How Are Income and Education Related to the Prevention and Management of Diabetes?

Abstract

Objective: To understand whether income and education have unique contributions in the prevention and management of diabetes, as measured by Hba1c levels among diagnosed and undiagnosed individuals. Method: Data were obtained from the Nashville Stress and Health Study (N = 1,189). Ordinary least squares regression analyses were used to examine the associations among education, annual household income, diabetes onset, and diabetes management (Hba1c). Results: Education, but not income, was inversely associated with Hba1c in undiagnosed participants. Income, but not education, was inversely associated with Hba1c among diagnosed participants. Discussion: These results support the idea that education and income provide non-fungible social and economic resources that vary in their consequences for Hba1c level depending on whether someone has been diagnosed with diabetes. The social resources provided by education may be more important in delaying the onset of disease, whereas resources provided by income may be more important for disease management.

Keywords

diabetes prevention management fundamental cause SES disparities

Introduction

Diabetes mellitus type 2 (diabetes) is the seventh leading cause of death in the United States and an estimated 9.4% of the U.S. population met the criteria for the disease in 2015 (Centers for Disease Control and Prevention [CDC], 2017). Diabetes prevalence is socially patterned, as individuals with lower socioeconomic status (SES) disproportionately bear its burden (Beckles, Chou, & Centers for Disease Control Prevention, 2013). For instance, individuals with less than a high school education have nearly double the rate of diabetes compared with those with a college degree (CDC, 2017).

Although we know people with lower SES are more likely to develop diabetes and less likely to manage it successfully, we know less about how individual dimensions of SES are related to both the onset and management of diabetes (Adler & Newman, 2002; Cutler, Lleras-Muney, & Vogl, 2008). Education and income may provide different types of non-interchangeable resources. For this reason, education and income may differ in their associations with diabetes prevention and management. For example, education may confer cognitive and psychosocial resources (critical thinking skills, sense of control) that help individuals to avoid the onset of disease (Herd, Goesling, & House, 2007; Mirowsky & Ross, 2015), whereas income might provide more tangible resources acquired through the marketplace (e.g., health care access, testing supplies, medications) that enable individuals to manage their diabetes. Informed by these ideas, this research tests how income and education are associated with the onset and management of diabetes by using a community-based sample of Black and White adults (ages 22-69) living in Nashville-Davidson County, Tennessee.

Our research builds upon the influential work of Herd et al. (2007) who provided evidence that the relationship between SES and disease varied by both dimension of SES and the stage of disease progression. They assessed multiple outcomes by using a measure of seven chronic conditions, including diabetes, arthritis, heart disease, and cancer. They found education was protective over the onset of disease, whereas income, but not education, was protective over the progression of disease. Here, we focus on one of their outcomes (diabetes) and in doing so overcome a key limitation of their study—self-reports of disease progression.

Part of the difficulty in addressing the potentially differing links between SES and diabetes onset and management is accurately measuring diabetes management. While self-reports of diabetes prevalence are reasonably accurate (Goldman, Lin, Weinstein, & Lin, 2003; Jackson et al., 2014), self-reports of diabetes management are notoriously prone to measurement error (Gonzalez & Schneider, 2011). Social desirability, patient memory, patient literacy, and numeracy all contribute to measurement error in self-reports of disease management. Moreover, given the literacy and numeracy issues inherent in self-reports, measurement error might vary by SES, with less accuracy at lower levels of educational attainment and household income. In the current study, we attempt to overcome the limitation by utilizing a measure of glycated hemoglobin (Hba1c) that is not vulnerable to self-report bias.

Background

In their classic work, Link and Phelan (1995) argued that macro-level social structures shape a person’s exposure to more proximate, individual-level risk factors of disease, such as stress, smoking, poor nutrition, and lack of exercise. As Link and Phelan stated, low SES increases individuals’ “risk of risks” of disease (Link & Phelan, 1995, p. 80). Higher SES is protective because it confers a broad set of flexible resources that can be deployed to avoid health risks (Link, Phelan, Miech, & Westin, 2008). Exposure to risk is thought to persist even as the most common types of diseases change over time. Social conditions, such as low SES and its associated lack of social and material resources, are therefore “fundamental causes” of disease (Link & Phelan, 1995, p. 80).

Furthermore, fundamental cause theory (FCT) postulates that the link between SES and disease will be dependent upon the type of disease in question. Diseases for which humans have effective prevention and intervention strategies will be more influenced by SES than diseases for which they do not. For example, conditions such as heart disease and diabetes are more determined by SES inequalities than pancreatic cancer or Lou Gehrig’s disease (Link & Phelan, 1995). As there are many social mechanisms that influence diabetes prevention and management, FCT predicts that higher SES will be protective against the onset of diabetes and aid in its management when it develops. Educational attainment and household income represent two important dimensions of SES that may provide the resources elaborated on by FCT. Educational attainment is usually completed in early adulthood, does not change, and acts as a resource across most of adulthood prior to the ages at which people are most at risk of developing diabetes. Income is a resource that tends to increase across adulthood and peak, on average, in middle age when people are most likely to experience the onset of diabetes (Herd et al., 2007).

Education and the Prevention of Diabetes

Education provides several unique benefits and resources that tend to delay the onset of chronic conditions such as diabetes, whereas income may have an important, but less primary role in prevention. Education is conferred early in the life course, before major health events typically occur. It shapes different dimensions of peoples’ lives throughout adulthood including their occupation, income, diet, leisure activities, and social networks (Mirowsky & Ross, 2015). Furthermore, education provides intrinsic and psychosocial resources that span the life course, such as planning skills, critical thinking, and self-efficacy (Leganger & Kraft, 2003; Mirowsky & Ross, 2015) that are not provided by income alone.

