Combination of BMI and Waist Circumference for Identifying Cardiovascular Risk Factors in Whites

Transcription

1 Combination of BMI and Waist Circumference for Identifying Cardiovascular Risk Factors in Whites Shankuan Zhu,* Stanley Heshka,* ZiMian Wang,* Wei Shen,* David B. Allison, Robert Ross, and Steven B. Heymsfield* Abstract ZHU, SHANKUAN, STANLEY HESHKA, ZIMIAN WANG, WEI SHEN, DAVID B. ALLISON, ROBERT ROSS, AND STEVEN B. HEYMSFIELD. Combination of BMI and waist circumference for identifying cardiovascular risk factors in whites. Obes Res. 2004;12: Objective: BMI (kilograms per meters squared) and waist circumference (WC) (measured in centimeters) are each associated with the risk of developing cardiovascular disease (CVD). Therefore, a combination of the two may be more effective in identifying subjects at risk than either alone. The present study sought to identify the combination of BMI and WC that has the strongest association with CVD risk factors in whites. Research Methods and Procedures: Subjects were 8712 white men and women from the Third National Health and Nutrition Examination Survey. The optimal combination of BMI and WC was developed using logistic regression models with BMI and WC as predictors and CVD risk factors as outcomes. The combined measure of BMI and WC using current cut-off points was also examined. Sensitivity, specificity, and receiver operating characteristics curves were compared between the combined measures and BMI alone. Results: For white men, the optimal combination of BMI Received for review April 29, Accepted in final form January 26, The costs of publication of this article were defrayed, in part, by the payment of page charges. This article must, therefore, be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. *New York Obesity Research Center, St. Luke s-roosevelt Hospital Center and Institute of Human Nutrition, College of Physicians and Surgeons, Columbia University, New York; Injury Research Center, Department of Family and Community Medicine, Medical College of Wisconsin, Wisconsin; Department of Biostatistics, Section on Statistical Genetics and Clinical Nutrition Research Center, University of Alabama, Birmingham, Alabama; and School of Physical and Health Education, Queens University, Kingston, Ontario. Address correspondence to Shankuan Zhu, Injury Research Center, Department of Family and Community Medicine, Medical College of Wisconsin, 8701 Watertown Plank Road, Milwaukee, WI szhu@mcw.edu Copyright 2004 NAASO and WC for identifying CVD risk factors was 0.68 BMI 0.32 WC. This combination generated a score that better estimated the odds of having CVD risk factors than either alone. For white women, WC alone largely determined the likelihood of having CVD risks. The combination of BMI and WC using current cut-off points may provide an improved measure of CVD risk. Combined measures showed a higher sensitivity or a shorter distance in receiver operating characteristic curves in the identification of CVD risk factors. Discussion: Combined measures of BMI and WC may provide a higher overall test performance for CVD risk factors and may be useful in some ethnic groups as an improved means of screening subjects for further evaluation in the clinical setting. Key words: BMI, waist circumference, cardiovascular disease, risk factors Introduction Healthy weight, overweight, and obesity are defined according to the magnitude of BMI, expressed as weight in kilograms divided by the square of height in meters. According to the World Health Organization s (WHO s) 1 definition, overweight is classified as a BMI 25 kg/m 2, and obesity is defined as a BMI 30 kg/m 2 for both men and women (1). Although BMI correlates well with measures of adiposity (2) and is associated with both mortality (3) and some chronic diseases (1,4,5), other factors contribute independently to risk (6,7). Specifically, many studies have reported 1 Nonstandard abbreviations: WHO, World Health Organization; WC, waist circumference; CVD, cardiovascular disease; NHANES III, Third National Health and Nutrition Examination Survey; HDL, high-density lipoprotein cholesterol; TG, triglyceride; LDL, low-density lipoprotein cholesterol; O BMI WC, optimal scaling combination; C BMI WC, combination measure using current BMI and WC cut-off points; OR, odds ratio; ROC, receiver operating characteristics; CI, confidence interval. OBESITY RESEARCH Vol. 12 No. 4 April

2 that after controlling for BMI, increased intra-abdominal adipose tissue is strongly associated with metabolic and cardiovascular risk and a variety of chronic diseases (8 14). BMI does not account for the wide variation in body fat distribution that exists at any level of relative body size (1,4). Waist circumference (WC), however, compensates for this limitation of BMI, by bringing regional fat into consideration (15 21). Janssen et al. (7,21) have shown that WC has an independent association with cardiovascular disease (CVD), indicating the potential utility of using WC in addition to BMI in clinical practice and suggesting that a combination of BMI and WC may be preferable to BMI alone for obesity risk assessment. Such a combination measure would require little extra cost or equipment and could increase the clinician s ability to identify individuals at high risk for diseases