Abstract
Objectives. We investigated whether racial disparities in the prevalence of type 2 diabetes exist beyond what may be attributable to differences in socioeconomic status (SES) and other modifiable risk factors.
Methods. We analyzed data from 34331 African American and 9491 White adults aged 40 to 79 years recruited into the ongoing Southern Community Cohort Study. Participants were enrolled at community health centers and had similar socioeconomic circumstances and risk factor profiles. We used logistic regression to estimate the association between race and prevalence of self-reported diabetes after taking into account age, SES, health insurance coverage, body mass index, physical activity, and hypertension.
Results. Multivariate analyses accounting for several diabetes risk factors did not provide strong support for higher diabetes prevalence rates among African Americans than among Whites (men: odds ratio [OR]=1.07; 95% confidence interval [CI]=0.95, 1.20); women: OR=1.13, 95% CI=1.04, 1.22).
Conclusions. Our findings suggest that major differences in diabetes prevalence between African Americans and Whites may simply reflect differences in established risk factors for the disease, such as SES, that typically vary according to race.
Members of racial and ethnic minority groups in the United States, including African Americans, suffer disproportionately from many chronic diseases, including type 2 diabetes (hereafter “diabetes”).1–3 Prevailing statistics suggest that African American adults are 50% to 100% more likely to have diabetes than are Whites,3–8 with evidence that diabetes precursors may even be more common in African American than in White children.9,10 Reasons for racial disparities in diabetes prevalence are not clear, but behavioral, environmental, socioeconomic, physiological, and genetic contributors have all been postulated.3,8,11
Because of the high prevalence of diabetes in the African American community, it has been suggested that African Americans may be more susceptible to the disease compared with Whites through direct genetic propensity or unfavorable gene–environment interactions.11 The fact that diabetes prevalence rates among Whites exceeded those among African Americans through at least the first half of the 20th century12 has led to the hypothesis that modern lifestyle factors (especially those that promote obesity) may have a greater effect on African Americans than on Whites.11,13
However, treating race as an etiological factor has been the subject of debate,14–16 and it has been argued that despite some genotypic delineations, race largely represents a complex mixture of behavioral, environmental, and social exposures.17,18 In comparison with Whites, African Americans often are poorer, have less education, are more likely to live in distressed households and communities, are less able to access quality health care, and have a less favorable risk factor profile for many diseases.18–20 Because socioeconomic (and associated environmental) differences between racial groups are so pervasive, attempts to isolate an effect of race will typically involve substantial confounding,16 resulting in difficulty estimating the relative contributions of genetic and environmental factors.
There have been several attempts to evaluate whether the disparity between African Americans and Whites with regard to diabetes can be attributed to factors other than racial background.7,13,21–27 Studies involving nationally representative sampling frames7,21,23–25,27 provided the platform for many of these analyses, which poses a challenge in that the average African American is of substantially lower socioeconomic status (SES) than the average White American. Because racial disparities persisted in these studies after adjustment for known diabetes risk factors, including some measures of SES, a possible genetic explanation has been invoked for the residual association, although the precise biological mechanisms remain speculative. Many of the studies conducted to evaluate the underlying reasons for racial disparities in diabetes prevalence have included fewer than 1000 each of African American men and women.7,13,21,23,25
Using the study population from the ongoing Southern Community Cohort Study (SCCS), which includes large numbers of African Americans and members of other racial/ethnic groups from generally similar socioeconomic circumstances, we had a unique opportunity to evaluate racial disparities in diabetes in a context in which confounding by extraneous factors related to race and SES would be limited by design. If racial disparities are driven by SES, one would expect little racial difference in diabetes prevalence rates within this population. We addressed the question of whether differences in diabetes prevalence between African Americans and Whites can be fully explained by SES or by adjustment for other correlates of diabetes risk.