These resources may aid prevention more effectively than income because they allow individuals to override the “default American lifestyle” (Mirowsky & Ross, 2015), which is an institutionalized culture in Western society in which unhealthy, chronic disease promoting behaviors are the norm, default lifestyle. Pathologies from the long-term engagement in these “default” behaviors accumulate, producing chronic poor health conditions, including diabetes. Mirowsky and Ross (2015) argue that overriding the default American lifestyle requires human capital in the form of insight, self-direction, knowledge, critical analysis, personal agency, and implementation of long-term strategies. Furthermore, they argue that these skills are developed primarily through education, not income. Higher levels of education also tend to be associated with more stable work schedules, greater autonomy, and higher prestige (Schieman & Plickert, 2008), all of which may be protective for stress exposure across decades of the adult life course preceding disease onset. Higher levels of education may decrease the likelihood of shift work, which has been associated with increased stress and weight gain, two risk factors for diabetes onset (Suwazono et al., 2008). The cumulative influence of education may be particularly important for diabetes, which tends to develop over the course of several years or decades (Koopman, Mainous, Diaz, & Geesey, 2005).

Those with higher levels of education are less likely to be diagnosed with diabetes and to engage in poor diabetes-related health behaviors (Borrell, Dallo, & White, 2006; CDC, 2017; Pampel, Krueger, & Denney, 2010). For this reason, education may also serve to increase proactive health behaviors among individuals who are at risk of diabetes. Moreover, education shares stronger associations with health behaviors than other dimensions of SES (Winkleby, Jatulis, Frank, & Fortmann,1992; Ley, Hamdy, Mohan, & Hu, 2014).

Income and the Management of Diabetes

The social, cognitive, and symbolic resources enabled by education may become secondary in importance to the financial flexibility that income provides once diabetes onset occurs. While onset tends to occur after decades of accumulation of almost imperceptible risk through a variety of potential mechanisms, diabetes is known to be manageable through medication, monitoring glucose, and diet (Gorst et al., 2015). Several market goods requiring considerable financial means are needed to successfully manage diabetes, including quality health care, medications and testing supplies, nutritional counseling, and healthy foods (White et al., 2015). In 2013, 13.4% of the U.S. population was uninsured for the full year (Smith & Medalia, 2014), and lower rates of coverage are disproportionately found among those with lower incomes (Berchick, Hood, & Barnett, 2018). Moreover, low-income persons are more often limited to community health care settings, which tend to be plagued by staffing shortages, care provision by medical residents and nonspecialists, lack of continuity of care, and time constraints on providers (Luftey & Freese, 2005). These factors may contribute to less provider time with the patient and poorer communication, which has been shown to have a negative effect on medication compliance and hence diabetes management (White et al., 2015). Providers tend to prescribe regimes with tighter diabetes control (i.e., blood sugar levels) to higher SES patients (Luftey & Freese, 2005). This is noteworthy because tighter diabetes control reduces the risk of long-term diabetes complications. Moreover, income may facilitate transportation to health care providers and pharmacies (Syed, Gerber, & Sharp, 2013).

Next, patients with low incomes often face barriers to accessing their medications and testing supplies. The average medical expenditure for persons diagnosed with diabetes is US$16,750 per year, with about US$9,600 directly related to diabetes (ADA, 2018). Approximately half of the adults with diabetes report financial strain, and 14% report cost-related nonadherence (CRN) to treatment regimes (Patel, Piette, Resnicow, Kowalski-Dobson, & Heisler, 2016). The most common reason for intentional nonadherence to medication regimes among individuals with chronic disease is skipping doses to make them last longer due to affordability issues (Gadkari & McHorney, 2012). Individuals with CRN of medications have higher Hba1c levels and poorer physical and mental health outcomes (Piette, Wagner, Potter, & Schillinger, 2004). Moreover, the percentage of patients who engage in CRN appears to occur on a gradient by insurance coverage: 9% of VA patients, 18% of patients with private insurance, 25% of Medicare patients, 31% of Medicaid patients, and 40% of uninsured patients (Piette, Wagner, Potter, & Schillinger, 2004). This can accelerate disease progression and introduce complications. It also suggests that CRN is a concern among insured patients, as even state-funded health care for low-income people requires non-negligible copayments in obtaining medication (Powell, Saloner, & Sabik, 2016). There are also income-based disparities to types of medications and testing supplies, as newer, more expensive medications and supplies have better side effect profiles and are more user-friendly, which may facilitate continual and consistent use of diabetes medications (e.g., Thulé & Umpierrez, 2014).

Next, nutritional counseling and diabetes education enables patients to understand which foods will best help them manage their blood sugar, as well as the most beneficial ways to prepare them. This service is more readily available to patients who can afford transportation and time away from work to attend these programs. For patients with elevated Hba1c levels in the prediabetes range, educational programs have been found to delay impaired glucose tolerance and progression to diabetes (Kawahara et al., 2008).

Finally, the consumption of healthy low glycemic foods is an essential part of diabetes management and these types of foods tends to be relatively expensive. For instance, nutritious foods often cost more, and healthy perishable foods, such as fruits and vegetables, are less available and tend to have inflated costs in low-income areas (Hendrickson, Smith, & Eikenberry, 2006). Indeed, evidence shows that access to diabetes-managing foods is a significant barrier for individuals with low incomes (Breland, McAndrew, Gross, Leventhal, & Horowitz, 2013). Also, people may cope with the stress of poverty by consuming foods that are anathema to proper diabetes maintenance. Living in poverty may lead to repeated activation of the hypothalamus–pituitary–adrenal (HPA) axis which is linked to an increased appetite for calorically dense, high-fat, and high-sugar foods (Adam & Epel, 2007).

Furthermore, among the poor, food insecurity is also a particularly important obstacle for diabetes management. Food-insecure adults often shift their diets to cheap high-calorie, low-nutrient foods to maintain their caloric intake. Food insecurity has been linked to decreased self-efficacy, increased diabetes distress, decreased blood glucose monitoring, and increased hypoglycemia-related emergency department visits among diabetes patients (Ippolito et al., 2017; Seligman, Davis, Schillinger, & Wolf, 2010).

