m

American Journal of Epidemiology Copyright O 1992 by The Johns Hopkins University School of Hygiene and Public Health Al rights reserved

Vol 136, No. 12 Printed in U.S.A.

Obesity and Body Fat Distribution in Relation to the Incidence of Non-lnsulin-dependent Diabetes Mellitus A Prospective Cohort Study of Men in the Normative Aging Study Patricia A. Cassano,12 Bernard Rosner,2 Pantel S. Vokonas,3 and Scott T. Weiss 24

The relation between the abdominal accumulation of body fat, total-body adiposity, and blood glucose level and the risk of non-insulin-dependent diabetes mellitus was evaluated prospectively among 1,972 male participants in the Department of Veterans Affairs Normative Aging Study cohort. The risk of non-insulin-dependent diabetes mellitus was assessed by means of the proportional hazards model; 226 cases of diabetes occurred among the 1,972 men (mean age at entry, 41.9 years; range, 22-80 years) over 35,496 person-years of observation. The relation of body mass index to diabetes risk was partly explained by body fat distribution; after adjusting for age, the ratio of abdominal circumference to hip breadth, and cigarette smoking, men in the top tertile for body mass index had a 1.3-fold greater risk of diabetes than did men in the lowest tertile (95% confidence interval 0.9-1.8). Moreover, after adjusting for age, body mass index, and cigarette smoking, men in the top tertile for the ratio of abdominal circumference to hip breadth had a 2.4-foW greater risk of diabetes than did men in the lowest tertile (95% confidence interval 1.7-3.7). When blood glucose was analyzed as a continuous outcome variable, the findings were consistent, i.e., a positive association with abdominal fat independent of total-body adiposity. These results confirm previous reports of a prospective relation between abdominal adiposity and the risk of diabetes and provide prospective evidence of a relation between Wood glucose levels and both body fat distribution and obesity. Am J Epidemiol 1992;136:1474-86. blood glucose; body composition; diabetes mellitus, non-insulin-dependent; obesity

Non-insulin-dependent diabetes mellitus (NIDDM) is an important cause of morbidity and mortality in the United States (1). Current information suggests that the pathogenesis of this condition involves the inter-

play of several factors. Studies of twins and of high-risk population subgroups such as the Pima Indians provide strong evidence for the importance of a genetic predisposition to susceptibility (2). Life-style factors,

Received for publication January 27, 1992, and in final form June 19, 1992 Abbreviations: LRSaa, likelihood ratio statistic with 2 degrees of freedom; NIDDM, non-msiiin-dependent diabetes mellitus. 1 Division of Nutritional Sciences, Cornell University, Ithaca, NY. 2 Channing Laboratory, Department of Medicine, Harvard Medical School and Brigham and Women's Hospital, Boston, MA. 3 The Normative Aging Study, Department of Veterans Affairs Outpatient Clinic, and Section of Preventive Medicine and Epidemiology and the Evans Memorial Department of Clinical Research, Department of Medicine, Boston University School of Medicine, Boston, MA. * Pulmonary Division, Department of Medicine, Harvard Medical School, and Beth Israel Hospital, Boston, MA.

Reprint requests to Dr S. T. Weiss, Channing Laboratory, 180 Longwood Avenue, Boston, MA 02115. Supported by grant HL37871 from the National Heart, Lung, and Blood Institute, National Institutes of Health (NBH), Bethesda, MD, and by the Medical Research Service, Department of Veterans Affairs This work was done in part while P A. C. was a Research Fellow in Medicine at the Channing Laboratory, Department of Medicine, Harvard Medical School and Bngham and Women's Hospital; she was supported in part by NIH institutional research award HL07427. The authors thank Marianne Mora and llene Brill for their assistance with computer programming. Also, the authors acknowledge Dr. Lewis Landsberg for his insights and critical input throughout this project and Dr Mark Segal for his statistical advice in the early phase of these analyses

1474 Downloaded from https://academic.oup.com/aje/article-abstract/136/12/1474/198582 by University of Durham user on 20 March 2018

Body Fat Distributor!, Blood Glucose, and NIDDM

including physical activity, cigarette smoking, stress, diet, and obesity, also may be important contributors to both blood glucose level and diabetes risk (2, 3). Some of these factors may involve a heritable component; for instance, a genetic propensity may contribute to both the degree of obesity and the distribution of body fat. It is hypothesized that, early in the natural history of diabetes, obesity plays a role in the elevation of levels of circulating insulin through stimulation of the pancreatic beta cells to increase insulin secretion and through desensitization of peripheral tissue to insulin action (2, 3). Body fat distribution also may be related to metabolic aberrations, including increased insulin resistance, which lead in turn to an increased risk of diabetes (2, 3). The net result is a hyperinsulinemic state that may be followed by exhaustion of the pancreatic beta cells, leading to a failure of insulin-mediated glucose uptake, a pathologic state characterized by elevated fasting levels of glucose and insulin in the blood, and eventually a decreased insulin response to glucose challenge (3). The total proportion of disease attributable to obesity is quite high, with estimates ranging up to 92 percent (4). In early studies by Vague (5), an "index of masculine differentiation" based on skinfold ratios (sum of cervical/sacral and triceps/thigh) suggested the importance of fat distribution in relation to diabetes, and the prevalence of diabetes was noted to increase with "android" body build among both men and women. Since then, several cross-sectional studies have assessed a positive association between fat distribution and glucose tolerance or diabetes prevalence (6-8). Total-body adiposity (represented by body mass index or relative weight), central or abdominal adiposity (represented by the ratio of waist to hip circumferences), and upper-body fat (represented by skinfold ratios, usually subscapular/ triceps) were generally positively and independently related to glucose tolerance and diabetes prevalence. Although cross-sectional studies provide evidence for the importance of body fat distribution in the pathophysiology of dia-