that correlate with excess adiposity. However, a more recent study brings into question the potential utility of a combination measure (22). The study reports that, with the exception of hypertension in men, BMI makes no independent contribution to predicting CVD risk once WC has been taken into account. Thus, although it has been demonstrated frequently that WC makes an independent contribution when added to BMI, initial observations suggest that the converse has not been investigated carefully. The aim of the present study was to investigate whether a combined measure of BMI and WC was preferable to BMI or WC alone for the identification of CVD risk factors in a specific subsample, whites, of the Third National Health and Nutrition Examination Survey (NHANES III). Our hypothesis was that a combined measure of BMI and WC would have a better overall test performance for detecting CVD risk factors compared with BMI or WC alone. Research Methods and Procedures Study Population Subjects were from NHANES III, which was conducted by the National Center for Health Statistics during the period 1988 to Detailed information on NHANES III is presented elsewhere (23). We initially restricted the study to 10,103 whites 20 years of age with available anthropometric and medical examination data. Additionally, we excluded 1026 subjects who consumed food or beverages within 6 hours before the venipuncture, 161 women who were pregnant or lactating at baseline, and 204 subjects with abmi 40 kg/m 2. Of the remaining 8712 subjects, there were 4290 men and 4422 women. The following information was used in the present study: anthropometric and demographic information including age, height, weight, BMI, WC, smoking and drinking status, physical activity, economic status, education level, and menopausal status; and medical examination data including systolic and diastolic blood pressure, total cholesterol, highdensity lipoprotein cholesterol (HDL), triglycerides (TGs), serum glucose, and medication for diabetes or hypertension at baseline. The measurement procedures for anthropometric and laboratory tests are described elsewhere (24,25). Definition of CVD Risk Factors We defined the CVD risk factors according to The Practical Guide: Identification, Evaluation, and Treatment of Overweight and Obesity in Adults released by the NIH in 2000 (6): 1) elevated low-density lipoprotein (LDL)-cholesterol [LDL 160 mg/dl (4.14 mm); LDL was calculated as LDL total cholesterol HDL 0.2 TG]; 2) low HDL-cholesterol [HDL 35 mg/dl (0.91 mm) for men and 45 mg/dl (1.17 mm) for women]; 3) high blood pressure [systolic blood pressure 90 mm Hg or diastolic blood pressure 140 mm Hg]; and 4) high glucose [serum glucose 125 mg/dl (6.94 mm)]. Subjects with one or more of the above conditions or currently under medication treatment for diabetes or hypertension were considered as having CVD risk factors. Derivation of Combined BMI and WC Measures Two types of BMI and WC combinations were developed in the present study: one was an optimal scaling combination (O BMI WC ), and the other was a combination measure using current BMI and WC cut-off points (C BMI WC ). Optimal Scaling Combination of BMI and WC. The O BMI WC was expressed as O BMI WC a BMI BMI a WC WC. The proportion coefficients of BMI (a BMI ) and WC (a WC ) were calculated as a BMI BMI /( BMI WC ) and a WC WC /( BMI WC ), respectively, where BMI and WC were the coefficient parameters of BMI and WC derived from a logistic regression model. To obtain coefficients, we entered BMI and WC as continuous predictor variables in a logistic regression model in which the dependent variable was the presence or absence of CVD risk factors and covariates were age, smoking and drinking habits, physical activity, education and income levels, and menopausal status for women. To assess the effects of CVD and diabetes history on the associations of BMI and WC with CVD risk factors, we ran the analysis again excluding subjects who had a CVD or diabetes history but did not have any of the four CVD risk factors at baseline. A CVD or diabetes history included self-reported type 2 diabetes, hypertension, or a history of heart attack, congestive heart failure, or stroke. Combination Using Current Existing BMI and WC Cutoff Points. Our previous study (26) revealed the importance of assessing WC as an indicator of obesity-associated health risk. The results displayed the relationship of CVD risk factors with BMI and WC, in particular, the points at which the risks associated with a given WC correspond to those for BMIs of 25 and 30 kg/m 2. These correspondence points suggest that a WC of 90 cm for men and 83 cm for women, 634 OBESITY RESEARCH Vol. 12 No. 4 April 2004