The SCCS is a prospective epidemiological cohort study with ongoing participant enrollment across the southeastern United States.28 For the present analysis, we included cohort members enrolled from the beginning of the study (March 2002) until January 2006. These participants were enrolled in person at 48 community health centers located in both urban and rural areas across the states of Alabama, Arkansas, Florida, Georgia, Kentucky, Louisiana, Mississippi, North Carolina, South Carolina, Tennessee, Virginia, and West Virginia. Community health centers are government-funded health care facilities that provide basic health services primarily to low-income individuals.29 Nationally, approximately 70% of community health center patients live at or below the federal poverty level. African American and White participants were enrolled at the same community health centers.
The strategy used to enroll participants was to randomly approach people entering the community health centers (e.g., patients, individuals accompanying patients, community residents seeking other services offered by the community health centers) who appeared to be between the ages of 40 and 79 years and determine their eligibility for and interest in study participation. In addition to the age criterion, participants were required to speak English and to not have undergone treatment for cancer within the preceding year. The primary focus of the SCCS is determining reasons for racial disparities in cancer risk.
Participants completed a comprehensive, in-person, baseline interview covering various aspects of health and behavior, including personal and family medical history, diet, exercise, tobacco use, medication use, social support, psychological well-being, and access to health services. During this interview, participants were asked, “Has a doctor ever told you that you have had diabetes or high blood sugar?” Participants responding yes were asked follow-up questions regarding their age at first diagnosis and use (and names) of prescription medications taken to manage their diabetes. Women were specifically asked not to include gestational diabetes in their reporting.
The variables used to estimate SES were total household income in the previous year (less than $15000, $15000–$24999, $25 000–$49999, $50 000–$99999, $100 000 or more), highest level of education completed, and type of job held for the longest period of time during the participant’s adult life (reported in 20 broad categories, including “never worked” and “housewife”). We used Nam–Powers–Boyd (NPB) occupational status scores30 (on a scale from 1 [lowest] to 100 [highest], representing the socioeconomic standing of an occupation) for the occupational categories by assigning each category the average of the scores for its individual job examples. In the case of a small number of participants (n = 371; 0.8%) for whom we were unable to code longest-held job with our categories, or for whom this information was missing, we substituted the score for their current occupation.
Participants self-reported their current weight, their weight at age 21 years, the most they had ever weighed (not including weight during pregnancy), and their height. Because current weight may reflect weight adjustments (in either direction) after a diagnosis of diabetes,31 we chose to focus on participants’ reports of the most they had ever weighed and their weight at age 21 years (to account for long-term overweight or obesity). We calculated body mass index (BMI; weight in kilograms divided by height in meters squared) for each of these 2 weight measures; we defined overweight as a BMI of 25 kg/m2 or higher, obesity as a BMI of 30 kg/m2 or higher, and extreme obesity as a BMI of 40 kg/m2 or higher32 (categories are not mutually exclusive).
Participants also reported on leisure time physical activities they had engaged in during their 30s (amount of time per week). These activities included both moderate-level sports (e.g., bowling, dancing, golfing, and softball) and vigorous sports (e.g., jogging, aerobics, bicycling, tennis, swimming, weightlifting, and basketball).
We included in our cross-sectional analyses participants who identified themselves as either only African American or only White (n = 43 899; 97% of the cohort). We excluded an additional 77 participants (0.2%) who were missing self-reported information on diabetes status, leaving 43 822 (34331 African American, 9491 White) participants to serve as our study population.
We used multivariate logistic regression analyses to estimate measures of association (odds ratios [ORs] and 95% confidence intervals [CIs]). The following factors were selected a priori as covariates and modeled via the categories shown in Table 1: age at interview, educational level, total household income, NPB score, health insurance coverage, current BMI, highest BMI, BMI at age 21 years, and hypertension. Physical activity (in minutes), also selected a priori, was modeled with continuous variables. The (approximate) quartiles used for NPB scores were gender specific, with cutoffs for men of 22, 30, and 42.5 and cutoffs for women of 22, 30, and 56.
