Body mass index and obesity-related metabolic disorders in
Taiwanese and US whites and blacks: implications for definitions of
overweight and obesity for Asians
1–3Wen-Harn Pan, Katherine M Flegal, Hsing-Yi Chang, Wen-Ting Yeh, Chih-Jung Yeh, and Wen-Chung Lee ABSTRACT
Background: Recommendations based on scanty data have been made to lower the body mass index (BMI; in kg/m2
) cutoff for obesity in Asians.
Objective: The goal was to compare relations between BMI and metabolic comorbidity among Asians and US whites and blacks. Methods: We compared the prevalence rate, sensitivity, specific-ity, predictive values, and impact fraction of comorbidities at each BMI level and the BMI-comorbidity relations across ethnic groups by using data from the third National Health and Nutrition Exam-ination Survey and the Nutrition and Health Survey in Taiwan (1993–1996).
Results: For most BMI values, the prevalences of hypertension, diabetes, and hyperuricemia were higher for Taiwanese than for US whites. In addition, increments of BMI corresponded to higher odds ratios in Taiwanese than in US whites for hypertriglyceride-mia (P⫽ 0.01) and hypertension (P ⫽ 0.075). BMI-comorbidity relations were stronger in Taiwanese than in US blacks for all comorbidities studied. BMIs of 22.5, 26, and 27.5 were the cutoffs with the highest sum of positive and negative predictive value for Taiwanese, US white, and US black men, respectively. The same order was observed for women. For BMIs⬎27, ⬎85% of Tai-wanese, 66% of whites, and 55% of blacks had at least one of the studied comorbidities. However, a cutoff close to the median of the studied population was often found by maximizing sensitivity and specificity. Reducing BMI from⬎25 to ⬍25 in persons in the United States could eliminate 13% of the obesity comorbidity studied. The corresponding cutoff in Taiwan is slightly⬍24. Conclusion: These data suggest a possible need to set lower BMI cutoffs for Asians, but where to draw the line is a complex issue. Am J Clin Nutr 2004;79:31–9.
KEY WORDS BMI, definitions, obesity, overweight, ethnic-ity, Asians, diabetes mellitus, hypertension, hyperuricemia, dys-lipidemia
INTRODUCTION
Obesity has become an increasing public health problem internationally (1). Attention has been given to the adverse health consequences of a moderate increase in body mass index (BMI) in different ethnic groups (2). For example, it was suggested that a moderate increase in BMI makes South Asians more prone to insulin resistance and related diseases (3–7).
In early 1999, a recommendation was made to set criteria for Asians defining overweight as a BMI (in kg/m2) ofⱖ23 and obesity as a BMI ofⱖ 25 on the basis of the scanty data for Asians (8) that were available at that time. Taiwan recently adopted BMIs of 24 and 27 as the cutoffs for overweight and obesity, respectively. Whether to have separate definitions of overweight and obesity for minorities in multiethnic countries such as the United States is a related and even more compli-cated issue. Comparative studies are needed to delineate the differences in BMI-disease relations between minority race ethnic groups and whites, because the World Health Organi-zation definitions for overweight and obesity were established by using data primarily from whites. In particular, the ration-ales and the implications for selecting BMI cutoffs to define overweight and obesity for nonwhite populations should be carefully laid out.
The objective of the present study was to compare the relation of BMI to several obesity-related metabolic disorders in different ethnic groups. The data for Taiwanese Asians came from the Nutrition and Health Survey in Taiwan (NAHSIT), which was carried out during 1993–1996 (9). The data for US non-Hispanic whites and non-Hispanic blacks came from the third National Health and Nutrition Examination Survey (NHANES III), which was carried out in the United States during 1988–1994. Issues on how to define overweight and obesity for Asians and across ethnic groups were examined by using information on the relative risk, sensitivity, specificity, predictive values, and impact fraction for the selected meta-bolic disorders at different BMI cutoffs.
1From the Division of Epidemiology and Public Health, Institute of Biomedical Sciences, Academia Sinica, Taipei, Taiwan (W-HP, W-TY, and C-JY); the Division of Nutritional Sciences, Institute of Agricultural Chemistry, National Taiwan University, Taipei, Taiwan (W-HP); the Grad-uate Institute of Epidemiology, College of Public Health, National Taiwan University, Taipei, Taiwan (W-HP, C-JY, and W-CL); the National Center for Health Statistics, Centers for Disease Control and Prevention, Hyatts-ville, MD (KMF); and the Division of Health Policy, National Health Research Institute, Taipei, Taiwan (H-YC).
2Supported by grants DOH-83-FS-41, DOH-84-FS-11, DOH-85-FS-11, and DOH-86-FS-11 from the Department of Health, Republic of China.