Hypotheses

Given theory and past evidence, we hypothesized two relationships with regard to education, income, diabetes prevention, and diabetes management. First, we hypothesized that educational attainment would share an inverse association with the onset, but not management, of diabetes. Second, we hypothesized that household income would share a positive association with the management of, but not onset, of diabetes.

Data, Method, and Analysis

Data

The Nashville Stress and Health Study (NSAHS) is a random sample of non-Hispanic Black and White men and women aged 25 to 65 years living in greater Nashville-Davidson County, Tennessee. The NSAHS was completed for the purpose of improving our understanding of Black–White and SES health disparities. These are ideal data to test our hypotheses because it provides information on both doctor-diagnosed diabetes and glucose management, Hba1c. This allows us to gauge both diagnostic status and deficits in disease management.

Data were gathered between April 2011 and January 2014 with the approval of the Vanderbilt University IRB. This analysis uses secondary data from NSAHS. Simple random sampling was used to draw 199 block groups within the county. Survey Sampling Corporation then provided 7,000 randomly selected addresses from these block groups. Due to budget constraints, 6,490 of these addresses received household screens for eligibility. Of those, 1,375 households refused, 1,411 households were ineligible due to age or race restrictions of the study, and 676 households were vacant or nonresidences. This left an eligible pool of 3,028 households, from which 2,400 households were randomly sampled. Only one individual per household was selected. Interviews were completed on 1,267 respondents (response rate = 66%). Previously published studies using these data excluded the 15 respondents with Hispanic or Other origin, but we included these given that racial/ethnic differences were not the focus of this study. The substantive findings presented in this study do not hinge on this decision, however. Interviews were conducted either in the respondent’s home or on Vanderbilt University campus. Trained interviewers conducted the interviews and were matched to respondents based on race. Respondents were offered US$50 to participate in the survey interview.

Separate clinical appointments were scheduled for measurements, blood draws, and urine sample collection. Of the 1,252 participants who completed the interview, 30 participants declined a clinical appointment, 44 agreed to a partial clinical appointment but declined a blood draw, and four participants completed a blood draw, but had insufficient Hba1c in their samples. This yielded 78 participants (6.4% of the full sample) without a valid Hba1c measure. They were excluded from our analyses. Independent t tests indicated that there was no significant difference in mean income or educational attainment between those with and without valid Hba1c measures for this excluded subsample who did not have Hba1c values and those who did have Hba1c values. Our final sample, therefore, included 1,189 participants. Twenty-seven participants (approximately 2.3%) had missing values for household income. These missing values were replaced using a regression-based imputation procedure (Landerman, Land, & Pieper, 1997).

Key Measures

Dependent Variables

Hba1c was based on assays taken from a venous blood sample. Hba1c gives the average level of blood glucose in the last 3 months (0%-17%). It is the primary and most common assessment tool used in diabetes clinical care and population research (National Institutes of Health [NIH], 2018). Hba1c is a measure of the percent of hemoglobin that is bound to a glucose molecule. The lifespan of hemoglobin is approximately 90 days; therefore, Hba1c can be used to provide a 3-month average of blood glucose. Hba1c levels of greater than 5.7% are indicative of prediabetes (Carson, Reynolds, Fonseca, & Muntner, 2010), and an Hba1c level of 6.5% or greater is currently the established threshold for a diabetes diagnosis (Pippitt, Li, & Gurgle, 2016).

To distinguish diabetes onset versus management, we utilized a self-report of whether the respondent had been diagnosed with diabetes using the following survey question: “Have you ever been diagnosed with diabetes by a doctor?” Participants could respond to this question with “Yes” or “No.” We then divided the sample into two groups based on diagnostic status reported by the respondent: participants who have not been diagnosed with diabetes by a physician (the undiagnosed group; n = 1,031) and participants who have been diagnosed with diabetes by a physician (the diagnosed group; n = 158).

Covariates

Education was measured by the highest degree completed. The following question from the survey was used: “What is the highest grade of school or year of college you successfully completed?” After participants provided a numeric grade level, the interviewer was prompted to ask whether the participant received a degree corresponding with the reported grade level. Responses were measured by the highest degree reported. For example, if a participant indicated that the highest grade completed was 12th, the interviewer was prompted to ask, “Did you graduate from high school?” Response choices were as follows: 1 = Yes, 2 = GED, 3 = No. An indicator variable was created for each degree level, with 1 = Yes and 0 = No. The following categories were used: “Less than high school,” “High school,” “Some college,” “College degree,” and “Graduate degree.”

To capture income, the following question was asked of respondents: “Please tell me the number that gives the best estimate of your total annual household income before taxes.” All ranges of values were then recoded at the midpoints with the exception of the highest income range, which does not have a midpoint and was therefore set at its starting value. The following household income values were generated from this: 0 = “No income,” 1 = “US$2,500,” 2 = “US$7,500,” 3 = “US$12,500,” 4 = “US$17,500,” 5 = “US$22,500,” 6 = “US$30,000,” 7 = “US$40,000,” 8 = “US$50,000,” 9 = “US$60,000,” 10 = “US$70,000,” 11 = “US$80,000,” 12 = “US$90,000,” 13 = “US$105,000,” 14 = “US$125,000,” 15 = “US$135,000+.” Finally, income was divided by 10,000 to improve interpretability of the regression results. We also dichotomized income at the 40th percentile (US$30,000 or less) to give a measure of whether the respondent had a low income or not.

Age was measured by the question, “How old are you?” Responses were given in years. Gender was indicated by whether the respondent identified as a man or woman. Race was measured with the following question: “With which of the groups shown do you most closely identify? 1 = Non-Hispanic White, 2 = Hispanic, 3 = African American, 4 = Other.”

