Diabetologia

, Volume 53, Issue 7, pp 1322–1330

Anthropometric measures and glucose levels in a large multi-ethnic cohort of individuals at risk of developing type 2 diabetes

  • L. de Koning
  • H. C. Gerstein
  • J. Bosch
  • R. Diaz
  • V. Mohan
  • G. Dagenais
  • S. Yusuf
  • S. S. Anand
  • for the EpiDREAM Investigators
Article

DOI: 10.1007/s00125-010-1710-3

Cite this article as:
de Koning, L., Gerstein, H.C., Bosch, J. et al. Diabetologia (2010) 53: 1322. doi:10.1007/s00125-010-1710-3

Abstract

Aims/hypotheses

We determined: (1) which of BMI, waist circumference, hip circumference and WHR has the strongest association and explanatory power for newly diagnosed type 2 diabetes and glucose status; and (2) the impact of considering two measures simultaneously. We also explored variation in anthropometric associations by sex and ethnicity.

Methods

We performed cross-sectional analysis of 22,293 men and women who were from five ethnic groups and 21 countries, and at risk of developing type 2 diabetes. Standardised anthropometric associations with type 2 diabetes and AUC of glucose status from OGTT (AUCOGTT) were determined using multiple regression. Explanatory power was assessed using the c-statistic and adjusted r2.

Results

An increase in BMI, waist circumference or WHR had similar positive associations with type 2 diabetes, AUCOGTT and explanatory power after adjustment for age, sex, smoking and ethnicity (p < 0.01). However, using BMI and WHR together resulted in greater explanatory power than with other models (p < 0.01). Associations were strongest when waist circumference and hip circumference were used together, a combination that had greater explanatory power than other models except for BMI and WHR together (p < 0.01). Results were directionally similar according to sex and ethnicity; however, significant variations in associations were observed among these subgroups.

Conclusions/interpretation

The combination of BMI and WHR, or of waist circumference and hip circumference has the best explanatory power for type 2 diabetes and glucose status compared with a single anthropometric measure. Measurement of waist circumference and hip circumference is required to optimally identify people at risk of type 2 diabetes and people with elevated glucose levels.

Keywords

Anthropometric measuresEthnicityOral glucose tolerance testType 2 diabetes

Abbreviations

AUCOGTT

AUC of glucose status

DREAM

Diabetes Reduction Assessment with Ramipril and Rosiglitazone Medication

EpiDREAM

Epidemiological arm of the DREAM study

Introduction

Type 2 diabetes mellitus is a major risk factor for blindness, limb amputation, cardiovascular disease and death. Among the preventable risk factors for type 2 diabetes, overweight (BMI ≥ 25 kg/m2) and obesity (BMI ≥ 30 kg/m2) are regarded as the most important [1]. Currently more than 2 billion people over the age of 15 in the world (∼29% of total population) are overweight or obese, a figure expected to rise to over 3 billion (∼40%) by 2030 [2, 3]. Most of this increase will occur in low-income countries where sedentary lifestyles and high-energy diets are being adopted [1]. By 2030, the prevalence of type 2 diabetes is projected to rise to 7.3%, afflicting over 380 million people worldwide [4, 5].

Traditionally, BMI has been used to define overweight and obesity, and is predictive of cardiometabolic risk including incident type 2 diabetes [6] and cardiovascular events such as myocardial infarction [7]. However, BMI has some limitations. First, it does not distinguish between individuals with high muscle mass, excess fat or abdominal obesity. A preponderance of abdominal and visceral fat is strongly associated with insulin resistance, type 2 diabetes and high lipids [8]. Recently, in a large international case–control study, measures of abdominal obesity (waist circumference and WHR) were more strongly associated with myocardial infarction than BMI [7]. Second, BMI may not be appropriate to use in all individuals, as the association of BMI with cardiometabolic risk seems to vary according to sex [9, 10] and ethnicity [6, 11, 12].

The primary objective of this investigation was to determine which of the anthropometric measures BMI, waist circumference, hip circumference and WHR has the strongest association and best explanatory power for glucose levels in a large, ethnically diverse cohort of individuals at risk of developing type 2 diabetes. The second objective was to measure the impact of considering an additional anthropometric measure. The third was to explore whether associations differ by sex and ethnicity.