We evaluated additional variables as potential confounders, but they were found not to alter the main results by more than 5% and were not included in the final model. These variables were marital status, smoking status, and 2 measures of social support (participants’ reports of how many close friends or relatives would help with their emotional problems if needed and how many people they could ask for help in an emergency or with lending them money). Income adjusted for household size (determined by dividing the midpoint of the reported income category by the total number of people reported to be living in a given household) was also computed and substituted in the final model for the income variable, but it was not found to alter the results.
Participants’ mean age at enrollment was 51.2 years (SD = 8.7). The majority (61%) reported a household income below $15000 per year, and one third reported less than 12 years of schooling (Table 1). At the time of the baseline interview, 73% of participants were overweight, 44% were obese, and 11% were extremely obese. The prevalence of obesity was significantly (P< .001) higher among women than among men and significantly (P< .001) higher among African American than among White women. On the basis of the participants’ highest reported weight (and computed highest BMI), we found that a large percentage of each group (43% of African American men, 54% of White men, 69% of African American women, and 61% of White women) had been obese at some point in their lives. In general, we observed that the socioeconomic and other factors included in Table 1 had relatively similar race-specific distributions within each gender.
Overall, 9223 (21%) of the participants reported having been diagnosed with diabetes, and of these individuals, 86% reported taking diabetes medication, including insulin (Table 1). Differences in the reported prevalence of diabetes between African Americans and Whites were modest. Among women, African Americans were more likely to report diabetes than were Whites (24% vs 21%), whereas the converse was true for men (20% for Whites vs 17% for African Americans).
Diabetes prevalence rates in relation to factors previously shown to have significant associations with the disease (age, educational level, income, BMI) are reported in Table 2. As expected, the prevalence of diabetes increased with increasing age and BMI, and with decreasing education and income. Diabetes prevalence rose 8-fold from a low of 5% among participants whose highest BMI was less than 25 kg/m2 to 40% among those whose highest BMI was 40 kg/m2 or greater. Among participants who had ever been obese, the prevalence of diabetes varied little according to race or gender (30% among African American women, 29% among White women, 28% among African American men, and 30% among White men).
The prevalence of diabetes was inversely related to educational level, particularly among women, and overall it was 1.6 times higher among participants with less than 9 years of education than among those who had graduated from college (Table 2). Similarly, among participants in the lowest income category (less than $15000 per year), the prevalence of diabetes was 1.4 times higher than among participants with a household income of $50 000 per year or more; however, there were variations in the relationship between income and diabetes in each gender–race stratum, and the general trend of prevalence rising with decreasing income did not hold for African American men.
After adjustment for age, we observed no association between race and diabetes among men (for African Americans relative to Whites, OR = 0.92; 95% CI = 0.83, 1.01) and a modest excess among African American women in comparison with White women (OR = 1.39; 95% CI = 1.29, 1.49; Table 3). After further adjustment for educational level, income, NPB score, health insurance coverage, highest BMI, BMI at age 21, hypertension, and physical activity, there was still no significant difference between African American and White men, and the difference for women had been attenuated and remained only marginally significant (OR = 1.13; 95% CI = 1.04, 1.22). To avoid any potentially biasing effects of including individuals with type 1 diabetes, we repeated the analyses excluding participants who reported their age at first diagnosis as younger than 30 years. The results (Table 3) were nearly identical to those of our main analyses.
Table 4 shows the strong association between (highest) BMI and diabetes estimated from multivariate regression models run separately for each of the 4 gender and race groups. The strong trend of increasing risk across increasing categories of BMI was seen in all groups but tended to be more enhanced among Whites, although race × BMI interaction terms in gender-specific regression models were statistically significant only for women. We used regression models containing these interaction terms to estimate the effect of race on diabetes prevalence at various levels of BMI. Among women, the odds ratio for the effect of race was highest at the lowest BMI level (ORs = 2.03, 1.41, 1.26, 1.19, and 0.85 for women whose highest BMI was less than 25 kg/m2, 25–29.99, 30–34.99, 35–39.99, and 40 kg/m2 or higher, respectively), but no significant interaction was observed among men (ORs = 1.03, 1.42, 1.34, 0.97, and 0.60 for highest BMI of less than 25 kg/m2, 25–29.99 kg/m2, 30–34.99 kg/m2, 35–39.99 kg/m2, and 40 kg/m2 or higher, respectively).
