Skip Navigation



Epidemiologic Reviews Advance Access published online on May 10, 2007
Epidemiologic Reviews, doi:10.1093/epirev/mxm008
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
29/1/115    most recent
mxm008v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrowRequest Permissions
Google Scholar
Right arrow Articles by Vazquez, G.
Right arrow Articles by Silventoinen, K.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Vazquez, G.
Right arrow Articles by Silventoinen, K.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

Epidemiologic Reviews Copyright © 2007 by the Johns Hopkins Bloomberg School of Public Health All rights reserved; printed in U.S.A.

Article

Comparison of Body Mass Index, Waist Circumference, and Waist/Hip Ratio in Predicting Incident Diabetes: A Meta-Analysis

Gabriela Vazquez1, Sue Duval1, David R. Jacobs, Jr.1 and Karri Silventoinen2

1 Division of Epidemiology and Community Health, School of Public Health, University of Minnesota, Minneapolis, MN
2 Department of Public Health, University of Helsinki, Helsinki, Finland

Correspondence to Gabriela Vazquez, Division of Epidemiology and Community Health, School of Public Health, University of Minnesota, 1300 South Second Street, Suite 300, Minneapolis, MN 55454 (e-mail: vazquez{at}epi.umn.edu).

accepted for publication February 19, 2007.


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 RESEARCH DESIGN AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Body mass index, waist circumference, and waist/hip ratio have been shown to be associated with type 2 diabetes. From the clinical perspective, central obesity (approximated by waist circumference or waist/hip ratio) is known to generate diabetogenic substances and should therefore be more informative than general obesity (body mass index). Because of their high correlation, from the statistical perspective, body mass index and waist circumference are unlikely to yield different answers. To compare associations of diabetes incidence with general and central obesity indicators, the authors conducted a meta-analysis based on published studies from 1966 to 2004 retrieved from a PubMed search. The analysis was performed with 32 studies out of 432 publications initially identified. Measures of association were transformed to log relative risks per standard deviation (pooled across all studies) increase in the obesity indicator and pooled using random effects models. The pooled relative risks for incident diabetes were 1.87 (95% confidence interval (CI): 1.67, 2.10), 1.87 (95% CI: 1.58, 2.20), and 1.88 (95% CI: 1.61, 2.19) per standard deviation of body mass index, waist circumference, and waist/hip ratio, respectively, demonstrating that these three obesity indicators have similar associations with incident diabetes. Although the clinical perspective focusing on central obesity is appealing, further research is needed to determine the usefulness of waist circumference or waist/hip ratio over body mass index.

body fat distribution • body mass index • diabetes mellitus, type 2 • meta-analysis • obesity • waist-hip ratio

BMI, body mass indexCI, confidence intervalRR, relative riskSD, standard deviationWC, waist circumferenceWHR, waist/hip ratio


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
RESEARCH DESIGN AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Obesity has become a major worldwide epidemic affecting more than 300 million people. It is an important risk factor for diabetes mellitus, type 2, a chronic disorder of carbohydrate, fat, and protein metabolism. From the clinical perspective, visceral adipose tissue is known to generate diabetogenic substances (1) and, as such, may be more informative than total fat for diagnostic evaluation. The standard epidemiologic translation of these important clinical facts uses anthropometric measures. Waist circumference and waist/hip ratio have been used as measures of central obesity (where visceral adipose tissue is stored), and body mass index (kg/m2) has been used as a measure of general obesity (2).

Clinical evidence suggests that the association of diabetes with central obesity is stronger than the association with general fat. Studies using computed tomography and magnetic resonance imaging have provided further evidence to support that central obesity, visceral adipose tissue, and upper-body nonvisceral fat are the major contributors to the metabolic complications (3–6). Central obesity has been associated with decreased glucose tolerance, alterations in glucose insulin homeostasis, reduced metabolic clearance of insulin, and decreased insulin-stimulated glucose disposal.

In addition, studies that have analyzed the association of anthropometric measures and abdominal visceral fat have found waist circumference to be a better measure of central obesity because it is a better predictor of abdominal visceral fat obtained with computed tomography than is waist/hip ratio, and it can be easily measured and interpreted (2, 7–10). However, waist circumference cannot distinguish abdominal subcutaneous fat, total abdominal fat, and total body fat, and it is strongly correlated with body mass index. Body mass index has been shown to be a good indicator of general fatness (fat areas in the arm, thigh, and waist using computed tomography scans), muscularity (muscle area in the thigh), and frame size (bone area in thighs) (11).

As expected, epidemiologic studies have demonstrated that these three obesity indicators are strong and consistent predictors of diabetes mellitus, type 2. However, despite the clear, clinical difference between visceral and other forms of fat, little epidemiologic difference would be expected in the relations of diabetes with body mass index versus waist circumference. From a statistical perspective, the two measures yield similar information, with the correlation coefficient typically about 0.8 (12). Several studies have shown that waist circumference is a better predictor of diabetes mellitus, type 2, than is body mass index, but these findings are inconclusive (13–15), while other studies provide evidence that waist/hip ratio has a positive effect independent of body mass index (16–18). In addition, the ability of these obesity indicators to predict diabetes may differ by ethnicity, age, and sex (19–22). For example, among Asian populations, central obesity has been shown to be a more consistent predictor of diabetes than is total obesity (18, 23), while general obesity has been shown to be a better predictor among White US populations and Europeans (24, 25).

To study the magnitude of the association among different obesity indicators in multiethnic populations comprising studies worldwide, we performed a meta-analysis of published studies that reported the association between obesity and incident diabetes. Additionally, we explored if the associations differed by region and other population characteristics. Finally, we investigated if the study designs and model assumptions contributed to the heterogeneity of the reported results.


   RESEARCH DESIGN AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 RESEARCH DESIGN AND METHODS
RESULTS
 DISCUSSION
 REFERENCES
 
Data sources

A PubMed (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi) search of published articles from 1966 to 2004 was performed during April 2002 (search 1) and updated in December 2004 (search 2) to select relevant publications. Refer to figure 1 for selection criteria.