Methods

Written consent was obtained from all participants prior to initiating this study, which was approved by the Ethics Committees at participating institutions. The study complied with the declaration of Helsinki. Between July 4 2001 and August 15 2003, 24,595 men and women aged 30 years or over at 191 centres from 21 countries (Electronic supplementary material [ESM] Table 1) were screened for entry into the Diabetes Reduction Assessment with Ramipril and Rosiglitazone Medication (DREAM) trial [13]. Clinical centres screened individuals who had an increased risk of type 2 diabetes as defined by family history, ethnicity, gestational diabetes and abdominal obesity. All individuals underwent an 8 h fast and a 75 g OGTT. On the basis of a single test, fasting venous plasma glucose ≥7.0 mmol/l (126 mg/dl) or a 2 h glucose ≥11.1 mmol/l (200 mg/dl) was defined as being compatible with a diagnosis of type 2 diabetes [14]. There were no inclusion criteria for BMI. Screened persons with impaired fasting glucose or impaired glucose tolerance were asked to participate in the DREAM trial [13]. Trial participants and non-participants became part of an epidemiological arm of DREAM called EpiDREAM, which is the cohort used for this analysis.

Anthropometric measures including weight (kg), height (m), waist circumference (cm) and hip circumference (cm) were taken using a standardised protocol. Standing height was measured to the nearest 0.1 cm with the participant looking straight ahead in bare feet and with his/her back against a wall. Weight was measured to the nearest 0.1 kg in light clothing. Waist and hip circumference were measured in duplicate using a non-flexible tape measure with an attached spring balance exerting a weight of 750 g. Waist circumference was assessed at the smallest diameter between the costal margin and iliac crest. Hip circumference was assessed at the level of the greater trochanters. Averages of the two measures were used in all analyses.

Information on ethnicity, medical history and smoking status was collected by self-administered questionnaire.

Statistical analysis

For this analysis we excluded participants (n = 1,847) who: (1) could not be assigned to one of five ethnic groups (Aboriginal, African, South Asian, European and Latin American); (2) had missing data (n = 373); or (3) had been previously diagnosed with type 2 diabetes (n = 82). This left 22,293 participants. Student’s t test, one-way ANOVA and the χ2 test were used to evaluate differences in participant characteristics. Simple linear and logistic regression were used to test for trends between anthropometric measures and participant characteristics.

Multiple logistic and linear regression were used to determine standardised (per 1 SD) associations of anthropometric measures (BMI, waist circumference, hip circumference, WHR) with type 2 diabetes and a continuous measure of glucose status defined as the AUC (line) of fasting and 2 h post-load plasma glucose values, i.e. AUC of glucose status (AUCOGTT), which was calculated (mmol/l × h) as fasting + 2 h glucose, after adjusting for age, sex, smoking status and ethnicity. To validate the use of the AUCOGTT, we repeated our analyses using fasting and 2 h glucose levels separately.

Significant differences in standardised associations were defined to have occurred when the 95% CIs of ORs or beta coefficients did not overlap. We determined the model explanatory power for type 2 diabetes by calculating the c-statistic for receiver operating characteristic curves and used the Mann–Whitney test to test for differences between models arising from the same dataset [15]. Explanatory power for linear regression was evaluated using the adjusted R2 statistic.

To determine the impact of considering an additional anthropometric measure on association strength and explanatory power, an extra term was included in the models. This was only done if the correlation between the two measures was less than 0.80. Models containing waist circumference or hip circumference with WHR were not tested because of difficulties in interpreting the coefficients. To test whether anthropometric associations were consistent across levels of the second measure, multiplicative terms were included in the models (e.g. waist circumference × hip circumference). To illustrate the effect of considering two measures simultaneously, measures were collapsed into thirds (tertiles), and ORs and AUCOGTT levels were calculated for the nine categories. Analyses were repeated among men, women and ethnic groups.

To test for interactions with sex, multiplicative terms were included in the regression models (e.g. waist circumference × sex). To test for interactions with ethnicity, log-odds ratios and beta coefficients (adjusted for age, sex and smoking) were evaluated over the same interval (1 SD) and compared using Cochran’s Q test for heterogeneity [16]. When there was significant ethnic heterogeneity, we defined significant intra-ethnic differences as occurring when the 95% CIs of ORs or beta coefficients did not overlap. SAS version 9.1 (SAS, Cary, NC, USA) and Microsoft Excel version 9.0 (Redmond, WA, USA) were used for all data analyses.