In this large study of adults with similar socioeconomic circumstances and risk factor profiles, we found little evidence of a higher prevalence of diabetes among African Americans than among Whites. Even before we had controlled for BMI and other known determinants of the disease, we observed only a modest excess of diabetes among African Americans and only among women.
A social gradient in diabetes risk has been well documented both in the United States5,33–36 and in other Westernized countries.37–40 The factors underlying this gradient may include fetal or infant malnutrition,41–43 chronic stress,44,45 depression and other psychosocial factors,37,38,46 obesity,24,47 inactivity,24 and lack of access to preventive health care. Controlling for SES is problematic because SES stands as a proxy for a myriad of (often unmeasured) confounders, is difficult to quantify, and is prone to a high level of measurement error.16 Because confounding by SES can be intractable in the analysis phase of a study, strategies designed to limit it in the design phase may be more effective.
Despite our advantage of having a study design that produced a population closely “matched” across racial groups in terms of SES and our efforts to quantify SES, residual confounding by SES was still, in all likelihood, a factor in our findings, possibly accounting for the small residual racial effect we observed among women. Racial disparities in diabetes are often reported to be stronger among women than among men,3,6,22,23,25 and it may be that SES is a stronger confounder among women than among men. We found some evidence for the latter possibility, with education and income showing a stronger relation with diabetes among women, both crudely (Table 2) and in our final multivariate regression models (data not shown).
Our overall finding of nearly equal rates of diabetes among African Americans and Whites is contrary to the results of practically all published epidemiological studies on this subject.7,13,21–27 Although the findings from these previous studies are somewhat mixed, a common conclusion has been that racial differences in diabetes prevalence cannot be fully explained by established risk factors. We believe, however, that few investigations have overcome the confounding inherent in studies of race and disease. Indeed, these studies noted striking differences in several important confounders (e.g., measures of obesity or central adiposity, education, income, occupational status, or physical activity) between African Americans and Whites in their samples,7,13,21–26 and adjustment for SES often involved adjustment for education only.7,22–25
In a recent investigation that undertook a more comprehensive evaluation of the effects of confounding by SES and other variables, an initial African American excess in diabetes prevalence among women of 76% was eliminated (OR=1.04) after adjustment for poverty income ratio (i.e., income divided by the federal poverty line for a given family size), a number of examination-related variables (e.g., length of fast, time of day), body size variables, and measures of physical activity, diet, smoking, and alcohol consumption.21 In the same study, however, adjusting for the identical set of variables did not negate the effect of race among men.21
It has been suggested that, in terms of diabetes risk, obesity may have a more detrimental effect among African Americans than among Whites.7,23,25 Such a finding was reported in a pair of investigations involving data from the National Health and Nutrition Examination Survey (NHANES)7,25; in one of these studies, the strongest effect of obesity was observed among African American women.7 We did not find supportive evidence for an interaction in this direction, and in fact we observed the effect of obesity to be greater among White women than among African American women. Our finding is consistent with clinical evidence indicating that upper body obesity is more strongly associated with a diabetes-promoting metabolic profile among nondiabetic White women than among African American women.48–50
Our results raise the possibility that any racial differences in diabetes among women may be greatest at low BMI levels, with the racial gap disappearing as BMI increases. This has been noted elsewhere,23 but others have reported the opposite7 (i.e., similar diabetes prevalence rates among African Americans and Whites at ideal body weights, with a racial disparity growing with increasing percentage of desirable weight). One interpretation of our finding would be that if there is a race-based disparity among women, it may be more pronounced in women of normal weight; the reason may be that obesity has a greater effect on Whites, as we observed, or that the disparity is overshadowed in general by the strong influence of obesity on diabetes. Given the large number of comparisons made in this analysis, it is also possible that our finding of an interaction between race and BMI among women arose by chance.