Figure 1
View larger version (18K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIGURE 1. Study selection diagram for the meta-analysis of studies published from 1966 to 2004. NIDDM, non-insulin-dependent diabetes mellitus; WC, waist circumference; WHR, waist/hip ratio.

 
Study selection

The search was limited to articles published in English. Study selection was performed with two levels of study screening. At the first level, abstracts were examined by two independent reviewers, using the following criteria: diabetes as the outcome, at least one indicator of abdominal obesity as the exposure or as a confounding factor, and follow-up study. Full manuscripts were then obtained for all publications accepted at level 1 screening. For level 2 screening, we verified that relevant data were available and that multiple publications describing the same study population were entered only once in the meta-analysis. When a study had multiple publications, the latest reference in which relevant data were available was used.

Data extraction

From each study, we retrieved study population characteristics (age range, gender, geographic area, ethnicity, inclusion criteria, incident diabetes rate, and mean and variability measure for each obesity indicator reported); study design characteristics (sampling design, follow-up time, number of visits, and sample size); diabetes assessment (epidemiologic criteria and collection instrument); and model assumptions (obesity indicator representation, parameter of association, level of covariate adjustment, and subgroup analysis). Estimates of magnitude of association and variability, that is, standard errors or 95 percent confidence intervals, were extracted over the entire study samples or by subgroups. Several publications reported associations with different levels of covariate adjustment, all of which were extracted.

Statistical analysis

The publications retrieved used different representations of the association of diabetes with the obesity indicator (continuous, categorical, and baseline means for diabetes cases and noncases). The measures were transformed to calculate a log-linear slope with diabetes risk per 1-standard deviation (pooled across all studies) increase of the obesity indicator. To transform measures based on categorical representations, the obesity indicators were assumed to be normally distributed, and all values in a category were assigned to the median value. This assumption was checked by repeating the analysis using a gamma distribution, revealing no differences in the individual study associations to the second decimal place. For studies with three or more categories of the obesity indicator, the slope was estimated with the method of Greenland and Longnecker (26) (using STATA's glst function; StataCorp LP, College Station, Texas), which calculates a weighted linear regression of the natural logarithm (log) of the relative risks across anthropometric categories, taking into account the correlation between estimates. For binary representation of the obesity indicator, classified according to the median value, the log relative risks were divided by the interquartile range. For studies with three or more categories reporting the association of only two categories, the log relative risks were divided by the distance between the median values of the reported categories. For studies with only baseline means and standard errors, the log relative risks were calculated from a simulation of cases and noncases assuming a normal distribution. Where the study measures of association were stratified by subgroups, stratum-specific estimates were pooled using a fixed-effects model, weighted by the inverse subgroup variance.

Studies were classified by region for comparative analysis. Degree of obesity and diabetes incidence were compared across populations. Region-specific means and standard deviations were pooled from individual studies to describe the degree of obesity measured by body mass index, waist circumference, and waist/hip ratio. Region-specific incident diabetes rates were calculated as geometric means.

To assess the association between the three measures of obesity and incident diabetes, we calculated pooled estimates across all studies for the relative risk using a random-effects model. Relative risks (RRs) were expressed per standard deviation (computed across all studies) of each indicator (body mass index (BMI), waist circumference (WC), waist/hip ratio (WHR)). The difference between RRBMI and RRWC (or RRBMI and RRWHR) was evaluated only in studies that included both of each pair of indicators. The analysis was performed using RRBMI and RRWC as repeated dependent variables within each study in SAS, version 9.1.3, PROC MIXED (27), following the method of van Houwelingen et al. (28). Within-study variances of the relative risks were fixed for each study using the PARMS (or parameter) statement, which declares the parameters and specifies their initial values. A similar analysis was done for RRBMI and RRWHR.

I2 was used to describe the proportion of total variation among study-specific estimates that is due to heterogeneity (29). Heterogeneity was investigated ecologically by comparing pooled relative risks of subgroups defined by region, gender, mean body mass index, mean age, incident diabetes rate, and criteria used to define the target sample. Tests for differences between subgroups were performed unadjusted for other study level characteristics using meta-regression. Further exploration of the heterogeneity of the relative risk due to design characteristics and model assumptions was performed only for body mass index, including body mass index representation, reported parameter of association, outcome assessment, level of covariate adjustment, and follow-up time.

Additionally, funnel plots with pseudo 95 percent confidence intervals were used for visual assessment of publication bias, and the trim and fill method (30) was used to estimate any possible publication bias. All meta-analyses except the bivariate analyses were performed using the STATA, version 8.2, statistical package (31).


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 RESEARCH DESIGN AND METHODS
 RESULTS
DISCUSSION
 REFERENCES
 
Figure 1 presents a flow diagram outlining the systematic review process. An initial search generated a list of 432 publications (290 in April 2002 and 142 in December 2004). After review of the abstracts, 47 publications (36 from search 1 and 11 from search 2) met inclusion criteria. From those 47 publications, we included data from 29 publications, reporting on 32 distinct study populations (13, 16–18, 24, 32–56). Burke et al. (47) included a subset of Mexican origin subjects from the San Antonio Heart Study (13) to compare them with those from the Mexico City Diabetes Study. Although this study contained a subset of subjects already in our study sample, the study was retained because the majority of the sample targeted a different population. In addition, Edelstein et al. (56) reported the result of six studies, four not previously included. Only studies concerned with body mass index, waist circumference, and waist/hip ratio were included in the present analysis. Other anthropometric indicators reported in the publications were the subscapular/triceps skinfold thickness ratio, waist/thigh ratio, subscapular skinfold thickness, triceps skinfold thickness, skinfold thickness sum, hip circumference, and waist/height ratio.

