Type II diabetes (diabetes) is a potent risk factor for cardiovascular disease (CVD). Centers for Disease Control and Prevention (CDC) 2010 data show that almost 26 million Americans are diabetic, of whom 7.0 million remain undiagnosed [1]. Risk factors for diabetes include physical inactivity, family history of diabetes, impaired glucose metabolism and race/ethnicity. Although unhealthy lifestyle increases the risk of diabetes, and lower socioeconomic status (SES) is associated with higher CVD risk, whether diabetes, a CVD risk equivalent is similarly related to SES is uncertain [2].

To date, a majority of studies about SES and diabetes have been cross-sectional in nature. For example, low education level and occupational position were associated with a three-fold risk of prevalent diabetes in a cross-sectional cohort of middle aged men and women [3]. Similar findings were noted in the KORA Survey 2000 among elderly women [4]. In a prospective analysis utilizing the National Health and Nutrition Evaluation Study (NHANES) 1 Epidemiologic Follow up Study (NHEFS) data, higher income and education were associated with lower rates of diabetes in women [5]. Results from the prospective Whitehall II Study evaluating the association of social position with diabetes also indicate that among men and women lower civil service employment grade was related to at least a 2-fold increase in diabetes risk, while data from the geographically localized Alameda County Study also showed decreased diabetes risk with increasing years of education [6], [7].

Besides the relative paucity of prospective data about SES and diabetes, there remains a lack of comprehensive information about the biologic mediators of any potential relationship. Thus, although factors such as obesity, older age, family history of diabetes, hypertension, abnormal lipid and other CVD biomarker levels are linked to the development of diabetes, whether or not these same factors mediate any relationship between SES and incident diabetes is not known [8], [9]. Current public health strategies related to diabetes care support physical activity, diet, insulin and oral medications to control glucose levels. However, SES which although not traditionally thought of as modifiable risk factor for disease after a certain age is arguably a potentially modifiable via early implementation of targeted public health strategies for vulnerable populations such as provision of safe, clean space for physical activity as well as educational and job opportunities aimed at improving SES disparities. Because the effect of SES on diabetes development has not been adequately examined, we sought to 1) evaluate the association between education and income and incident diabetes; and 2) examine significant mediators of any potential relationship in a large United States cohort of female health professionals in an effort to inform public health strategies.

During a median follow-up period of 12.3±1.9 years, 1,262 participants developed diabetes (overall incidence was 4.5/1000 person-years). As shown in Table 1, compared to women with <2 years of HPE, women with doctorates were less likely to be obese (BMI, 24.3±4.2 versus 27.0±5.3 kg/m2 – not in table), have a history of hypertension and be current smokers. The most educated women were more likely exercise at least 4 times weekly and consume daily alcohol than the least educated women.

Table 2 shows the distribution of baseline lipid, inflammatory biomarker and hemoglobin A1c levels according to education and income levels. TC, LDL-C, and triglyceride levels decreased with higher education and income levels (p-trend<0.001 for each), whereas HDL-C levels increased with education and income categories (p-trend<0.001 for each). HsCRP, s-ICAM-1, and fibrinogen levels were all lower with higher categories of education and income (p-trend<0.001 for each).

Table 3 depicts the hazard ratios for diabetes according to education and income categories. We observed an inverse relationship between increasing education and incident diabetes. Adjustment for age, race/ethnicity, and family history (Model 2) was associated with relative risk reductions of 28%, 38%, 46%, and 63% with 2–<4 years of HPE, BS degree, MS degree, and doctorate respectively (p-trend<0.001). In a similar multivariable model (not shown in the Table) containing both education and income as independent variables, we observed a comparable trend; namely 21%, 30%, 38%, and 55%, relative risk reductions associated with 2–<4 years HPE, BS degree, MS degree, and doctorate degrees respectively (p-trend<0.001).

In separate Cox regression models that considered each risk factor, one at a time, and adjusted for age, race/ethnicity, and family history of diabetes, there was minimal attenuation noted in the HR comparing the most educated to the least educated women based on education and income respectively (Appendix S1 A/B); body mass index contributed the largest effect of the individual factors. For education, when lipid, inflammatory, and behavioral mediators were considered simultaneously in a fully adjusted model, the HR comparing the most with the least educated women was substantially attenuated; increasing from HR 0.37 (95%CI 0.27–0.52) in the base model (same as model 2, Table 3) to HR 0.95 (95%CI 0.68–1.33) [p-trend 0.96]. In a similar evaluation, the HR comparing the highest with the lowest income level was substantially attenuated; increasing from HR 0.38 (95%CI 0.27–0.53) in the base model to HR 1.04 (95%CI 0.74–1.46), [p-trend 0.42]. As the inverse relation between SES and diabetes was essentially entirely explained by the mediators examined, we sought to evaluate the proportion of the relationship explained by each set of potential mediators (Figure 1).

Table 4 demonstrates hazard ratios associated with incident diabetes and education/income based on control for certain biomarkers of inflammation, lipids, behavioral characteristics and a fully adjusted model separately.

Finally, we found no evidence of an interaction between the effect of SES with BMI or physical activity (education_BMI p-value>0.7; education_exercise p-value>0.9; income_BMI p-value>0.2; and income_exercise p-value>0.5).

In this prospective cohort of female health professionals, we observed a progressive decrease in diabetes with increasing levels of education and income. Our results indicate that lower SES is associated with increased diabetes risk in women, even in this relatively well-educated cohort of health professionals. Notably, one concern about many SES-health analyses has been their focus on the lower boundaries of SES thereby possibly missing the existence of “the gradient” at all levels of SES in middle aged and older women. We also sought to understand the mechanism by which SES is associated with a decreased hazard of diabetes. Our data suggests that a majority of this effect is mediated by measured lipid, inflammatory, and behavioral characteristics. In particular, behavioral factors significantly affected the relationship between education/income and diabetes risk. Thus, at a public health level, interventions targeting these behaviors could substantially impact the risk of diabetes. Our data represent one of the few comprehensive, prospective studies of SES and incident diabetes. Additionally, our results extend previous work by assessing a large group of apparently healthy women at baseline and by examining potential mediators of the relationship between SES and diabetes.

