Serum Insulin, Obesity, and the Incidence of Type 2 Diabetes in Black and White Adults: The Atherosclerosis Risk in Communities Study: 1987–1998

This investigation was conducted in the Atherosclerosis Risk in Communities (ARIC) Study. A probability sample of black and white men and women aged 45–64 years was recruited from the metropolitan areas Forsyth County, NC, Washington County, MD, and Jackson, MS (black residents only) and from the suburbs of Minneapolis, MN. A detailed description of the response rates, study design, and methods is available (11). At baseline (1987–1989), 15,792 individuals were examined. Participants were excluded for the following reasons: neither black nor white race, missing insulin or BMI data, fasting for <8 h, prevalent diabetes, or BMI indicating underweight (<18.5 kg/m²). A total of 13,287 participants (1,930 black women, 5,395 white women, 1,205 black men, and 4,757 white men) were included in this analysis.

Participants underwent clinical examinations at baseline and at three subsequent examinations on 3-year cycles (examination 2: 1990–1992; examination 3: 1993–1995; examination 4: 1996–1998). All measurements were collected according to standardized protocols common to all ARIC study sites (12).

Participants were asked to fast overnight (>8 h) before the clinic examination and to refrain from drinking alcohol and caffeine or smoking the day of the examination. Blood was drawn from seated participants and shipped to a central laboratory for assay. Serum insulin was measured by radioimmunoassay using an Insulin Kit (Cambridge Medical Diagnosis, Billerica, MA). Serum glucose was measured by a hexokinase/glucose-6-phosphate dehydrogenase method on a Coulter DACOS device. Homeostasis model assessment was calculated as the product of fasting insulin (microunits per milliliter) and fasting glucose (micromoles per liter) divided by 22.5 (13). Fasting serum insulin and homeostasis model assessment were used as markers of insulin resistance (14). Total cholesterol and triglycerides were measured in plasma with enzymatic methods. HDL cholesterol was measured after dextran-magnesium precipitation (15). Blood pressure was measured three times from seated participants; the average of the last two measurements was used in this study. Hypertension was defined as systolic blood pressure ≥140 mmHg, diastolic blood pressure ≥90 mmHg, or use of medications to lower blood pressure in the 2 weeks before the clinic examination.

BMI was calculated as the ratio of weight (kilograms) to standing height (meters) squared (kg/m²); participants with a BMI ≥30 were classified as obese. To calculate waist-to-hip ratio, waist girth was measured at the umbilicus, and hip girth was measured as the largest diameter around the gluteal muscles.

Type 2 diabetes was defined according to the American Diabetes Association criteria (16) as any of the following: a fasting serum glucose level of 7 mmol/l (≥126 mg/dl), a nonfasting glucose level ≥11.1 mmol/l (200 mg/dl), self-reported use of medications for diabetes, or a self-reported previous physician diagnosis. Participants who were diagnosed with diabetes at examinations 2, 3, or 4 were considered to have incident diabetes. Follow-up time was calculated as the difference between the dates of clinic examination at diagnosis and the baseline clinic examination.

Age, race, sex, education level, physical activity, and diet were assessed at baseline. Education was categorized based on the highest grade level completed, and participants with less than a high school education (grade 12) were compared with participants with a high school education or higher. Alcohol consumption was categorized as current, former, or never. Current cigarette smokers were compared with participants who smoked <100 cigarettes in a lifetime. The modified questionnaire by Baecke et al. (17) was used to assess leisure time physical activity (e.g., gardening) and sports-related physical activity (e.g., jogging) on a five-point scale from one (low) to five (high). The modified 61-item food frequency questionnaire by Willett et al. (18) was used to assess diet. Prevalent coronary heart disease was defined as a history of coronary artery bypass surgery, balloon angioplasty, or myocardial infarction based on electrocardiograph or physician diagnosis (19). Medication use was identified and defined by coding all reported medications, vitamins, and supplements used in the 2 weeks before the clinic examination.

All analyses were performed separately by sex. The distribution of baseline characteristics was compared by race using t tests (means) and χ² tests (proportions). The distribution of insulin was skewed so values were log-transformed for analyses and back-transformed to geometric means for presentation. BMI was divided into six categories: 18.5 to <22, 22 to <25, 25 to <28, 28 to <31, 31 to <34, and ≥ 34 kg/m². Means of fasting insulin (95% CI) by racial group and BMI category were calculated from an ANOVA model including an interaction term between race and BMI category. In a separate model, the interaction between race and obesity (BMI ≥30 kg/m²) was examined. All models were adjusted for BMI as a continuous variable.

