indices of obesity and assess the respective associations with type 2 diabetes, hypertension and dyslipidaemia.
Design and setting. A national sample of 11 247 Australians aged ≥25 years was examined in 2000 in a cross-sectional survey.
Main outcome measures. The examination included a fasting blood sample, standard 2-h 75-g oral glucose tolerance test, blood pressure measurements and questionnaires to assess treatment for dyslipidaemia and hypertension. BMI, waist circumference and WHR were measured to assess overweight and obesity.
Results. The prevalence of obesity amongst Australian adults defined by BMI, waist circumference and WHR was 20.8, 30.5 and 15.8% respectively. The unadjusted odds ratio for the fourth vs. first quartile of each obesity measurement showed that WHR had the strongest relationship with type 2 diabetes, dyslipidaemia (women only) and hypertension. Following adjustment for age, however, there was little difference between the three measures of obesity, with the possible exceptions of hypertension in women, where BMI had a stronger association, and dyslipidaemia in women and type 2 diabetes in men, where WHR was marginally superior.
Conclusions. Waist circumference, BMI and WHR identified different proportions of the population, as measured by both prevalence of obesity and cardiovascular disease (CVD) risk factors. Whilst WHR had the strongest correlations with CVD risk factors before adjustment for age, the three obesity measures performed similarly after adjustment for age. Given the difficulty of using age-adjusted associations in the clinical setting, these results suggest that given appropriate cut-off points, WHR is the most useful measure of obesity to use to identify individuals with CVD risk factors.
Whilst precise, sophisticated techniques for measuring body fat distribution and body composition are available [1, 2], they are generally not appropriate outside specific research settings. Simple anthropometric measurements have been used as surrogate measurements of obesity and have more practical value in both clinical practice and for large-scale epidemiological studies. Body mass index (BMI), which relates weight to height, is the most widely used and simple measure of body size, and is frequently used to estimate the prevalence of obesity within a population [3, 4]. BMI has been found to be consistently associated with an increased risk of cardiovascular disease (CVD) and type 2 diabetes , yet this measurement does not account for variation in body fat distribution and abdominal fat mass, which can differ greatly across populations and can vary substantially within a narrow range of BMI . Excess intra-abdominal fat is associated with greater risk of obesity-related morbidity than is overall adiposity [7, 8]. Thus, measurements of waist circumference and waist–hip ratio (WHR) have been viewed as alternatives to BMI, with both measures regularly used in the clinical and research settings. Waist circumference has been shown to be the best simple measure of both intra-abdominal fat mass and total fat [9, 10].
Several studies in adults have reported a stronger positive association between cardiovascular risk factors such as hypertension, and lipid and glucose concentrations, with abdominal adiposity (measured by waist circumference or WHR) than with overall adiposity (as measured by BMI) [11–17], although BMI has also been reported as being one of the most important risk factors for type 2 diabetes . Despite the fact that a close relationship is apparent between abdominal adiposity and risk of CVD, the current waist circumference cut-off points suggested by the World Health Organization (WHO) are not based on associations with CVD risk factors, but rather on their correlation with corresponding values of BMI [6, 19].
Few large-scale, population-based studies exist where it has been possible to examine all three measures of obesity in relation to their associations with a range of CVD risk factors. The AusDiab study, a large representative sample of the Australian adult population is one such study. Here BMI, waist circumference and WHR are compared as indices of obesity and assess their respective associations with type 2 diabetes, hypertension and dyslipidaemia.
A detailed description of the methodology is reported elsewhere . A stratified cluster sample of the national population was drawn from 42 randomly selected areas (Census Collector Districts, CDs) across Australia (six CDs in each of the six states and the northern territory). CDs containing fewer than 100 persons aged 25 years or over, those classified as 100% rural, or those containing more than 10% Aboriginal or Torres Strait Islander populations were excluded. Within each CD, all homes were approached, and adults aged 25 years or over who were usual residents were invited to attend the survey. The survey took place between May 1999 and December 2000, and consisted of a short household interview, followed by a biomedical examination at the local survey centre. Householders not available at the initial door knock interview were approached again on up to four more occasions. Of those homes with at least one eligible adult where a response was obtained, 70% took part in the household questionnaire. The final survey sample (those attending the biomedical examination) included 11 247 adults, representing 55% of those completing the household interview.
Following an overnight fast of at least 8 h, a blood sample was collected for plasma glucose, HDL cholesterol, LDL cholesterol and triglycerides. All participants, except those with diabetes who were taking oral hypoglycaemic agents or insulin, then had an oral glucose tolerance test (OGTT).