Previous work such as the cross-sectional analysis by Agardh and colleagues revealed that fathers' middle income position and lower educational levels were more associated with diabetes in women than in men [3]. Based on these findings, the authors posited that social position is more persistently influenced by family background in women than in men. Prospective work from the NHANES involving 11,069 subjects (∼62% women) showed a lower risk of diabetes among women with increasing education [5]. Notably in NHANES, even after control for behavioral factors such as body size, physical activity, diet, smoking, and alcohol use, the SES-incident diabetes relationship was not entirely explained. Consistent with our findings, Maty et al. found that study participants with <12 years education had 50% excess risk of incident diabetes compared with those with more education (HR = 1.5; 95%CI 1.11–2.04) [7]; income was also not associated with increased diabetes risk. In the Maty et al. study, measures of adiposity also significantly impacted the education-diabetes relationship. Unlike our data, neither of these aforementioned studies adjusted for lipid and inflammatory factors.

Our analysis reveals that the relationship between education and diabetes was most affected by behavioral factors. BMI explained the majority of the SES- DM association explaining 32% of the education and 39% of the income effects respectively. Indeed, lower educational and financial resources are in part associated with more risky health behaviors, lower levels of social support and more adverse physical and environmental exposures [22]. For example, inadequate dietary, housing, transportation options can lead to weight gain with resultant dyslipidemia as well as chronic psychological stress. Specifically, poor housing and transportation options might be associated with unsafe physical neighborhood environments that include higher levels of violence, lack of sidewalks and parks, factors that would decrease the likelihood of resident recreational physical activity. Experimental evidence suggests that at the biological level these experiences which likely occur throughout the lifespan contribute to the development of insulin resistance, excessive inflammation, dysregulation of the hypothalamic-pituitary axis and sympathetic system overdrive over time [23].

The importance of lifestyle factors such as BMI in diabetes risk was also demonstrated in data from the Nurses' Health Study [20]. Indeed, at the molecular level, higher BMI is associated with elevated plasma levels of free fatty acids (FFA) which promote peripheral (muscle) insulin resistance [24], clinically deleterious lipid levels, and increased inflammation. For example, our research team has previously shown that markers of inflammation such as hsCRP are significant predictors of incident diabetes [25]. Likewise, in other work, Rathmann et al. demonstrated that adjustment for BMI and waist circumference resulted in a null association between increased CRP and low SES (p = 0.23) [26]. Extending previous work, our analysis here also investigates the impact of additional inflammatory markers including sICAM-1 and fibrinogen on the SES-diabetes risk relationship. However, these measured inflammatory biomarkers only partially explain our SES- diabetes relationship, an observation that is likely a consequence of the ability of these acute phase response biomarkers to capture only certain aspects of the inflammatory cascade.

Strengths of the present study include its prospective design, sample size, and evaluation of traditional and non-traditional risk indicators. The availability of long-term follow-up with a large number of confirmed events also enhanced our analysis. However, limitations also warrant discussion. First, the study population consisted of predominantly white, middle-aged female health professionals. This cohort consists of a low number of women from other race/ethnic groups, particularly African Americans and Hispanics, groups that have a much higher prevalence of diabetes raising the speculation that our findings could be more pronounced in these groups. To date, the largest prospective study of African-American women related to incident Type 2 diabetes showed that among 46, 382 women participating in the Black Women's Health Study (BWHS), women with ≤12 years relative to ≥17 years of education had a self-reported diabetes incident rate ratio of 1.28 (95% CI 1.15–1.43); for household income <$15,000 compared to >$100,000, the incident rate ratio for diabetes was 1.57. Similar to our study, the most important mediator in the relationship between SES and diabetes was body mass index. Notably, although BWHS participants have a broader education range in WHS, a majority of participants are college educated [27]. Second, a single baseline plasma measurement of each biomarker was utilized and thus we were unable to evaluate the effects of changes in plasma levels of inflammatory markers over time. Residual confounding by obesity and other unmeasured factors such as psychological stress is also possible. Third, we used BMI rather than waist circumference as the measure of obesity, however BMI and waist circumference both have similar demonstrated ability to predict diabetes [28]. Moreover individual body weight may change over time. Fourth, since it is plausible that diabetes risk factors accumulate over the lifespan, the WHS assessed only certain adult measures of SES [3], [29]. Fifth, we utilized education and income as measures of SES. Other measures of SES such as neighborhood composition/environment or occupation need to be incorporated into other longitudinal studies. Despite the high correlation between education and income and prior suggestions that education is a more robust measure of SES compared to other SES indicators, different measures of SES likely reflect different aspects of social stratification. Moreover, in the current study, it is possible that characteristics of retired women who would have lower income but potentially greater wealth than younger women could confound any multivariate income-diabetes relationship.

In summary, we found an inverse association between incident diabetes and increasing education and income levels in a large cohort of initially healthy, female health professionals. Our data are consistent with the hypothesis that these relationships are almost entirely mediated by lipid, inflammatory, and particularly by behavioral factors. Our findings extend existing data that suggest socioeconomic disadvantage at low SES boundaries predicts health risk to higher SES boundaries, an area where research is lacking. Finally, our results support the need for public health programs specifically targeting the obesity epidemic as a crucial means to decrease the incidence of diabetes even among well educated and affluent populations.