Measures for Ancillary Analyses

We also utilized several additional covariates as robustness checks and ancillary analyses. The interview question for the chronic conditions measure was as follows: “This part of the interview is about any health problems you may have had. Have you ever had . . .” This question is followed by a series of chronic conditions for which the respondent answered “Yes” or “No.” If the respondent answered “Yes,” the question followed with whether the condition started during the past 12 months and whether it was diagnosed by a physician. The chronic conditions included serious or repeated pneumonia; other lung problems such as asthma, emphysema, or bronchitis, tuberculosis; arthritis; diabetes; high blood pressure; stomach problems or ulcers; kidney problems; migraines; stroke; high cholesterol; pancreatitis; anemia; and liver problems. We summed the number of chronic conditions reported by each respondent.

The NSAHS included five interview questions used to measure functional limitations. These questions asked, “Are you able to . . .” and followed with activities of daily living, such as “Reach up and get a five-pound object from just above your head?,” “Bend down to pick up an object from the floor?,” “Stoop or crouch down?,” “Walk a quarter of a mile?,” and “Sit for more than two hours?” Answers included “Easily,” “With some difficulty,” “With much difficulty,” and “Unable to do.” These variables were summed to produce our measure of functional limitations.

We calculated body mass index (BMI) with participants’ height and weight measures using the established CDC (2019) formula: weight (pounds)/(height [inches])² × 703. Seven participants had missing values for BMI, and regression imputation was used for these cases.

Analytical Strategy

First, we separated the sample into undiagnosed and diagnosed groups and tested for compositional differences in education, income, and other demographic characteristics between the two groups. Second, we tested for bivariate associations between all key study variables by diagnostic status group. We then employed separate ordinary least squares multivariable models for each group. These regressions tested for the association between Hba1c and both education and income after adjusting for statistical controls. Three models were estimated for each group. Model 1 included education and demographic controls and was used to assess the association between educational attainment and Hba1c. Model 2 included income and controls and was used to assess the association between income and Hba1c. Model 3 tested for key associations after adjusting for all covariates. Models 4 to 6 repeated these analyses for those with a diabetes diagnosis. All models adjusted for clustering by census tract. Finally, because many of our theoretical links between income and Hba1c concerned poverty-related issues, we re-ran our analyses using our measure of low income to test for a nonlinear relationship between income and Hba1c.

We then ran several ancillary analyses. Among those with a diabetes diagnosis, reverse causation was possible as elevated Hba1c may have led to increased chronic health conditions and functional limitations and, in turn, led to diminished income. We tested this possibility by including the number of chronic conditions and functional limitations in our final model (i.e., Model 6 in Table 2) and assessed how the income coefficient changed. To address the possibility that education and income operated differently between men and women and Blacks and Whites, we estimated our final models separately for these four demographics. To determine whether key associations varied by race or gender, we used a t test to compare regression coefficients across groups. Furthermore, as BMI is a theoretical pathway in the proposed relationships between income, education, and Hba1c, we tested for mediation via BMI using the SEM command in Stata.

Next, we conducted a sensitivity analysis by reclassifying respondents in the undiagnosed group who had Hba1c levels of 6.5% or higher into the diagnosed group and re-running our final multivariable model. We then compared how these results differed from those reported in our primary analysis. Finally, we tested to what extent our results were robust to our regression-based imputation strategy by replicating our analyses after employing listwise deletion and comparing the results to those reported in our study.

Results

Sample

The racial composition of the overall sample was 49.12% (n = 584) African American, 49.62% (n = 590) Non-Hispanic White, and 1.26% (n = 15) Hispanic or Other. The proportion of the overall sample who were men was 47.43%, and the proportion who were women was 52.57%. The average age of the overall sample was 45.92 years (SD = 11.47 years).

Table 1 provides descriptive statistics for the focal dependent and independent variables in our analysis by diagnostic group. Overall, those who have not been diagnosed with diabetes are more highly educated and have higher incomes than their counterparts with a diagnosis. The mean Hba1c of the undiagnosed group is 5.42% and is within the normal range. As expected, the diagnosis group has a higher mean Hba1c, 7.16%. A two-sample t test revealed there was a statistically significant difference in the scores for Hba1c between the undiagnosed group (M = 5.42, SD = 0.63) and diagnosed group (M = 7.16, SD = 1.78); (t = −23.34, p = .00). The full distribution of Hba1c for each group is provided in histograms in Appendix A. A chi-square test of independence also showed a significant difference between groups in educational attainment, as 25.99% of those who had not been diagnosed had a high school education and 34.18% of those who had been diagnosed had a high school education (χ² = 4.65; p = .03). Approximately 44% of the undiagnosed group had a college degree or higher, compared with 37% of the diagnosed group that had this level of education, although this difference was not statistically significant. The average household income of the undiagnosed group was US$56,320 (SD = 3,993) and was US$46,831 (SD = 3,605) for the diagnosed group (t = 2.82; p < .01). Notably, 22 individuals in the undiagnosed group had Hba1c levels above the clinical cutoff range of 6.5 or higher.

Table 1.

Descriptive Statistics for Undiagnosed (n = 1,031) and Diagnosed (n = 158) Participants.

	Undiagnosed			Diagnosed
	M/percentage	(SD)/no. of cases	Minimum/Maximum	M/Percentage	(SD)/no. of cases	Minimum/Maximum	p value
Hba1c	5.42	(0.63)	3.3/7.4	7.16	(1.78)	4.7/3.4	.00^a
Less than high school	10%	99		11%	17		.65
High school	26%	268		34%	54		.03^a
Some college	18%	181		17%	27		.89
College degree	29%	304		28%	45		.80
Graduate degree	15%	150		9%	14		.05^a
Income (in US$)	56,320	(3.99)	0/135,000+	46,831	(3.61)	0/135,000+	.00^a
Low income (below 30,000)
Yes	477	46%		62	39%		.00^a
No	554	54%		96	61%		.00^a
Race
Black	47%	484		63%	100		.00^a
White	52%	532		37%	58		.00^a
Other race	1.1%	12		0%	0		.17
Gender
Male	49%	504		38%	60		.01^a
Female	51%	527		62%	98		.01^a
Age in years	45.10	(11.43)	22/68	51.27	(10.24)	27/69	.00^a
Observations		1,031			158

Significance was assessed based on a two-sample t test or a chi-square test of independence.