Table 1 provides a summary of the characteristics of the included studies, and table 2 provides a detailed description of each study in the meta-analysis. Among the studies included, four targeted men, three targeted women, and 25 targeted both genders; the age range was 20–80 years. All studies analyzed the progression from nondiabetes to diabetes. However, several targeted more restricted populations; for example, the Mauritius study (40) targeted subjects with normal glucose tolerance, five studies targeted subjects with impaired glucose tolerance, and four studies included only participants free of chronic conditions. The studies were classified in three regions based on geographic proximity: Europe (nine studies), United States (12 studies), and Asia (four studies). Seven studies (36, 40, 44, 45, 47, 48, 55) were excluded from regional comparisons because they differed from the core population of each region or were located in other geographic areas. The study designs differed in sampling design, follow-up time, number of follow-up visits, and sample size. Diabetes assessment relied on different instruments and diagnostic criteria (57–60). In terms of the analysis, differences relied on the representation of the obesity–diabetes association, statistical approach, level of covariate adjustment, and level of stratification.


View this table:
[in this window]
[in a new window]

  		TABLE 1. Summary of characteristics of studies between 1985 and 2004 included in the meta-analysis that reported the association of body mass index, waist circumference, or waist/hip ratio and incident diabetes

 

View this table:
[in this window]
[in a new window]

  		TABLE 2. Description of characteristics of studies between 1985 and 2004 included in the meta-analysis that reported the association of body mass index, waist circumference, or waist/hip ratio and incident diabetes

 
Table 3 provides descriptive information for body mass index (kg/m2), waist circumference (cm), and waist/hip ratio by region. Information reported in publications was sometimes unadjusted and sometimes adjusted for age, sex, or other factors, and in two studies not reported. Pooled means for body mass index, waist circumference, and waist/hip ratio were 25.8 (standard deviation (SD): 4.3), 87.2 (SD: 11.6), and 0.84 (SD: 0.07), respectively. People from Asian studies were identified as being the leanest: 24.2 (SD: 3.1) kg/m2 and 79.7 (SD: 8.6) cm for body mass index and waist circumference, respectively, while studies from Europe and the United States reported significantly higher values: 26.3 (SD: 3.4) kg/m2 and 92.1 (SD: 9.9) cm for European studies and 26.6 (SD: 4.5) kg/m2 and 88.2 (SD: 10.9) cm for studies in the United States. Body mass index and waist circumference, but not waist/hip ratio, reflected regional differences in degree of obesity between individuals from different regions (p values for geographic differences were 0.02 for both body mass index and waist circumference, using an F test for overall difference). Correlations between obesity indicators at the study level were similar for those observed at the individual level (13, 33): 0.88, 0.34, and 0.44 for body mass index–waist circumference, body mass index–waist/hip ratio, and waist circumference–waist/hip ratio, respectively.


View this table:
[in this window]
[in a new window]

  		TABLE 3. Incident diabetes rate, mean and standard deviation of body mass index, waist circumference, and waist/hip ratio by region and overall for studies between 1985 and 2004 included in the meta-analysis

 
Incident diabetes rates were highest for people from US studies (13.5 new cases per 1,000 person-years) and lowest for people from Asian studies (5.2 new cases per 1,000 person-years). They appear to vary according to differences in inclusion criteria and other individual characteristics of the study populations, including the level of obesity. However, diabetes rates should be interpreted with caution because they were reported at various levels of adjustment.

Figures 2, 3, and 4 and table 4 present the pooled and study-level relative risk for each of the obesity indicators. The pooled estimates of RRBMI (n = 32) (SD: 4.3) were 1.92 (95 percent confidence interval (CI): 1.70, 2.17). Among the 18 studies that included both body mass index and waist circumference, RRBMI was 1.72 (95 percent CI: 1.47, 2.02), and RRWC (SD: 11.6) was 1.87 (95 percent CI: 1.62, 2.15). In the 25 studies that included both body mass index and waist/hip ratio, RRBMI was 1.98 (95 percent CI: 1.70, 2.30) and RRWHR (SD: 0.84) was 1.82 (95 percent CI: 1.55, 2.13). Neither comparison, body mass index and waist circumference (p = 0.50) or body mass index and waist/hip ratio (p = 0.73), revealed a significant difference in the magnitude of association. Heterogeneity was present for all obesity indicators: I2 > 0.90.


Figure 2
View larger version (18K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIGURE 2. Forest plot of the association between body mass index (BMI) and incident diabetes (95% confidence interval) for 32 studies published between 1985 and 2004. The pooled relative risk per standard deviation (SD: 4.3) (plotted as diamond) is 1.87 (95% confidence interval: 1.67, 2.10). NAS, Normative Aging Study; NHANES I, First National Health and Nutrition Examination Survey; SAHS, San Antonio Heart Study; IOWA, Iowa Women's Health Study; N-NCCS, Northern Native Canadian Cohort Study; FINRISK, Finland study; MHPS, Male Health Professionals Study; NHS, Nurses' Health Study; ARIC, Atherosclerosis Risk in Communities Study; IRAS, Insulin Resistance Atherosclerosis Study; JACDS, Japanese-American Community Diabetes Study; MCDS, Mexico City Diabetes Study; SA, San Antonio; EPIC, European Prospective Investigation into Cancer and Nutrition; KIHDRF, Kuopio Ischemic Heart Disease Risk Factor Study; BLSA, Baltimore Longitudinal Study of Aging; RBS, Rancho Bernardo Study; SLVDS, San Luis Valley Diabetes Study.

 

Figure 3
View larger version (10K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIGURE 3. Forest plot of the association between waist circumference (WC) and incident diabetes (95% confidence interval) for 18 studies published between 1985 and 2004. The pooled relative risk per standard deviation (SD: 11.6) (plotted as diamond) is 1.87 (95% confidence interval: 1.58, 2.20). NAS, Normative Aging Study; SAHS, San Antonio Heart Study; IOWA, Iowa Women's Health Study; MHPS, Male Health Professionals Study; NHS, Nurses' Health Study; IRAS, Insulin Resistance Atherosclerosis Study; JACDS, Japanese-American Community Diabetes Study; MCDS, Mexico City Diabetes Study; SA, San Antonio; KIHDRF, Kuopio Ischemic Heart Disease Risk Factor Study; BLSA, Baltimore Longitudinal Study of Aging; RBS, Rancho Bernardo Study; SLVDS, San Luis Valley Diabetes Study.

