Depressive Symptoms, Antidepressant Use, and the Incidence of Diabetes in the Black Women’s Health Study

Abstract

OBJECTIVE

To assess the relationship of depressive symptoms and use of antidepressants with incident type 2 diabetes in prospective data from a large cohort of U.S. African American women.

RESEARCH DESIGN AND METHODS

The Black Women’s Health Study (BWHS) is an ongoing prospective cohort study. We followed 35,898 women from 1999 through 2011 who were without a diagnosis of diabetes and who had completed the Center for Epidemiologic Studies Depression Scale (CES-D) in 1999. CES-D scores were categorized as <16, 16–22, 23–32, and ≥33, which reflected increasingly more depressive symptoms. We estimated incidence rate ratios (IRRs) and 95% CIs for incident diabetes using Cox proportional hazards models. The basic multivariable model included age, time period, family history of diabetes, and education. In further models, we controlled for lifestyle factors and BMI. We also assessed the association of antidepressant use with incident diabetes.

RESULTS

Over 12 years of follow-up, there were 3,372 incident diabetes cases. Relative to CES-D score <16, IRRs (95% CI) of diabetes for CES-D scores 16–22, 23–32, and ≥33 were 1.23 (1.12–1.35), 1.26 (1.12–1.41), and 1.45 (1.24–1.69), respectively, in the basic multivariate model. Multiple adjustment for lifestyle factors and BMI attenuated the IRRs to 1.11 (1.01–1.22), 1.08 (0.96–1.22), and 1.22 (1.04–1.43). The adjusted IRR for antidepressant use was 1.26 (1.11–1.43). Results were similar among obese women.

CONCLUSIONS

Both depressive symptoms and antidepressant use are associated with incident diabetes among African American women. These associations are mediated in part, but not entirely, through lifestyle factors and BMI.

Introduction

The prevalence of type 2 diabetes among African American women is double that of white women and nearly 40% greater than that of African American men (1). Obesity is the strongest modifiable risk factor for diabetes (2) and has its highest prevalence among African American women (3), who are also less likely to exercise than people of other ethnic groups (4). Previous studies have also identified a moderately strong association between depression and incident diabetes (5–13), although whether this relationship is mediated fully by obesity and lifestyle factors is not clear. Higher levels of inflammation (14) and dysregulation of the hypothalamic-pituitary-adrenal axis (15) are proposed to lie on the causal pathway between depression and incident diabetes and may explain some of the residual risk after multivariable adjustment that has been noted in prior work (7).

Despite the disproportionate burden of diabetes among African American women, prospective studies have shed little light on the relationship between depression and incident diabetes among African American women. Everson-Rose et al. (16) reported that in a 3-year multiethnic study, the adjusted association between depression and incident diabetes was strongest among African Americans. However, other studies conducted in the U.S. provide very few or no data on African Americans (5,6). Still other studies that include African Americans in larger numbers do not report adjusted estimates separately for African American women (7–9,11,13,17–19). In addition, although the risk of diabetes associated with antidepressants may be at least as high as that associated with depressive symptoms (13,20–22), we are unaware of investigations that explore the association between antidepressant use and incident diabetes among African American women.

In the current study, we assessed the relationship of depressive symptoms and antidepressant medication use to incident type 2 diabetes using prospective data from a large cohort of African American women. We also examined whether obesity and lifestyle factors are potential mediators of the relationship.

Research Design and Methods

Study Population

We used data from the Black Women’s Health Study (BWHS), an ongoing prospective cohort study of African American women from across the U.S. The BWHS was designed to investigate the determinants of health and disease in African American women. In 1995, women enrolled by responding to questionnaires mailed to subscribers of Essence magazine (a popular publication targeted to African American women), to members of several African American professional organizations, and to friends or relatives of early respondents. The final cohort included 59,000 women, aged 21–69 years at baseline in 1995 (23). The study protocol was approved by the institutional review boards of Boston University and Howard University, Washington, DC.

Participants completed a baseline questionnaire that included information on adult height, current weight, demographic characteristics, reproductive history, medical history, medication use, smoking, alcohol use, and diet. Every 2 years, mailed questionnaires collect updated information on lifestyle factors, other exposures, and medical problems, including diabetes. After eight cycles of follow-up to date, follow-up of the baseline cohort is 80%.

