In total, 17 prospective and 35 cross-sectional studies in adults aged 18–74 years, with the aim of comparing betweenbody mass index (BMI), waist circumference (WC) and waist-to-hip ratio (WHR) in their relation to the incidence and prevalence of type II diabetes, were reviewed. Among these studies, only a few have used C-statistic, paired homogeneity test or log-likelihood ratio test for formally comparing the differences. Five prospective studies, in which formal statistic tests have been made, came out with inconsistent findings: two results were in favour of WC in Mexicans African Americanss, respectively, one result was in favour of BMI in Pima Indians, and no difference was found in the other 2 studies. Among the 11 cross-sectional studies that have formally tested the differences, most found a higher odds ratio or slightly larger area under the ROC curve (AUC) for WC than for BMI. A meta-analysis based on the individual data of the Asian cohorts using a paired homogeneity test showed, however, that there was no difference in odds ratio between BMI and WC in Chinese, Japanese, Indian, Mongolian and Filipino men. In conclusion, all studies included in this review showed that either BMI or WC (WHR) predicted or was associated with type II diabetes independently, regardless of the controversial findings on which of these obesity indicators is better.
Diabetes increases dramatically worldwide (Zimmet et al., 2001) as a consequence of changes in lifestyle, including physical inactivity and unhealthy diet. Physical inactivity and obesity have been well recognized as major lifestyle-related risk factors for diabetes. In the light of evidence that the onset of diabetes can be prevented or delayed through lifestyle intervention, including weight reduction and increasing physical activity (Pan et al., 1997; Tuomilehto et al., 2001; Diabetes Prevention Program Research Group, 2002; Kosaka et al., 2005; Ramachandran et al., 2006), lifestyle intervention to prevent non-communicable diseases including diabetes has been included in the 2008–2013 Action Plan (World Health Organization, 2008). One of the objectives of this plan is to develop simple strategies to identify those at risk and provide them with early lifestyle interventions. As the glucose test is invasive, relatively expensive, time consuming and not easy to apply to mass-screening programmes, several other diagnostic tools, including obesity indicators such as waist circumference (WC) and body mass index (BMI), have been proposed and applied in diabetes prevention programmes in recent years (Rolka et al., 2001; Lindstrom and Tuomilehto, 2003; Schulze et al., 2007). However, controversial opinions exist on which of the obesity measures, WC (waist-to-hip ratio (WHR)) or BMI, is more strongly associated with the increased risk of type II diabetes. In this article, studies for comparing between BMI, WC and WHR in their relation to the incidence and prevalence of type II diabetes in adults were reviewed.
Materials and methods
Publications of studies with the aim of comparing between BMI, WC and WHR in their relation to the incidence and prevalence of type II diabetes in adults were eligible for inclusion. A few studies that used unstandardized methods were excluded from this review because the results obtained could not be compared directly. For example, we have excluded studies that showed that an odds ratio for a BMI 28 kg/m2 was higher (or lower) than the odds ratio for a WC 94 cm, but did not mention why the cutoff values for BMI and WC were chosen and whether they are comparable, and there was no formal statistical test to support their conclusions also. Studies that have used quintiles or quartiles or s.d. changes in BMI and WC were included.
Data sources and limitations
The published articles related to these topics were searched from the PubMed from 1975 onwards or obtained through conferences and colleagues, and were reviewed by two independent researchers (RN and QQ). Most of the studies included individuals aged 18–74 years, except for one Chinese study (Woo et al., 2002). Participants in this Chinese study were older than 70 years and were selected from a list of recipients of Old Age and Disability Allowance. In different studies, waist has been measured in different anatomic locations, and diabetes was defined on the basis of either a previous history of diabetes or fasting plasma glucose or fasting plasma glucose plus 2-h post-challenge glucose test levels. Most of the studies are population based with random sampling approaches and a few are hospital based with participants coming for the health check-up as indicated in Supplementary Table 1.