Results

Participant characteristics

Participant characteristics are listed in Table 1. Of the 22,293 individuals, 60.0% were women and 47.1% were of non-white ethnicity. The prevalence of newly diagnosed type 2 diabetes was 14.0% and mean AUCOGTT was 13.2 mmol/l × h. Mean AUCOGTT among those with type 2 diabetes was significantly higher than in healthy participants (20.9 vs 12.0 mmol/l × h; p < 0.01). Anthropometric measures were significantly and positively associated with type 2 diabetes and AUCOGTT, as well as with age, male sex, smoking, African ethnicity and European ethnicity (data not shown; p < 0.01 for trends). Anthropometric measures were significantly (p < 0.01) correlated with each other (BMI with hip circumference r = 0.82, waist circumference with hip circumference r = 0.76, BMI with waist circumference r = 0.75, waist circumference with WHR r = 0.57, BMI with WHR r = 0.14, hip circumference with WHR r = −0.08).
Table 1

Characteristics of participants in EpiDREAM

Characteristic

Overall

Women

Men

Aboriginal

African

South Asian

European

Latin American

Per cent (n)

100 (22,293)

60.0 (13,374)

40.0 (8,919)

12.1 (2,697)

6.5 (1,458)

20.3 (4,535)

52.9 (11,795)

8.1 (1,808)

Age (years)

52.2 (11.4)

52.1 (11.3)

52.4 (11.5)

52.0 (11.6)

53.9 (11.0)

44.7 (9.2)

55.1 (10.8)

51.3 (11.4)

Post-menopausal women, % (n)

28.5 (5,054)

37.9 (5,054)

 

32.2 (799)

37.2 (406)

14.7 (563)

31.7 (2,799)

31.9 (487)

Smoker, % (n)a

44.3 (9,861)

36.6 (4,899)

55.6 (4,962)

55.4 (1,495)

37.5 (546)

15.8 (714)

52.8 (6,217)

49.1 (889)

Never smoked, % (n)

55.8 (12,432)

63.4 (8,475)

44.4 (3,957)

44.6 (1,202)

62.6 (912)

84.3 (3,821)

47.3 (5,578)

50.8 (919)

BMI (kg/m2)

29.9 (6.2)

30.6 (6.6)

28.9 (5.3)

31.4 (6.2)

32.3 (7.0)

26.3 (4.5)

30.5 (6.1)

30.9 (6.0)

Obese, % (n)

42.8 (9,533)

47.6 (6,362)

35.6 (3,171)

53.3 (1,438)

57.6 (840)

17.6 (799)

47.0 (5,546)

50.3 (910)

Waist circumference (cm)

95.2 (14.4)

92.6 (14.4)

99.0 (13.4)

98.3 (14.6)

97.6 (14.5)

89.3 (11.2)

96.4 (14.9)

95.3 (13.6)

Hip circumference (cm)

106.8 (13.3)

108.6 (14.3)

104.1 (11.1)

109.6 (13.2)

111.4 (14.3)

100.1 (10.2)

108.1 (13.4)

107.0 (12.7)

WHR (units)

0.892 (0.089)

0.853 (0.080)

0.951 (0.068)

0.897 (0.086)

0.877 (0.077)

0.894 (0.082)

0.892 (0.095)

0.891 (0.084)

Type 2 diabetes, % (n)

14.0 (3,112)

12.3 (1,650)

16.4 (1,462)

9.9 (267)

18.2 (265)

14.1 (640)

14.5 (1712)

12.6 (228)

AUCOGTT (mmol/l × h)

13.2 (4.3)

13.0 (3.9)

13.6 (4.7)

12.9 (3.8)

14.0 (4.7)

13.2 (5.4)

13.3 (3.8)

12.9 (4.0)

Values are means (SD) for continuous variables, unless otherwise indicated

aCurrent or previous smoker

All characteristics differed significantly by sex (p < 0.01 t test); similarly, significant inter-ethnic differences were seen among all characteristics (p < 0.01 ANOVA)

Aboriginal: Indigenous North American, Indigenous South American or Australian Aborigine; African: Black; South Asian: Indian, Sri Lankan, Pakistani, Bangladeshi; European: White; Latin American: Mixed European and Native South American