Bivariate Relationships

A summary of the bivariate correlations among education, income, controls, and Hba1c levels is provided in Appendix B. Among those who have not been diagnosed, Hba1c was significantly correlated to the following focal variables: high school education (r = .157, p < .001), graduate degree (r = −.091, p < .01), and income (r = −.099, p < .01). Among those who have been diagnosed, Hba1c was only significantly correlated with income (r = −.169, p < .05).

OLS Regression Results

Table 2 provides the results of a series of ordinary least squares regression models that control for race, gender, and age. The first three models in the table show the OLS results for participants who have not been diagnosed with diabetes. Model 1 shows that education was inversely associated with Hba1c. Compared with those with a college degree, participants with a high school education had 0.134 higher Hba1c levels (p < .05) holding all other variables constants. Model 2 shows that income was unrelated to Hba1c. Model 3 shows that the association between education and Hba1c persisted after the inclusion of household income to the model (β = .127; p < .05). The magnitude of this association was modest. For example, it was 48.3% of the size of the association between race and Hba1c (β = .263; p < .001)—a well-known demographic predictor of Hba1c. Respondents with less than a high school education, some college, or a graduate degree had similar levels of Hba1c as those with a college degree.

Table 2.

Hba1c Levels by Education and Income for Undiagnosed and Diagnosed Participants With and Without a Diabetes Diagnoses.

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
	b/(SE)	b/(SE)	b/(SE)	b/(SE)	b/(SE)	b/(SE)
Less than high school	−0.01 (0.05)		−0.02 (0.05)	0.54 (0.70)		0.21 (0.71)
High school	0.13** (0.06)		0.13* (0.05)	0.08 (0.35)		−0.11 (0.37)
Some college	−0.02 (0.04)		−0.02 (0.04)	0.28 (0.38)		0.23 (0.38)
Graduate degree	−0.04 (0.03)		−0.04 (0.04)	−0.44 (0.59)		−0.43 (0.62)
Age	0.01*** (0.00)	0.01*** (0.00)	0.01*** (0.00)	−0.01 (0.01)	0.01 (0.02)	0.01 (0.01)
Male	0.08* (0.04)	0.09** (0.04)	0.09* (0.04)	0.04 (0.30)	0.10 (0.28)	0.04 (0.30)
Black	0.26*** (0.04)	0.26*** (0.04)	0.26*** (0.04)	−0.01 (0.26)	−0.36 (0.35)	−0.38 (0.35)
Income 10,000		−0.01 (0.00)	−0.01 (0.00)		−0.11** (0.05)	−0.11* (0.05)
Constant	4.68*** (0.06)	4.71*** (0.06)	4.70*** (0.06)	7.21*** (0.79)	7.63*** (0.79)	7.65*** (0.84)

Note. Referent groups: College degree, White and Other race, unmarried, and female.

p < .05. **p < .01. ***p < .001.

Models 4 to 6 in Table 2 show results of the OLS regression for those who have been diagnosed with diabetes. Model 4 shows that education was unrelated to Hba1c. Model 5 shows that respondents with higher incomes tended to have lower levels of Hba1c. For every US$10,000 increase in income, Hba1c level decreased by 0.111 (p < .05). Model 6 shows this association persisted after including all study covariates.

We found support for a nonlinear relationship between income and Hba1c when we re-ran our models with our dichotomous measure of low income (see Table 3) instead of our continuous measure of income. Those with a low income had Hba1c scores that were 0.85 higher than their non-disadvantaged counterparts (p < .05). Having a low income was unrelated to Hba1c among the undiagnosed group. We also included linear and quadratic income terms in our models (not shown). These models provided no support for a quadratic relationship between income and Hba1c.

Table 3.

Hba1c Levels by Education and Dichotomized Income for Undiagnosed and Diagnosed Participants.

	No diabetes diagnosis			Diabetes diagnosis
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
	b/(SE)	b/(SE)	b/(SE)	b/(SE)	b/(SE)	b/(SE)
Less than high school	−0.01 (0.05)		−0.02 (0.05)	0.54 (0.70)		0.24 (0.67)
High school	0.13** (0.06)		0.13* (0.05)	0.08 (0.35)		−0.04 (0.35)
Some college	−0.02 (0.04)		−0.02 (0.04)	0.28 (0.38)		0.21 (0.36)
Graduate degree	−0.04 (0.03)		−0.04 (0.04)	−0.44 (0.59)		−0.40 (0.61)
Age	0.01*** (0.00)	0.01*** (0.00)	0.01*** (0.00)	−0.00 (0.02)	0.00 (0.01)	0.00 (0.01)
Male	0.08* (0.04)	0.09** (0.04)	0.09* (0.04)	0.04 (0.30)	0.05 (0.27)	−0.00 (0.30)
Black	0.26*** (0.04)	0.27*** (0.04)	0.26*** (0.04)	−0.01 (0.26)	−0.26 (0.28)	−0.27 (0.28)
Low income (below 30,000)		0.04 (0.05)	0.02 (0.05)		0.90** (0.34)	0.85* (0.34)
Constant	4.68*** (0.06)	4.66*** (0.06)	4.67*** (0.06)	7.21*** (0.79)	6.76*** (0.71)	6.82*** (0.71)

Note. Referent groups: College degree, White and Other race, unmarried, and female.

p < .05. **p < .01. ***p < .001.

Robustness Checks and Ancillary Analyses

We performed several robustness checks and ancillary analyses. First, we included measures of number of chronic conditions and functional limitations in our final model. The income coefficient did not substantively change when these measures were included, suggesting that these variables were not influencing the relationship between income and Hba1c. Next, we tested whether any of our key associations varied by race or gender. They did not. Next, we reassigned the 22 respondents who had high Hba1c levels to the diagnosed group and retested whether income was associated with Hba1c for the undiagnosed (n = 1,008) and diagnosed groups (n = 154). Education was no longer a predictor of Hba1c for the undiagnosed group, but income remained inversely associated with Hba1c for the diagnosed group (p < .05).