3 Figure 1: Criteria for a combined measure of BMI and WC using current existing cut-off points. F, female; M, male. equivalent in risk to a BMI of 25 kg/m 2, may represent an action level for limiting future weight gain, whereas a WC of 100 cm for men and 93 cm for women, equivalent to a BMI of 30 kg/m 2, suggest the need for risk reduction and weight loss. Thus, in the present study, we used a WC of 90 cm for men and 83 cm for women and 100 cm for men and 93 for women as thresholds. Figure 1 shows the criteria for threshold levels based jointly on BMI and WC cut-off points. According to the classification of BMI for normal, overweight, and obesity (1) and WC for clinical threshold levels I and II (26), the normal range of C BMI WC was defined as BMI 25 kg/m 2 and WC 90 cm for men and 83 cm for women; clinical threshold level I was defined as BMI 25 kg/m 2 and 30 kg/m 2 or WC 90 cm and 100 cm for men and WC 83 cm and 93 cm for women; and clinical threshold level II was defined as BMI 30 kg/m 2 or WC 100 cm for men and WC 93 cm for women. Model Covariates. Physically active, moderately active, and physically inactive were defined based on subject physical activity intensity rating scores obtained from participating in one of the following activities during the past 1 month: walking, jogging or running, riding bicycle, swimming, doing aerobics or aerobic dancing, other dancing, doing calisthenics, doing garden or yard work, lifting weights, or other activities. The physical activity intensity rating scores were defined as the ratio of activity metabolic rate to resting metabolic rate (25). The physically inactive category included subjects with a total intensity rating score of 3.5. The physically active category was defined as a total intensity rating score of The point at which the total intensity rating score equals 3.5 and 12.5 corresponds to the 20th and 60th percentiles in the study samples, respectively. Smoking was categorized as current, past, and never smoker. Past smokers were those who reported that they had smoked at least 100 cigarettes during their lifetime but did not currently smoke cigarettes. Drinking was categorized as heavy, moderate, never drank, and unknown. Heavy drinkers were subjects who OBESITY RESEARCH Vol. 12 No. 4 April

4 responded that they ever drank five or more alcoholic beverages almost every day or drank beer, wine, or hard liquor one time per day during the past month. Moderate drinkers consumed less alcohol than the heavy drinkers, and never drinkers were those who did not drink beer, wine, or hard liquor during the past month. Education level was divided into three categories: 8 years, 8 to 12 years, and more than 12 years of education. Economic status was divided into three categories according to the previous year s household income: $15,000 or less, $15,001 to $25,000, and over $25,000. Menopausal status was determined from the response to an interview item on the cessation of menses for at least 12 months. Statistical Analysis The percentages of subjects with metabolic risk factors were compared between men and women using an adjusted Wald test (27). The prevalence of CVD risk factors was calculated according to a three three cross-tabulation of BMI defined as a BMI of 25, 25 to 29.9, and 30 kg/m 2 and WC categorized as a WC of 90 cm for men and 83 cm for women, 90 to 99.9 cm for men and 83 to 92.9 cm for women, and 100 cm for men and 93 cm for women. Separate logistic regression models were applied to calculate the odds ratio (OR) of having CVD risk factors vs. no CVD risk factors for subjects in each cell of the three three cross-tabulation compared with subjects in the normal weight range cell. We developed two exponential equations for calculating the OR for BMI or O BMI WC using logistic regression models: OR i EXP [ (X i X ref )], where EXP represents the exponent, X i is a specific BMI or O BMI WC,X ref is the reference point of BMI or O BMI WC, and is the coefficient parameter of BMI or O BMI WC derived from logistic regression models, which were also adjusted for age, smoking, alcohol consumption, physical activity, education level, economic status, and menopausal status. The reference point was set at the 25th percentile for BMI or O BMI WC in the male and the female study samples (i.e., O BMI WC 54.0 and BMI 23.6 kg/m 2 in men and O BMI WC 69.0 and BMI 21.8 kg/m 2 in women). These reference values were chosen because the BMI values corresponding to the 25th percentile in the study population are considered to have the lowest risk of death from any cause (28,29). Values of O BMI WC were then identified that carried the same OR as each BMI unit in the BMI range of 16 to 40 kg/m 2.O BMI WC thresholds were identified where ORs for O BMI WC corresponded to those seen at BMI values of 25 and 30 kg/m 2 (i.e., the cut-off points for overweight and obesity defined by WHO). In addition, the interaction term between BMI and WC was added into the above-mentioned logistic models for men and women to test whether BMI and WC interacted in terms of a relationship with CVD risks. We then conducted sensitivity and specificity analyses to compare sensitivity and specificity at two different cut-off points for BMI, O BMI WC, and C BMI WC in the identification of CVD risk factors. Sensitivity refers to the probability of correctly diagnosing subjects with CVD risk factors by the test (i.e., BMI, O BMI WC,orC BMI WC ), whereas specificity is a measure of probability of correctly identifying subjects with no CVD risk factors by the test (30). The cut-off points of the test were first set at action level I (i.e., 25 kg/m 2 ) for BMI, 46 for O BMI WC, which corresponds to risk at BMI of 25 kg/m 2, and 25 kg/m 2 BMI and 90 cm WC for C BMI WC for men. For women, the cut-off points of the test were set at 25 kg/m 2 for BMI, 83 for O BMI WC, and 25 kg/m 2 BMI and 83 cm WC for C BMI WC. The cut-off points were