The prevalence of diabetes in the SCCS is higher than the prevalence in the general population of the southeastern states. According to the Centers for Disease Control and Prevention’s Behavioral Risk Factor Surveillance System, 2004 age-specific diabetes prevalence rates in the 12 SCCS enrollment states were in the range of 7% to 12% in the 45- to 54-year age group, 13% to 19% in the 55- to 64-year age group, and 14% to 22% among individuals 65 years or older51; by contrast, we found rates of 19%, 31%, and 34%, respectively. We believe that the reason for the high prevalence of diabetes in the SCCS is that participant enrollment takes place in impoverished communities and within community health centers in which people seek care for their diabetes. Thus, our prevalence estimates do not lend themselves to generalization outside of our study population, but internal comparisons between subgroups of participants (in this case, by race) remain valid.
Systematic inaccuracies in reporting of diabetes diagnoses may have obscured actual racial differences in our study, but our collection of data in a standardized fashion across racial groups should have minimized this possibility. Nondifferential misclassification can also dampen true differences in reported outcomes but would have to be substantial to generate a null finding. Some confounders included as covariates in our analyses (e.g., participants’ weight at age 21 years and physical activity in their 30s) involved recall over a long period of time, but we would not expect race-specific differences in associated reporting errors.
Although exclusion of undiagnosed disease was a limitation of our study, it is not a likely reason for our null findings, given that self-reports have been used in other studies documenting strong racial disparities.4,5,26 An estimated 29% to 44% of diabetes cases in the United States are undiagnosed.4,6 However, African Americans in our study should not have been less likely than Whites to be diagnosed because of differential access to health care, because recruitment within community health centers ensured that all cohort members had essentially equal access to primary health care, and there were few racial differences in income level or type of health insurance coverage. Moreover, a recent analysis of NHANES (1999–2000) data revealed similar rates of undiagnosed diabetes among African Americans and non-Hispanic Whites according to fasting plasma glucose test results.4
On the basis of familial aggregation, twin studies, and recent advances in identifying molecular markers of risk, it is clear that diabetes is a genetically influenced disease.52 However, success in identifying genetic underpinnings of racial disparities in diabetes prevalence has been elusive. Although the existence of a “thrifty genotype” (a genetic adaptation to feast and famine cycles) has been posited since the 1960s,53 it has yet to be characterized. Furthermore, the notion that the thrifty genotype would affect African Americans more than other groups such as European Americans has been called into question.54 Genome-wide scans have uncovered some candidate markers of diabetes risk in affected African American families,55,56 and racial variations in the adiponectin57 and other genes58,59 involved in insulin sensitization or resistance have been noted; however, the contribution of these polymorphisms to racial disparities, or to diabetes risk in general, has not been firmly established.
Our results suggest that major differences in diabetes prevalence between African Americans and Whites are unlikely to be tied to race per se; rather, it is likely that they are linked to differences in established risk factors for diabetes that typically vary between African Americans and Whites. Our findings do not discount the possibility of race-specific differences in the pathogenesis or pathophysiological characteristics of diabetes60–62 or the possibility of racial differences in the molecular etiology of diabetes, but they do seem to refute the position that there is an intractable diabetes excess among African Americans unexplainable by nongenetic risk factors. Curtailing rising trends in obesity and improving the economic conditions of disadvantaged groups in the United States may be the key to controlling diabetes across all racial groups.