3Address reprint requests to W-H Pan, Institute of Biomedical Sciences, Academia Sinica, Taipei 115-29, Taiwan. E-mail: pan@ibms.sinica.edu.tw.
Received October 7, 2002.
Accepted for publication June 27, 2003.
31 Am J Clin Nutr 2004;79:31–9. Printed in USA. © 2004 American Society for Clinical Nutrition
at National Taiwan Univ. Hospital on April 21, 2009
www.ajcn.org
SUBJECTS AND METHODS
Nutrition and Health Survey in Taiwan, 1993–1996
NAHSIT was an island-wide survey in Taiwan of noninsti-tutionalized residents agedⱖ4 y selected through a multistage, complex sampling design. Details on the design and operation of the survey have been published elsewhere (9). A sample of 9961 participants was obtained through door-to-door inter-views, which corresponds to a 74% response rate. Of these, 4956 were adults aged ⱖ20 y, of whom 52.2% (2585) had complete clinical data of interest for this study. Blood pressure was measured by trained technicians using standard sphygmo-manometers at each participant’s home after the participant had rested forⱖ5 min. An appropriately sized cuff was used for each participant. Two blood pressure measurements were made 30 s apart with the arm at the level of the heart. If the 2 measurements were⬎10 mm Hg apart, a third measurement was made. The 2 closest blood pressure values were averaged to obtain mean blood pressure. Other clinical data were ob-tained from a physical examination carried out in temporary clinics set up in the neighborhood of the survey sites.
Anthropometric measurements were carried out after the subjects had removed their shoes and heavy clothes. The sub-jects were asked to wear an examination gown if their apparel was not appropriate for taking measurements. Body weight was measured to the nearest 0.1 kg, and body height was measured to the nearest 1 mm. BMI was calculated from measured weight and height as weight (kg)/height (m2). Fasting whole blood glucose was measured immediately after the blood draw-ing by the glucose oxidase method with use of a glucose analyzer (model 23A; YSI, Yellow Springs, OH) in blood samples collected into sodium fluoride tubes. Fasting serum uric acid, triacylglycerol, and cholesterol values were measured with a Hitachi 747 analyzer (Hitachi, Tokyo) within 1 mo of the blood draw in collaboration with the clinical chemistry laboratory of the affiliated hospital of National Taiwan Uni-versity Medical School.
Third National Health and Nutrition Examination Survey
NHANES III (1988–1994) is the seventh in a series of health examination surveys conducted by the National Center for Health Statistics of the Centers for Disease Control and Pre-vention. With the use of a complex, multistage sampling de-sign, the survey aimed to provide estimates of the health and nutritional status of the civilian noninstitutionalized population of the United States agedⱖ2 mo. Details of the survey design and questionnaires are published in the NHANES III Plan and
Operation reference manual (10). A complete description of
the survey and laboratory procedures is also available (11). In this articles, the terms white and black will be used to refer to non-Hispanic whites and non-Hispanic blacks, respectively.
Definitions for selected metabolic disorders
The same definitions were applied to data collected at base-line in both surveys. Hypertension was defined as use of an antihypertensive medication, having a systolic blood pressure
ⱖ140 mm Hg, or having a diastolic blood pressure ⱖ90 mm
Hg. Diabetes mellitus was defined as use of insulin or
hypo-glycemic agents, having a fasting plasma glucose concentration
ⱖ7.0 mmol/L (NHANES III), or having a fasting whole blood
glucose concentrationⱖ6.1 mmol/L (NAHSIT) (12). Hyper-cholesterolemia was defined as use of cholesterol-lowering medication or having a serum cholesterol concentrationⱖ6.21 mmol/L. Hypertriglyceridemia was defined as a serum triacyl-glycerol concentrationⱖ2.26 mmol/L. Hyperuricemia was de-fined as a serum uric acid concentration ⱖ392.6 mol/L (women) orⱖ458mol/L (men).
Statistical analysis
All analyses were carried out with SUDAAN version 7.5 (Research Triangle Institute, Research Triangle Park, NC), which was used to account for the effect of the complex sampling design in statistical testing and in deriving means and SEs. Missing values were handled separately in the analysis for each metabolic disorder. The sample sizes are listed in Table 1. For comparison across populations by BMI, the survey respon-dents were grouped into 6 BMI categories: 16–19.9, 20–21.9, 22–23.9, 24–25.9, 26–29.9, and 30–39.9; BMI values⬍16 or
ⱖ40 were excluded. For other analyses, all BMI data were
included. In comparing the prevalence of hypertension, diabe-tes, hyperuricemia, and dyslipidemia between populations, the data were standardized by age and sex to the 1980 US popu-lation, and design effect–adjusted SEs were computed. Two-sample t tests were then used to compare the rates either for overall comparison or for comparison at each BMI group. In comparing the distribution of age, sex, and BMI across popu-lations, a chi-square test was used. P valuesⱕ 0.05 were used to indicate statistical significance in general, but a Bonferroni-corrected P value of 0.017 was used for the situation of 3 pairwise multiple comparisons.