then reset at action level II (i.e., 30 kg/m 2 ) for BMI, 54 for O BMI WC, which corresponds to risk at BMI of 30 kg/m 2, and 30 kg/m 2 BMI and 100 cm WC for C BMI WC for men. For women, the corresponding values were 30 kg/m 2 for BMI, 94 for O BMI WC, and 30 kg/m 2 BMI and 93 cm WC for C BMI WC. The Euclidean distance from a point on a receiver operating characteristic (ROC) curve to maximum sensitivity and maximum specificity (i.e., when both sensitivity and specificity equal 100%) was also calculated for each of the above-mentioned cut-off points. The distances at each corresponding cut-off point for BMI and the combined measures were compared, and the shorter distance indicated greater sensitivity and specificity [i.e., smaller value of square root of sum of (1 sensitivity) 2 and (1 specificity) 2 ] (31). To investigate how the selection of different outcome criteria and clustering of several risk factors (i.e., metabolic syndrome) affect the relative weights of BMI and WC, we used the metabolic syndrome risk factors as an outcome and reestimated the optimal scaling of BMI and WC. Four risk factors were selected from five metabolic syndrome criteria according to the Adult Treatment Panel III (excluding abdominal obesity measured by WC) as follows (32): 1) high TGs (TGs 150 mg/dl); 2) low HDL-cholesterol (HDL 40 mg/dl for men and 50 mg/dl for women); 3) high blood pressure (systolic 130 mm Hg or diastolic 85 mm Hg); and 4) high fasting plasma glucose (fasting plasma glucose 110 mg/dl). Subjects at risk were defined as those with one or more, two or more, or three or more of these metabolic syndrome risk factors in separate analyses. Because nonstandardized regression coefficients would differ if measurement units were changed, our optimal combination equation would hold true only for BMI in kilograms per meter squared and for WC in centimeters. To verify the relative importance of each variable in the association with risk factors across changes in measurement units, we also computed and examined the standardized regression coefficients. A standardized coefficient is computed when the original raw scores on the variables have been converted to z scores before the regression. 636 OBESITY RESEARCH Vol. 12 No. 4 April 2004

5 Table 1. Subject characteristics Characteristics Men [sample size 4290; weighted size (million) 58.32] Women [sample size 4422; weighted size (million) 59.98] Mean 95% CI Mean 95% CI Age (years) to to 48.5 Height (cm) to to Weight (kg) to to 67.8 BMI (kg/m 2 ) to to 25.9 WC (cm) to to 87.9 All analyses were conducted using Stata statistical software (version 7.0 for Windows; Stata Corporation, College Station, TX) to calculate weighted data with adjustments for the complex NHANES III sample design to produce nationally representative estimates. Statistical significance was set at p 0.05 (two-tailed), and the stability of the estimates is shown by 95% confidence intervals (CIs). Men and women were analyzed separately throughout. Results The baseline characteristics of subjects and the prevalence of CVD risk factors are presented in Tables 1 and 2. In men, the coefficient for BMI was (95% confidence interval, to ; p 0.030) and for WC was (95% CI, to ; p 0.010). In women, the coefficient for WC was (95% CI, to ; p 0.001); however, the coefficient for BMI was not significant ( 0.119; 95% CI to ; p 0.490). Additional interaction terms between BMI and WC were found not significant for both men and women (men: , 95% CI to , p 0.636; women: , 95% CI to , p 0.656). The equations for calculating O BMI WC are presented in Table 3. In men, the cut-off points for O BMI WC that corresponded to the ORs of BMI at 25 and 30 kg/m 2 were 46.3 and 53.6, respectively. In women, because BMI did not make an independent contribution to the CVD risks when WC was present in the model, a combination measure was not warranted. Thus, at BMI values of 25 and 30 kg/m 2, the corresponding WC cut-off points showing the same ORs as BMI in women were 83 and 94 cm, respectively. Appendix 1 shows values of O BMI WC for men corresponding to ORs of a BMI range from 16 to 40 kg/m 2 and for women ORs of WC corresponding to a BMI range from 16 to 40 kg/m 2. A nomogram illustrating how values of O BMI WC are related to BMI risks for men is provided in the Appendix. After excluding subjects with a history of CVD or diabetes who did not have any metabolic risk factors at base- Table 2. Prevalence of metabolic risk factors Parameter Men Women Percentage 95% CI Percentage 95% CI High LDL to to 19.3 Low HDL to to 25.8 High blood pressure to * 15.1 to 18.9 High glucose to to 3.8 Hypertension medication to to 15.2 Diabetes medication to to 2.0 Subjects with one or more above items to to 50.8 * p 0.05, p 0.01, p 0.001, statistical comparisons for prevalence of metabolic risk factors between men and women, by adjusted Wald test. OBESITY RESEARCH Vol. 12 No. 4 April