1475

betes, the temporal relation between variables cannot be assessed. One prospective study examined upper-body adiposity (subscapular/triceps ratio) and found a positive relation between this ratio and the risk of NIDDM (9). Other prospective studies of men and women in Gothenburg, Sweden (10-12), demonstrated that the ratio of waist to hip circumference, an index of centripetal adiposity, was related to the risk of NIDDM. In men, this relation was independent of total-body adiposity but did not persist after adjusting for baseline levels of glucose (II). Among women, this relation also was independent of total-body adiposity and did persist after adjusting for baseline levels of glucose (12). In a 2-year follow-up study of US women (13), the waist/hip ratio was confirmed as a predictor of the self-reported incidence of diabetes, but no information on blood glucose levels was available. Using data from the Normative Aging Study, an ongoing longitudinal study of men in the greater Boston area, we sought to confirm the earlier prospective reports of a relation between the accumulation of abdominal fat and the risk of NIDDM, independent of total-body adiposity. In addition, we examined the relation, over an average of 18 years of follow-up, between total-body adiposity, body fat distribution, and both the fasting serum glucose level and the glucose level 2 hours after administration of an oral glucose load. MATERIALS AND METHODS Study design and data collection

The study population comprised 2,280 men aged 20-80 years at enrollment in 1963 into the Department of Veterans Affairs Normative Aging Study cohort. Details of this interdisciplinary study have been described elsewhere (14, 15). Of the male volunteers enrolled, 98 percent were Caucasian, the majority were veterans, and all met screening criteria based on current medical status and medical history. Only subjects who were initially healthy were enrolled; for example, men with diabetes at screening were ineligible, and thus no prevalent cases

Downloaded from https://academic.oup.com/aje/article-abstract/136/12/1474/198582 by University of Durham user on 20 March 2018

1476

Cassanoetal.

were expected in the cohort at baseline. Neither obesity nor body fat distribution was considered in the eligibility decision. Once enrolled, subjects were examined every 3-5 years, and these examinations continue at present. The detailed information on participants collected at each visit over two and a half decades (with up to eight visits/participant during the follow-up period) included the results of an extensive physical examination, laboratory findings, anthropometric values, and questionnaire data. Blood glucose levels were measured after both an overnight fast and 2 hours after a 100-g glucose load (hereafter referred to as 2-hour glucose levels). From the beginning of the study through January 1970, serum glucose levels were determined by manual methods as described by Folin and Wu (16). Thereafter, automated methods were used, with some changes over time in both the equipment (Technicon Auto-analyzer; Technicon Instruments Corp., Tarrytown, New York, from 1970 through 1973; Technicon AA II after 1973) and the specific assay used (the ferricyanide method, followed by a modified neocuproine method, and finally, through 1987, the glucose oxidase method). Although these changes were not systematically evaluated at the time they were instituted, retrospective evaluation of the most important change, from manual to automated methods, suggested that glucose values were fairly consistent through this transition. At each examination, serum glucose values were used to classify men into one of three groups according to criteria defined by the World Health Organization (17): 1) NIDDM, a fasting blood glucose level of > 140 mg/dl and/or a 2-hour postchallenge level of >200 mg/dl; 2) impaired glucose tolerance, a fasting serum glucose level of < 140 mg/dl and a 2-hour postchallenge glucose level between 141 and 199 mg/ dl; and 3) normal glucose tolerance, all other levels. Although the study used a 100-g glucose load (defined in the protocol and adhered to throughout the study) as opposed to the 75-g load on which the World Health Organization criteria (17) are based, these two loads are known to result in similar 2-

hour post-challenge plasma glucose values among healthy subjects (18). A subject was classified as having NIDDM if this disease was diagnosed by a physician involved in the study or if, at any examination during the study, the criteria based on blood values were met. For subjects with physiciandiagnosed NIDDM, the time of onset of the condition was recorded by the physician. For subjects with NIDDM diagnosed solely on the basis of blood glucose levels, the time of onset was taken to be the year in which the elevated blood values were first detected. Diabetes was viewed as a chronic condition; that is, once a subject met the criteria defining NIDDM, he was considered a diabetic for the remainder of the follow-up period. The diagnosis of NIDDM on the above bases introduced some potential for misclassification. Of 226 men with NIDDM, 13 percent had a physician's diagnosis recorded in the study files but no evidence of elevated glucose values at any examination, 26 percent had both a physician's diagnosis and glucose values that met the World Health Organization criteria for NIDDM on at least one examination, and 60 percent were diagnosed on the basis of glucose values alone. In the latter group (n = 136), 43 percent had their first elevated glucose level at their last examination, 27 percent had a single elevated glucose level and normal glucose values thereafter, and 31 percent had elevated values documented at multiple examinations. (More than half of the men in the last category had elevated blood glucose values at four examinations.) This information was used in analyses to increase the stringency of case definition. For instance, among men diagnosed as having NIDDM on the basis of laboratory values alone, cases were limited to those with repeated glucose elevations, and the effects of total-body adiposity and fat distribution were reevaluated. Anthropometric measurements were made at each examination by trained personnel using well-defined body landmarks and standard equipment, including an anthropometer (G. P. M., Gneupel, Switzerland), a Lange skin caliper (Cambridge Scientific Industries, Cambridge, Maryland),