 

Figure 4
View larger version (15K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIGURE 4. Forest plot of the association between waist/hip ratio (WHR) and incident diabetes (95% confidence interval) for 25 studies published between 1985 and 2004. The pooled relative risk per standard deviation (SD: 0.07) (plotted as diamond) is 1.88 (95% confidence interval: 1.61, 2.19). SAHS, San Antonio Heart Study; IOWA, Iowa Women's Health Study; N-NCCS, Northern Native Canadian Cohort Study; FINRISK, Finland study; MHPS, Male Health Professionals Study; NHS, Nurses' Health Study; ARIC, Atherosclerosis Risk in Communities Study; EPIC, European Prospective Investigation into Cancer and Nutrition; KIHDRF, Kuopio Ischemic Heart Disease Risk Factor Study; BLSA, Baltimore Longitudinal Study of Aging; RBS, Rancho Bernardo Study; SLVDS, San Luis Valley Diabetes Study.

 

View this table:
[in this window]
[in a new window]

  		TABLE 4. Pooled incident diabetes relative risk per standard deviation in the obesity indicator and measure of between-study heterogeneity for studies between 1985 and 2004 included in the meta-analysis

 
Figures 5, 6, and 7 present the pooled RRBMI, RRWC, and RRWHR stratified by study-level characteristics. Differences in the obesity–diabetes relative risks were marked between groups defined by study incident diabetes rate and inclusion criteria for all three obesity indicators. Those studies that targeted a sample of subjects with a higher diabetes rate or with impaired glucose tolerance presented a shallower obesity–diabetes relative risk. When pooled relative risks between obesity indicators were compared, some modest differences were found by region and age. RRBMI and RRWC were similar for the three regions but were shallower for RRWHR-Asia (1.4 (95 percent CI: 1.1, 1.7)) compared with RRWHR-Europe (1.9 (95 percent CI: 1.7, 2.2)) and RRWHR-United States (1.7 (95 percent CI: 1.4, 2.2)). In addition, in studies where the mean age was less than 50 years, RRWHR (2.1 (95 percent CI: 1.7, 2. 6)) was higher than RRBMI (1.7 (95 percent CI: 1.4, 2.0)) and RRWC (1.6 (95 percent CI: 1.4, 1.9)). For studies where the mean age was greater than or equal to 50 years, RRWHR (1.7 (95 percent CI: 1.5, 2.0)) was weaker than RRBMI (2.0 (95 percent CI: 1.7, 2.3)) and RRWC (2.0 (95 percent CI: 1.6, 2.7)).


Figure 5
View larger version (11K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIGURE 5. Pooled relative risk (95% confidence interval) for body mass index (BMI) with incident diabetes from the meta-analysis of studies published between 1985 and 2004, stratified by study-level population characteristic: gender, diabetes rate, inclusion criteria, age, general obesity, and region. DM, diabetes mellitus; NDM, non-diabetes mellitus; NCHD, non-coronary heart disease; IGT, impaired glucose tolerance; NGT, normal glucose tolerance; SD, standard deviation.

 

Figure 6
View larger version (10K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIGURE 6. Pooled relative risk (95% confidence interval) for waist circumference (WC) with incident diabetes from the meta-analysis of studies published between 1985 and 2004, stratified by study-level population characteristic: gender, diabetes rate, inclusion criteria, age, general obesity, and region. DM, diabetes mellitus; NDM, non-diabetes mellitus; NCHD, non-coronary heart disease; IGT, impaired glucose tolerance; BMI, body mass index; SD, standard deviation.

 

Figure 7
View larger version (10K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIGURE 7. Pooled relative risk (95% confidence interval) for waist/hip ratio (WHR) with incident diabetes from the meta-analysis of studies published between 1985 and 2004, stratified by study-level population characteristic: gender, diabetes rate, inclusion criteria, age, general obesity, and region. DM, diabetes mellitus; NDM, non-diabetes mellitus; NCHD, non-coronary heart disease; IGT, impaired glucose tolerance; NGT, normal glucose tolerance; BMI, body mass index; SD, standard deviation.

 
Figure 8 presents the pooled RRBMI estimates by study design and model characteristic groups. Differences in study design, outcome assessment, and model assumption may explain some of the heterogeneity of the relative risks across studies. However, none of the characteristics alone seems to play an important role in the level of heterogeneity. In addition, it is not possible to differentiate the impact of each component with a multivariable approach because of the scarcity of studies in each category. Analysis was repeated for RRWC and RRWHR, and results were consistent with those found with RRBMI.


Figure 8
View larger version (10K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIGURE 8. Pooled relative risk (RR) (95% confidence interval) for body mass index (BMI)-incident diabetes from the meta-analysis of studies published between 1985 and 2004, stratified by study design features: level of adjustment, diagnostic criteria, frequency of diagnosis assessment, statistical model, and exposure representation. Full adj, adjustment for additional confounders; Simple adj, age and sex adjusted; SR-reg, self-reported or hospital registry; ADA 97, American Diabetes Association 1997 criteria; WHO 99, World Health Organization 1999 criteria; WHO 85, World Health Organization 1985 criteria; 1–2 v/5y, 1–2 visits within 5 years; ≥3 v/5y, three or more visits within 5 years; OR, odds ratio; Bl-means, baseline means; Multipar, multiparameter model; SD, standard deviation.

 
The funnel plots and trim and fill method showed no publication bias for RRBMI and RRWC (figures not shown). The filled RRWHR was 1.7 (95 percent CI: 1.5, 2.0) after imputing three missing studies according to the trim and fill method compared with the observed RRWHR (1.9 (95 percent CI: 1.6, 2.2)).