Questions about depressive symptoms were first asked in the 1999 questionnaire, and thus the present analyses are based on follow-up from 1999 through 2011. We excluded women with diabetes in 1999 (i.e., baseline in the present analysis) (n = 4,312), with diabetes diagnosed at age ≤30 years (n = 28), or with missing data on depressive symptoms at baseline (n = 16,564). We also excluded women with history of cancer (n = 1,424), myocardial infarction (n = 427), coronary artery bypass graft surgery (n = 146), and/or stroke (n = 430) at baseline. This resulted in a final analytic sample of 35,898 women.

Ascertainment of Depressive Symptoms and Antidepressant Use

Depressive symptoms were ascertained in 1999 (baseline in the present analyses) and 2005, using the Center for Epidemiologic Studies Depression Scale (CES-D). The CES-D is a 20-item self-report psychiatric instrument (24). It was designed by the National Institute of Mental Health Center for Epidemiologic Studies to identify symptoms of depression in community settings; it is not a diagnostic tool for clinical depression. The CES-D has been shown to be a reliable and valid tool for use in identifying the presence and severity of depressive symptoms in community settings and among African American women (24–26) and in the BWHS with a Cronbach α coefficient = 0.74, indicating high internal consistency (27). Respondents use a 4-point scale to indicate how often a symptom was experienced over the past week. Symptoms include those such as “I felt hopeful about the future”, “I felt that people disliked me”, and “I was bothered by things that don’t usually bother me.” Participants can select “Rarely or none of the time (0)”, “Some or a little of the time (1)”, “Occasionally or a moderate amount of the time (2)”, or “Most or all of the time (3).” Total scores range from 0 to 60, with higher scores indicating more depressive symptoms. A cut point of ≥16 has traditionally been used in the literature to indicate the presence of depressive symptoms (24). At this level, the CES-D has a moderate to high sensitivity for major depression (60–100%) but a lower positive predictive value (24–48%) (28). In one community validation study, a cut point of 23 was used to distinguish “probable” from “possible” depression (29). In a study among pregnant African American women, a cut point of ≥33 was used (30). In concordance with previous work in the BWHS (31), we classified CES-D scores into four levels: <16, 16–22, 23–32, and ≥33. We used a score <16 to indicate few depressive symptoms, with each successively higher cut point indicating an increasing number of depressive symptoms.

Use of antidepressant medication was ascertained by the question “Do you take any of the following medications or vitamins at least 3 days a week?” Patients who marked yes for “Antidepressants (Prozac, Zoloft, Elavil, etc.)” were classified as using antidepressant medications. Questionnaires also included the open-ended question “Please list all other medications or supplements that you currently take at least 3 days a week.” Participants who listed any antidepressant were classified as current users of antidepressant medication.

Ascertainment of Diabetes

The 1995 questionnaire and all biennial follow-up questionnaires asked participants if they had been diagnosed by a physician as having diabetes and when it was first diagnosed. Reported cases of diabetes were classified as incident if there was no report of a previous diagnosis. Participants were also asked in each questionnaire cycle whether they took injections or pills for diabetes; a positive response was also considered evidence of diabetes.

To assess the validity of self-reported diabetes, we had previously examined the accuracy of self-reported diabetes among 227 randomly selected women who consented to release of medical records from their physicians and whose providers replied to the request. The diagnosis of type 2 diabetes was confirmed by the medical record in 218 (96%) of the 227 reported cases. Of the nine remaining participants, two had type 1 diabetes, one had metabolic syndrome, one had steroid-induced diabetes, two had gestational diabetes, and three did not have diabetes.

Covariates

Weight information was obtained from the baseline questionnaire (1999) and updated in biennial follow-up questionnaires. Adult height was reported in 1995. Current weight and 1995 height were used to calculate current BMI (kg/m²). In a validation study of anthropometric measures conducted among 115 BWHS participants, Spearman correlations for self-reported versus technician-measured weight and height were 0.97 and 0.93, respectively (32). Data on vigorous physical activity, television watching, smoking, and alcohol drinking were obtained from the baseline questionnaire for the present analysis (1999) and updated in follow-up questionnaires. First-degree family history of diabetes was ascertained in 1995 and 1999. Educational attainment was asked in the 1995 and 2003 questionnaires. Information on energy intake was taken from 1995 and 2001 food frequency questionnaires (33,34).

Analysis

We compared age-adjusted baseline characteristics across the four CES-D score categories by computing means of continuous variables and proportions of categorical variables in each group.