Logistic regression was used to compare the prevalence rate ratios (odds ratios) per increment of BMI for each comorbidity outcome across racial groups. NHANES and NAHSIT data were pooled in the analysis, with race, sex, age, BMI, and BMI
⫻ race as the independent variables. We assumed that all
primary sampling units were “sampled without replacement.” However, the primary sampling units of NHANES were sam-pled with replacement. We therefore made the selection prob-ability of the NHANES primary sampling units ⬍0.05 to mimic the original sampling process.
Sensitivity, specificity, and positive and negative predictive values (13) relative to each of the selected conditions (hyper-tension, diabetes, hypercholesterolemia, hypertriglyceridemia, and hyperuricemia) were calculated every one-half unit of BMI from 20 to 30. A t test was used to compare the difference in predictive values as described previously.
The impact fraction was calculated by using data on the BMI distribution and on the risk ratios for each population. Logistic regression models that included linear and quadratic terms for BMI as a continuous variable were used to generate monotonic risk ratio estimates. The impact fraction (14, 15) for a selected cutoff was calculated as the excess prevalence of the condition associated with BMIs greater than the selected cutoff divided by the total prevalence of the condition. This is an estimate of the fraction of cases that could be removed if persons with BMI values above the cutoff were shifted to below the cutoff. In the impact fraction calculation, a subject was considered as a case if he or she had one or more of the selected metabolic conditions. In calculating sensitivity, specificity, predictive
at National Taiwan Univ. Hospital on April 21, 2009
www.ajcn.org
values, and the impact fraction of having at least one disorder, data with missing values for any variable were deleted.
RESULTS
Sample size and characteristics of the studied populations
The Taiwanese population studied included a lower percent-age of older persons than did the US white population studied (Table 1). The Taiwanese population had lower proportions of persons in the high-BMI groups than did both the US white and the US black populations, and the US white population had lower proportions than did the US black population. Taiwanese and US whites had lower crude prevalence rates of hyperten-sion and diabetes mellitus than did US blacks. The prevalence of hypercholesterolemia was lower in the Taiwanese than in either the US whites or the US blacks. However, the prevalence of hyperuricemia was much higher in the Taiwanese than in either the US whites or the US blacks.
Age- and sex-standardized prevalence estimates by BMI
For most of the high-BMI groups, the age- and sex-standard-ized prevalence rates of hypertension and hyperuricemia for Taiwanese were higher than those for US whites (Figure 1). The higher the BMI value, the greater the prevalence differ-ence. In the BMI range of 30–40, the differences between the Taiwanese and the other groups in the prevalences of hyper-tension and hyperuricemia were ⬎20 percentage points. In low-BMI groups, the Taiwanese had lower prevalences of hypertension than did the US blacks, but the prevalence rates increased faster with BMI for the Taiwanese than for the US blacks.
The prevalence of hypercholesterolemia was in general lower for the Taiwanese than for the US whites and the US blacks in most BMI groups. The Taiwanese and the US whites had similar prevalences of hypertriglyceridemia in every BMI group, whereas prevalences were lower for the US blacks. In most BMI groups, the age- and sex-standardized rates of
dia-TABLE 1
Sample size and weighted proportions of selected characteristics for Taiwanese, US whites, and US blacks1
Variable Taiwanese US whites US blacks
Sample size (n) 3047 6706 4542 Age [n (%)] 20–44 y 1118 (64.4) 2375 (53.2) 2697 (64.5) 45–64 y 1292 (24.4) 1796 (27.6) 1080 (22.8) ⱖ65 y 637 (11.2) 2535 (19.2) 765 (12.6) Chi-square test a b a Sex [n (%)] Men 1432 (51.0) 3152 (48.6) 2091 (45.1) Women 1615 (49.0) 3554 (51.4) 2451 (54.9) Chi-square test a a b BMI, in kg/m2[n (%)] ⬍16 22 (0.9) 16 (0.12) 13 (0.2) 16–19.9 414 (18.0) 493 (7.8) 326 (7.0) 20–21.9 527 (22.5) 837 (13.2) 447 (9.7) 22–22.9 322 (13.0) 503 (8.0) 293 (6.7) 23–23.9 351 (12.1) 534 (8.2) 292 (7.0) 24–25.9 603 (16.8) 1169 (17.4) 638 (14.6) 26–29.9 613 (12.7) 1738 (24.2) 1150 (25.4) 30–39.9 193 (3.8) 1269 (18.7) 1139 (24.0) ⱖ40 2 (0.2) 147 (2.5) 244 (5.2) Chi-square test c b a Median Men 22.8 26.0 25.8 Women 22.4 24.6 27.6 Prevalence [n (%)] Hypertension2 1070/30173(23.3)b 2450/6696 (25.0)b 1546/4536 (30.7)a Hypercholesterolemia4 415/2823 (11.6)c 1728/6483 (22.4)a 815/4160 (18.1)b Hypertriglyceridemia5 394/2617 (10.9)b 686/3812 (16.8)a 198/2594 (7.0)c Hyperuricemia6 750/2820 (21.2)a 685/6385 (9.2)c 538/4091 (12.2)b Diabetes7 290/2665 (5.7)b 278/2981 (6.6)b 214/1902 (9.6)a
1Proportions are weighted to the target populations that the Taiwanese, US white, and US black samples originated from. A chi-square test was used to compare distributions of age, sex, BMI, and disease prevalence across populations. Values with different letters are significantly different, P⫽ 0.017 after Bonferroni correction. For disease prevalence, a, b, and c represent significant differences in descending order.
