Depression is a common comorbidity among people with diabetes. In a recent meta-analysis, the prevalence of depression among people with diabetes was about twice as high as that among those without diabetes (1
). Depression among people with diabetes reduces quality of life and is associated with morbidity, mortality, and health care costs (2
While there is a substantial body of research on comorbid depression among individuals with diabetes, significant gaps exist in the literature. First, little is known about this association among the rural elderly. This population constitutes >25% of the total U.S. elderly population and has limited access to health care, particularly specialty health care (5
), mental health services, and other resources necessary for appropriate chronic disease management (6
). Second, there is limited information on comorbid depression among ethnic minorities with diabetes, particularly African Americans (9
) and Native Americans. Diabetes is more common among ethnic minorities (10
), so the public health impact of depression in this population may be substantial.
The current study has two major aims: to assess the prevalence of depressive symptoms among a sample of older African Americans, Native Americans, and whites with diabetes living in rural communities; and to determine the demographic and health characteristics associated with depressive symptoms in this sample. Comparisons of the prevalence and correlates of depressive symptoms in this sample were therefore made to relevant literature.
The ELDER (Evaluating Long-term Diabetes Self-management Among Elder Rural Adults) Study, a 4-year study funded by the National Institute on Aging and the National Center for Minority Health and Health Disparities, is a population-based cross-sectional survey designed to comprehensively assess the self-care strategies of older rural adults (aged ≥65 years) with diagnosed diabetes and the impact of these strategies on diabetes control. Participants for the study were selected from two counties in central North Carolina. These counties were selected because they are largely rural, they have large numbers of ethnic minorities and individuals living below the poverty line, and the investigative team had previously developed strong ties in these communities. The study began in 2001, with recruitment of participants conducted from May through October 2002. The study was approved by the institutional review board of the Wake Forest University School of Medicine.
The study recruited a random sample of community-dwelling older adults with diabetes, including African-American, Native-American, and white men and women. A sampling frame was selected using claims records from the Centers for Medicare and Medicaid Services (CMS). Individuals were included in the sampling frame if they were ≥65 years of age, a resident of one of the study counties, and had at least two outpatient claims for diabetes (coded 250 [diabetes] in the International Classification of Diseases, 9th Revision) in 1998–2000. Random samples of men and women were selected. Letters were sent from CMS and from the study team requesting participation in the study. The letters were followed with a phone call or a personal home visit from an interviewer on the study team to further assess eligibility (confirming diabetes status and residence in study counties and that the individual is physically and mentally able to participate in the survey) and willingness to participate in the study.
The final sample consisted of 701 individuals. Of the 1,222 people contacted, 313 were disqualified because they reported that they did not have diabetes (n = 118), lived out of the study counties (n = 51), lived in a nursing home (n = 84), were not ≥65 years of age (n = 2), did not speak English (n = 1), or failed the Mini-Mental State Exam (n = 5), and 52 were deceased. We were unable to assess the eligibility of an additional 122 people because a surrogate refused participation in study (n = 48), they were physically (n = 8) or mentally (n = 14) unable to respond to eligibility questions, or they could not be located (n = 52). For those who met the eligibility criteria after contact, 86 were not interviewed because they refused participation (n = 74) or study staff determined that the participant was physically (n = 6) or mentally (n = 6) unable to participate. The overall response rate for known eligible participants was 89% (701 of 787). A total of 696 participants were used for this analysis. Three participants who did not fit the ethnic categories and an additional two for whom the Center for Epidemiologic Study of Depression (CES-D) scale score could not be calculated due to missing data were excluded. Because of missing data on specific interview items, the sample sizes for some analyses are reduced.