aScore range = 1 (lowest socioeconomic ranking) to 100 (highest socioeconomic ranking); gender-specific approximate quartiles. bBody mass index is weight in kilograms divided by height in meters squared; calculated from participants’ self-reports of their highest weight. cMeasure was self-reported (“Has a doctor ever told you that you have had high blood pressure?”). dDefined to participants with the examples of bowling, dancing, golfing, and softball. eDefined to participants with the examples of jogging, aerobics, bicycling, tennis, swimming, weightlifting, and basketball. fMeasure was self-reported (“Has a doctor ever told you that you have had diabetes or high blood sugar?”). aBody mass index is weight in kilograms divided by height in meters squared; calculated from participants’ self-reports of their highest weight. Note. OR = odds ratio; CI = confidence interval. aAdjusted for age, educational level, household income, Nam–Powers–Boyd occupational status score, health insurance coverage, highest body mass index, body mass index at age 21, hypertension, time per week engaged in moderate sports in 30s, and time per week engaged in vigorous sports in 30s. See “Methods” section for details about measures used. bLogistic regression model included only participants reporting an age at diagnosis of diabetes of 30 years or older and participants not reporting a diagnosis of diabetes.These restrictions were applied to exclude probable cases of type 1 diabetes. Note. CI = confidence interval. Odds ratios were estimated from separate logistic regression models run for each of the 4 gender and race groups. They were adjusted for age, educational level, household income, Nam–Powers–Boyd occupational status score, health insurance coverage, body mass index at age 21, hypertension, time per week engaged in moderate sports in 30s, and time per week engaged in vigorous sports in 30s. See “Method” section for details about measures used. aBody mass index is weight in kilograms divided by height in meters squared; calculated from participants’ self-reports of their highest weight.
Men, No. (%) Women, No. (%) Characteristic African American White African American White Total, No. 14 236 (100) 3 165 (100) 20 095 (100) 6 326 (100) Age at interview, y 40–44 4 236 (29.8) 805 (25.4) 5 275 (26.3) 1 297 (20.5) 45–49 3 730 (26.2) 700 (22.1) 4 742 (23.6) 1 346 (21.3) 50–54 2 871 (20.2) 550 (17.4) 3 892 (19.4) 1 106 (17.5) 55–59 1 582 (11.1) 414 (13.1) 2 543 (12.7) 1 020 (16.1) 60–64 948 (6.7) 342 (10.8) 1 684 (8.4) 776 (12.3) 65–69 482 (3.4) 193 (6.1) 1 022 (5.1) 409 (6.5) ≥70 387 (2.7) 161 (5.1) 937 (4.7) 372 (5.9) Educational level Less than 9th grade 1 306 (9.2) 390 (12.3) 1 590 (7.9) 594 (9.4) 9th–11th grade 3 720 (26.1) 656 (20.7) 4 864 (24.2) 1 315 (20.8) High school/vocational school 5 955 (41.8) 1 174 (37.1) 7 924 (39.4) 2 540 (40.2) Some college or junior (2-year) college 