2Blood pressureⱖ140 mm Hg/90 mm Hg or taking antihypertensive medication. 3Number of cases/total number of individuals with complete data.
4Fasting blood cholesterolⱖ6.21 mmol/L or taking lipid-lowering medication. 5Fasting blood triglyceridemiaⱖ2.26 mmol/L or taking lipid-lowering medication.
6Fasting blood uric acidⱖ458mol/L for men and ⱖ392.6 mol/L for women or taking gout medication. 7Fasting whole blood glucoseⱖ6.1 mmol/L, fasting plasma glucose ⱖ 7.0 mmol/L, or taking diabetes medication.
at National Taiwan Univ. Hospital on April 21, 2009
www.ajcn.org
betes were similar for the Taiwanese and the US blacks but were lower for the US whites.
Odds ratios per unit change in BMI for selected metabolic conditions
The odds ratios for each of the selected conditions were calculated for each increment of BMI (Table 2). Increments of BMI corresponded to significantly higher odds ratios in the Taiwanese than in the US blacks for all 5 metabolic conditions studied. When the Taiwanese and the US whites were com-pared, significance and borderline significance were shown for hypertriglyceridemia and hypertension, respectively.
Sensitivity and specificity of BMI cutoffs in predicting selected metabolic disorders
Shown in Table 3 are the BMI cutoffs corresponding with the maximum sum of sensitivity and specificity for screening
subjects with the studied condition. For the Taiwanese, the best cutoffs were BMIs between 23.0 and 23.8. For the US whites (25.5–27.9) and the US blacks (26.3–29.0), the values for these cutoffs were much higher. The order of the cutoffs corresponded with that of the median BMI in the Taiwanese (women: 22.4; men: 22.8), the US whites (women: 24.6; men: 26.0), and the US blacks (women: 27.6; men: 25.8). These BMI cutoffs for screen-ing persons havscreen-ing at least one of the studied conditions in Taiwanese (women: 22.6; men: 22.8), US whites (women: 24.9; men: 26.0), and US blacks (women: 28.0; men: 25.9) coincided closely with the median BMI value for the group.
Positive and negative predictive values at various BMI cutoffs
As the BMI cutoffs increased, positive predictive value increased and negative predictive value decreased gradually for
FIGURE 1. Age- and sex-standardized prevalences of various obesity-related comorbidities by BMI and ethnic group (F, Taiwanese;䡲, US whites; E, US blacks).*Significant difference between US whites and Taiwanese, P⬍ 0.017 (after Bonferroni correction).#Significant difference between US blacks and Taiwanese, P⬍ 0.017 (after Bonferroni correction).
at National Taiwan Univ. Hospital on April 21, 2009
www.ajcn.org
all 3 ethnic groups (Figure 2). For both men and women, for most BMI cutoffs, the positive predictive value for identifying subjects who had at least one of the selected conditions was significantly higher for the Taiwanese than for the US whites or the US blacks. At low BMI values, the negative predictive value tended to be lowest for the US blacks, both male and female, but no significant difference was detected. At higher BMI values, the negative predictive value was lower for the Taiwanese than for the US whites and the US blacks, with significant differences in males only. The BMI cutoffs at which positive and negative predictive value were equal to each other were⬇22.5, 26.0, and 27.5 for Taiwanese, US white, and US black men, respectively. For Taiwanese, US white, and US
black women, these BMI cutoffs were⬇24.8, 28.5, and some-where⬎30, respectively.