Incidence rates of diabetes (per 1,000 person-years) and incident rate ratios were calculated using Poisson regression. Cox proportional hazards regression models (adjusted for discrete failure time) were used to estimate the relative risk of incident diabetes by insulin level and racial group (20). To test our a priori hypothesis that the relationship between fasting insulin and incident diabetes differed between nonobese black and white participants, an interaction term between race and fasting insulin was entered into the regression models. A significant change in the maximum likelihood χ ² indicated statistical interaction.

The role of covariates on the relationship between fasting insulin and incident diabetes was evaluated by entering covariates into regression models. A change in the relative risk for insulin of >10% was accepted as an indication of a statistically important confounder. To account for multiple testing, statistical significance was denoted at P < 0.01. All analyses were performed using SAS version 8.1 (SAS Institute, Cary, NC).

At baseline, a higher proportion of black women had less than a high school education, their mean caloric intake was higher, their physical activity levels were lower, and their prevalence of hypertension was nearly double that of white women (Table 1). Black women were twice as likely to be obese than white women, and mean fasting insulin was significantly higher. The same characteristics differed significantly between black and white men, with the notable exceptions of BMI and insulin.

In each category of BMI, black women had higher fasting insulin than white women (race and BMI category interaction term, P = 0.0003) (Fig. 1A). In individuals with BMIs from 22 to 31 kg/m², differences in mean insulin were 15–20 pmol/l higher in black women than in white women. In the highest BMI category (≥34 kg/m²), means of insulin converged (absolute difference 5.8 pmol/l, black versus white) and 95% CIs overlapped, suggesting little difference between racial groups. Differences in fasting insulin when comparing obese (BMI ≥30 kg/m²) and nonobese (BMI <30 kg/m²) black and white women were also highly significant (race and obesity interaction term, P < 0.0001). Mean fasting insulin was 19.7 pmol/l higher among nonobese black women and 7.5 pmol/l higher among obese black women compared with white women.

Mean fasting insulin was similar among black and white men at each BMI category (Fig. 1B). Neither the interaction term for race and the six-level BMI category nor the term for race and obesity was significant (BMI categories, P = 0.2620; obesity, P = 0.5120). Therefore, all modeling of the relationship between fasting insulin and diabetes was restricted to women.

Over an average of 8.7 years of follow-up (SD 1.9), 750 women (10.2%) and 729 men (12.4%) were diagnosed with diabetes. In the full sample of women, the incidence rate ratio was double among black women compared with white women. This pattern was equally strong in nonobese women but less so in obese women (Table 2). The magnitude of the racial difference was slightly smaller among all men and nonobese men compared with women, and the rate of diabetes did not differ between obese black and white men.

Among nonobese women, the risk of developing diabetes was 1.37 (95% CI 1.22–1.54) for a 1-SD increase in insulin (60 pmol/l) for black women and 1.32 (1.24–1.41) for white women. This association was homogeneous by racial group (χ² = 0.24, P = 0.6273). Similarly, the association between insulin and incident diabetes did not differ between black (1.21 [95% CI: 1.02–1.44]) and white (1.33 [1.26–1.40]) nonobese men (χ² = 1.16, P = 0.2811). There was a positive monotonic relationship between fasting insulin and the relative risk of developing diabetes for both black and white women (Fig. 2A) and men (Fig. 2B).

In the absence of interaction between race and insulin among nonobese women and men, racial groups were pooled. We evaluated demographic (race, age, and education), dietary (% total calories from carbohydrates, dietary fiber), and physical activity as potential confounders of the relationship between insulin and incident diabetes. No variables were confounders according to the criteria of a 10% change in either sex. The pooled relative risk of developing diabetes did not change by > 1% with the inclusion of any of the covariates entered singly or in groups into the models for women (1.31 [95% CI 1.26–1.36]) or men (1.31 [1.24–1.38]) per SD increase in insulin.

In this population-based sample, nonobese black women, previously thought to be at a lower risk for developing diabetes (compared with their obese counterparts), demonstrated evidence of insulin resistance. Among nonobese women, the magnitude of association between hyperinsulinemia and diabetes was statistically equivalent between black and white women.

In the National Health and Nutrition Examination Survey (NHANES) I, Epidemiologic Follow-Up Study, the incidence of diabetes is only elevated in nonobese (BMI <30 kg/m²) black adults compared with white adults, but also the risk of diabetes is equivalent between races among the people who are not obese (BMI ≤30 kg/m²) (10). The authors suggest that greater visceral adiposity at a lower BMI in black adults may explain this association. However, previous research in this sample (21) and others (6) does not support this theory. In our sample, waist-to-hip ratio was smaller among black men and women compared with their white counterparts at lower body weights (data not shown). This finding suggests that another mechanism may be responsible for the racial disparity in diabetes incidence at lower body weights.