Downloaded from https://academic.oup.com/aje/article-abstract/136/12/1474/198582 by University of Durham user on 20 March 2018

Body Fat Distribution, Blood Glucose, and NIDDM

and a cloth tape measure. While the subjects were clad in underclothes, approximately 46 anthropometric measurements were made. Reliability data were not available for the anthropometric values, but such error is expected to be random and would thus bias our findings toward the null value. Of direct relevance to the current study, the following measurements were made: 1) height in inches (later converted to meters) against a wall chart; 2) weight in pounds (later converted to kilograms); 3) hip breadth in centimeters, measured at the widest breadth of the hips while the subject was seated, with the anthropometer on the greater trochanter and without compression of the soft tissues; and 4) abdominal circumference in centimeters, at the level of the umbilicus without compression of the skin. Body mass index was calculated as weight in kilograms over height squared in meters, and the ratio of abdominal circumference to hip breadth was computed. We have discussed the use of hip breadth and provided justification for its use elsewhere (15). Other data were available for the control of covariates, including, for example, age, alcohol intake, cigarette smoking, and the use of antihypertensive medications. Data analysis

Graphic methods, including smoothing techniques (19), were used initially to examine the shape of the relation between anthropometric measurements and blood glucose levels. In addition, Pearson's product moment correlations were computed in an examination of the correlation structure of repeated measures of blood glucose over time. Multivariate analysis used two complementary approaches. Fasting and 2-hour blood glucose values were dichotomized into diabetic versus normal and impaired glucose tolerance, as defined above, and information on the diagnosis of diabetes by a physician also was used to identify diabetics. The time to diabetes constituted the "survival time" in Cox proportional hazards regression models (20). For these analyses, baseline

1477

values of the covariates were examined in relation to the risk of NIDDM over the follow-up period of the study. Similar analyses examined the risk of impaired glucose tolerance versus normal glucose tolerance over follow-up. Fasting and 2-hour glucose levels also were considered as continuous outcome variables; the use of a multiple regression model described by Rosner (21) allowed for the intraclass correlation between repeated measures for the same individual. Examination of the correlation matrix of repeated measures of the outcome (both fasting and 2-hour blood glucose) showed that the magnitude of the correlations was fairly stable over time, with values ranging from 0.3 to 0.6 for fasting glucose and 0.4 to 0.7 for 2hour glucose. Thus, a constant correlation model was assumed, and fitting proceeded using the generalized linear model with intraclass correlation.- Both fasting and 2-hour glucose values were transformed (natural logarithm) prior to analysis, and examination of the residuals from regression models with and without such transformation confirmed its appropriateness. Study population

Among the 2,280 men originally enrolled in the study, 41 subjects with prevalent diseases, for whom the health-screening criteria had been relaxed, were excluded from further consideration. An additional 267 men who dropped out of the study early could not be included because anthropometric data were lacking. For most of the study period (prior to 1984), anthropometric examinations were carried out separately from laboratory and physical examinations, and the first anthropometric examination was conducted after the baseline study examination. In creating a longitudinal data file for the current analyses, we linked anthropometric data to laboratory/physical examination data with the constraint that they had to be either concurrent or before the laboratory data. Thus, anthropometric data were not available at the baseline, and the first visit considered here was the visit after

Downloaded from https://academic.oup.com/aje/article-abstract/136/12/1474/198582 by University of Durham user on 20 March 2018

1478

Cassano et al.

the baseline examination. As mentioned above, 267 early dropouts had no anthropometric data, leaving 1,972 men (approximately 88 percent of the eligible cohort) available for the main analyses. In those analyses that considered time to diabetes, a further four subjects who developed NIDDM after enrollment but before the first anthropometric examination were excluded. At the last study visit considered here (conducted before or during November 1987), the current status of the 1,972 men providing information for these analyses was as follows: 78.7 percent (n = 1,551) had participated fully, 7.4 percent (n = 146) had completed questionnaires only, 3.7 percent (n = 72) had dropped out or been lost, and 10.3 percent (n = 203) had died. The average length of follow-up was 18 years (range, 026 years), with a study total of 35,496 person-years. In order to check for bias resulting from selective depletion of the cohort, the subgroup of men with no anthropometric data was compared with the rest of the cohort on the basis of information available from the baseline study visit. Men in the excluded subgroup were, on average, approximately 2 years younger than the continuing participants (mean ± standard deviation, 39.6 ± 10.7 and 41.9 ± 9.2 years, respectively), but average fasting and 2-hour glucose values were quite similar for the two groups. The groups could not be compared with regard to body size or body fat distribution as these variables were missing for the subgroup without anthropometric data. RESULTS