Adjusted associations of anthropometric measures with type 2 diabetes and AUCOGTT

Among individual metrics, an increase in BMI, waist circumference, hip circumference or WHR was associated with a significant increase in the odds of type 2 diabetes and AUCOGTT after adjusting for age, sex, smoking and ethnicity (Fig. 1). The associations for hip circumference were significantly weaker. For type 2 diabetes, the model containing waist circumference had the highest explanatory power, which was significantly greater than the model containing BMI (Fig. 1a). Similarly, the model containing waist circumference had the highest explanatory power for AUCOGTT (Fig. 1b).
https://static-content.springer.com/image/art%3A10.1007%2Fs00125-010-1710-3/MediaObjects/125_2010_1710_Fig1_HTML.gif
Fig. 1

Standardised associations of anthropometric measures with type 2 diabetes (a) and AUCOGTT (b). Odds ratios (a) and beta coefficients (b) represent changes in the odds of type 2 diabetes (T2D) or AUCOGTT for one SD increase in anthropometric measures independently of age, sex, smoking, ethnicity and other anthropometric measures. The + symbol indicates: additionally adjusted for anthropometric measure after symbol. SDs: BMI 6.2 kg/m2, waist circumference (WC) 14.4 cm, hip circumference (HC) 13.3 cm, WHR 0.09. c-Statistic significantly higher for B than A (p < 0.01). p < 0.02 for c-statistic higher than model containing BMI alone; **p < 0.01 for c-statistic lower than model containing BMI alone; p < 0.02 for c-statistic higher than all models. Standardised beta coefficients and p values for interaction terms: (1) BMI × waist circumference: type 2 diabetes β = −0.06, p < 0.01; AUCOGTTβ = −0.13 mmol/l × h, p < 0.01; (2) BMI × WHR: type 2 diabetes β = −0.02, p = 0.36; AUCOGTTβ = −0.05 mmol/l × h, p = 0.03; and (3) WC × HC: type 2 diabetes β = −0.02, p = 0.19; AUCOGTTβ = −0.04 mmol/l × h, p = 0.03

Due to the high correlation between BMI and hip circumference (r = 0.82), the only additional models that were tested included BMI + waist circumference, BMI + WHR and waist circumference + hip circumference. Models containing BMI and WHR had the greatest explanatory power (T2D: p < 0.01) (Fig. 1). Participants in the highest tertile combination of BMI and WHR had a fourfold greater odds of type 2 diabetes (AUCOGTT 14.6 mmol/l × h) than participants in the lowest combination (AUCOGTT 12.0 mmol/l × h) (Fig. 2a and c). Models containing waist circumference and hip circumference had higher explanatory power (T2D: p < 0.01) than other models except those containing BMI and WHR (Fig. 1). Participants in the highest tertile of waist circumference and lowest tertile of hip circumference had a 2.25-fold higher odds of type 2 diabetes (AUCOGTT 14.2 mmol/l × h) than those in the lowest tertile of waist circumference and highest tertile of hip circumference (AUCOGTT 10.9 mmol/l × h) (Fig. 2b and d). Using BMI and waist circumference together had little impact on explanatory power; however, BMI associations were attenuated compared with waist circumference. Interactions between anthropometric measures in the same model (i.e. BMI × waist circumference, BMI × WHR, waist circumference × hip circumference) showed that associations with AUCOGTT were slightly weaker at higher levels of the second measure (p < 0.03) (Fig. 1). The associations between BMI and type 2 diabetes, and waist circumference and type 2 diabetes were somewhat reduced when at the higher level of the other measure (p < 0.01).
https://static-content.springer.com/image/art%3A10.1007%2Fs00125-010-1710-3/MediaObjects/125_2010_1710_Fig2_HTML.gif
Fig. 2

Odds ratios for type 2 diabetes and mean AUCOGTT levels by tertiles of anthropometric measures, stratified (a, c) by BMI and WHR, and (b, d) by waist circumference (WC) and hip circumference (HC). Type 2 diabetes (T2D) cases with non-cases are shown for tertiles as indicated (c, d). Mean AUCOGTT levels are for 52-year-old, non-smoking European men. Data are presented as n/n (mean AUCOGTT in mmol/l × h). All associations adjusted for age, sex, smoking and ethnicity

Findings were similar for fasting glucose and 2 h glucose (ESM Table 2) among men, women and ethnic groups (data not shown).