Age- and time-stratified Cox proportional hazards regression was used to calculate incidence rate ratios (IRRs) and 95% CIs for the association of level of depressive symptoms and use of antidepressants with incident type 2 diabetes. Person-years of follow-up were calculated from baseline (1999) to the year of diagnosis of type 2 diabetes, loss to follow-up, death, or end of follow-up (2011), whichever came first. We used the Andersen-Gill approach to update time-varying exposures and covariates over follow-up. Depressive symptoms were updated in 2005; use of antidepressants and all the time-varying covariates were updated at every 2-year questionnaire cycle. To our basic multivariate model (adjusted for age, questionnaire cycle, healthcare utilization, first-degree family history of diabetes, and years of education), we added a group of lifestyle factors (vigorous physical activity [h/week], television watching [h/day], smoking [never use, past use, or current use], alcohol drinking [never use, past use, or current use], and dietary energy intake [kilocalories/day]) to assess whether symptoms of depression and use of antidepressants affect risk of diabetes through these behaviors. We further adjusted to determine whether BMI (<25, 25–29, 30–34, 35–39, and ≥40 kg/m²) may lie on the causal pathway from depressive symptoms and use of antidepressants to incident diabetes. We restricted an analysis to obese women (BMI ≥30 kg/m²), while adjusting for lifestyle factors, to test the hypothesis that symptoms of depression and use of antidepressants affect risk of type 2 diabetes through other pathways beyond lifestyle factors and BMI. For this analysis we adjusted for BMI as a continuous variable. We also assessed use of antidepressants within each category of CES-D score to assess their joint association with incident diabetes.

Results

At baseline, of the 35,898 women included in the study, 71.5% scored <16 on the CES-D, 15.3% had CES-D score 16–23, 9.1% had CES-D score 23–33, and 4.1% had CES-D score ≥33 (Table 1). Baseline CES-D score was positively associated with BMI, caloric intake, antidepressant use, first-degree family history of diabetes, current smoking, current alcohol consumption, and television watching. CES-D score was inversely correlated with age, level of education, being married, vigorous exercise, and healthcare utilization.

Table 1.

Age-adjusted baseline (1999) characteristics by CES-D score categories^*

Characteristic	CES-D score categories
	<16	16–22	23–32	≥33
Number of women	25,663	5,508	3,268	1,459
Mean age, years (SD)	42.1 (10.3)	39.9 (9.4)	39.2 (8.8)	38.6 (8.4)
Mean energy intake, kcal/day (SD)	1,420 (644)	1,481 (681)	1,521 (700)	1,598 (739)
Mean BMI, kg/m2 (SD)	28.2 (6.2)	29.3 (6.5)	29.7 (6.8)	29.9 (6.8)
Education ≤12 years, %	14	18	21	23
Currently married, %	46	39	35	33
Antidepressant use, %	2	6	10	20
Family history of diabetes, %	36	38	39	39
Current smokers, %	12	17	19	24
Current alcohol drinkers, %	27	33	33	35
Vigorous exercise ≥5 h/week, %	10	7	7	6
Television watching ≥5 h/day, %	16	22	25	33
Use healthcare options,** %	96	95	95	92

*All of the characteristics vary significantly across CES-D group at the α = 0.05 level.

**Based on having had a mammogram, Pap smear, or routine blood test in the course of a physical exam, over the past 2 years at baseline.

Over 12 years with a total of 359,407 person-years of follow-up, there were 3,372 incident diabetes cases (Table 2). The numbers of cases per 1,000 person-years were 8.8 for CES-D <16, 10.4 for CESD-D 16–22, 10.8 for CES-D 23–32, and 13.0 for CES-D ≥33. According to use of antidepressants, the numbers of cases per 1,000 person-years were 9.0 for nonusers and 15.4 for users.

Table 2.

IRR for diabetes according to CES-D score categories and use of antidepressant medication in the BWHS, 1999–2011

	CES-D score categories				Antidepressant use
	<16	16–22	23–32	≥33	No	Yes
All BMI
Cases/person-years	2,306/261,795	549/52,538	339/31,365	178/13,709	3,079/340,419	293/18,988
IRR
Basic multivariate*	1.00 (reference)	1.23 (1.12–1.35)	1.26 (1.12–1.41)	1.45 (1.24–1.69)	1.00 (reference)	1.43 (1.26–1.62)
+Lifestyle factors†	1.00 (reference)	1.18 (1.07–1.30)	1.17 (1.04–1.32)	1.31 (1.12–1.53)	1.00 (reference)	1.35 (1.19–1.53)
+BMI	1.00 (reference)	1.11 (1.01–1.22)	1.08 (0.96–1.22)	1.22 (1.04–1.43)	1.00 (reference)	1.26 (1.11–1.43)
BMI ≥30 kg/m2
Cases/person-years	1,550/93,203	380/21,932	260/13,871	137/6,357	2,105/126,564	222/8,799
IRR‡	1.00 (reference)	1.06 (0.95–1.19)	1.10 (0.96–1.26)	1.24 (1.03–1.48)	1.00 (reference)	1.26 (1.09–1.46)

*Basic covariates: age, questionnaire cycle, healthcare utilization, family history of diabetes, and years of education.