To examine the possibility that the decision to use regression-based imputation for missing income values might produce different results than listwise deletion, we re-ran the analysis without the 27 respondents with missing income values. Across both the original analysis and the analysis that used listwise deletion for missing income, the following similarities in results were found: (a) all statistically significant relationships among focal variables remained, (b) no new statistically significant relationships emerged among focal variables, and (c) all coefficients remained in the same direction. This suggested that the decision to use regression-based imputation, rather than listwise deletion, did not substantively influence our findings.

Finally, we tested whether BMI mediated any of our focal associations. For the undiagnosed group, the total association between high school education on Hba1c was .131 (p < .01). The direct association between high school education and Hba1c was .127 (p < .01). The indirect association between high school education and Hba1c that operated through BMI was .003 and was not statistically different from zero. This suggested that BMI was not a significant mediator in the relationship between high school education and Hba1c. For the diagnosed group, the total association between income (measured continuously) and Hba1c was −.11 (p < .05). The direct association between income and Hba1c was −.11 (p < .05). The indirect association between income and Hba1c that operated through BMI was −.00 and was not statistically different from zero. This suggested that BMI was not a significant mediator in the relationship between income and Hba1c. Similar results were found with low income.

Conclusion

This study tested two central hypotheses. First, we tested the hypothesis that educational attainment would share an inverse association with Hba1c among individuals who were never diagnosed with diabetes but not among those diagnosed with diabetes. We found support for this hypothesis, as those with a college degree had lower levels of Hba1c than those with a high school degree among the undiagnosed group. This association was not found among those in the diagnosed group. These results were consistent with the findings of Herd et al. (2007), who argued that education, as opposed to income, is most closely linked with the health behaviors, cognitive, and psychosocial resources that are advantageous in avoiding the onset of chronic disease.

Second, we tested the hypothesis that household income would share an inverse association with Hba1c among those previously diagnosed with diabetes but not among those that had never been diagnosed. This hypothesis was also supported as income was inversely associated with Hba1c level in the diagnosed group but not the undiagnosed group. These results were also consistent with the work of Herd et al. (2007), which found that income is more important in the management of chronic diseases and speculated that it may influence one’s ability to obtain the necessary resources to manage the disease after onset. Furthermore, our analyses suggested that this association was disproportionately driven by those with incomes below US$30,000.

Our results point toward three important implications. First, these data suggest that the prevention of diabetes appears closely linked to educational attainment. The resources conferred by education appear to be important in delaying the cumulative adverse health events that precipitate the onset of diabetes. While our investigation did not assess which education-specific resources are important in delaying onset, it did suggest that the aggregate effect of these resources is different from the aggregate effect of the resources available from income in the time period preceding diabetes onset. Consistent with FCT, these data suggest that targeted intervention programs to prevent diabetes onset may be inadequate at addressing disparities, as there may be more persistent distal structural barriers that increase poor health behaviors, work and life distress, and other circumstances that increase the likelihood of onset of diabetes in individuals with lower educational levels.

Second, education and income appear to confer unique resources that are utilized differently depending on the stage of disease progression. Similarly, it is important for the onset and management of diabetes to be treated as two separate constructs, as failing to do so may lead to inaccurate results. For example, research that examines the connection between SES and Hba1c, but fails to account for diabetes status may underestimate the effects of educational attainment, especially in samples with large numbers of people who are diagnosed with diabetes.

Third, our finding that low income operates as a restraint on the effective management of diabetes is particularly meaningful, as many medications and monitoring tools have become more costly over time (Hua et al., 2016; Landon, Zaslavsky, Souza, & Ayanian, 2018; Lipska et al., 2014). For instance, the price of insulin has tripled in the last decade and this increase in cost has been linked to medical noncompliance (Herkert et al., 2018). The rising cost of effective diabetes management mixed with our study results portend to a future where income-based disparities in the effective treatment of diabetes will widen.

Surprisingly, we found that those who had less than a high school education did not have Hba1c levels that differed from their college-educated counterparts. This may have been due to limited statistical power for this educational group due to its low total number of participants in the overall sample (n = 116), and low number of participants by diagnosis (diagnosed = 17; undiagnosed = 99). It is also plausible that mortality selection occurred disproportionately among those without a high school education.

We showed that 22 individuals in our study surpassed the clinical Hba1c threshold (>6.5%) but did not report ever having been diagnosed with diabetes. This phenomenon points to the complexity in the relationship between SES, Hba1c, and diagnosed diabetes as the majority of these 22 individuals had low levels of educational attainment and income. Low education and income may have influenced diagnosis rates by leading people to see health care providers less regularly or by having poorer access to health care providers and/or diagnostic assessments.

There are several limitations to this study. First, we could not differentiate between type 1 and types 2 diabetes in the data. However, type 1 diabetes has very low rates, 2.6 to 3.4 individuals per 1,000 in the United States (Menke et al., 2013), so approximately two to three individuals in our sample may have had type 1 instead of type 2. Our results were unlikely to have been due to measurement error in diabetes diagnosis, however, given these small numbers. Also, type 1 diabetes requires substantial financial resources to manage, so we would expect the relationship between income and Hba1c to be similar for those with type 1 and type 2 diabetes. Next, our data were from the Nashville-Davidson County area, and external validity beyond this area remains a concern. Conversely, however, restricting data to a single city increases the internal validity of the study by holding constant all the complexities of place and geography. Replicating the study in other locations would serve to address external validity concerns.

Next, the subsample of individuals who were diagnosed with diabetes was relatively small (n = 158). Given that diabetes affects approximately 9.4% of the general population (ADA, 2017), a sample of 158 indicates diabetes diagnoses are slightly overrepresented in this sample (i.e., 9.4% of 1,187 is 111). A small sample size could have affected the analysis in a few important ways. First, it limits the statistical power available for the diagnosed group. It can lead to overestimates of effect size, as well as decreasing external validity. However, these limitations apply only to the diagnosed group. Next, reverse causality is possible in the association between income and Hba1c among those with diabetes. This limitation is tempered, however, by our ancillary analyses that showed that our key findings did not differ after adjusting for number of chronic conditions and functional limitations.