The results of 17 prospective studies and 35 cross-sectional studies, including large international collaborative studies (counted as one study), comparing between BMI and WC, are summarized and presented in Supplementary Tables 2 and 3. The relative risk (RR) or odds ratio for diabetes was estimated using either Cox regression analysis or logistic regression analysis corresponding to either a 1 s.d. increase (in BMI, WC, WHR, and so on) or dichotomous variables (top quintile or quartile vs the lowest quintile or quartile). Areas under the ROC curves (AUCs) with their 95% confidence intervals were reported in some of the studies. The differences between AUCs or between relative risks (or odds ratios) were formally tested in only a few studies (Supplementary Table 3).
Main findings from the prospective study
The follow-up length among the 17 prospective studies ranged from 3 to 15 years. One of these studies applied the paired homogeneity test, 3 compared the AUCs by applying the DeLong method for correlated data (C-statistic), 1 study reported results of both the log-likelihood ratio test and C-statistic, 12 studies did not perform any formal statistical test to compare between BMI and WC (or WHR). Therefore, the results from the 12 studies provide less convincing information to this review. Incidence of diabetes was defined on the basis of elevated fasting (7.8 or 7.0 mmol/l) and/or elevated post-load 2-h glucose level (11.1 mmol/l) in 11 studies, and on the basis of physician's judgement in 6 studies.
As shown in Supplementary Table 1, in some studies the relative risks or the AUCs for predicting development of diabetes were higher for WC than for BMI, but in other studies BMI was higher than WC. The variations in observations were independent of age, gender, ethnicity, diagnostic criteria for diabetes and methods in anthropometric measures. Two facts, however, need to be addressed when interpreting the findings: (1) the 95% confidence intervals for BMI and WC overlapped in all these studies and (2) no formal statistical tests were carried out in most of these studies. The five studies in which formal statistic tests have been applied produced different findings (Supplementary Table 3). The result was in favour of WC in Mexican Americans (Wei et al., 1997) and African Americans (Stevens et al., 2001), but in favour of BMI in Pima Indians (Tulloch-Reid et al., 2003) and the White American men (Stevens et al., 2001), and no difference was found in either the study (Diabetes Prevention Program Research Group, 2006) or the Mauritius Indian studies (Nyamdorj et al., 2009).
The San Antonio Heart Study among Mexican Americans in the United States (Wei et al., 1997), aged 25–64 years and followed up for 7 years, showed that WC was a better risk predictor for type II diabetes than BMI, but there was no difference between BMI and WHR. The AUC was statistically significantly larger for WC than for BMI in both women (P<0.001) and men (P=0.012) (Supplementary Table 1), but there was no difference between the WC and the WHR (P=0.07 in women and P=0.13 in men). The log-likelihood ratio test showed that the addition of BMI to the model with WC did not improve the model prediction based on the WC alone (P=0.53), but it improved prediction based on the WHR alone (P=0.0021); WC improved the model prediction based on either BMI alone (P=0.0006) or WHR alone (P=0.0004), and WHR improved the model prediction beyond that based on BMI alone (P=0.0017), but not based on WC alone (P=0.21) (Supplementary Table 1).
The Pima Indian study (Tulloch-Reid et al., 2003) is a population-based 5-year follow-up study among individuals >18 years. Applying the DeLong method for correlated data, the study showed that the AUC was slightly but statistically significantly larger for BMI than for either WC or WHR.
Again applying the DeLong method, data from the Diabetes Prevention Program (Diabetes Prevention Program Research Group, 2006) of the United States were analysed to compare the risk of each of the obesity indices. The mean age of individuals included in this intervention trial was 54 years, with a follow-up length of 3 years. Compared with BMI, there was no difference in the AUCs between WC and BMI and between WHR and BMI in the groups with placebo, with Metformin and with lifestyle interventions.
In the Atherosclerosis Risk in Community study among 12 814 Americans aged 45–64 years with a follow-up of 9 years, the AUC for prediction of type II diabetes for WC was significantly higher than for BMI in African men and women, but the AUC for WHR was lower than that for BMI in White men (Stevens et al., 2001).