6 Table 3. Optimal combination equations and the equations for calculating ORs for the presence of metabolic risk factors Logistic regression derived equations Men Women Optimal combination equations O BMI WC 0.68BMI 0.32 WC O BMI WC 0.00BMI 1.00 WC Odds ratio equations BMI OR BMI Exp [ (BMI 23.6)] OR BMI Exp [ (BMI 21.8)] O BMI WC OR OBMI WC Exp [0.1061(O BMI WC 44.2)] OR OBMI WC Exp [ (O BMI WC 76.2)] line, the optimal combination equation for men was O BMI WC 0.67 BMI 0.33 WC and for women was O BMI WC 0.20 BMI 0.80 WC. The optimal scaling of BMI and WC from regression analyses when using metabolic syndrome risk factors as an outcome was as follows: 1) subjects with one or more of four metabolic syndrome risk factors, 0.54 BMI 0.46 WC for men and 0.22 BMI 0.78 WC for women; 2) subjects with two or more of four metabolic syndrome risk factors, 0.64 BMI 0.36 WC for men and 0.45 BMI 0.55 WC for women; and 3) subjects with three or more of four metabolic syndrome risk factors, 0.73 BMI 0.27 WC for men and 0.34 BMI 0.66 WC for women. The standardized regression coefficients for men ranged from 0.30 to 0.49 for BMI and from 0.70 to 0.51 for WC, varying with the number of risk factors in the regression. The standardized regression coefficients for women ranged between 0.10 and 0.17 for BMI and 0.90 and 0.83 for WC. In all of these models (except for the models in which subjects with one or more of four metabolic syndrome risk factors were considered as an outcome), the contribution of BMI remained statistically significant (p 0.05 or p 0.01), even when WC was included in the logistic regression for men, but it was not significant for women. The prevalence and OR of having one or more CVD risk factors according to BMI and WC categories are shown in Table 4. The consistency in classification by BMI and WC categories (i.e., the percentage of subjects with the same results using BMI or WC divided by all subjects tested) was 67.9% for men and 69.5% for women. The inconsistencies in classification by BMI and WC categories were most frequently observed in the BMI 25 kg/m 2 and WC 90 to 100 cm (83 to 93 cm for women) category and the BMI of 25 to 29.9 kg/m 2 and WC 100 cm (93 cm for women) category. Among men, the lowest and highest prevalence of CVD risk factors were found in the BMI 30 kg/m 2 and WC 90 cm category (0.0%) and BMI 30 kg/m 2 and WC 100 cm category (69.6%), respectively. Given the same category of BMI or WC, ORs increased (approximately) along with the increase in WC or BMI. The highest OR was found in the BMI 30 kg/m 2 and WC of 90 to 100 cm category compared with the reference group (i.e., BMI 25 kg/m 2 and WC 90 cm). Among women, the lowest and highest prevalence were found in the BMI 25 kg/m 2 and WC 83 cm category (24.3%) and BMI 25 kg/m 2 and WC 93 cm category (82.0%), respectively. Given the same category of WC, the prevalence of CVD risk factors did not clearly increase along with the increase in BMI, whereas given the same category of BMI, the prevalence of CVD risk factors increased along with the increase of WC. The sensitivity and specificity of BMI, O BMI WC, and C BMI WC in the identification of subjects having one or more CVD risk factors at two clinical threshold levels are presented in Table 5. At action level I, compared with BMI, the sensitivities were highest for C BMI WC, followed by O BMI WC as the second highest (WC for women). The specificities were highest for O BMI WC for men, followed by BMI alone as the second highest; for women, the specificities were highest for BMI alone, followed by WC as the second highest. At action level II, sensitivities were highest for C BMI WC, but specificity was best using BMI alone. The minimum distance in ROC curves was shortest for women using WC and for men using O BMI WC at action level I. At action level II, the minimum distance in ROC curves for both men and women was observed for C BMI WC. Discussion Our results identified combinations of continuous measures of BMI and WC for identifying metabolic risk factors in white men and levels of WC in white women that corresponded to risk at a BMI range of 25 to 30 kg/m 2. The incorporation of WC in the calculation of risk offers the possibility of better identifying patients at risk for CVD than is offered using current guidelines. The currently recommended classification of healthy weight, overweight, and obesity developed by WHO (1) 638 OBESITY RESEARCH Vol. 12 No. 4 April 2004

7 Table 4. Prevalence and ORs of having one or more metabolic risk factor(s) according to BMI and WC categories BMI (kg/m 2 ) WC (cm) <25 25 to 29.9 >30 Men 90 Subject number (million)* Subject percentage Prevalence (%) Odds ratio (CI) 1.0 (reference) 2.4 (1.22 to 4.83) 90 to 100 Subject number (million) Subject percentage Prevalence (%) Odds ratio (CI) 1.3 (0.97 to 1.73) 2.7 (2.07 to 3.56) 6.5 (2.36 to 18.08) 100 Subject number (million) Subject percentage Prevalence (%) Odds ratio (CI) 1.9 (0.70 to 5.28) 3.3 (2.20 to 4.84) 5.6 (3.95 to 7.81) Women 83 Subject number (million) Subject percent (%) Prevalence (%) Odds ratio (CI) 1.0 (reference) 2.5 (1.36 to 4.74) 4.0 (0.34 to 46.85) 83 to 93 Subject number (million) Subject percent (%) Prevalence (%) Odds ratio (CI) 2.3 (1.55 to 3.38) 3.1 (2.46 to 3.90) 1.9 (0.94 to 3.72) 93 Subject number (million) Subject percent (%) Prevalence (%) Odds ratio (CI) 6.6 (2.73 to 16.00) 4.9 (3.28 to 7.28) 8.0 (5.36 to 11.93) * Weighted subject number. Percentage of subjects who fit the BMI and WC category. The prevalence of metabolic risk factors. ORs were calculated using logistic regression models at which dependent variable was defined as having metabolic risk factors and adjusted for the covariates. The reference group was set at BMI 25 kg/m 2 and WC 90 cm for men and BMI 25 kg/m 2 and WC 83 cm for women. was established based on the empirical relationship between BMI and mortality or morbidity. Many studies have reported that obesity, as defined by BMI, is consistently related to increased blood pressure and unfavorable lipid profiles and glucose metabolism (3,33,34), and several have shown an additional benefit of considering WC (2,6,7,35,36). In the present study, 10% of subjects with BMI 25 kg/m 2 had a WC that exceeded action level I, and OBESITY RESEARCH Vol. 12 No. 4 April