Plasma glucose, serum HDL cholesterol and triglycerides were determined enzymatically using an Olympus AU600 analyser (Olympus Optical Co Ltd, Tokyo, Japan).
Glucose tolerance was classified according to the WHO criteria . Participants who reported a history of physician-diagnosed diabetes and who were (i) taking oral hypoglycaemic tablets or insulin injections, or (ii) had a fasting plasma glucose (FPG) level ≥7.0 mmol l−1 or 2 h plasma glucose (2 hPG) level ≥11.1 mmol l−1, were classified as having known diabetes mellitus (KDM). Subjects not reporting diabetes and who had FPG ≥7.0 mmol l−1 or 2 hPG ≥11.1 mmol l−1 were classified as having newly-diagnosed diabetes mellitus (NDM). For those without KDM, FPG <7.0 mmol l−1 and 2 hPG ≥7.8 mmol l−1 but <11.1 mmol l−1 indicated impaired glucose tolerance (IGT); impaired fasting glucose (IFG) was defined as FPG ≥6.1 mmol l−1 and <7.0 mmol l−1, with 2 hPG <7.8 mmol l−1, and normal glucose tolerance was defined as FPG <6.1 mmol l−1 and 2 hPG <7.8 mmol l−1. Those individuals who were classified as having type 1 diabetes, based on their age of diagnosis of diabetes, age when started insulin and BMI , were not included in any analysis involving diabetes.
Height was measured to the nearest 0.5 cm without shoes using a stadiometer. Each participant stood with heels, buttocks and shoulders resting lightly against the backing board so that the Frankfort plane (a line connecting the superior border of the external auditory meatus with the infraorbital rim) was horizontal (i.e. parallel to the floor). Weight was measured after removal of shoes and when wearing light clothing only, using a mechanical beam balance, and was recorded to the nearest 0.1 kg. BMI was calculated as weight (kg)/height (m)2. Those with a BMI of 25.0–29.9 kg m−2 were classified as overweight, whilst those with a BMI ≥ 30.0 kg m−2 were classified as obese . Waist circumference was measured using a steel measuring tape, with measurements made halfway between the lower border of the ribs, and the iliac crest in a horizontal plane. Hip circumference was measured at the widest point over the buttocks. For each of waist and hip circumference, two measurements to the nearest 0.5 cm were recorded. If the variation between the measurements was greater than 2 cm, a third measurement was taken. The mean of the two closest measurements was calculated. Men with a waist circumference 94–101.9 cm and women with a waist circumference 80–87.9 cm were classified as overweight, whilst men with a waist circumference ≥102.0 cm and women with a waist circumference ≥88.0 cm were classified as obese . WHR was obtained by dividing the mean waist circumference by the mean hip-circumference. Men with a WHR 0.90–0.99 and women with a WHR 0.80–0.84 were classified as overweight, whilst men with a WHR ≥ 1.00 and women with a WHR ≥ 0.85 were classified as obese [6, 23].
Blood pressure was measured in a seated position after the participant had rested for at least 5 min. After measurement of the circumference of the mid-upper arm, a cuff of suitable size was applied to the participant's exposed upper arm (the arm not used for blood collection), which was supported by the table at heart level. In the state of Victoria only, blood pressure was measured with a standard mercury sphygmomanometer using the first and fifth Korotkoff sounds, recorded to the nearest 2 mmHg. In all other states, blood pressure was measured using a Dinamap semi-automatic oscillometric recorder (Critikon, Tampa, FL, USA), where three readings were taken at 1-min intervals. Based on a comparison study [conducted on every 20th person in the last six states surveyed (n = 469)] of readings using the sphygmomanometer and the Dinamap, an adjustment was made to all diastolic blood pressure readings recorded in the state using the sphygmomanometer. To obtain the final measure of blood pressure, the mean of the first two readings was calculated, unless the difference between these readings was greater than 10 mmHg, in which case the mean of the two closest of three measurements was used. Participants were classified as hypertensive if they were on treatment for hypertension, had a mean systolic reading ≥140 mmHg or a mean diastolic reading ≥90 mmHg. Dyslipidaemia was defined as HDL cholesterol < 1.0, triglycerides ≥2.0, or if participants reported lipid lowering medication was being taken [24, 25].