   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 RESEARCH DESIGN AND METHODS
 RESULTS
 DISCUSSION
REFERENCES
 
The association of body mass index, waist circumference, and waist/hip ratio with incident diabetes was confirmed in our study by the significant pooled estimates of the relative risk. When comparing the associations in the subset of studies with both body mass index and waist circumference, the pooled RRWC was modestly stronger than the RRBMI. When comparing the RRBMI and RRWHR, RRBMI was modestly stronger. None of these differences was statistically significant.

A recent meta-analysis of the association of body mass index and incident diabetes found similar results (61), although this study differed in some analytical aspects in estimating study-level relative risk, study selection, and target population. Because our meta-analysis focused on studies reporting an additional measure of central obesity, we have fewer studies included in our analyses.

Ford et al. (12) support the use of waist circumference as a measure of obesity to predict health risk. Among their arguments are that waist circumference has been shown to be a good or better predictor than body mass index of the metabolic syndrome, diabetes, cardiovascular disease, and all-cause mortality; it provides information about health risk in addition to body mass index; and it is conceptually easy to measure, although it does require some training and standardization. However, others have noted that substitution of body mass index by waist circumference as an indicator of risk for cardiovascular disease and diabetes may be an oversimplification (2, 11, 62). Some counterarguments are that waist circumference is strongly correlated to body mass index (r ~ 0.8) (12, 13, 33, 63); waist circumference does not differentiate between subcutaneous fat and visceral fat; it has not been shown that a consistent association exists between waist circumference with visceral fat after adjustment for age and body mass index; and body fat distribution is different across racial, sex, and age (2, 10, 62, 64, 65) strata.

Other indicators have been suggested to describe fat distribution associated with abdominal obesity (2). For example, the subscapular/triceps skinfold ratio has been used to describe central versus peripheral obesity. The waist/hip ratio and the waist/thigh ratio have been used to identify upper versus lower body obesity. In addition, other indices, such as waist/height ratio, conicity index, and abdominal to mid-thigh girth, have been developed on the basis of a variety of criteria. However, ratios are more difficult to interpret biologically, are less sensitive to weight gain, and have statistical limitations (66). Because relatively few studies have considered these indicators, we did not include them in our meta-analysis.

In our analysis, we included waist/hip ratio because it was the most common obesity-related predictor of diabetes after body mass index and it has a weaker correlation with body mass index (r ~ 0.4) (13, 33) than does waist circumference. However, some have argued against the use of waist/hip ratio as a measure of obesity because of its ambiguous biologic interpretation, its lesser sensitivity to weight gain, its greater variability across age, sex, and ethnic groups, and its greater computational complexity and interpretation in a public health context (2).

In the present analysis, we have compared the associations across the obesity indicators by focusing on the difference between risk ratios. Several other strategies are possible to compare the performance of disease markers, such as measures of predictive power (likelihood measures) or measures of discriminatory performance, such as the area under the receiver-operating characteristic curve. For example, Stevens et al. (15) found that waist circumference had better discriminatory performance for diabetes than did body mass index or waist/hip ratio.

A potential problem that arises in meta-analyses of observational data, such as this one, is that the findings may appear to be very precise but are simply reinforcing biases present in individual studies. In addition, although an individual study may report an "un-confounded" estimate of an association, associations may differ between studies because of different distributions of the obesity indicator and the confounding variables or because of differences in the mechanism in biologic action across populations (67).

The necessity to investigate heterogeneity in meta-analyses of observational studies is well recognized (68). We investigated the role of study-level characteristics and found only two characteristics that affected the strength of the association: incident diabetes rate and study inclusion criteria. A smaller relative risk in studies with a higher incident diabetes rate or higher baseline glucose levels may be explained by study-level confounding or a different mechanism of biologic action between obesity and diabetes in those populations where diabetes is more prevalent. We found few differences in the relative risk of the three obesity indicators for each population group. Only for Asia do body mass index and waist circumference seem to have a stronger association than does waist/hip ratio with incident diabetes. We hypothesized that associations may be heterogeneous, reflecting different underlying causes of overweight, genetic predisposition, and obesity distribution.

When comparing differences between obesity indicators across study-level characteristics, we found that the present analysis had several limitations. Comparisons of waist circumference and body mass index or of waist/hip ratio and body mass index were not based on the same set of studies. In the case of body mass index, the estimated relative risk for the full set of studies was different from the estimated relative risk based on the subset of studies used for the pairwise comparisons with waist circumference and waist/hip ratio. The differences seen in the estimated relative risk between subsets may be due to differences in the study characteristics (study-level confounding) and random variation. When we performed ecologic comparisons, studies included in the analysis were not the same for all three obesity indicators, potentially introducing study-level confounding. Sparseness of studies in each category did not allow us to further analyze the heterogeneity with a multivariable approach. Additional heterogeneity may be derived from diversity of design features, clinical characteristics, and model assumptions. Importantly, study-level characteristics are ecologic and may not reflect the relative risks between subgroups formed across individuals within a study.

Although it is important to summarize the existing literature, the present meta-analysis suffers from having to work with the data as they were reported. In addition, for the relative risk to be obtained, assumptions had to be made regarding the distribution of the three anthropometric indicators and the correlation of the estimates across anthropometric categories. A more detailed approach can be performed with meta-analysis of data on individual participants.

The Collaborative Study of Obesity and Diabetes in Adults (CODA) project has been established to answer some of these questions. This project has collected data on individual participants from 37 studies worldwide (69). Preliminary results presented at the American Diabetes Association's 64th Scientific Sessions (63) were similar to those found in the present literature-based meta-analysis, that waist circumference is a slightly better predictor of diabetes than is body mass index: RRWC = 2.1, 95 percent CI: 1.9, 2.3; RRBMI = 1.9, 95 percent CI: 1.7, 2.0. The estimated relative risks from the data on individual participants were slightly higher than those observed in this literature-based meta-analysis. We hypothesize that the difference in the estimated relative risk may be explained because different sets of studies were analyzed in each project, producing random variation and study-level confounding, in addition to any bias introduced by our analytical approach.