For the entire study cohort, the average age at entry was 41.9 years, and 36 percent of the subjects were current cigarette smokers at entry into the study. In light of all of the follow-up data obtained at all examinations, three groups were defined: 1) 1,312 (66.5 percent) men had persistently normal glucose tolerance; 2) 434 (22.0 percent) men were classified as having impaired glucose tolerance at one or more visits; and 3) 226 (11.5 percent) men were classified as having NIDDM at one or more visits. The charac-

teristics of participants at entry into the study were somewhat different across these three groups (table 1). Compared with those with normal glucose tolerance, men who were ultimately diagnosed with NIDDM were somewhat older at entry, had greater total-body adiposity, and had higher average glucose levels both after fasting and 2 hours postchallenge. In terms of these characteristics, the group with impaired glucose tolerance was intermediate between the normal and the NIDDM groups. Because some of the differences in other variables may have resulted from the age differences between groups, age-adjusted means also were considered. There was little or no change in any of the means after adjusting for age. Exploratory data analysis was undertaken to examine the linearity of the relation between the measures of glucose and the ratio of abdominal circumference to hip breadth. Nonparametric smoothing techniques (19) were used, and no important nonlinearities were evident. There was a fairly steady increase in both fasting and 2-hour glucose values with increasing ratios of abdominal circumference to hip breadth (figure 1). Predictors of diabetes incidence

The cohort was split into tertiles by both body mass index and the ratio of abdominal circumference to hip breadth according to the values obtained at entry into the study. (Intertertile values were 24.59 and 26.88 for body mass index and 2.48 and 2.61 for the ratio of abdominal circumference to hip breadth). The age-adjusted incidence rates of NIDDM within these subgroups were calculated as the number of events divided by the number of person-years at risk. The rate of NIDDM was highest among men in the upper tertile for both body mass index and the ratio of abdominal circumference to hip breadth (figure 2). Among men in the upper two tertiles for the ratio of abdominal circumference to hip breadth, the rate of NIDDM was highest in the top tertile of body mass index. Moreover, the rate of NIDDM consistently increased by tertile of the ratio of abdominal circumference to hip

Downloaded from https://academic.oup.com/aje/article-abstract/136/12/1474/198582 by University of Durham user on 20 March 2018

Body Fat Distribution, Blood Glucose, and NIDDM

1479

TABLE 1. Characteristics of the cohort members (n : 1,972)* at entry, grouped by outcome status: Normative Aging Study cohort, Boston, 1963-1987 Normal glucose tolerance ( n - 1,312)

Variable

Age (years) Height (m) Weight (kg) Body mass Index (weight (kg)/height (m)2) Abdominal circumference (cm) Hip breadth (cm) Ratio of abdominal circumference to hip breadth Fasting serum glucose (mg/dl) Two-hour postfoad serum glucose (mg/ dl)

Cigarette smoking Never Former Current

Impaired glucose tolerance

NIDDMf

40.9 ± 9.2* 1.8 ±0.1 78.8 ±10.3

44.1 ±8.9 1.7 ±0.1 80.9 ±10.1

44.2 ± 8.6 1.8 ±0.1 82.7 ±12.0

25.4 ± 2.7

26.5 ±2.8

26.9 ± 3.4

92.5 ± 8.0 36.6 ± 2.3

95.5 ±8.1 36.9 ±2.3

97.0 ± 9.4 37.0 ± 2.4

2.5 ± 0.2

2.6 ±0.2

2.6 ± 0.2

97.3 ± 9.4

99.5 ±10.1

100.9 ±10.5

101.5 ± 17.5

110.6 ±19.8

113.5 ±20.7

n

%

n

%

n

%

375 443 491

28.7 33.8 37.5

143 149 141

33.0 34.4 32.6

51 98 76

22.7 43.6 33.8

• The value for n may vary sightly because of mtesing data, t NIDDM, rcn-insulin-dependent diabetes meStus. | Mean ± standard deviation.

breadth, across all levels of body mass index. Consideration of these rates suggested there was some evidence for effect modification using an additive model, and the incidence attributable to the joint effect of body mass index and the ratio of abdominal circumference to hip breadth was 23.9/104. Thus, among men who were in the top tertile for both body mass index and the ratio of abdominal circumference to hip breadth, approximately 20 percent of the incidence of NIDDM (23.9/104) may be attributable to the joint effect of these two factors. In an examination of the risk of diabetes in a multivariate model, we considered the time to diabetes with the use of the Cox proportional hazards model (20). Both totalbody adiposity and body fat distribution at entry into the study were assessed in relation to the risk of NIDDM. Adjusting for age alone, we noted that an increased body mass index was related to an increased risk of