Previous research detected higher insulin among black children compared with white children (6–9,22). In a study of 73 black and white children (5–10 years old), fasting and postchallenge insulin was higher among black children even after total body fat, intra-abdominal fat, and subcutaneous fat were controlled (7). In the Insulin Resistance and Atherosclerosis Study, black adults aged 40–69 years demonstrated evidence of insulin resistance when compared with white adults at a similar body weight, independent of diabetes status (9,22). Lovejoy et al. (6) confirmed these results in a small (n = 59) sample of black and white women. Thus, in this large population, evidence of insulin resistance at a lower body weight in black women confirms previous research.

Each of the above-mentioned studies reports effects in both men and women. The sample presented by Lovejoy et al. (6) was restricted to women, whereas Gower et al. (7) and Karter et al. (9) demonstrated differences in both sexes, and Haffner et al. (22) controlled for sex in their analysis. We did not have an a priori hypothesis that the relationship between body weight and insulin would differ by sex. Rather, our decision was prompted by racial differences in energy expenditure, metabolism, and other factors associated with high insulin that may be more pronounced among women.

In a national sample, the proportion of physically inactive women was higher among black and Hispanic women than among white women (23). Weyer et al. (24) found that sleeping metabolic rate and 24-h energy expenditure was lower among black women compared with white women; less noticeable differences were detected among men. The striking excess of obesity among U.S. black women compared with white women (25) and the absence of such a difference between black and white men (26) may be a manifestation of these differences.

Interpreting differences in disease risk by race is complicated because of the close relationship between socioeconomic status and race (27). In this sample, there were marked socioeconomic differences as measured by education between black and white participants (Table 1). Less educated individuals may be more likely to engage in detrimental health behaviors, which may place them at increased risk for disease. However, previous research has shown that the racial disparity in diabetes cannot be explained fully by differences in socioeconomic status (4). Similarly, in this study, adjustment for education, physical activity, and dietary components does not attenuate the relationship between race or high fasting insulin and diabetes. Any epidemiologic study is limited in its ability to accurately measure socioeconomic status; thus, the possibility of residual confounding remains. However, given the consistency of these results across other study populations (6–9,22), it is equally plausible that there is a real biologic or genetic basis for susceptibility to insulin resistance in black women that should be investigated.

Primary strengths of this study include the large population sample, the longitudinal design, and the ability to assess the risk of developing diabetes among individuals with a combination of baseline risk factors. However, some potential limitations may affect these results. This epidemiologic study was restricted to a surrogate marker of insulin resistance: fasting insulin. In previous research, fasting insulin was highly correlated with Bergman’s insulin sensitivity index (Spearman’s r = ∼0.6) in people free of diabetes (14). At the time of insulin measurement, our sample was restricted to individuals without diabetes.

The possibility of geographic confounding by race arises in this study because the majority of black participants (89%) are from one study site: Jackson, MS. To address this concern, we conducted a secondary analysis that restricted comparisons to the Forsyth County, NC, site, which includes both black (n = 374, 59% women) and white (n = 3,148, 54% women) participants. The direction of the association was similar for all analyses, although the ability to ascertain statistical significance was limited because of sample size.

Because hyperinsulinemia is an equally strong predictor of diabetes in both racial groups, increased susceptibility to insulin resistance in black women may be one explanation for the excess incidence of diabetes in nonobese black women. However, hyperinsulinemia among nonobese black men is not a likely explanation for the excess risk in that group compared with white men; sex differences in the relationship between obesity and insulin resistance should be investigated. Future research should evaluate whether additional factors—genetic, environmental, or the combination of both—explain higher fasting insulin levels among black women compared with white women, independent of obesity, and whether insulin-sensitizing therapies and behavioral modifications are effective in reducing the burden of type 2 diabetes among black women.

This study was supported by National Heart, Lung, and Blood Institute (NHLBI) ARIC Contracts N01-HC-55015, N01-HC-55016, N01-HC-55018, N01-HC-55019, N01-HC-55020, N01-HC-55021, and N01-HC-55022 and by National Institutes of Health/NHLBI NRSA (National Research Service Award) Training Grant 5T32HL07034-26.

The authors thank the staff and participants in the ARIC Study for their important contributions.