In conclusion, despite the largely unexplained heterogeneity of relative risk, the present study demonstrated consistently strong associations of body mass index, waist circumference, and waist/hip ratio with incident diabetes. Although the clinical appeal of use of a measure of visceral fat is undeniable, the statistical reality is that waist circumference and body mass index are very highly correlated and likely to behave similarly in diabetes prediction. Waist/hip ratio, despite lower correlation with body mass index and waist circumference, appears to have the same ability to predict diabetes as do both body mass index and waist circumference. Additional insight into these issues will be gained by a meta-analysis of data on individual participants, such as the Collaborative Study of Obesity and Diabetes in Adults project is currently undertaking.


   ACKNOWLEDGMENTS
 
This study was supported by the Division of Diabetes Translation, US Centers for Disease Control and Prevention, contract 200-2001-00061.

Conflict of interest: none declared.


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 RESEARCH DESIGN AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 

  1. DeFronzo RA. Pathogenesis of type 2 diabetes mellitus. Med Clin North Am (2004) 88:787–835.[CrossRef][ISI][Medline]
  2. Molarius A, Seidell JC. Selection of anthropometric indicators for classification of abdominal fatness—a critical review. Int J Obes Relat Metab Disord (1998) 22:719–27.[CrossRef][ISI][Medline]
  3. Despres JP, Moorjani S, Lupien PJ, et al. Regional distribution of body fat, plasma lipoproteins, and cardiovascular disease. Arteriosclerosis (1990) 10:497–511.[Abstract/Free Full Text]
  4. Despres JP, Moorjani S, Tremblay A, et al. Relation of high plasma triglyceride levels associated with obesity and regional adipose tissue distribution to plasma lipoprotein-lipid composition in premenopausal women. Clin Invest Med (1989) 12:374–80.[ISI][Medline]
  5. Peiris AN, Mueller RA, Smith GA, et al. Splanchnic insulin metabolism in obesity. Influence of body fat distribution. J Clin Invest (1986) 78:1648–57.[ISI][Medline]
  6. Jensen MD. Is visceral fat involved in the pathogenesis of the metabolic syndrome? Human model. Obesity (Silver Spring) (2006) 14(suppl):20S–4S.
  7. Rankinen T, Kim SY, Perusse L, et al. The prediction of abdominal visceral fat level from body composition and anthropometry: ROC analysis. Int J Obes Relat Metab Disord (1999) 23:801–9.[CrossRef][ISI][Medline]
  8. Pouliot MC, Despres JP, Lemieux S, et al. Waist circumference and abdominal sagittal diameter: best simple anthropometric indexes of abdominal visceral adipose tissue accumulation and related cardiovascular risk in men and women. Am J Cardiol (1994) 73:460–8.[CrossRef][ISI][Medline]
  9. Goodman-Gruen D, Barrett-Connor E. Sex differences in measures of body fat and body distribution in the elderly. Am J Epidemiol (1996) 143:898–906.[Abstract/Free Full Text]
  10. Hill JO, Sidney S, Lewis CE, et al. Racial differences in amounts of visceral adipose tissue in young adults: the CARDIA (Coronary Artery Risk Development in Young Adults) Study. Am J Clin Nutr (1999) 69:381–7.[Abstract/Free Full Text]
  11. Seidell JC, Bjorntorp P, Sjostrom L, et al. Regional distribution of muscle and fat mass in men—new insight into the risk of abdominal obesity using computed tomography. Int J Obes (1989) 13:289–303.[ISI][Medline]
  12. Ford ES, Mokdad AH, Giles WH. Trends in waist circumference among U.S. adults. Obes Res (2003) 11:1223–31.[ISI][Medline]
  13. Wei M, Gaskill SP, Haffner SM, et al. Waist circumference as the best predictor of noninsulin dependent diabetes mellitus (nIDDM) compared to body mass index, waist/hip ratio and other anthropometric measurements in Mexican Americans—a 7-year prospective study. Obes Res (1997) 5:16–23.[ISI][Medline]
  14. Janssen I, Katzmarzyk PT, Ross R. Waist circumference and not body mass index explains obesity-related health risk. Am J Clin Nutr (2004) 79:379–84.[Abstract/Free Full Text]
  15. Stevens J, Couper D, Pankow J, et al. Sensitivity and specificity of anthropometrics for the prediction of diabetes in a biracial cohort. Obes Res (2001) 9:696–705.[ISI][Medline]
  16. Ohlson LO, Larsson B, Bjorntorp P, et al. Risk factors for type 2 (non-insulin-dependent) diabetes mellitus. Thirteen and one-half years of follow-up of the participants in a study of Swedish men born in 1913. Diabetologia (1988) 31:798–805.[ISI][Medline]
  17. Lundgren H, Bengtsson C, Blohme G, et al. Adiposity and adipose tissue distribution in relation to incidence of diabetes in women: results from a prospective population study in Gothenburg, Sweden. Int J Obes (1989) 13:413–23.[ISI][Medline]
  18. Cassano PA, Rosner B, Vokonas PS, et al. Obesity and body fat distribution in relation to the incidence of non-insulin-dependent diabetes mellitus. A prospective cohort study of men in the Normative Aging Study. Am J Epidemiol (1992) 136:1474–86.[Abstract/Free Full Text]
  19. Resnick HE, Halter JB, Valsania P, et al. Differential effect of BMI on diabetes risk among black and white Americans. Diabetes Care (1998) 21:1828–35.[Abstract]
  20. Nakagami T, Qiao Q, Carstensen B, et al. Age, body mass index and type 2 diabetes—associations modified by ethnicity. Diabetologia (2003) 46:1063–70.[CrossRef][ISI][Medline]