Interactions with sex

Significant interactions with sex were observed for waist circumference and hip circumference. For an equivalent increase in waist circumference, women had a significantly greater increase in the odds of type 2 diabetes than men (Table 2). In models containing waist circumference, an increase in hip circumference was associated with a greater decrease in the odds of type 2 diabetes among men. Results for AUCOGTT were qualitatively similar (ESM Table 3).
Table 2

Interactions of anthropometric measures and sex with type 2 diabetes

Measure

Men

Women

p value for interaction

OR

95% CI

p value

OR

95% CI

p value

BMI

1.33

1.25, 1.42

<0.01

1.34

1.28, 1.40

<0.01

0.97

WC

1.28

1.21, 1.37

<0.01

1.48

1.41, 1.54

<0.01

<0.01

HC

1.14

1.07, 1.22

<0.01

1.22

1.17, 1.28

<0.01

0.11

WHR

1.39

1.29, 1.50

<0.01

1.40

1.33, 1.48

<0.01

0.83

BMI + WCa

1.11

0.92, 1.35

0.27

1.23

1.06, 1.42

<0.01

0.26

WC + BMIa

1.22

1.13, 1.32

<0.01

1.39

1.28, 1.50

<0.01

<0.01

BMI + WHRa

1.21

1.13, 1.30

<0.01

1.29

1.23, 1.36

<0.01

0.12

WHR + BMIa

1.27

1.18, 1.37

<0.01

1.35

1.28, 1.43

<0.01

0.20

WC + HCa

1.58

1.46, 1.71

<0.01

1.87

1.73, 2.03

<0.01

<0.01

HC + WCa

0.67

0.61, 0.73

<0.01

0.80

0.74, 0.86

<0.01

<0.01

Odds ratios represent sex-specific changes in the odds of type 2 diabetes for one SD increase in anthropometric measures independently of age, smoking, ethnicity and other anthropometric measures

Standard deviations: BMI 6.2 kg/m2, waist circumference (WC) 14.4 cm, hip circumference (HC) 13.3 cm, WHR 0.09

aAdditionally adjusted for second term after +

Results for AUCOGTT were qualitatively similar and are shown in ESM Table 2

Interactions with ethnicity

Ethnic heterogeneity was identified among several BMI and hip circumference associations with type 2 diabetes (Fig. 3) and AUCOGTT (data not shown). However there were few significant ethnic comparisons. In general, people of Aboriginal origin had the strongest associations of both BMI and hip circumference with type 2 diabetes and/or AUCOGTT, whereas people of African origin had the weakest associations. Among Aboriginals, BMI, hip circumference and BMI adjusted for WHR were significantly more strongly associated with type 2 diabetes and AUCOGTT than among Africans and South Asians. Associations between BMI and type 2 diabetes among Europeans were significantly stronger than among Africans.
https://static-content.springer.com/image/art%3A10.1007%2Fs00125-010-1710-3/MediaObjects/125_2010_1710_Fig3_HTML.gif
Fig. 3

Ethnic-specific standardised associations exhibiting significant overall ethnic heterogeneity by Cochran’s Q test. Odds ratios represent changes in the odds of type 2 diabetes (T2D) for one SD increase in anthropometric measures independently of age, sex, smoking and other anthropometric measures. Standard deviations: BMI 6.2 kg/m2, waist circumference (WC) 14.4 cm, hip circumference (HC) 13.3 cm, WHR 0.09. Results for AUCOGTT were qualitatively similar. The + symbol indicates: additionally adjusted for anthropometric measure after symbol. Black squares, Aboriginal; white diamonds, African; triangles, South Asian; circles, European; black diamonds, Latin American

Discussion

This investigation demonstrates that waist circumference and hip circumference are required to optimally identify people with elevated glucose levels in an ethnically diverse, high-risk population. The combination of BMI and WHR, or of waist circumference and hip circumference has the best explanatory power for type 2 diabetes and glucose status compared with a single anthropometric measure. These findings were generally consistent among men, women and different ethnic groups, although some sex- and ethnicity-related variations were observed.