Impact fraction for lowering BMIs from above cutoffs to below cutoffs
The impact fraction is plotted against chosen BMI cutoffs in
Figure 3. At high BMI cutoffs, the impact fraction was much
smaller for the Taiwanese than for the US whites. If the BMI cutoff were lowered further, however, the relative impact on disease prevention would become progressively greater for the Taiwanese. At a BMI of 21.5, the impact fraction for the Taiwanese reached the level of that for the US blacks. At no BMI in the figure did the impact fraction for the Taiwanese
TABLE 2
Odds ratios (95% CI) per unit change in BMI for hypertension, diabetes, hypercholesterolemia, hypertriglyceridemia, and hyperuricemia1
Metabolic disorder Taiwanese US whites US blacks
Hypertension 1.24 (1.11, 1.39) 1.11 (1.09, 1.14) 1.07 (1.05, 1.09) P — 0.07512 0.01153 Diabetes mellitus 1.13 (1.09, 1.17) 1.15 (1.13, 1.17) 1.08 (1.05, 1.10) P — 0.4439 0.0305 Hypercholesterolemia 1.10 (1.05, 1.15) 1.05 (1.04, 1.07) 1.02 (1.00, 1.04) P — 0.1111 0.0067 Hypertriglyceridemia 1.20 (1.14, 1.26) 1.11 (1.09, 1.14) 1.05 (1.03, 1.07) P — 0.0100 0.0001 Hyperuricemia 1.16 (1.12, 1.19) 1.13 (1.10, 1.15) 1.09 (1.06, 1.11) P — 0.1336 0.0006
1A logistic regression model was used to estimate the odds ratios for BMI-disease relations in each race group (race, sex, age, and race⫻ BMI) and to examine the differences between racial groups on the magnitude of the relations (race, sex, age, BMI, and race⫻ BMI). BMI and age were continuous variables, and the others were dummy variables.
2P value corresponding to the comparison between theˆ of US whites and that of Taiwanese. 3P value corresponding to the comparison between theˆ of US blacks and that of Taiwanese.
TABLE 3
BMI cutoffs (in kg/m2) balancing sensitivity (Se) and specificity (Sp) in predicting hypertension, hyperuricemia, diabetes, hypertriglyceridemia, and hypercholesterolemia1
Men Women
Se Sp Balancing point Se Sp Balancing point
Taiwanese (n⫽ 1202 M, 1365 F) Hypertension 0.63 0.6 23.3 0.68 0.70 23.7 Hyperuricemia 0.61 0.62 23.3 0.59 0.58 23.0 Diabetes 0.62 0.63 23.7 0.66 0.65 23.8 Hypertriglyceridemia 0.62 0.65 23.5 0.61 0.61 23.4 Hypercholesterolemia 0.55 0.53 23.0 0.60 0.58 23.0 Any disorder 0.67 0.68 22.8 0.63 0.64 22.6 US whites (n⫽ 1330 M, 1507 F) Hypertension 0.60 0.62 26.6 0.64 0.63 25.6 Hyperuricemia 0.64 0.62 26.9 0.67 0.67 26.9 Diabetes 0.67 0.67 27.6 0.74 0.72 27.9 Hypertriglyceridemia 0.66 0.64 26.7 0.68 0.67 26.3 Hypercholesterolemia 0.57 0.57 26.4 0.61 0.61 25.5 Any disorder 0.64 0.64 26.0 0.67 0.66 24.9 US blacks (n⫽ 798 M, 978 F) Hypertension 0.59 0.59 26.3 0.6 0.60 28.2 Hyperuricemia 0.66 0.67 27.3 0.65 0.60 29.0 Diabetes 0.62 0.60 26.8 0.65 0.58 28.9 Hypertriglyceridemia 0.61 0.61 26.8 0.55 0.54 28.1 Hypercholesterolemia 0.62 0.60 26.6 0.57 0.56 28.4 Any disorder 0.61 0.61 25.9 0.58 0.58 28.0
1The sample sizes listed are for the analysis combining all disorders. The sample sizes for the individual disorders are the same as those in Table 1.
at National Taiwan Univ. Hospital on April 21, 2009
www.ajcn.org
surpass that for the US whites. At a BMI of 25, the World Health Organization cutoff for overweight, the impact fraction for whites was⬇13%. The corresponding BMI cutoff in the Taiwanese for a 13% impact fraction was slightly⬍24. DISCUSSION
We found that the Taiwanese had much higher age-standard-ized prevalences of hypertension, hyperuricemia, and diabetes than did the US whites at most BMI values. In addition, increments of BMI corresponded with higher odds ratios in the Taiwanese than in the US whites for hypertriglyceridemia, hypertension, hypercholesterolemia, and hyperuricemia, al-though differences in the last 2 were not significant. Signifi-cantly stronger BMI-comorbidity relations were also shown in the Taiwanese than in the US blacks for all of the metabolic disorders studied.