  21. Abate N, Chandalia M. The impact of ethnicity on type 2 diabetes. J Diabetes Complications (2003) 17:39–58.[CrossRef][ISI][Medline]
  22. Haffner SM, Mitchell BD, Hazuda HP, et al. Greater influence of central distribution of adipose tissue on incidence of non-insulin-dependent diabetes in women than men. Am J Clin Nutr (1991) 53:1312–17.[Abstract/Free Full Text]
  23. Boyko EJ, Fujimoto WY, Leonetti DL, et al. Visceral adiposity and risk of type 2 diabetes: a prospective study among Japanese Americans. Diabetes Care (2000) 23:465–71.[Abstract]
  24. Chan JM, Rimm EB, Colditz CA, et al. Obesity, fat distribution, and weight gain as risk factors for clinical diabetes in men. Diabetes Care (1994) 17:961–9.[Abstract]
  25. Spiegelman D, Israel RG, Bouchard C, et al. Absolute fat mass, percent body fat, and body-fat distribution: which is the real determinant of blood pressure and serum glucose? Am J Clin Nutr (1992) 55:1033–44.[Abstract/Free Full Text]
  26. Greenland S, Longnecker MP. Methods for trend estimation from summarized dose-response data, with application to meta-analysis. Am J Epidemiol (1992) 135:1301–9.[Abstract/Free Full Text]
  27. SAS Institute, Inc. SAS OnlineDoc 9.1.3 (2003) Cary, NC: SAS Institute, Inc. (http://support.sas.com/onlinedoc/913/docMainpage.jsp).
  28. van Houwelingen HC, Arends LR, Stijnen T. Tutorial in biostatistics. Advanced methods in meta-analysis: multivariate approach and meta-regression. Stat Med (2002) 21:589–624.[CrossRef][ISI][Medline]
  29. Higgins J, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat Med (2002) 21:1539–58.[CrossRef][ISI][Medline]
  30. Duval S, Tweedie R. A nonparametric "trim and fill" method of accounting for publication bias in meta-analysis. J Am Stat Assoc (2000) 95:89–98.[CrossRef][ISI]
  31. StataCorp LP. Stata statistical software: release 8.0 (2003) College Station, TX: StataCorp LP.
  32. Feskens EJM, Kromhout D. Cardiovascular risk factors and the 25-year incidence of diabetes mellitus in middle-aged men. Am J Epidemiol (1989) 130:1101–8.[Abstract/Free Full Text]
  33. Ohlson LO, Larsson B, Svardsudd K, et al. The influence of body fat distribution on the incidence of diabetes mellitus: 13.5 years of follow-up of the participants in the study of men born in 1913. Diabetes (1985) 34:1055–8.[Abstract]
  34. Lipton RB, Liao Y, Cao G, et al. Determinants of incident non-insulin-dependent diabetes mellitus among Blacks and Whites in a national sample. The NHANES I Epidemiological Follow-up Study. Am J Epidemiol (1993) 138:826–39.[Abstract/Free Full Text]
  35. Folsom AR, Kushi LH, Anderson KE, et al. Associations of general and abdominal obesity with multiple health outcomes in older women: the Iowa Women's Health Study. Arch Intern Med (2000) 160:2117–28.[Abstract/Free Full Text]
  36. Young TK, Harris SB. Risk of clinical diabetes in a northern Native Canadian cohort. Arctic Med Res (1994) 53:64–70.[Medline]
  37. Mykkanen L, Kuusisto J, Pyorala K, et al. Cardiovascular disease risk factors as predictors of type 2 (non-insulin-dependent) diabetes mellitus in elderly subjects. Diabetologia (1993) 36:553–9.[ISI][Medline]
  38. Carey VJ, Walters EE, Colditz GA, et al. Body fat distribution and risk of non-insulin-dependent diabetes mellitus in women: the Nurses' Health Study. Am J Epidemiol (1997) 145:614–19.[Abstract/Free Full Text]
  39. Schmidt MI, Duncan BB, Sharrett AR, et al. Markers of inflammation and prediction of diabetes mellitus in adults (Atherosclerosis Risk in Communities Study): a cohort study. Lancet (1999) 353:1649–52.[CrossRef][ISI][Medline]
  40. Boyko EJ, Chitson P, de Courten M, et al. Features of the metabolic syndrome predict higher risk of diabetes and impaired glucose tolerance. A prospective study of Mauritius. Diabetes Care (2000) 23:1242–8.[Abstract/Free Full Text]
  41. Snijder MB, Dekker JM, Visser M, et al. Associations of hip and thigh circumferences independent of waist circumference with the incidence of type 2 diabetes: the Hoorn Study. Am J Clin Nutr (2003) 77:1192–7.[Abstract/Free Full Text]
  42. Festa A, D'Agostino RB, Tracy RP, et al. Elevated levels of acute-phase proteins and plasminogen activator inhibitor-1 predict the development of type 2 diabetes: the Insulin Resistance Atherosclerosis Study. Diabetes (2002) 51:1131–7.[Abstract/Free Full Text]
  43. Shin CS, Lee HK, Koh CS, et al. Risk factors for the development of NIDDM in Yonchon County, Korea. Diabetes Care (1997) 20:1842–6.[Abstract]
  44. Sargeant LA, Bennett FI, Forrester TE, et al. Predicting incident diabetes in Jamaica: the role of anthropometry. Obes Res (2002) 10:792–8.[ISI][Medline]
  45. McNeely MJ, Boyko EJ, Shofer JB, et al. Standard definitions of overweight and central adiposity for determining diabetes risk in Japanese Americans. Am J Clin Nutr (2001) 74:101–7.[Abstract/Free Full Text]
  46. Wang SL, Pan WH, Hwu CM, et al. Incidence of NIDDM and the effects of gender, obesity and hyperinsulinaemia in Taiwan. Diabetologia (1997) 40:1431–8.[CrossRef][ISI][Medline]
  47. Burke JP, Williams K, Haffner SM, et al. Elevated incidence of type 2 diabetes in San Antonio, Texas, compared with that of Mexico City, Mexico. Diabetes Care (2001) 24:1573–8.[Abstract/Free Full Text]
  48. Warne DK, McCance DR, Charles MA, et al. Comparison of body size measurements as predictors of NIDDM in Pima Indians. Diabetes Care (1995) 18:435–9.[Abstract]