Finally, because 22 respondents had elevated Hba1c levels indicative of diabetes but did not report a diagnosis, our results have the potential for bias due to this misclassification. This limitation is mitigated, however because our sensitivity analyses showed income was inversely related to Hba1c among the diagnosed group regardless of how we classified these 22 respondents. While education was no longer inversely associated with Hba1c among the nondiagnosed group after we removed these 22 cases, we contend this finding is consistent with our conceptual framework as long as these 22 were not being treated for and actively managing their diabetes. In short, we would expect that low education would lead to elevated Hba1c over time and that a small number of people would not be aware that they have crossed the threshold from nondiabetic to diabetic. By removing them from the analysis of the nondiagnosed group, we ostensibly removed the most influential cases in our analysis. To the extent that these 22 cases had been diagnosed with diabetes but did not report it, our results regarding education among the nondiagnosed group will be biased. This latter concern, however, should be interpreted in light of past research that consistently finds that those with lower levels of education are more likely to develop diabetes than their more educated counterparts (e.g., Beckles et al., 2013; CDC, 2017).

Future research might focus on learning more about the influence of specific resources of education and income on diabetes onset and management and investigate how non-interchangeable SES resources operate vis-à-vis other chronic conditions (e.g., hypertension and chronic obstructive pulmonary disease). Similar to diabetes, other chronic conditions tend to develop following the slow accruement of risk, are stratified by education, and require financial resources to successfully manage. In short, this study provided evidence that SES-based social determinants of health vary based on individual dimension of SES and whether disease onset has occurred.

Footnotes

Appendix A

Appendix B

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Kristin Allen

References

Adam

T. C.

Epel

E. S.

(2007). Stress, eating and the reward system. Physiology & behavior, 91(4), 449-458.

Adler

N. E.

Newman

(2002). Socioeconomic disparities in health: Pathways and policies. Health Affairs, 21, 60-76.

American Diabetes Association. (2017). Statistics about diabetes. Retrieved from https://www.diabetes.org/resources/statistics/statistics-about-diabetes

American Diabetes Association. (2018). Economic costs of diabetes in the US in 2017. Diabetes Care, 41, 917-928.

Beckles

G. L.

Chou

C. F.

, & Centers for Disease Control Prevention. (2013). Diabetes—United States, 2006 and 2010. MMWR Surveillance Summaries, 62(Suppl. 3), 99-104. Retrieved from https://www.cdc.gov/mmwr/preview/mmwrhtml/su6203a17.htm

Berchick

E. R.

Hood

Barnett

J. C.

(2018). Health insurance coverage in the United States: 2017 (Current Population Reports P60-264). Washington, DC: U.S. Government Printing Office.

Borrell

L. N.

Dallo

F. J.

White

(2006). Education and diabetes in a racially and ethnically diverse population. American Journal of Public Health, 96, 1637-1642.

Breland

J. Y.

McAndrew

L. M.

Gross

R. L.

Leventhal

Horowitz

C. R.

(2013). Challenges to healthy eating for people with diabetes in a low-income, minority neighborhood. Diabetes Care, 36, 2895-2901.

Carson

A. P.

Reynolds

Fonseca

V. A.

Muntner

(2010). Comparison of A1C and fasting glucose criteria to diagnose diabetes among US adults. Diabetes Care, 33, 95-97.

10.

Centers for Disease Control and Prevention. (2017). National Diabetes Statistics Report, 2017. Atlanta, GA: Centers for Disease Control and Prevention, U.S. Dept of Health and Human Services.

11.

Centers for Disease Control and Prevention. (2019). About adult BMI. Retrieved from https://www.cdc.gov/healthyweight/assessing/bmi/adult_bmi/index.html#Definition

12.

Cutler

D. M.

Lleras-Muney

Vogl

(2008). Socioeconomic status and health: Dimensions and mechanisms (No. w14333). Cambridge, MA: National Bureau of Economic Research.

13.

Gadkari

A. S.

McHorney

C. A.

(2012). Unintentional non-adherence to chronic prescription medications: How unintentional is it really? BMC Health Services Research, 12, Article 98.

14.

Goldman

Lin

I. F.

Weinstein

Lin

Y. H.

(2003). Evaluating the quality of self-reports of hypertension and diabetes. Journal of Clinical Epidemiology, 56, 148-154.

15.

Gonzalez

J. S.

Schneider

H. E.

(2011). Methodological issues in the assessment of diabetes treatment adherence. Current Diabetes Reports, 11, 472-479.

16.

Gorst

Kwok

C. S.

Aslam

Buchan

Kontopantelis

Myint

P. K.

. . .Mamas

(2015). Long-term glycemic variability and risk of adverse outcomes: A systematic review and meta-analysis. Diabetes Care, 38, 2354-2369.

17.

Hendrickson

Smith

Eikenberry

(2006). Fruit and vegetable access in four low-income food deserts communities in Minnesota. Agriculture and Human Values, 23, 371-383.

18.

Herd

Goesling

House

J. S.

(2007). Socioeconomic position and health: The differential effects of education versus income on the onset versus progression of health problems. Journal of Health and Social Behavior, 48, 223-238.

19.

Herkert

Vijayakumar

Luo

Schwartz

J. I.

Rabin

T. L.

DeFilippo

Lipska

K. J.

(2018). Cost-related insulin underuse among patients with diabetes. JAMA Internal Medicine, 179, 112-114.

20.

Hua

Carvalho

Tew

Huang

E. S.

Herman

W. H.

Clarke

(2016). Expenditures and prices of antihyperglycemic medications in the United States: 2002-2013. The Journal of the American Medical Association, 315, 1400-1402.

21.