To account for the clustering and stratification of the survey design, and to adjust for nonresponse, the data were weighted to match the age and gender distribution of the 1998 estimated residential population of Australia aged 25 years and over unless otherwise stated. The weighting factor was based on the probability of selection in each cluster. Therefore all figures relate to the total 1998 Australian population aged 25 years and over. Prevalence rates and 95% confidence intervals were calculated using Stata Statistical Software Release 6.0 (1999) (StataCorp., College Station, TX, USA), accounting for the clustered and stratified nature of the survey design. Quartiles of BMI, waist circumference and WHR were obtained separately for male and female subjects.
Logistic regression was used to obtain crude and age-adjusted odds ratios (with 95% confidence intervals) of type 2 diabetes, hypertension and dyslipidaemia, according to each quartile of BMI, waist circumference and WHR. Individuals with type 1 diabetes were excluded from all analyses involving diabetes.
The prevalence of obesity and overweight by BMI, waist circumference and WHR is given in Table 1. Using BMI, 59.8% of Australian adults were considered to be either overweight or obese, with one-third of them (20.8% overall) falling into the obesity category. When using waist circumference, 30.5% of Australian adults were obese, whilst only 15.8% were obese when classification is based on WHR. Differences between genders in the prevalence of obesity were observed for the three different measures. Amongst male subjects, obesity as defined by WHR accounted for only 9.5% of the total, compared with 26.8% for waist circumference and 19.3% for BMI (Table 1). In female subjects, however, the obese groups defined by both BMI and WHR were of a similar magnitude (22%), with the prevalence of obesity defined by waist circumference being considerably higher (34.1%).
Table 1. Prevalence (weighted %) of obesity by body mass index, waist circumference and waist–hip ratio
Body mass index
M, male subjects; F, female subjects.
Amongst those in the obese category, the prevalence of each of type 2 diabetes, hypertension and dyslipidaemia in both male and female subjects was highest when WHR was used to define obesity (Table 2). Amongst male subjects, despite the prevalence of type 2 diabetes, hypertension and dyslipidaemia being higher in the group defined as obese by WHR, this group actually contained the fewest subjects with each risk factor when compared with the obese groups defined by BMI and waist circumference. This was because of the lower prevalence of obesity when defined by WHR.
Table 2. Prevalence of type 2 diabetes, hypertension and dyslipidaemia by three measures of obesity
Type 2 diabetes
Data are weighted percentages (actual n).
BMI, body mass index; WC, waist circumference; WHR, waist–hip ratio.
Over 75% of obese male subjects and over 65% of obese female subjects had at least one cardiovascular risk factor (type 2 diabetes, hypertension or dyslipidaemia) (Table 3). Even in the group defined as being of normal weight by the cut-off points used, the presence of at least one of these factors was common, although female subjects had a slightly lower prevalence of risk factors than male subjects in all groups. For both male and female subjects, there was little difference in the prevalence of one or more risk factors between the three measures of obesity. Amongst both obese male and female subjects, the prevalence of having all three CVD risk factors (type 2 diabetes, hypertension and dyslipidaemia) was between 6.9 and 9.8% for all three measures of obesity, with the prevalence highest for those defined as obese by WHR (9.8 and 9.1% for male and female subjects, respectively).
Table 3. Prevalence of one or more risk factors (type 2 diabetes, hypertension, dyslipidaemia) by obesity category
Data are weighted percentages (actual n).
BMI, body mass index; WC, waist circumference; WHR, waist–hip ratio.
The respective correlations between the measures of obesity and systolic blood pressure, triglycerides, HDL, fasting and postload glucose are given in Table 4. For female subjects, the strongest correlations with all five risk factors were with waist circumference, although the difference between the different measures is not great in most cases. WHR, however, showed the strongest correlations with three of the five conditions displayed amongst male subjects. Only HDL (for which BMI showed a stronger correlation) and fasting blood glucose (for which waist circumference had a stronger correlation) are the exceptions.
Table 4. Correlation of three measures of obesity with risk factors
Data are correlation coefficients.
P-value < 0.001 for all correlation values.
FPG, fasting plasma glucose; 2HPG, 2 h plasma glucose; M, male subjects; F, female subjects.
The crude and age-adjusted associations between type 2 diabetes, hypertension and dyslipidaemia, with waist circumference, WHR and BMI are shown in Fig. 1. WHR had the strongest relationship with type 2 diabetes in men, however, age adjustment attenuated this correlation, with waist circumference and WHR showing similar associations following adjustment. Both WHR and waist circumference were superior to BMI in their association with type 2 diabetes. WHR also showed the strongest relationship with type 2 diabetes in women, although again, age adjustment reduces the strength of the correlation, with all three measures becoming equal once adjusted. Adjusted odds ratios (OR) for type 2 diabetes in male subjects increased at a lower level of waist circumference [OR for second quartile = 2.07 (95% CI, 0.9–4.5)] and WHR [OR for second quartile = 2.17 (1.0–4.8)] than in female subjects [OR for WC second quartile = 0.99 (0.4–2.3); OR for WHR second quartile = 1.0 (0.4–2.4)].