NIDDM (likelihood ratio statistic with 2 degrees of freedom (LRS2df) = 20.5, p < 0.001). There was an approximately twofold increase in the age-adjusted risk of NIDDM for men in the upper tertile for body mass index (>26.9 kg/m2) over that for men in the lowest tertile (figure 3). Men in the middle tertile were indistinguishable from men in the lowest tertile in terms of their risk of diabetes. Adjusting for body fat distribution attenuated the effect of body mass index (LRS2df = 6.44, p = 0.04), and a 30 percent greater risk of NIDDM was evident for men in the highest tertile for body mass index than for those in the lowest. Further adjustment for cigarette smoking or alcohol intake had little or no effect on these estimates. Body fat distribution at entry into the study, as represented by the ratio of abdominal circumference to hip breadth, was a strong predictor of NIDDM risk over the study period (figure 4). Adjusting for age

Downloaded from https://academic.oup.com/aje/article-abstract/136/12/1474/198582 by University of Durham user on 20 March 2018

1480

Cassano et al.

m110108Fasting Glucose

106-

104102"

100

40

60

SO

100

40

60

80

100

1401

130"

2 hour Qluoose

no100 0

20

Percentie of the ratio of abdominal circumference to hip breadth

FIGURE 1. Smoothed curves of the relation between the plasma glucose level (mg/dl) and the percentile of the ratio of abdominal circumference (cm) to hip breadth (cm) among men in the Normative Aging Study, Boston, Massachusetts, 1963-1987.

alone, we found a highly statistically significant relation of the ratio of abdominal circumference to hip breadth to the risk of NIDDM (LRSMT = 40, p < 0.001), and the risk of NIDDM was approximately threefold and 1.6-fold higher for men in the upper and middle tertiles, respectively, than for men in the lowest tertile. In models that adjusted for total-body adiposity as represented by body mass index, there was little or no change in the magnitude of the coefficients for the ratio of abdominal circumference to hip breadth (LRSiar = 25.2, p < 0.001). Similarly, the effect of the ratio of abdominal circumference to hip breadth was little changed by further adjustment for cigarette smoking and alcohol intake. When models were extended in a consideration of the possibility of effect modification (multiplicative) between body mass index and the ratio of abdominal circumference to hip breadth, no such effect was found.

Time-dependent covariates also were considered to allow assessment of current values of both fat distribution and adiposity in relation to risk. In models that allowed both body mass index and the ratio of abdominal circumference to hip breadth to vary with time, there was little or no change in the estimated regression coefficients for either variable (risk ratios for abdomen/hip ratio, tertile II vs. tertile I, 1.3 (95 percent confidence interval 0.9-1.9); tertile III vs. tertile I, 2.0 (95 percent confidence interval 1.42.9); x2 = 15, p = 0.001). Similarly, in models that considered the baseline values of both the ratio of abdominal circumference to hip breadth and body mass index, while allowing the incremental change in abdomen/hip ratio to vary with time, there was little or no change in the coefficients, and the coefficient for the variable representing the change in the abdomen/hip ratio was not statistically significant. Both current and former smokers were contrasted with nonsmokers in these models. The risk of NIDDM was about 1.5fold higher among both current and former smokers than among nonsmokers (table 2). There was little or no relation of alcohol intake (in one of five categories) to the risk of NIDDM, and final models did not adjust for this variable. Additional models that adjusted for baseline fasting or 2-hour glucose levels demonstrated an increased risk of diabetes in men with higher initial values. The hazard ratios were 1.03 (95 percent confidence interval 1.02-1.04) for glucose (mg/ dl) and 1.02 (95 percent confidence interval 0.95-1.09) for 2-hour glucose (mg/dl). However, the effect estimates for the ratio of abdominal circumference to hip breadth were changed very little by adjusting for either of these variables. The above analyses were refined in several ways (table 2). A series of models was considered, with different definitions for the outcome and comparison groups, in an examination of the robustness of these findings with regard to misclassification bias. When a stricter definition of diabetes was used (limited to cases diagnosed by a physician and/or by elevated blood glucose levels on

Downloaded from https://academic.oup.com/aje/article-abstract/136/12/1474/198582 by University of Durham user on 20 March 2018

Body Fat Distribution, Blood Glucose, and NIDDM

1481

FIGURE 2. Age-adjusted incidence of non-insulin-dependent diabetes mellrtus (per 10,000 person-years) by tertiies for the ratio of abdominal circumference (cm) to hip breadth (cm) and body mass index (weight (kg)/height (mf) among men in the Normative Aging Study, Boston, Massachusetts, 1963-1987.

ISO j

1.8

1,7-

13

1.3-

i

1.2

\

0.9

0.9-

0.8

! ,

0.5-

22

.

24

,

0.6 ,

,—

26

28

30

body mass Index FIGURE 3. Risk of non-insulin-dependent diabetes meintus by body mass index at study entry (weight (kg)/hetght (mf) shown for intertertile values among men in the Normative Aging Study, Boston, Massachusetts, 1963-1987. O, age-adjusted values only; • , values further adjusted for the ratio of abdominal circumference to hip breadth and cigarette smoking status at study entry; bars, 95% confidence interval for adjusted values that extend from the fined circles (•).

Downloaded from https://academic.oup.com/aje/article-abstract/136/12/1474/198582 by University of Durham user on 20 March 2018

1482

Cassano et aJ.