In a recent study based on the data of the Mauritius Non-communicable Disease Survey of 3945 Indians and Creoles, aged 25–74 years with a maximum follow-up length of 11 years, Nyamdorj et al. showed that the risk size (relative risk) of BMI did not differ from that of WC or WHR when the paired homogeneity test was performed (Supplementary Table 3).
Main findings from the cross-sectional study
In total, 35 cross-sectional studies on the topic of interest were reviewed and the results are summarized in Supplementary Table 2. Among these studies, 2 performed the paired homogeneity test, 8 the C-statistic test 11 fitted all variables of interest in the same model but did not show the changes in the model prediction as compared with the nested model and 14 did not run any test.
The point estimate of the odds ratio or the AUC for prevalent diabetes was higher for WC or WHR than for BMI in most of the comparisons, but the confidence intervals were overlapped for most of the reports (Supplementary Table 2). The variations were not explained by differences in ethnicity, age and gender. Studies that tested the differences in the AUCs and the strength of the odds ratio of the risk factors are summarized and presented in Supplementary Table 3. One study ON Australian aboriginal people and Torres Strait islanders (Wang et al., 2007), applying the DeLong method, revealed that the AUC was larger for WHR than for others (BMI, WC and waist-to-stature ratio), but the data were not shown, and hence the study is not included in Supplementary Table 3. As shown in Supplementary Table 3, most of these cross-sectional studies revealed that the AUC was slightly larger for WC or for WHR than for BMI. The odds ratio was also stronger for WC or WHR than for BMI in a study made by Huxley et al. (2008) using the paired homogeneity test. However, a meta-analysis based on the individual data of the Diabetes Epidemiology: Collaborative Analysis of Diagnostic Criteria for Diabetes in Asia study applying the paired homogeneity test showed that there was no difference in odds ratio in Chinese, Japanese, Indian, Mongolian and Filipino men (Nyamdorj et al., 2008, 2009). The BMI was found to be inferior to WC, WHR or the waist-to-stature ratio in only the Filipino women.
Meta-analysis based on AUC data
Recently, a meta-analytic study by Lee et al. (2008) has analysed the data of AUCs on the basis of one prospective and eight cross-sectional studies. The combined AUC for predicting type II diabetes in men was 0.672 (95% confidence interval 0.646, 0.697) for BMI, 0.701 (0.670, 0.732) for WC, 0.721 (0.664, 0.778) for WHR and 0.726 (0.698, 0.754) for waist-to-height ratio. The corresponding figures for women were 0.693 (0.629, 0.757), 0.744 (0.695, 0.794), 0.748 (0.687, 0.810) and 0.756 (0.700, 0.811), respectively. The test for heterogeneity between each of the abdominal obesity measures with BMI showed significant differences between BMI and waist-to-height ratio in only men (P<0.01). The criteria for diagnosing diabetes and the method of measuring WC were not described.
The evidence based on the prospective studies equally favoured all anthropometric measures of BMI, WC, WHR and the waist-to-stature ratio. But most of the cross-sectional studies showed that WC or WHR discriminate better the cases with diabetes from those without, as compared with BMI. As the number of the prospective studies is limited and covered only limited ethnic groups, the evidence obtained is less convincing and difficult to generalize. The cross-sectional study itself provides only possible association or evidence. Nevertheless, all these studies have shown that either BMI or WC predicted or was associated with increased diabetes risk, independent of other factors.
Conflict of interest
The authors declare no conflict of interest.
The earlier version of this paper was prepared as a background paper for the WHO Expert Consultation on waist circumference and waist–hip ratio (Geneva, 8–11 December 2008). We owe our sincere thanks to all experts who gave comments to improve the paper. This work has been financially supported by the Academy Finland (118492).
About this article
Supplementary Information accompanies the paper on European Journal of Clinical Nutrition website (http://www.nature.com/ejcn)