Face-to-face interviews were conducted by local, trained interviewers. Participation in the study involved a 1.5-h interview. Interview data were recorded on paper forms, entered into EpiInfo (version 6.0; Centers for Disease Control and Prevention, Atlanta, GA), and analyzed using SAS statistical software (version 8.2; SAS Institute, Cary, NC). The survey instrument included well-established standardized scales as well as items developed and pilot tested by the investigators. Variables used in this report included sex, ethnicity, age, marital status, highest level of formal education, annual household income, Medicaid status, number of people living in the home, duration of diabetes, BMI, glycemic control (HbA1c
), current use of diabetes medications, total number of prescription medications, and history of chronic health conditions. The quality of life measure was the Physical Component Summary (PCS) score sub-scale of the SF-12 (11
). Higher scores for this measure indicate higher physical functioning. Glycemic control was assessed by measurement of HbA1c
from fingerstick blood samples collected in a capillary tube, stored in the AccuBase A1c kit (Diabetes Technologies, Thomasville, GA), and shipped to Premiere Laboratories (Kansas City, MO) for HbA1c
assessment. A second tube was collected on a random sample of 10% of participants for quality control. The intraclass coefficient for this analysis was 0.996 (95% CI 0.994-0.998, n
The outcome variable was assessed by the CES-D, a 20-item self-report depressive symptoms scale (12
). The version of the CES-D scale used was validated in the Duke Established Populations for Epidemiological Studies of the Elderly (EPESE) (13
), in which the response categories for symptoms experienced in the previous week were modified from the original Likert scale (all of the time, most of the time, some of the time, and none of the time) to “yes” and “no” responses. This approach was chosen by the investigative team after careful consideration of the difficulty experienced by older adults, particularly those with low levels of formal education, in responding to Likert-type questions. Because the Duke EPESE cohort is very similar to the one in the current study (half rural and over half African American), the CES-D modification validated in this major study of elderly was judged to be appropriate. “Yes” responses were scored as 1 and “no” responses were scored as 0, with a range of scores from 0 to 20. A value ≥9 was used to define the threshold for significant depressive symptoms. This has been determined by Blazer et al. (13
) (through comparisons with the Yale EPESE sample and the Iowa 65+ Rural Health Study) to be equivalent to the cut point of ≥16 traditionally used with the original scoring method, which produces scores of 0–60.
Demographic and health characteristics and CES-D scores were summarized using counts and percentages. Age was treated as a continuous variable. Marital status was dichotomized as married or not married. Education was categorized as less than high school, a high school degree, or at least some college. Poverty status was categorized into three groups: receiving Medicaid, not receiving Medicaid and annual household income <$25,000/year, and not receiving Medicaid and annual household income ≥$25,000. Number of people in the household was classified as one, two, or three or more. Because of the collinearity between marital status and number of people in the household, these variables were combined to form a living arrangement variable: living alone, living with others and unmarried, and living with others and married. Duration of diabetes was treated as a continuous variable. Diabetes medication was categorized as none, on oral agents, or on insulin with or without oral agents. The total number of prescription medications and chronic conditions were dichotomized as above or below the median value in the sample for each variable (≤5 or >5 for both variables). The physical component score was treated as a continuous variable. CES-D scores were dichotomized as 0–8 and 9–20 (depressive symptoms) based on the previous classification of Blazer et al. (13
Associations between CES-D and independent variables were evaluated for statistical significance using regression modeling. Multiple logistic regression models were used to evaluate potential predictors of CES-D. Significance tests were performed for sex × ethnicity interactions, controlling for sex, ethnicity, age, level of education, living arrangement, poverty status, diabetes duration (decades), BMI, HbA1c, diabetes medication group, number of prescription medications, number of chronic conditions, and physical functioning (PCS ÷ 10). If a sex × ethnicity term was significant (P ≤ 0.05), then significance tests were performed among the three ethnic groups for all pairwise comparisons of odds ratios (ORs) for practitioner use in women versus men. If a sex × ethnicity term was nonsignificant, then the interaction term was dropped from the model and significance tests were performed for main effects of sex and ethnicity. If an ethnicity term was significant, then significance tests were performed for all pairwise comparisons among the three ethnic groups. Pairwise comparison results for the effects of potential predictors having more than two groups were evaluated using Bonferroni's method.