3.83.7 i

3.4-

1

3.02.6
50 percent of relevant examinations). Men who used antihypertensive agents at any time (or, for diabetics, before diagnosis) had a 1.4-fold increase in diabetes risk (95 percent confidence interval 0.982.02). Furthermore, men who had used such

Downloaded from https://academic.oup.com/aje/article-abstract/136/12/1474/198582 by University of Durham user on 20 March 2018

Body Fat Distribution, Blood Glucose, and NIDDM

1483

TABLE 2. Hazard ratios for the risk of non-insulin-dependent diabetes and impaired glucose tolerance for men: Normative Aging Study cohort, Boston, 1963-1987 Model specification - outcome vs. comparison group Variable*

Age (years) Body mass index Tertile II (vs. I) Tertile III (vs. I) Ratio of abdominal circumference to hip breadth Tertile II (vs. I) Tertile III (vs. I) Cigarette smoking Current (vs. non-) Former (vs. non-)

NIDDMf vs. normal, Strtcter§ NIDOM vs. Stricter NIDOM vs. normaM IGTt4 normal, IGT|

IGT vs. normal#

Hazard ratio

95% Clt

Hazard ratio

95% a

Hazard ratio

95% a

Hazard ratio

95% Cl

1.03

1.01-1.04

1.02

0.99-1.04

1.02

1.00-1.04

1.03

1.02-1.05

0.8 1.3

0.6-1.2 0.9-1.8

1.2 1.5

0.7-2.1 0.9-2.5

1.3 1.8

0.8-2.2 1.1-3.0

1.3 1.8

1.0-1.8 1.3-2.3

1.5 2.5

1.0-2.2 1.7-3.7

1.6 3.0

0.9-2.9 1.7-5.3

1.7 3.4

1.0-3.0 1.9-5.9

1.2 1.6

1.0-1.6 1.3-2.2

1.5 1.7

1.0-2.1 1.2-2.4

1.1 1.5

0.7-1.8 0.9-2.3

1.1 1.5

0.7-1.8 1.0-2.4

1.0 0.9

0.8-1.3 0.7-1.2

* Note: aP variables isted are entered into the equation simuitaneousty. t NIDDM, non-lnsuarKJepenclent dtebetes meflltus; IGT, Impaired giucose tolerance; Cl, confidence interval, i n = 221 cases and 1,734 censored. § Stricter NIDDM was defined as a physician's diagnosis and/or £2 examinations with blood glucose values of £140 mg/dl fasting or £200 mg/dl postchaBenge. | n = 118 cases and 1,734 censored. I n - 118cases and 1,301 censored. # n - 4 1 7 cases and 1,301 censored

medications relatively often (as reported at >50 percent of the eligible examinations) had a twofold increase in diabetes risk (95 percent confidence interval 1.4-3.3). However, the coefficient estimates for both adiposity and fat distribution were little changed in the models including antihypertensive use. Finally, in models excluding diabetics who had ever used antihypertensive medications (n = 35, with 188 cases left for consideration), the effect estimates for adiposity and fat distribution were little changed. Men in the top tertile for both body mass index and ratio of abdominal circumference to hip breadth are considered to be at highest risk for NIDDM. Among the 19 percent of the population in this group (365 of 1,968), 34 percent of the NIDDM cases arose (76 of 226 cases). The age-adjusted rate of NIDDM for men in the upper tertile on both body mass index and the ratio of abdominal circumference to hip breadth (119.4/104) was compared with that for the remainder of the population (54.6/104), yielding a relative risk estimate of 2.2. The population attributable risk percentage was then calculated

(22) using this relative risk estimate and 0.19 for the proportion of the population exposed. The population attributable risk was approximately 20 percent; that is, 20 percent of the cases in the population as a whole could be attributed to the top third of the population distribution for body mass index and ratio of abdominal circumference to hip breadth. The health screening of participants at the time of enrollment may partly explain why our estimate of the population attributable risk (20 percent) is lower than that reported in other studies (4). Longitudinal changes in glucose tolerance

The relation of totaJ-body adiposity and body fat distribution to fasting and 2-hour blood glucose values was examined in multivariate regression models that permitted the correlation between repeated measurements for individuals across examinations. Fasting and 2-hour glucose levels were considered in separate models to allow for possible differences in their importance. In models with natural logarithm 2-hour glucose as the out-

Downloaded from https://academic.oup.com/aje/article-abstract/136/12/1474/198582 by University of Durham user on 20 March 2018

1484

Cassano et al.

come, a positive relation was found between the ratio of abdominal circumference to hip breadth and glucose. Adjusting for body mass index attenuated this relation, yielding a coefficient that was approximately half the unadjusted value but still highly statistically significant (table 3). The relation was positive; thus, across all examinations, men with higher values for the ratio of abdominal circumference to hip breadth were predicted to have higher 2-hour blood glucose levels. Results for fasting glucose values were very similar. These analyses are consistent with the results from proportional hazards models. The relation of age to glucose tolerance also was assessed across all examinations. Little or no relation was found in the "longitudinal" regression model. However, according to ordinary least-squares regression analysis of the first observation for each study participant, age accounted for about 2 percent of the variance in 2-hour glucose levels, and glucose values increased by about 4 mg/dl per decade of age. Other estimates from cross-sectional studies for the effect of age have ranged from 6 to 9 mg/dl per decade, with R2 values as high as 20 percent (23). We may have seen less of a relation with age in this study as a result of the health screening of participants at the time of enrollment.