8 Table 5. Sensitivity, specificity, and distances at ROC curves in identification of metabolic risk factor BMI BMI WC (optimal scaling) BMI WC (current cut-offs) Men Action Level I Cut-off points BMI 25 BMI WC 56 BMI 25/WC 90 Subject (%)* Prevalence (%) 30.5/ / /56.9 Sensitivity (%) Specificity (%) ROC distance Action Level II Cut-off points BMI 30 BMI WC 64 BMI 30/WC 100 Subject (%)* Prevalence (%) 42.9/ / /66.8 Sensitivity (%) Specificity (%) ROC distance Women Action Level I Cut-off points BMI 25 WC 83 BMI 25/WC 83 Subject (%)* Prevalence (%) 32.8/ / /63.7 Sensitivity (%) Specificity (%) ROC distance Action Level II Cut-off points BMI 30 WC 94 BMI 30/WC 93 Subject (%)* Prevalence (%) 42.3/ / /72.5 Sensitivity (%) Specificity (%) ROC distance * Percent of subject in equal or greater than cut-off point (action level) group. Prevalence of CVD risk factors: less than cut-off point (action level) group/equal or greater than cut-off point (action level) group. ROC distance square root of [(1 sensitivity) 2 (1 specificity) 2 ]. 10% of subjects with a BMI of 25 to 29.9 kg/m 2 showed a WC that exceeded action level II (Table 4). These subjects had higher ORs for CVD risk factors, indicating that a proportion of subjects with elevated risk were not identified by BMI alone. Using the current cut-off points of BMI and our previously published cut-off points for WC, we found that BMI and WC, taken jointly, had a greater sensitivity in identifying subjects at risk, although with slightly lower specificity than BMI cut-off points alone (Tables 4 and 5). These findings support the use of combined BMI and WC measures in white men and the use of WC in white women as improved clinically applicable indices of cardiovascular risk. The proportion coefficients of optimal combination scaling were obtained from BMI and WC coefficient parameters derived directly from regression models. The selected combination had the best regression model fit based on a greater log likelihood than any other combination in relation to CVD risk factors. The proportion of BMI or WC for O BMI WC in men was approximately two-thirds for BMI 640 OBESITY RESEARCH Vol. 12 No. 4 April 2004

9 and one-third for WC, given the present measurement units: BMI in kilograms per meter squared and WC in centimeters. In contrast, in women, BMI represented only 14% of O BMI WC, and the coefficient was not statistically significant, suggesting agreement with Janssen et al. (22), who have found that BMI does not contribute to identifying women with CVD risk factors once WC is taken into account. Although BMI and WC were weighted approximately two-thirds and one-third, respectively, for men in the present measurement units, the standardized coefficients, in standardized units, showed that, also for men, variation in WC had a greater impact on the likelihood of risk factors than variation in BMI. This is consistent with our previously reported finding that blood pressure, LDL, and serum glucose correlate more highly with WC than with BMI in both men and women (26). For women, again, the standardized coefficients for BMI were not significant and confirmed that the impact of WC on risk factors was far greater than the impact of BMI. We controlled for potentially confounding effects of socioeconomic status and lifestyle factors in the logistic regression analyses. It is well known that older age, lack of exercise, smoking, alcohol abuse, low economic and education levels, and postmenopausal status are all associated with a significantly increased risk of diabetes and other chronic diseases (34,37,38). Simultaneous adjustment for these factors in the regression models allowed for the potential assessment of each effect independent of the others. In addition, the optimal combination of BMI and WC did not change substantially after excluding subjects with a history of CVD or diabetes. Using cut-off points for BMI and WC jointly identified more men at metabolic risk than using WC or BMI cut-off points alone; however, specificities were higher for BMI compared with the combined measures. This indicates that some men with elevations in CVD risk factors may be missed if BMI cut-offs, rather than the proposed combined measure, are used as a screening test. On the other hand, fewer people would be incorrectly identified as having elevations in CVD risk factors. However, in a comparison of BMI, O BMI WC, and C BMI WC by ROC curves, O BMI WC performed slightly better overall for both men and women at action level I than either C BMI WC or BMI alone, whereas for action level II, C BMI WC showed better overall test performance than O BMI WC or BMI alone for both men and women (Table 5). Combination measures require very little in terms of additional measurements or equipment