In both male and female subjects, WHR had the strongest association with hypertension before age adjustment. Following correction for age, however, BMI appeared to be superior to both waist circumference and WHR (Fig. 1). Age did not appear to confound the relationship between obesity (by any measure) and dyslipidaemia in male subjects, although the same was not true in female subjects, with age adjustment attenuating the associations between each of WHR, waist circumference and BMI with dyslipidaemia. For both male and female subjects, WHR was the strongest predictor of having at least one of diabetes, hypertension or dyslipidaemia, although all measures performed equally well after adjustment for age.
The definitions of overweight and obesity recommended by the WHO (BMI >25 and 30 respectively) are a result of the relationship between BMI with morbidity and mortality outcomes . BMI is the most frequently used measure of obesity because of the robust nature of the measurements of weight and height, and the widespread use of these measurements in population health surveys. BMI does not, however, take into account the proportion of weight related to increased muscle or the distribution of excess fat within the body, both of which affect the health risks associated with obesity.
Individuals with a similar BMI can vary considerably in their abdominal-fat mass, with premenopausal women typically having half the abdominal-fat mass of men . For this reason, a measure of obesity that takes into account the increased risk of obesity-related illness because of the accumulation of abdominal fat is desirable. WHR was previously acknowledged as the clinically accepted method of identifying patients with excess abdominal fat accumulation. However, more recently, waist circumference alone has been suggested as being a more practical measure of intra-abdominal fat mass and total body fat. Indeed, waist circumference has been found in some studies to be more closely correlated with the level of abdominal visceral adipose tissue than is WHR [14, 27, 28].
Waist circumference correlates closely with both BMI and WHR and has been shown to reflect the level of risk for CVD and other chronic diseases , although the level of risk varies between population groups. In a comparison of the utility of various anthropometric measures in identifying CVD risk factors in a Hong Kong population , Ho et al. found that BMI and waist circumference proved most effective for men, whilst waist circumference and WHR were preferable for women. This study showed an overlap between BMI and waist circumference, with the relationship between BMI and the metabolic variables studied disappearing after adjustment for waist circumference and vice versa. This finding leads to speculation that waist circumference may provide a useful index reflecting general and central obesity.
We investigated the correlations between each of three anthropometric measures of obesity (BMI, WHR and waist circumference), and CVD risk factors such as type 2 diabetes, hypertension and dyslipidaemia in a predominantly Europid or White Caucasoid population. Using conventional cut-off points for overweight and obesity, these three commonly used anthropometric measures clearly identify different subpopulations, as measured by both prevalence of obesity and prevalence of CVD risk factors amongst those defined as obese. In order to identify true differences between the three measures in their ability to identify individuals at greatest risk of CVD, a standardized method of comparison needs to be used, rather than the conventionally used arbitrary cut-off points for obesity. For this reason, the risk of diabetes, hypertension and dyslipidaemia by obesity status presented in Fig. 1 is based on quartiles of BMI, waist circumference and WHR. From the OR in Fig. 1, it is clear that obesity is strongly associated with each of hypertension, type 2 diabetes and dyslipidaemia, as is often reported.
We observed that the unadjusted associations with each of the CVD risk factors in both male and female subjects were strongest for obesity when defined using WHR. Dyslipidaemia in male subjects is the only exception, with none of the obesity measures showing a particularly stronger association. After adjustment for age, however, the strength of the associations with each of hypertension, type 2 diabetes and dyslipidaemia (in male subjects only) is similar for each of the three obesity measures. Amongst female subjects, WHR maintains its stronger association with dyslipidaemia even after age adjustment.