Several studies have compared the associations of anthropometric measures with type 2 diabetes. In a meta-analysis of 32 prospective studies, BMI, waist circumference and WHR had associations with incident type 2 diabetes that were not significantly different [17]. In a review examining variation in the c-statistic (17 prospective studies, 35 cross-sectional studies), Qiao and Nyamdorj found that no measure had consistently higher explanatory power for type 2 diabetes risk [18]. In our study, BMI, waist circumference and WHR had nearly identical associations with type 2 diabetes and AUCOGTT, and similar explanatory power. However, associations and explanatory power changed when an extra measure was included in the models. For example, when BMI and WHR were included in the same model, associations with type 2 diabetes became weaker, whereas explanatory power rose significantly. This probably occurred because BMI and WHR both provide information for predicting type 2 diabetes and glucose status (general obesity, abdominal shape), and are weakly correlated with each other (correlation of BMI with WHR r = 0.14) [19].

When waist circumference and hip circumference were included in the same model, associations for waist circumference became significantly stronger and hip circumference was inversely associated with type 2 diabetes and glucose status. This was accompanied by a significant increase in explanatory power. Waist and hip circumference have opposite relationships with cardiometabolic risk. However, due to their high correlation (r = 0.76 in EpiDREAM), these only become apparent after mutual adjustment [2024]. Increased waist circumference is likely to be associated with cardiometabolic risk by the presence of visceral fat and is balanced by hip circumference, which is associated with leg muscularity, hip subcutaneous fat and oestrogen [23]. Use of WHR to control for this confounding may not be appropriate because absolute levels of waist and hip circumference can differ between individuals with the same WHR. This can result in misclassification if both waist and hip circumference are determinants of risk [25]. WHR also has lower correlations with visceral fat [26]. In our study, the model containing both waist and hip circumference had stronger associations as well greater explanatory power than the model containing WHR. However, waist and hip circumference together had significantly lower explanatory power than the model containing BMI and WHR, which may be due to the high correlation between waist circumference and hip circumference. Nevertheless, the simplest strategy to achieve a significant improvement in explanatory power is to use waist circumference and hip circumference rather than BMI and WHR. Using both waist and hip circumference does not require the collection of weight and height, or any calculation. Interestingly, the choice of two measures may be less critical among large individuals, as we observed a slight weakening of associations at higher levels of other anthropometric measures.

Our results indicate that an increase in waist circumference is associated with greater risk among women and that an increase in hip circumference is associated with a lower risk among men. While the mechanism behind these findings is unknown, women have lower muscle mass than men, so an increase in waist circumference could be associated with a proportionally greater increase in visceral fat compared with men. Also an equivalent amount of visceral fat in women is more strongly correlated with cardiometabolic risk than in men [27]. Another possibility is that a shift towards android obesity in women indicates a change in hormonal state as is observed in post-menopausal women and in women with polycystic ovarian syndrome or Turner’s syndrome, which can elevate type 2 diabetes risk. Oestrogen replacement among post-menopausal women is associated with a lower incidence of type 2 diabetes [28, 29]. The reason for the interaction with sex with regard to hip circumference is unclear. Hip circumference captures a measure of muscularity in the gluteal region, so an increase in hip circumference could be associated with a greater increase in protective muscle mass in men than in women, as men are naturally more muscular than women. Interestingly, controlling for BMI did not change these results (data not shown).

We found some ethnic variation in the associations. In particular, BMI and hip circumference had the strongest associations among Aboriginal people and the weakest among Africans. However, differences among BMI associations disappeared after adjusting for waist circumference. Previous studies indicate that Africans possess greater lean muscle mass than other groups for the same BMI and that Aboriginal people are susceptible to visceral fat accumulation [3033]. Therefore, an increase in BMI could reflect ethnic-specific changes in body composition. Several studies report ethnic differences in fat distribution and visceral fat content [12, 34, 35], which may explain ethnic differences in cardiometabolic risk [11, 12]. However we were not able to directly assess visceral fat in our study. Moreover, because the CIs overlapped for most associations, which were derived from the analysis of multiple subgroups, our results should be interpreted with caution.

Our study has several strengths. First, participants were sampled from 21 countries, with non-whites making up more than 50% of participants. Second, the sample size was large, allowing us to detect small differences in risk, which could be very important given the widespread use of anthropometric measures in type 2 diabetes risk assessment. Third, type 2 diabetes was newly detected and therefore participants could not have initiated behavioural (e.g. dietary) changes in response to a diagnosis of type 2 diabetes that might have altered anthropometric measurements and type 2 diabetes risk. Fourth, unlike meta-analytic comparisons of anthropometric measures, which make assumptions about the variation in measures across studies, we made comparisons in the same population.