and will identify more patients who might potentially benefit from weight management based on CVD risk factors. For example, a man with a BMI of 24 kg/m 2 and a WC of 100 cm has a combination measure (O BMI WC ) of 0.68 BMI 0.32 WC 48.3 (formula in Table 3). In Appendix 1, we can quickly establish that this corresponds to an OR of 1.45, which is equivalent to a BMI of 26 kg/m 2. This man s risk moves from the normal to the overweight range for BMI (i.e., 25 and 30 kg/m 2 ). The improvement in sensitivity of the combined measure may make it a better index for clinical and screening purposes and one that is both practical and economical. Patients at high risk according to the combination measures should ideally have further evaluations of blood pressure and serum lipid and glucose levels. Establishing a subject as high risk might also serve as motivation for adherence to prescribed diets or exercise. Obviously, when we adopt new criteria for the purpose of screening, the therapeutic benefits and cost-effectiveness of combined measures should also be taken into consideration. A simple nomogram illustrating how a subject s risk can be quickly estimated in the clinical setting based on the combined BMI and WC measure for men is provided in Appendix 2. The principal limitation of this study is the cross-sectional relationship between predictors and CVD risks used here. Future longitudinal studies are required to confirm the prospective validity of this combination index. In this study, we examined only white subjects because the reference values of BMI and WC vary among different race-ethnicity groups. For example, BMI at the 25th percentile was 23.6, 23.0, and 24.0 kg/m 2 for white, black, and Hispanic men, respectively, and 21.8 for white women, but 23.5 for black and Hispanic women; WC was 87.6 cm for white men, but 81.4 and 85.9 cm for black and Hispanic men, respectively, and 76.2 cm for white women and 80.4 and 80.0 cm for black and Hispanic women, respectively. Therefore, simply adjusting for race-ethnicity in regression models would not necessarily reflect accurately the OR estimates that would be obtained in a study using race-ethnic-specific cut-offs. Hence, separate studies for other racial-ethnic populations seem in order. Moreover, the therapeutic benefits and costeffectiveness of the combined measure and validation of the optimal scaling equations are important topics for future research. Given that the outcomes used in this study were primarily cardiovascular- and diabetes-related risk factors, one may raise the possibility that other outcomes may affect the results. It is possible that other risks of obesity, such as cancer or cholelithiasis, may have different relationships with BMI and WC than either CVD or diabetes. However, CVD and diabetes are the most prevalent obesity-related conditions, and a primary goal of weight management is heart disease prevention. It is also possible that selection of different outcome criteria, such as more than one CVD risk factor rather than any of four CVD-related risk factors, could affect the relative weights of BMI and WC. The strengths of this study are several. Using NHANES data enabled us to describe the relationship of BMI, WC, and the combined measure of BMI and WC with CVD risks in the U.S. white population. The large sample size and the OBESITY RESEARCH Vol. 12 No. 4 April

10 high quality of the anthropometric measurements and laboratory data reduced the potential biases and measurement errors. Adjustment for age, socioeconomic status, and lifestyle factors in the regression models allowed us to assess the independent relationships of BMI and WC with CVD risks and the combined measure of BMI and WC with CVD risks. In conclusion, our results identify an optimal combination of BMI and WC that is more strongly related to CVD risk factors than BMI or WC alone in white men and provide WC cut-off points that correspond to ORs based on BMI for white women. These BMI and WC equivalents may be useful as an improved measure of obesity-related risk in clinical and field settings. Derivation of these potentially useful combination measures is needed for other ethnic groups. Acknowledgments This research was supported in part by NIH Grants P30DK056336, P01DK042618, and P30DK and by an unrestricted grant from Tanita and Pfizer Pharmaceutical Inc. We are grateful to Dr. Shine Chang for her helpful comments. References 1. World Health Organization. Obesity, preventing and managing the global epidemic: report of a WHO consultation on obesity. Geneva, Switzerland: World Health Organization; Heymsfield SB, Allison DB, Heshka S, Pierson RN Jr. Assessment of human body composition. In: Allison DB, ed. Handbook of Assessment Methods for Eating Behaviors and Weight-Related Problems. Thousand Oaks, CA: SAGE Publications, Inc.; 1995, pp Allison DB, Fontaine KR, Manson J, et al. How many deaths are attributable to obesity in the United States? JAMA. 1999;282: Michels KB, Greenland S, Rosner BA. Does body mass index adequately capture the relation of body composition and body size to health outcomes? Am J Epidemiol. 