2 301 (16.2) 597 (18.9) 3 909 (19.5) 1 197 (18.9) College 681 (4.8) 227 (7.2) 1 283 (6.4) 454 (7.2) Graduate school 271 (1.9) 120 (3.8) 519 (2.6) 225 (3.6) Unknown 2 (0.0) 1 (0.0) 6 (0.0) 1 (0.0) Total annual household income, $ < $15 000 9 002 (63.2) 1 910 (60.4) 12 280 (61.1) 3 724 (58.9) 15 000–24 999 3 126 (22.0) 694 (21.9) 4 753 (23.7) 1 245 (19.7) 25 000–49 999 1528 (10.7) 376 (11.9) 2 168 (10.8) 779 (12.3) 50 000–99 999 382 (2.7) 140 (4.4) 567 (2.8) 399 (6.3) ≥100 000 64 (0.5) 32 (1.0) 75 (0.4) 105 (1.7) Unknown 134 (0.9) 13 (0.4) 252 (1.3) 74 (1.2) Nam–Powers–Boyd occupational status scorea Quartile 1 4 992 (35.1) 967 (30.6) 4 090 (20.4) 1 388 (21.9) Quartile 2 2 511 (17.6) 363 (11.5) 5 608 (27.9) 1 322 (20.9) Quartile 3 2 620 (18.4) 694 (21.9) 4 397 (21.9) 1 866 (29.5) Quartile 4 3 230 (22.7) 929 (29.4) 4 640 (23.1) 1 433 (22.7) Unknown 883 (6.2) 212 (6.7) 1 360 (6.8) 317 (5.0) Health insurance coverage None 7 065 (49.6) 1 513 (47.8) 7 703 (38.3) 2 654 (42.5) Any private insurance 2 503 (17.6) 449 (14.2) 4 653 (23.2) 1 475 (23.3) Medicaid only 1 756 (12.3) 341 (10.8) 3 618 (18.0) 789 (12.5) Medicare only 1 205 (8.5) 426 (13.5) 1 942 (9.7) 679 (10.7) Military only 483 (3.4) 89 (2.8) 95 (0.5) 35 (0.6) Other combinations 1 162 (8.2) 340 (10.7) 1 963 (9.8) 655 (10.4) Unknown 62 (0.4) 7 (0.2) 121 (0.6) 39 (0.6) Current body mass index, kg/m2 < 20 663 (4.7) 120 (3.8) 589 (2.9) 298 (4.7) 20–24.99 4 836(34.0) 885 (28.0) 2 827 (14.1) 1 309 (20.7) 25–29.99 4 906 (34.5) 1 076 (34.0) 5 068 (25.2) 1 642 (26.0) 30–34.99 2 359 (16.6) 620 (19.6) 4 970 (24.7) 1 377 (21.8) 35–39.99 939 (6.6) 265 (8.4) 3 293 (16.4) 828 (13.1) ≥40 479 (3.4) 192 (6.1) 3 127 (15.6) 825 (13.0) Unknown 54 (0.4) 7 (0.2) 221 (1.1) 47 (0.7) Highest body mass index,b kg/m2 < 20 156 (1.1) 24 (0.8) 172 (0.9) 67 (1.1) 20–24.99 2 812 (19.8) 436 (13.8) 1 715 (8.5) 879 (13.9) 25–29.99 5 058 (35.5) 1 002 (31.7) 4 159 (20.7) 1 460 (23.1) 30–34.99 3 540 (24.9) 887 (28.0) 5 257 (26.2) 1 442 (22.8) 35–39.99 1 611 (11.3) 434 (13.7) 3 834 (19.1) 1 036 (16.4) ≥40 1 010 (7.1) 378 (11.9) 4 797 (23.9) 1 412 (22.3) Unknown 49 (0.3) 4 (0.1) 161 (0.8) 30 (0.5) Body mass index at age 21 years, kg/m2 < 20 2 360 (16.6) 509 (16.1) 5 542 (27.6) 2 139 (33.8) 20–24.99 7 154 (50.3) 1 561 (49.3) 8 935 (44.5) 2 793 (44.2) 25–29.99 3 207 (22.5) 790 (25.0) 2 986 (14.9) 713 (11.3) 30–34.99 723 (5.1) 175 (5.3) 1 080 (5.4) 328 (5.2) 35–39.99 183 (1.3) 51 (1.6) 349 (1.7) 136 (2.2) ≥40 103 (0.7) 32 (1.0) 263 (1.3) 116 (1.8) Unknown 506 (3.6) 47 (1.5) 940 (4.7) 101 (1.6) Hypertensionc No 7 346 (51.6) 1 602 (50.6) 7 431 (37.0) 3 077 (48.6) Yes 6 879 (48.3) 1 562 (49.4) 12 658 (63.0) 3 246 (51.3) Unknown 11 (0.1) 1 (0.0) 6 (0.0) 3 (0.1) Moderate sports activityd in 30s, h/wk 0 4 917 (34.5) 1 354 (42.8) 8 098 (40.3) 2 932 (46.4) 0.01–2.00 2 739 (19.2) 597 (18.9) 4 255 (21.2) 1 204 (19.0) 2.01–4.99 2 724 (19.1) 480 (15.2) 3 596 (17.9) 958 (15.1) ≥5 3 748 (26.3) 716 (22.6) 3 897 (19.4) 1 182 (18.7) Unknown 108 (0.8) 18 (0.6) 249 (1.2) 50 (0.8) Vigorous sports activitye in 30s, h/wk 0 4 520 (31.8) 1 499 (47.4) 11 038 (54.9) 3 486 (55.1) 0.01–2.00 3 096 (21.8) 611 (19.3) 3 977 (19.8) 1 151 (18.2) 2.01–4.00 2 397 (16.8) 357 (11.3) 2 131 (10.6) 620 (9.8) > 4 4 120 (28.9) 683 (21.6) 2 728 (13.6) 1 029 (16.3) Unknown 103 (0.7) 15 (0.5) 221 (1.1) 40 (0.6) Diabetesf No 11 858 (83.3) 2 546 (80.4) 15 178 (75.5) 5 017 (79.3) Yes 2 378 (16.7) 619 (19.6) 4 917 (24.5) 1 309 (20.7) Currently taking diabetes medication No 12 217 (85.8) 2 642 (83.5) 15 736 (78.3) 5 276 (83.4) Yes 2 018 (14.2) 523 (16.5) 4 356 (21.7) 1 050 (16.6) Unknown 1 (0.0) . . . 3 (0.0) . . . Men, % Women, % Characteristic African American White African American White Total, % Age, y 40–44 9.7 10.8 13.3 13.1 11.8 45–49 14.2 17.4 19.1 17.8 17.1 50–54 18.2 22.9 26.1 20.1 22.4 55–59 24.4 20.8 34.9 25.9 29.2 60–64 28.2 28.7 38.2 26.8 32.4 ≥65 30.0 28.3 39.2 26.4 33.7 Educational level Less than 9th grade 24.3 22.1 36.5 26.8 29.4 9th–11th grade 16.8 20.0 27.0 23.3 22.5 High school/vocational school 15.4 19.0 23.3 20.8 20.0 Some college or junior college 15.3 20.4 20.8 17.7 18.7 College or higher 18.1 16.4 19.9 15.5 18.3 Total annual household income, $ < 15 000 16.2 21.1 26.9 23.4 22.4 15 000–24 999 16.7 18.0 21.9 20.4 19.8 25 000–49 999 18.5 18.1 18.2 14.6 17.7 ≥50 000 19.5 12.2 18.2 9.9 15.6 Highest body mass index,a kg/m2 < 25 4.6 4.6 6.7 3.5 5.1 25–29.99 10.4 8.2 13.9 9.9 11.4 30–34.99 21.4 18.5 21.9 18.2 21.0 35–39.99 33.5 34.1 30.4 26.6 30.8 ≥40 40.8 54.0 38.5 41.5 40.2 Age Adjusted,OR (95% CI) Multivariate,a OR (95% CI) Multivariatea: Participants Diagnosed at 30 Years or Older,b OR (95% CI) Men White (Ref) 1 1 1 African American 0.92 (0.83, 1.01) 1.07 (0.95, 1.20) 1.04 (0.92, 1.18) Women White (Ref) 1 1 1 African American 1.39 (1.29, 1.49) 1.13 (1.04, 1.22) 1.14 (1.05, 1.24) Highest BMI,a kg/m2 African American Men,AOR (9% CI) White Men,AOR (9% CI) African American Women,AOR (9% CI) White Women,AOR (9% CI) < 25 (Ref) 1 1 1 1 25–29.99 2.09 (1.69, 2.60) 1.50 (0.87, 2.61) 1.75 (1.40, 2.18) 2.58 (1.72, 3.89) 30–34.99 4.12 (3.31, 5.14) 3.29 (1.92, 5.62) 2.89 (2.33, 3.57) 4.71 (3.16, 7.02) 35–39.99 6.92 (5.46, 8.76) 7.93 (4.52, 13.90) 4.29 (3.46, 5.33) 7.15 (4.76, 10.75) ≥40 8.26 (6.36, 10.73) 16.17 (8.98, 29.11) 5.92 (4.76, 7.36) 12.30 (8.18, 18.49)
The Southern Community Cohort Study was supported by the National Cancer Institute (NCI; grant R01 CA92447). D. G. Schlundt, M. S. Buchowski, and M. K. Hargreaves received partial funding from the NCI (grants P60-DK20593, HL67715, and 5P20-MD000516, respectively).
Human Participant Protection The Southern Community Cohort Study was approved by the institutional review boards of Vanderbilt University and Meharry Medical College. All participants provided written informed consent.