†Lifestyle factors: vigorous activity levels, daily hours of television watching, caloric intake, smoking, and alcohol consumption.

‡IRRs adjusted for all basic covariates + lifestyle factors + continuous BMI.

In the basic multivariate model, any level of CES-D score ≥16 was associated with a moderately increased diabetes risk (Table 2). Relative to CES-D scores <16, IRRs (95% CI) of incident diabetes for CES-D 16–22, 23–32, and ≥33 were 1.23 (1.12–1.35), 1.26 (1.12–1.41), and 1.45 (1.24–1.69), respectively. For all CES-D ≥16, we calculated an IRR of 1.27 (1.18–1.37). We observed attenuation of the IRRs after consecutive adjustment for lifestyle factors and BMI. The final multivariate model IRRs (95% CI) for diabetes were 1.11 (1.01–1.22) for CES-D 16–22, 1.08 (0.96–1.22) for CES-D 23–33, and 1.22 (1.04–1.43) for CES-D ≥33; the comparable result for CES-D ≥16 was 1.12 (1.03–1.20). Similar results were obtained when restricting the analysis to obese women. Use of antidepressant medication was significantly associated with risk of diabetes in all the models (Table 2). In the final multivariate model, use of antidepressants was associated with a 26% higher risk of diabetes. We observed the same result among obese women: IRR 1.26 (95% CI 1.09–1.46).

Table 3 shows the results of the joint analysis of symptoms of depression and use of antidepressants, using as reference group women with CES-D score <16 who were not taking antidepressants. Those on antidepressants with CES-D scores ≥16 were at a significantly increased risk of incident diabetes in every model (Table 3). Symptoms of depression in the absence of antidepressant use were associated with diabetes in the basic multivariate model and after adjustment for lifestyle factors. However, in the full multivariate model, the association was observed only for the group of women with the highest number of depressive symptoms (CES-D ≥33). We observed similar results when restricting the analysis to obese women.

Table 3.

IRR for diabetes according to combined categories of CES-D score and use of antidepressant medication in the BWHS, 1999–2011

	CES-D score categories
	<16		16–22		23–32		≥33
On antidepressants	No	Yes	No	Yes	No	Yes	No	Yes
All BMI
Cases/person-years	2,191/252,524	115/9,271	477/48,776	72/3,762	284/28,052	55/3,313	127/11,067	51/2,642
IRR
Basic multivariate*	1.00 (reference)	1.30 (1.08–1.57)	1.19 (1.08–1.32)	2.07 (1.64–2.63)	1.25 (1.10–1.42)	1.80 (1.38–2.36)	1.45 (1.21–1.74)	2.01 (1.52–2.66)
+Lifestyle factors†	1.00 (reference)	1.20 (1.00–1.45)	1.14 (1.03–1.26)	1.89 (1.50–2.40)	1.16 (1.02–1.32)	1.61 (1.23–2.11)	1.30 (1.09–1.56)	1.74 (1.32–2.30)
+BMI	1.00 (reference)	1.13 (0.94–1.37)	1.08 (0.97–1.19)	1.64 (1.30–2.08)	1.08 (0.95–1.23)	1.33 (1.02–1.75)	1.20 (1.00–1.44)	1.58 (1.19–2.09)
BMI ≥30 kg/m2
Cases/person-years	1,466/89,243	84/3,960	323/20,101	57/1,831	220/12,158	40/1,713	96/5,062	41/1,295
IRR‡	1.00 (reference)	1.14 (0.92–1.43)	1.02 (0.91–1.15)	1.67 (1.28–2.18)	1.12 (0.97–1.30)	1.21 (0.88–1.66)	1.19 (0.97–1.47)	1.67 (1.22–2.29)

*Basic covariates: age, questionnaire cycle, healthcare utilization, family history of diabetes, and years of education.

†Lifestyle factors: vigorous activity levels, daily hours of television watching, caloric intake, smoking, and alcohol consumption.