It has been known for some time that, after control for BMI, Asians have a higher prevalence of hypertension than do whites (16, 17), which indicates a greater effect of factors other than
obesity, either genetic or environmental. An extraordinarily high prevalence of hyperuricemia in Taiwan has only recently emerged (9, 18, 19). The extent of the change in the prevalence of hyperuricemia indicates the interplay of genes and environ-ment. The high incidence (20, 21), prevalence (22–26), and mortality (27) of diabetes mellitus and undiagnosed diabetes mellitus (28) have been well documented in recent years in Taiwanese. Short statue and central obesity (28) are among the risk factors. Consistent with the above literature, our results indicate that Taiwanese (or Asians) may respond more severely in either an absolute or a relative sense to the same degree of BMI elevation than do US whites or US blacks regarding the component disorders of the metabolic syndrome.
Studies have shown that given the same level of BMI, Chinese and many other Asian groups have a higher percentage of body fat than do whites (29, 30). Intrauterine malnutrition (31, 32), recent lifestyle associated with low physical activity and a low-fiber diet (33), and genetic differences are among the potential etiologic factors of this phenomenon. The issue of
FIGURE 2. Positive (PV⫹) and negative (PV-) predictive value for the presence of one or more of the selected conditions at a given BMI in Taiwanese (T), US whites (W), and US blacks (B).*Significant difference between the groups listed at the left in the table below each graph at the BMI indicated,
␣ ⫽ 0.017 after Bonferroni correction.
at National Taiwan Univ. Hospital on April 21, 2009
www.ajcn.org
how to define overweight and obesity in Asians and minority groups is not easily resolved by examining the relations between BMI and related disorders. In our study and in those of others (8, 34), there was no clear threshold in the BMI-comorbidity relations that would readily allow selection of a BMI cutoff. It is most likely that body fat, in interaction with age, sex, and other risk factors, increases the risk of metabolic disorders. We should know more about these interactions to simplify the strategy of defining obesity.
A few studies have used the approach of maximizing the sum of sensitivity and specificity. Ko et al (8) suggested using a BMI of 23 to define overweight in screening for diabetes, hypertension, dyslipidemia, and albuminuria. However, we found that the BMI distributions of the diseased and the healthy in our populations overlapped considerably (data not shown). Therefore, depending on the population studied, this approach will pick up a value close to the BMI median of that given population and designate around one-half of the population as overweight. It seems that the concept of balancing sensitivity and specificity in this setting of complex etiology is not as straightforward and applicable as in the case of clinical diag-nostics.
Positive predictive value is a function of the prevalence of a given disorder. For any single metabolic disorder in the present study, and particularly for young adults, the positive predictive
value was relatively low (data not shown). However, when several disorders were examined together, the positive predic-tive values for having one or more become reasonably good. Of every 10 subjects defined as overweight or obese by having a BMIⱖ24, 6-7 had one or more of the studied conditions. On the other hand, of every 10 subjects defined as not overweight by having a BMI⬍24, 3-4 still had one of the conditions. In general, the performance of BMI in screening the studied disorders ranked the best for the Taiwanese, next best for the US whites, and the poorest for the US blacks if one screened by using the BMI cutoffs that balanced positive and negative predictive value. Around 69% of Taiwanese and 63–65% of US whites could be accurately classified as having or not having the disorders included in this study. However, estimates of predictive values do not provide absolute guidance as to whether BMI cutoffs should be lower for Asians and, if so, how much lower. The choice of the best cutoff depends in part on the prevalence and the cost of the selected conditions, which may vary over time and with the kind of disorders studied. In addition, whether the best cutoff is the one with equivalent proportions of false positive and false negative values is also a complex issue.
The impact fraction of any given BMI corresponds to the percentage of persons that will be removed from the original diseased population if their BMIs are lowered to below the
FIGURE 3. Impact fraction for eradicating BMI values above cutoffs for the presence of one or more of the selected conditions at a given BMI in Taiwanese (F), US whites (䡲), and US blacks (E).
at National Taiwan Univ. Hospital on April 21, 2009
www.ajcn.org
cutoff. Because the percentage of persons with a high BMI is much less in Taiwan than in the United States, the impact fraction is also much less. A BMI cutoff slightly ⬍24 is required to achieve the same impact fraction (13%) that US whites can obtain at a BMI of 25. Nonetheless, one needs to know more about the cost-effectiveness of BMI reduction to truly understand its impact.