s o

i 2 8

c «

B

I

I c

£ o c o

DISCUSSION

Abdominal adiposity, as indexed by the ratio of abdominal circumference to hip breadth, was related to an increased risk of diabetes after adjustment for potentially confounding factors including age, cigarette smoking, and overall adiposity. Total-body adiposity, measured by body mass index, also was independently related to the risk of diabetes after adjustment for the effects of body fat distribution, age, and cigarette smoking. In models for which the definition of diabetes was more stringent, and in those excluding from the comparison group men who had evidence of impaired glucose tolerance, the same patterns were evident. Similar trends emerged when the risk of imDownloaded from https://academic.oup.com/aje/article-abstract/136/12/1474/198582 by University of Durham user on 20 March 2018

Body Fat Distribution, Blood Glucose, and NIDOM

paired glucose tolerance was considered as an outcome. These findings were consistent with those of earlier studies of this question. In analyses that considered blood glucose as a continuous outcome variable, the findings were consistent with the results of the proportional hazards analysis. These analyses revealed a positive effect of both fat distribution and total-body adiposity on glucose level across all examinations during the study. Although we have discussed possible shortcomings of the use of body mass index to represent total-body adiposity elsewhere (15), this issue deserves brief mention. As an index of adiposity, body mass index may be flawed in some populations because it is based on weight over height squared, and weight obviously reflects both fat and fatfree (muscle) mass. Men would attain high values on body mass index if they were either highly muscular or obese. In an assessment of the independent effect of body fat distribution, it is essential to adjust fully for the relation of adiposity to outcome to be sure that the effect of the ratio of abdominal circumference to hip breadth is not simply a matter of residual confounding (the ratio of abdominal circumference to hip breadth and body mass index are correlated, r = 0.56, in these data). A recent study that specifically considered the explanatory power of overall adiposity as assessed by hydrostatic weighing (considered the gold standard for estimating adiposity) found that this measure did not contribute to the prediction of either glucose or blood pressure level after body mass index had been accounted for in the model (24). This finding suggests that body mass index provides an excellent representation of total-body adiposity, although such conclusions obviously may depend in part on the characteristics of the study population, i.e., the level of physical conditioning. Unfortunately, data on physical activity are unavailable for our study population for the years before 1987, but these men are thought to be mostly sedentary. The current state-of-the-art pathophysiologic model that relates obesity and body fat distribution to the risk of diabetes

1485

has recently been reviewed by several authors (2, 3, 25- 28). Total-body adiposity may be related to diabetes via stimulation of the pancreas, an event that leads to increased secretion of insulin and at the same time engenders some resistance in peripheral tissue to the action of insulin. Although there is little evidence that body fat distribution, i.e., abdominal adiposity, per se is related to insulin secretion, it seems to bring about an increased resistance to insulin-mediated glucose uptake in muscle. In addition, abdominal adipocytes have higher rates of lipolysis that lead to a higher rate of turnover of free fatty acids; the resulting high levels of free fatty acids in the portal circulation are hypothesized to inhibit the hepatic extraction of insulin. Ultimately, this situation may contribute to both higher circulating levels of insulin and increased resistance to the effects of insulin, possibly through downregulation of the insulin receptor. Recent data (29) suggest the importance of a reduced suppression of hepatic glucose production in the early response to glucose challenge among persons with impaired glucose tolerance. Although it is hypothesized (29) that the reduced suppression of hepatic glucose output is caused by inadequate early beta cell response, fat distribution may also contribute to this effect through an effect on hepatic insulin resistance. Other factors are undoubtedly important, including the possible role of androgenic/ estrogenic hormone balance, which may contribute to an unfavorable distribution of body fat. The effects of genetic makeup and life-style on obesity, body fat distribution, and resistance to insulin involve a complicated web of interrelations that have yet to be clearly understood. Much uncertainty remains about the determinants of body fat distribution; a role for genetic factors is entirely consistent with the current evidence, and, in fact, the same genetic predisposition may lead to both insulin resistance and selective deposition of fat at abdominal sites. In two studies, researchers adjusted for baseline effects of glucose and insulin levels in assessing the prospective relation between body fat distribution and risk of diabetes (9, 11). In both reports, the relation of adiposity

Downloaded from https://academic.oup.com/aje/article-abstract/136/12/1474/198582 by University of Durham user on 20 March 2018

1486

Cassano et al.