Ippolito

M. M.

Lyles

C. R.

Prendergast

Marshall

M. B.

Waxman

Seligman

H. K.

(2017). Food insecurity and diabetes self-management among food pantry clients. Public Health Nutrition, 20, 183-189.

22.

Jackson

J. M.

DeFor

T. A.

Crain

A. L.

Kerby

T. J.

Strayer

L. S.

Lewis

C. E.

. . .Larson

J. C.

(2014). Validity of diabetes self-reports in the Women’s Health Initiative. Menopause, 21, 861-868.

23.

Kawahara

Takahashi

Inazu

Arao

Kawahara

Tabata

. . .Tanaka

(2008). Reduced progression to type 2 diabetes from impaired glucose tolerance after a 2-day in-hospital diabetes educational program: The Joetsu Diabetes Prevention Trial. Diabetes Care, 31, 1949-1954.

24.

Koopman

R. J.

Mainous

A. G.

Diaz

V. A.

Geesey

M. E.

(2005). Changes in age at diagnosis of type 2 diabetes mellitus in the United States, 1988 to 2000. The Annals of Family Medicine, 3, 60-63.

25.

Landerman

L. R.

Land

K. C.

Pieper

C. F.

(1997). An empirical evaluation of the predictive mean matching method for imputing missing values. Sociological Methods & Research, 26, 3-33.

26.

Landon

B. E.

Zaslavsky

A. M.

Souza

Ayanian

J. Z.

(2018). Trends in diabetes treatment and monitoring among Medicare beneficiaries. Journal of General Internal Medicine, 33, 471-480.

27.

Leganger

Kraft

(2003). Control constructs: Do they mediate the relation between educational attainment and health behaviour? Journal of Health Psychology, 8, 361-372.

28.

Ley

S. H.

Hamdy

Mohan

F. B.

(2014). Prevention and management of type 2 diabetes: Dietary components and nutritional strategies. The Lancet, 383, 1999-2007.

29.

Link

B. G.

Phelan

(1995). Social conditions as fundamental causes of disease. Journal of Health and Social Behavior, 35, 80-94.

30.

Link

B. G.

Phelan

J. C.

Miech

Westin

E. L.

(2008). The resources that matter: Fundamental social causes of health disparities and the challenge of intelligence. Journal of Health and Social Behavior, 49, 72-91.

31.

Lipska

K. J.

Ross

J. S.

Van Houten

H. K.

Beran

Yudkin

J. S.

Shah

N. D.

(2014). Use and out-of-pocket costs of insulin for type 2 diabetes mellitus from 2000 through 2010. The Journal of the American Medical Association, 311, 2331-2333.

32.

Luftey

Freese

(2005). Toward some fundamentals of fundamental causality: Socioeconomic status and health in the routine clinic visit for diabetes. American Journal of Sociology, 110, 1326-1372.

33.

Menke

Orchard

T. J.

Imperatore

Bullard

K. M.

Mayer-Davis

Cowie

C. C.

(2013). The prevalence of type 1 diabetes in the United States. Epidemiology, 24, 773-774.

34.

Mirowsky

Ross

C. E.

(2015). Education, health, and the default American lifestyle. Journal of Health and Social Behavior, 56, 297-306.

35.

National Institute of Health. (2018). The A1C test & diabetes. Retrieved from: https://www.niddk.nih.gov/health-information/diabetes/overview/tests-diagnosis/a1c-test

36.

Pampel

F. C.

Krueger

P. M.

Denney

J. T.

(2010). Socioeconomic disparities in health behaviors. Annual Review of Sociology, 36, 349-370.

37.

Patel

M. R.

Piette

J. D.

Resnicow

Kowalski-Dobson

Heisler

(2016). Social determinants of health, cost-related nonadherence, and cost-reducing behaviors among adults with diabetes: Findings from the national health interview survey. Medical Care, 54, 796-803.

38.

Piette

J. D.

Wagner

T. H.

Potter

M. B.

Schillinger

(2004). Health insurance status, cost-related medication underuse, and outcomes among diabetes patients in three systems of care. Medical care, 102-109.

39.

Pippitt

Gurgle

H. E.

(2016). Diabetes mellitus: Screening and diagnosis. American Family Physician, 93, 103-109.

40.

Powell

Saloner

Sabik

L. M.

(2016). Cost sharing in Medicaid: Assumptions, evidence, and future directions. Medical Care Research and Review, 73, 383-409.

41.

Schieman

Plickert

(2008). How knowledge is power: Education and the sense of control. Social Forces, 87, 153-183.

42.

Seligman

H. K.

Davis

T. C.

Schillinger

Wolf

M. S.

(2010). Food insecurity is associated with hypoglycemia and poor diabetes self-management in a low-income sample with diabetes. Journal of Health Care for the Poor and Underserved, 21, 1227-1233.

43.

Smith

J. C.

Medalia

(2014). Health insurance coverage in the United States: 2013 (Current Population Reports, P60-250). Washington, DC: U.S. Census Bureau.

44.

Suwazono

Dochi

Sakata

Okubo

Oishi

Tanaka

. . .Nogawa

(2008). A longitudinal study on the effect of shift work on weight gain in male Japanese workers. Obesity, 16, 1887-1893.

45.

Syed

S. T.

Gerber

B. S.

Sharp

L. K.

(2013). Traveling towards disease: Transportation barriers to health care access. Journal of Community Health, 38, 976-993.

46.

Thulé

P. M.

Umpierrez

(2014). Sulfonylureas: A new look at old therapy. Current Diabetes Reports, 14, 473.

47.

White

R. O.

Eden

Wallston

K. A.

Kripalani

Barto

Shintani

Rothman

R. L.

(2015). Health communication, self-care, and treatment satisfaction among low-income diabetes patients in a public health setting. Patient Education and Counseling, 98, 144-149.

48.

Winkleby

M. A.

Jatulis

D. E.

Frank

Fortmann

S. P.

(1992). Socioeconomic status and health: how education, income, and occupation contribute to risk factors for cardiovascular disease. American journal of public health, 82(6), 816-820.