One major limitation of the present study is that the compa-rability of 2 surveys is not easy to assess, although each followed stringent protocols. In addition, we did not address the issue of confounding factors such as smoking and drinking habits, physical activity, socioeconomic status, and meno-pausal status, because these issues are not only complex but also may not be satisfactorily addressed as a result of the differences in culture and questionnaire designs. However, the effect of these related systematic errors, if any, should not affect the results of relative risk per unit change of BMI, which was based on comparisons within surveys.
In conclusion, the results for odds ratios, predictive value, and impact fraction tend to support the use of lower BMI cutoffs for Taiwanese (or Asians). However, there is no easy way to determine how low the cutoffs should be. The BMI with equivalent positive and negative predictive value may be a realistic cutoff for defining overweight. After all, the major purpose of a cutoff is for screening. However, the impact (or cost) of false-positive and false-negative screening should be weighed when predictive value data are used. Another possi-bility is to take a pragmatic approach and consider the propor-tion of the populapropor-tion for which a country (or a society) can afford interventions. In Taiwan, one-quarter of the population has a BMI⬎25, one-third has a BMI ⬎24, and one-half has a BMI ⬎23. Consideration of national resources and of any possible negative impacts can help to determine which BMI cutoff to choose in terms of defining overweight. At the same time, a population approach to reduce overall BMI and waist circumference can be simultaneously implemented. Careful examination of the relations between body fat accumulation and distribution and BMI across ethnic groups is essential for disentangling this problem of defining obesity.
W-H Pan initiated and directed the study and was the main writer of the manuscript. KMF contributed to the analysis of the NHANES III data and the writing of the manuscript. H-YC, a biostatistitian, contributed to the analysis of the NAHSIT data and the selection of statistical methods for comparison between surveys. W-TY was extensively involved in the data collection for NAHSIT and carried out the data analyses. C-JY and W-CL contributed to the part of the article on the impact fraction.
REFERENCES
1. Obesity: preventing and managing the global epidemic. Report of a WHO consultation. World Health Organ Tech Rep Ser 2000;894:1– 253.
2. Kumanyika SK. Special issues regarding obesity in minority popula-tions. Ann Intern Med 1993;119:650 – 4.
3. Banerji MA, Faridi N, Atluri R, Chaiken RL, Lebovitz HE. Body composition, visceral fat, leptin, and insulin resistance in Asian Indian men. J Clin Endocrinol Metab 1999;84:137– 44.
4. Chandalia M, Abate N, Garg A, Stray-Gundersen J, Grundy SM. Relationship between generalized and upper body obesity to insulin resistance in Asian Indian men. J Clin Endocrinol Metab 1999;84: 2329 –35.
5. Knight TM, Smith Z, Whittles A, et al. Insulin resistance, diabetes, and risk markers for ischaemic heart disease in Asian men and non-Asian in Bradford. Br Heart J 1992;67:343–50.
6. Laws A, Jeppesen JL, Maheux PC, Schaaf P, Chen YD, Reaven GM. Resistance to insulin-stimulated glucose uptake and dyslipidemia in Asian Indians. Arterioscler Thromb 1994;14:917–22.
7. Marita AR, Desai A, Mokal R, Agarkar RY, Dalal KP. Association of insulin resistance to electrocardiographic changes in non obese Asian Indian subjects with hypertension. Endocr Res 1998;24:215–33. 8. Ko GT, Chan JC, Cockram CS, Woo J. Prediction of hypertension,
diabetes, dyslipidaemia or albuminuria using simple anthropometric indexes in Hong Kong Chinese. Int J Obes Relat Metab Disord 1999;23:1136 – 42.
9. Chang H, Pan W, Yeh W, Tsai K. Hyperuricemia and gout in Taiwan: results from the Nutritional and Health Survey in Taiwan (NAHSIT: 1993–96). J Rheumatol 2001;28:1640 – 6.
10. National Center for Health Statistics. Plan and operation of the Third National Health and Nutrition Examination Survey, 1988-94. Series 1: programs and collection procedures. Vital Health Stat 1 1994;Jul(32): 1– 407.
11. US Department of Health and Human Services (DHHS), National Center for Health Statistics. The third National Health and Nutrition Examination Survey, 1988-94. Reference manuals and reports [CD-ROM]. Hyattsville, MD: Center for Disease Control and Prevention, 1997.
12. Alberti KG, Zimmet PZ. Definition, diagnosis and classification of diabetes mellitus and its complications. Part 1: diagnosis and classifi-cations of diabetes mellitus provisional report of a WHO consultation. Diabet Med 1998;15:539 –53.