1998;147: Baumgartner RN, Heymsfield SB, Roche AF. Human body composition and the epidemiology of chronic disease. Obes Res. 1995;3: U.S. Department of Health and Human Service. The Practical Guide: Identification, Evaluation, and Treatment of Overweight and Obesity in Adults. NIH Publication No , October 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: Björntorp P. Visceral obesity: a civilization syndrome. Obes Res. 1993;1: Stevens J, Keil JE, Rust PF, et al. Body mass index and body girths as predictors of mortality in black and white women. Arch Intern Med. 1992;152: Folsom AR, Kaye SA, Sellers TA, et al. Body fat distribution and 5-year risk of death in old women. JAMA. 1993;269: Larsson B, Svardsudd K, Welin L, et al. Abdominal adipose tissue distribution, obesity, and risk of cardiovascular disease and death: 13 year follow up of participants in the study of men born in Br Med J. 1984;288: Vague J, Vague PH, Jubelin J, et al. Fat distribution, obesities and health: evolution of concepts. In: Bouchard C, Johnston FE, eds. Fat Distribution during Growth and Later Health Outcomes. New York: Alan R. Liss; 1988, pp Despres JP, Moorjani S, Lupien PJ, et al. Regional distribution of body fat, plasma lipoproteins, and cardiovascular disease. Arteriosclerosis. 1990;10: Reeder BA, Senthilselvan A, Despres JP, et al. The association of cardiovascular disease risk factors with abdominal obesity in Canada. CMAJ. 1997;157(suppl):S Han TS, Leer EM, Seidell JC, Lean MEJ. Waist circumference action levels in the identification of cardiovascular risk factors: prevalence study in a random sample. BMJ. 1995;311: 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 Card. 1994;73: Ross R, Leger L, Morris D, et al. Quantification of adipose tissue by MRI: relationship with anthropometric variables. J Applied Physiol. 1992;72: Lean MEJ, Han TS, Morrison CE. Waist circumference as a measure for indicating need for weight management. BMJ. 1995;311: Onat A. Waist circumference and waist-to-hip in Turkish adults: interrelation with other risk factors and association with cardiovascular disease. Int J Cardiol. 1999;70: Lean MEJ, Han TS, Deurenberg P. Predicting body composition by densitometry from simple anthropometric measurements. Am J Clin Nutr. 1996;63: Janssen I, Katzmarzyk PT, Ross R. Body mass index, waist circumference, and health risk: evidence in support of current NIH guidelines. Arch Intern Med. 2002;162: Janssen I, Katzmarzyk PT, Ross R. Waist circumference and not body mass index explains obesity-related health risk. Am J Clin Nutr. 2004;79: National Center for Health Statistics. The Third National Health and Nutrition Examination Survey, : Plan and Operations Procedures Manuals. Hyattsville, MD: Center for Disease Control and Prevention; Lohman TG, Roche AF, Martorell R (eds). Anthropometric Standardization Reference Manual. Champaign, IL: Human Kinetics Book; Centers for Disease Control and Prevention. The Third National Health and Nutrition Examination Survey (NHANES III ) Reference Manuals and Reports. Bethesda, MD: National Center for Health Statistics; Zhu SK, Wang Z, Heshka S, et al. Waist circumference and obesity-associated risk factors among whites in the Third National Health and Nutrition Examination Survey: clinical action thresholds. Am J Clin Nutr. 2002;76: OBESITY RESEARCH Vol. 12 No. 4 April 2004

11 27. Korn EL, Graubard BI. Analysis of Health Surveys. New York: John Wiley & Sons, Inc.; Stevens J, Cai J, Pamuk ER, et al. The effect of age on the association between body-mass index and mortality. N Engl J Med. 1998;338: Calle EE, Thun MJ, Peterlli JM, et al. Body-mass index and mortality in a prospective cohort of U.S. adults. N Engl J Med. 1999;341: Last JM. A Dictionary of Epidemiology. 4th ed. New York: Oxford University Press, Inc.; 2001, p Zweig MH, Campbell G. Receiver-operating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine. Clin Chem. 1993;39: Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults. Executive summary of the third report of the National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III). JAMA. 2001;285: Seidell JC, Deurenberg I. Obesity in Europe: prevalence and consequences for use of medical care. PharmacoEconomics. 1994;5(Suppl 1): Frank B, Manson JE, Stampfer MJ, et al. Diet, lifestyle, and the risk of type 2 diabetes mellitus in women. N Engl J Med. 2001;345: Kuczmarski RJ, Carrol MD, Flegal KM, Troiano RP. Varying body mass index cut-off points to describe overweight prevalence among U.S. adults: NHANES III (1988 to 1994). Obes Res. 1997;5: Chan JM, Rimm EB, Colditz GA, et al. Obesity, fat distribution, and weight gain as risk factors for clinical diabetes in men. Diabetes Care. 1994;17: Toth MJ, Tchernof A, Stes CK, Poehlman ET. Menopausal-related changes in body fat distribution. Ann NY Acad Sci. 2000; Han TS, Bijnen FCH, Lean MEJ, Seidell JC. (1998). Separate associations of waist and hip circumference with lifestyle factors. Int J Epidemiol. 1998;27: OBESITY RESEARCH Vol. 12 No. 4 April

12 Appendix 1. Optimal scaling of BMI and WC corresponding to BMI ranges of 16 to 40 kg/m 2 in relation to metabolic risk factors Men Women BMI (kg/m 2 ) OR O BMI WC * corresponding to OR O BMI WC corresponding to 95% OR CI OR O BMI WC corresponding to OR O BMI WC corresponding to 95% OR CI *O BMI WC 0.68BMI 0.32WC. O BMI WC WC. 644 OBESITY RESEARCH Vol. 12 No. 4 April 2004

13 Appendix 2: Nomogram for optimal scaling combination of BMI and WC for men. OBESITY RESEARCH Vol. 12 No. 4 April