and fat distribution to disease risk was fully explained by baseline levels of glucose and insulin, and the researchers concluded that the effects of abdominal adiposity are mediated through insulin resistance. In our study, Cox models that included baseline and 2-hour glucose levels showed statistically significant effects of these variables on diabetes risk. However, the effect estimates for body mass index and fat distribution showed little or no change. The lack of information on insulin levels at baseline precludes a full exploration of these issues based on these data, as insulin levels would be the best indicator of insulin resistance. In summary, our findings in this study provide further prospective evidence for a relation between both total body adiposity and body fat distribution and the risk of diabetes. REFERENCES 1. Kovar MG, Harris MI, Hadden WC. The scope of diabetes in the United States population. Am J Public Health I987;77:l 549-50. 2. Pedersen O. The impact of obesity on the pathogenesis of non-insulin-dependent diabetes mellitus: a review of current hypotheses. Diabetes Metab Rev l989;5:495-5O9. 3. Bjorntorp P. Abdominal obesity and the development of non-insulin-dependent diabetes mellitus. Diabetes Metab Rev l988;4:6l5-22. 4. Colditz GA, WiUett WC, Stampfer MJ, et al. Weight as a risk factor for clinical diabetes in women. Am J Epidemiol 1990; 132:501-13. 5. Vague J. The degree of masculine differentiation of obesities: a factor determining predisposition to diabetes, atherosclerosis, gout, and uric calculous disease. Am J Clin Nutr 1956;4:20-34. 6. Feldman R, Sender J, Siegelaub A. Difference in diabetic and nondiabetic fat distribution patterns by skinfold measurements. Diabetes 1969;18: 478-86. 7. Hartz A, Rupley DC, Kalkhoff RD, et al. Relationship of obesity to diabetes: influence of obesity level and body fat distribution. Prev Med 1983; 12: 351-7. 8. Haffner SM, Stern MP, Hazuda HP, et al. Do upper-body and centralized adiposity measure different aspects of regional body fat distribution? Relationship to non-insulin-dependent diabetes mellitus, lipids, and lipoproteins. Diabetes 1987; 36:43-51. 9. Haffner SM, Stern MP, Mitchell BD, et al. Incidence of type II diabetes in Mexican Americans predicted by fasting insulin and glucose levels, obesity, and body-fat distribution. Diabetes 1990;39: 283-8. 10. Ohlson L-O, Larsson B, Svardsudd K, et al. The influence of body fat distribution on the incidence

of diabetes mellitus. Diabetes 1985;34:1055-8. 11. Ohlson L-O, 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. 12. 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. 13. Kaye SA, Folsom AR, Sprafka JM, et al. Increased incidence of diabetes mellitus in relation to abdominal obesity in older women. J Clin Epidemiol 1991;44:329-34. 14. Bell B, Rose CL, Damon A. The Normative Aging Study: an interdisciplinary and longitudinal study of health and aging. Aging Hum Dev 1972; 3:5-17. 15. Cassano PA, Segal MR, Vokonas PS, et al. Body fat distribution, blood pressure, and hypertension: a prospective cohort study of men in the Normative Aging Study. Ann Epidemiol 1990; 1:33-48. 16. Folin O, Wu H. A system of blood analysis. Supplement 1. A simplified and improved method for determination of sugar. J Biol Chem 1920;41: 367-74. 17. World Health Organization Study Group on Diabetes Mellitus. Diabetes mellitus. Geneva: World Health Organization, 1985. (WHO technical report series 727). 18. National Diabetes Data Group. Classification and diagnosis of diabetes mellitus and other categories of glucose intolerance. Diabetes 1979^28:1039-57. 19. Segal MR, Weiss ST, Speizer FE. Smoothing methods for epidemiologic analysis. Stat Med 1988;7: 601-11. 20. Kalbfleish J, Prentice RL. The statistical analysis of failure time data. New York: John Wiley & Sons, Inc, 1980. 21. Rosner B. Multivariate methods in ophthalmology with application to other paired-data situations. Biometrics 1984;40:1025-35. 22. Rothman KJ. Modern epidemiology. Boston: Little, Brown and Company, 1986. 23. Shimokata H, Muller DC, Fleg JL, et al. Age as an independent determinant of glucose tolerance. Diabetes 1991;40:44-51. 24. Spiegelman D, Israel RG, Bouchard C, et al. Body fat distribution, fat mass, percent body fat: which is the real risk factor for diabetes and hypertension? (Abstract). Am J Epidemiol 1990; 132:806. 25. Bray GA. Obesity and diabetes. Acta Diabetol Lat 199O;27:81-8. 26. Landsberg L, Krieger DR. Obesity, metabolism, and the sympathetic nervous system. Am J Hypertens 1989;2(3Pt2):125S-32S. 27. Kissebah AH, Peiris AN. Biology of regional body fat distribution: relationship to non-insulindependent diabetes mellitus. Diabetes Metab Rev 1989,5:83-109. 28. Stern M, Haffner S. Body fat distribution and hyperinsulinemia as risk factors for diabetes and cardiovascular disease. Arteriosclerosis 1986;6: 123-30. 29. Mitrakou A, Kelley D, Mokan M, et al. Role of reduced suppression of glucose production and diminished early insulin release in impaired glucose tolerance. N Engl J Med 1992;326:22-9.

Downloaded from https://academic.oup.com/aje/article-abstract/136/12/1474/198582 by University of Durham user on 20 March 2018

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.

The relation between the abdominal accumulation of body fat, total-body adiposity, and blood glucose level and the risk of non-insulin-dependent diabe...
1MB Sizes 0 Downloads 0 Views