13. Ikeda M, Ishigaki T, Yamauchi K. Review of the new methodology of analyzing receiver operating characteristic curves. Medinfo 1998;9: 837– 40.
14. Lee WC. Characterizing exposure-disease association in human pop-ulations using the Lorenz curve and Gini index. Stat Med 1997;16: 729 –39.
15. Lee WC. Analysis of seasonal data using the Lorenz curve and the associated Gini index. Int J Epidemiol 1996;25:426 –34.
16. Jones DW. Hypertension in East Asia. Am J Hypertens 1995;8:111s– 4s.
17. Baba S, Pan WH, Ueshima H, et al. Blood pressure levels, related factors, and hypertension control status of Japanese and Americans. J Hum Hypertens 1991;5:317–32.
18. Chou P, Soong LN, Lin HY. Community-based epidemiological study on hyperuricemia in Pu-Li, Taiwan. J Formos Med Assoc 1993;92: 597– 602.
19. Lin KC, Lin HY, Chou P. Community based epidemiological study on hyperuricemia and gout in Kin- Hu, Kinmen. J Rheumatol 2000;27: 1045–50.
20. Tseng CH, Chong CK, Heng LT, Tseng CP, Tai TY. The incidence of type 2 diabetes mellitus in Taiwan. Diabetes Res Clin Pract 2000; 50(suppl):S61– 4.
21. Wang SL, Pan WH, Hwu CM, et al. Incidence of NIDDM and the effects of gender, obesity and hyperinsulinaemia in Taiwan. Diabeto-logia 1997;40:1431– 8.
22. Chen KT, Chen CJ, Gregg EW, Engelgau MM, Venkat Narayan KM. Prevalence of type 2 diabetes mellitus in Taiwan: ethnic variation and risk factors. Diabetes Res Clin Pract 2001;51:59 – 66.
23. Chen KT, Chen CJ, Gregg EW, Williamson DF, Narayan KM. High prevalence of impaired fasting glucose and type 2 diabetes mellitus in Penghu Islets, Taiwan: evidence of a rapidly emerging epidemic? Diabetes Res Clin Pract 1999;44:59 – 69.
24. Lu FH, Yang YC, Wu JS, Wu CH, Chang CJ. A population-based study of the prevalence and associated factors of diabetes mellitus in southern Taiwan. Diabet Med 1998;15:564 –72.
25. Chen HD, Shaw CK, Tseng WP, Chen HI, Lee ML. Prevalence of diabetes mellitus and impaired glucose tolerance in Aborigines and Chinese in eastern Taiwan. Diabetes Res Clin Pract 1997;38:199 –205. 26. Chou P, Li CL, Kuo HS, Hsiao KJ, Tsai ST. Comparison of the prevalence in two diabetes surveys in Pu-Li, Taiwan, 1987–1988 and 1991–1992. Diabetes Res Clin Pract 1997;38:61–7.
27. Lin RS, Lee WC. Trends in mortality from diabetes mellitus in Taiwan, 1960 –1988. Diabetologia 1992;35:973–9.
28. Pan WH, Yeh WT, Hwu CM, Ho LT. Undiagnosed diabetes mellitus in Taiwanese subjects with impaired fasting glycemia: impact of female sex, central obesity, and short stature. Chin J Physiol 2001;44:44 –51. 29. Deurenberg-Yap M, Schmidt G, van Staveren WA, Deurenberg P. The
paradox of low body mass index and high body fat percentage among
at National Taiwan Univ. Hospital on April 21, 2009
www.ajcn.org
Chinese, Malays and Indians in Singapore. Int J Obes Relat Metab Disord 2000;24:1011–7.
30. Deurenberg P, Yap M, van Staveren WA. Body mass index and percent body fat: a meta analysis among different ethnic groups. Int J Obes Relat Metab Disord 1998;22:1164 –71.
31. Gray P, Crowther N. Implications of the thrifty phenotype hypothesis for the health of societies undergoing acculturation–lessons for South African health planning. S Afr Med J 2001;91:325– 8.
32. Aros S, Cassorla F. [Possible perinatal determinants of morbidity in adult age. ] Rev Med Chil 2001;129:307–15(in Spanish).
33. Kromhout D, Bloemberg B, Seidell JC, Nissinen A, Menotti A. Physical activity and dietary fiber determine population body fat levels: the Seven Countries Study. Int J Obes Relat Metab Disord 2001;25:301– 6. 34. Stevens J, Juhaeri, Cai J, Jones DW. The effect of decision rules on the
choice of a body mass index cutoff for obesity: examples from African American and white women. Am J Clin Nutr 2002;75:986 –92.
at National Taiwan Univ. Hospital on April 21, 2009
www.ajcn.org