  49. Daimon M, Oizumi T, Saitoh T, et al. Decreased serum levels of adiponectin are a risk factor for the progression to type 2 diabetes in the Japanese population: the Funagata Study. Diabetes Care (2003) 26:2015–20.[Abstract/Free Full Text]
  50. Nauck MA, Meier JJ, Wolfersdorff AV, et al. A 25-year follow-up study of glucose tolerance in first-degree relatives of type 2 diabetic patients: association of impaired or diabetic glucose tolerance with other components of the metabolic syndrome. Acta Diabetol (2003) 40:163–72.[CrossRef][ISI][Medline]
  51. Chen KT, Chen CJ, Gregg EW, et al. Impaired fasting glucose and risk of diabetes in Taiwan: follow-up over 3 years. Diabetes Res Clin Pract (2003) 60:177–82.[CrossRef][ISI][Medline]
  52. Spranger J, Kroke A, Mohlig M, et al. Inflammatory cytokines and the risk to develop type 2 diabetes: results of the prospective population-based European Prospective Investigation into Cancer and Nutrition (EPIC)-Potsdam Study. Diabetes (2003) 52:812–17.[Abstract/Free Full Text]
  53. Laaksonen DE, Lakka HM, Niskanen LK, et al. Metabolic syndrome and development of diabetes mellitus: application and validation of recently suggested definitions of the metabolic syndrome in a prospective cohort study. Am J Epidemiol (2002) 156:1070–7.[Abstract/Free Full Text]
  54. Harding AH, Day NE, Khaw KT, et al. Dietary fat and the risk of clinical type 2 diabetes: the European Prospective Investigation of Cancer-Norfolk Study. Am J Epidemiol (2004) 159:73–82.[Abstract/Free Full Text]
  55. Rodriguez-Moran M, Guerrero-Romero F. Fasting glucose/insulin index and insulin levels 2-h after glucose load are predictors of the development of type 2 diabetes. Gac Med Mex (2000) 136:201–6.[Medline]
  56. Edelstein SL, Knowler WC, Bain RP, et al. Predictors of progression from impaired glucose tolerance to NIDDM: an analysis of six prospective studies. Diabetes (1997) 46:701–10.[Abstract]
  57. Report of the Expert Committee on the Diagnosis and Classification of Diabetes Mellitus. Diabetes Care (1997) 20:1183–97.[ISI][Medline]
  58. WHO Expert Committee on Diabetes Mellitus: second report. World Health Organ Tech Rep Ser (1980) 646:1–80.[Medline]
  59. Diabetes mellitus. Report of a WHO Study Group. World Health Organ Tech Rep Ser (1985) 727:1–113.[Medline]
  60. Definition, diagnosis and classification of diabetes mellitus and its complications. Part 1: diagnosis and classification of diabetes mellitus. Report of a WHO Consultation (1999) Geneva, Switzerland: World Health Organization. (http://www.staff.ncl.ac.uk/philip.home/who_dmc.htm).
  61. Hartemink N, Boshuizen HC, Nagelkerke NJD, et al. Combining risk estimates from observational studies with different exposure cutpoints: a meta-analysis on body mass index and diabetes type 2. Am J Epidemiol (2006) 163:1042–52.[Abstract/Free Full Text]
  62. Janssen I, Heymsfield SB, Allison DB, et al. Body mass index and waist circumference independently contribute to the prediction of nonabdominal, abdominal subcutaneous, and visceral fat. Am J Clin Nutr (2002) 75:683–8.[Abstract/Free Full Text]
  63. CODA Study group, Duval SJ. The association of BMI and waist circumference with diabetes in an international context: the CODA (Collaborative Study of Obesity and Diabetes in Adults) project. Presented at the American Diabetes Association 64th Scientific Sessions, Orlando, Florida, June 4–8, 2004.
  64. Molarius A, Seidell JC, Sans S, et al. Waist and hip circumferences, and waist-hip ratio in 19 populations of the WHO MONICA Project. Int J Obes Relat Metab Disord (1993) 23:116–25.[CrossRef]
  65. Duncan BB, Chambless LE, Schmidt MI, et al. Correlates of body fat distribution. Variation across categories of race, sex, and body mass in the Atherosclerosis Risk in Communities Study. Ann Epidemiol (1995) 5:192–200.[CrossRef][Medline]
  66. Allison DB, Paultre F, Goran MI, et al. Statistical considerations regarding the use of ratios to adjust data. Int J Obes Relat Metab Disord (1995) 19:644–52.[ISI][Medline]
  67. Egger M, Schneider M, Davey Smith G. Meta-analysis Spurious precision? Meta-analysis of observational studies. BMJ (1998) 316:140–4.[Free Full Text]
  68. Stroup DF, Berlin JA, Morton SC, et al. Meta-analysis of observational studies in epidemiology: a proposal for reporting. JAMA (2000) 283:2008–12.[Abstract/Free Full Text]
  69. Duval S, Vazquez G, Baker WL, et al. The Collaborative Study of Obesity and Diabetes in Adults (CODA) project: meta-analysis design and description of participating studies. Obes Rev (2007) 8:263–76.[CrossRef][Medline]

Add to CiteULike CiteULike   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us    What's this?


This article has been cited by other articles:


Home page
 DiabetesHome page
A. J.G. Hanley and L. E. Wagenknecht
Abdominal Adiposity and Diabetes Risk: The Importance of Precise Measures and Longitudinal Studies
Diabetes, May 1, 2008; 57(5): 1153 - 1155.
[Full Text] [PDF]

Home page
 Am J EpidemiolHome page
S. K. Kumanyika
The Obesity Epidemic: Looking in the Mirror
Am. J. Epidemiol., August 1, 2007; 166(3): 243 - 245.
[Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
29/1/115    most recent
mxm008v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrowRequest Permissions
Google Scholar
Right arrow Articles by Vazquez, G.
Right arrow Articles by Silventoinen, K.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Vazquez, G.
Right arrow Articles by Silventoinen, K.