Our study also has some limitations. The most important is that the EpiDREAM population is not a random sample, such that participant characteristics and associations reported by us may not be present in the general population. However, this analysis is internally valid and its results are clearly relevant to the substantial numbers of people at high risk of developing type 2 diabetes. Second, our study’s cross-sectional design means that causality cannot be inferred from this study alone. However, prospective investigations show that anthropometric measures are powerful predictors of type 2 diabetes [19]. Third, we did not adjust our associations for some health behaviours (e.g. diet), which may result in some confounding. Fourth, we used self-report to assign ethnicity, which may result in some misclassification. However, in a separate analysis where we genotyped EpiDREAM participants using a custom-made 50K single nucleotide polymorphism chip for variation in genes related to cardiometabolic risk [36], we found that self-reported ethnicity was strongly correlated with genotype cluster and hence ancestral origin determined by multidimensional scaling [37]. Fifth, we were not able to explore the biological basis of associations and interactions observed in this study.

Conclusions

The combinations of BMI and WHR, or of waist circumference and hip circumference have the best explanatory power for type 2 diabetes and glucose status compared with a single anthropometric measure. Measurement of both waist circumference and hip circumference is required to optimally identify people with elevated glucose levels.

Acknowledgements

This analysis was supported by a Frederick Banting and Charles Best Canada Graduate Scholarship Doctoral Research Award from the Canadian Institutes of Health Research to L. de Koning. EpiDREAM was supported by grants from the Heart and Stroke Foundation of Canada, the Canadian Institutes of Health Research, Sanofi-Aventis, GlaxoSmithKline and King Pharmaceuticals. We thank L. Dyal (Population Health Research Institute, Hamilton, ON, Canada) for statistical advice and database support. H. C. Gerstein holds the Population Health Research Institute Chair in Diabetes Research. S. S. Anand holds the Michael G. De Groote Heart and Stroke Foundation of Ontario Chair in Population Health Research and the Eli Lilly May Cohen Chair in Women’s Health. S. Yusuf holds a Heart and Stroke Foundation of Ontario Chair in Cardiovascular Research.

Duality of interest

The authors declare that there is no duality of interest associated with this manuscript.

Supplementary material

125_2010_1710_MOESM1_ESM.pdf (11 kb)
ESM Table 1Countries, centres and participants in EpiDREAM (PDF 11.4 kb)
125_2010_1710_MOESM2_ESM.pdf (32 kb)
ESM Table 2Standardised associations of anthropometric measures with fasting and 2 h post-load (OGTT) glucose (PDF 31.6 kb)
125_2010_1710_MOESM3_ESM.pdf (30 kb)
ESM Table 3Interactions between anthropometric measures and sex with AUCOGTT (PDF 29.5 kb)

Copyright information

© Springer-Verlag 2010

Authors and Affiliations

  • L. de Koning
    • 1
    • 3
  • H. C. Gerstein
    • 1
    • 2
    • 3
    • 7
  • J. Bosch
    • 1
    • 3
    • 7
  • R. Diaz
    • 4
  • V. Mohan
    • 5
  • G. Dagenais
    • 6
  • S. Yusuf
    • 1
    • 2
    • 3
    • 7
  • S. S. Anand
    • 1
    • 2
    • 3
    • 7
  • for the EpiDREAM Investigators
  1. 1.The Population Health Research InstituteMcMaster UniversityHamiltonCanada
  2. 2.The Department of MedicineMcMaster UniversityHamiltonCanada
  3. 3.The Department of Clinical Epidemiology and BiostatisticsMcMaster UniversityHamiltonCanada
  4. 4.Instituto Cardiovascular de RosarioRosarioArgentina
  5. 5.Dr Mohan’s Diabetes Specialties Center and Madras Diabetes Research FoundationChennaiIndia
  6. 6.Québec Heart Institute, Institut Universitaire de Cardiologie et de PneumologieUniversité LavalQuébecCanada
  7. 7.David Braley Cardiac, Vascular and Stroke Research InstituteHamiltonCanada