The difference between the patterns of association between the three obesity measures and the various CVD risk factors when adjusted and unadjusted for age is likely because of the tendency for obesity, as defined by WHR in particular, to identify older individuals. Having recognized that WHR appears to be identifying older individuals, the question then arises as to whether there is any value in considering OR adjusted for age with regard to the clinical setting. It is, of course, difficult in a clinical setting to take into account adjustment for age when making clinical treatment decisions for an individual patient. Without a complicated algorithm similar to the Framingham equation (used to calculate CVD risk), it is impossible to precisely take into account multiple variables (such as age and level of obesity) when determining levels of risk. For example, what is the differential in risk of CVD between a 60-year old who is classified as nonobese and a 50-year old who is classified as obese? Without a detailed knowledge of the interaction of obesity with age, it is difficult to ascertain which measure of obesity is most appropriate and also where the cut-off for obesity lies based on an increase in morbidity and mortality. Without the development of a complicated set of age-specific cut-offs, it may be better to simply be guided by an analysis unadjusted for age. Based purely on the results presented here, and ignoring the currently used cut-off points, it therefore appears better in clinical practice to use WHR to identify those patients who may be at increased risk of having risk factors for CVD.
However, as the results presented in Fig. 1 are based on quartiles of obesity, the conclusions presented above are not helpful when comparing the commonly used and accepted definitions of obesity as defined by WHR, waist circumference and BMI. For this reason, an analysis of the prevalence of each of the CVD risk factors is also presented. Table 2 shows that obesity defined by WHR carries the highest risk for each of type 2 diabetes, hypertension and dyslipidaemia in both men and women. However, the largest number with each of these CVD risk factors is detected amongst the group defined as obese by waist circumference (because of the much higher prevalence of obesity when using this measure).
When attempting to determine the superiority of one or other measure of obesity, measuring the prevalence of various disease-risk factors amongst those identified as obese, or how they correlate with obesity variables forms is only part of the study. Epidemiological data needs to be considered in conjunction with a consideration of the ease and accuracy of physical measurement. It is common clinical experience that hip measurements in the severely obese are difficult and unreliable. Both waist circumference and particularly WHR suffer from measurement error, and BMI is both simple and routinely measured. BMI would therefore appear to be the more appropriate choice in terms of ease of measurement. However, the problems with measurement of waist circumference and WHR are largely restricted to the extremely obese, for whom the assessment of further CVD risk factors is likely to be routine in any case. Therefore, considering that measurement of obesity in the clinical setting is usually conducted primarily to inform further investigations, there is no strong argument for any of the three measurements to be preferred based on ease of measurement.
We acknowledge that the AusDiab study was cross-sectional, and therefore causality cannot be determined from the associations observed. Furthermore, the level of response to the study should be considered, and the previously reported small differences between responders and nonresponders , when generalizing these results to the Australian population. In addition, although AusDiab was designed to provide estimates representative of the adult Australian population, the exclusion criteria used may have resulted in under-representation of indigenous and rural Australians.
In summary, this large, nationally representative population demonstrates that unadjusted for age, obesity as defined by WHR is superior in its ability to identify those with known risk factors for CVD.
As we have reported previously, the prevalence of obesity and overweight is escalating in Australia with a clear association between obesity and an increase in time spent in sedentary activities . The results presented here further highlight the consequences of the increased obesity prevalence observed, and suggest that the preferred measure of obesity to predict the presence of CVD risk in the clinical setting is WHR. Longitudinal follow-up of the AusDiab cohort is planned to assess the ultimate impact of the associations observed on CVD morbidity and mortality.
Conflict of interest statement
No conflict of interest was declared.
We are grateful to the following for their support of the study: The then Commonwealth Department of Health and Aged Care, Eli Lilly (Aust) Pty Ltd, Janssen–Cilag (Aust) Pty Ltd, Knoll Australia Pty Ltd, Merck Lipha s.a., Alphapharm Pty Ltd, Merck Sharp & Dohme (Aust), Roche Diagnostics, Servier Laboratories (Aust) Pty Ltd, SmithKline Beecham International, Pharmacia and Upjohn Pty Ltd, BioRad Laboratories Pty Ltd, Hitech Pathology Pty Ltd, the Australian Kidney Foundation, Diabetes Australia (Northern Territory), Queensland Health, South Australian Department of Human Services, Tasmanian Department of Health and Human Services, Territory Health Services, Victorian Department of Human Services and Health Department of Western Australia.
For their invaluable contribution to the field activities of AusDiab, we are grateful to Annie Allman, Adam Meehan, Claire Reid, Alison Stewart, Robyn Tapp and Fay Wilson.
Finally our thanks goes to the local collaborating centres, including Sir Charles Gairdner Hospital (Western Australia), the Prince of Wales Hospital (New South Wales), the Menzies Centre for Population Health Research (Tasmania), the Queen Elizabeth Hospital (South Australia), the Menzies School of Health Research (Northern Territory), Queensland Health, the Monash Medical Centre Department of Nephrology (Victoria), and the Centre for Eye Research Australia (Victoria).