‡IRRs adjusted for all basic covariates + lifestyle factors + continuous BMI.

Conclusions

In this prospective study of a large cohort of African American women, we found a positive association between depressive symptoms and risk of incident type 2 diabetes; the association became stronger with increased number of depressive symptoms. Although the relationship was largely explained by demographic factors, lifestyle factors, and BMI, a positive association remained among those with the highest level of depressive symptoms regardless of antidepressant use. We also found an increased risk of diabetes associated with antidepressant use, and this association remained at every level of elevated depressive symptoms. It is noteworthy that both the highest levels of depressive symptoms and use of antidepressants were associated with an increased risk even among obese women, suggesting that both exposures may affect type 2 diabetes risk beyond any effect on obesity.

Although there are several studies that assessed the relationship of depression to incident diabetes, we found only one that reported data from African American women separately; it included 696 African American women and 41 cases over a follow-up period of 3 years (16). The study reported a rate ratio of 2.56 (95% CI 1.27–5.15) of diabetes for women with CES-D ≥16 relative to women CES-D <16, after adjusting for age, waist circumference, and physical activity (16). In our study, the comparable estimate was smaller, 1.12 (1.03–1.20). Our results were based on a larger sample (3,372 cases) and a longer follow-up period (12 years), and we adjusted our estimates for a wide variety of well-established contributors to diabetes risk, including family history, multiple lifestyle factors, and BMI. To our knowledge, ours is the largest prospective study of the association of depressive symptoms and diabetes among African American women, who bear a very large burden of illness due to diabetes.

Our finding that the relationship between CES-D scale score and diabetes is strongest for those with the highest level of depressive symptoms is consistent with other studies (7,8,11) and may reflect that those with the highest level of depressive symptoms on the CES-D scale are the most likely to have clinical depression with its common sequelae of decreased energy expenditure and obesity. Much of the increase in diabetes risk was explained by adjustment for these and other strong risk factors for diabetes, as has been reported in prior studies (6,7,16). However, antidepressant use appeared to represent additional increased risk that was persistent across CES-D scores and that was not explained by demographics, lifestyle, or BMI.

Most studies investigating the association between antidepressant use and incident diabetes have noted a positive association between the two (20,22,35–39). Among women with elevated depressive symptoms in the Women’s Health Study, those who used antidepressants were at higher risk of incident diabetes compared with those who did not use them (13). In a recent meta-analysis, the use of antidepressant drugs was associated with an adjusted risk of diabetes of 1.68 (1.17–2.40), higher than our estimate of 1.26 (1.11–1.43), which was additionally adjusted for family history (21). Tricyclic antidepressants and selective serotonin reuptake inhibitors are both associated with weight gain (38,40). There is also some evidence that some, but not all, SSRIs induce cellular insulin resistance (41) and increased inflammatory markers (38). Other studies have found variation in the effects of different antidepressants on glucose homeostasis (42). More detailed examination of the association of specific antidepressants with incident diabetes is an important avenue for future work.

Notably, we found that the associations of both higher-level depressive symptoms and antidepressants with incident diabetes persisted even among women with obesity. The group on antidepressants represents those whose mood symptoms were significant enough to have come to medical attention. It is plausible that the persistent increased risk of diabetes among African American women in these groups represents the action of more severe depression through unmeasured pathways known to be associated with both depression and diabetes, such as inflammatory (14) or neuroendocrine alterations (15).

Our study has several strengths, including its large size, the long-term and high rate of follow-up, and information on multiple important diabetes risk factors and potential confounders. There are also some limitations. We relied on self-reported measures of depressive symptoms, which introduce the possibility of exposure misclassification. The CES-D, however, is a well-validated tool that is commonly used to assess the presence of depressive symptoms in epidemiologic studies (7,16). Since symptoms were measured before the occurrence of diabetes, misclassification is likely to have been random and would have tended to attenuate associations in the extreme categories. We also relied on a self-reported diabetes diagnosis rather than clinical data. However, in a validation study, self-report was found to have high (96%) sensitivity for diabetes diagnosis. The prevalence of undiagnosed diabetes among African American women is ∼4% (1), a degree of misclassification that is unlikely to have had a material effect on the estimates of risk that we calculated. In addition, we adjusted for healthcare utilization to account for any potential selection bias in the relationship between antidepressant use and diabetes. Although our results suggest that symptoms of depression and use of antidepressants affect risk of diabetes beyond behavior and BMI, we cannot rule out that the relationships we observed are due to residual confounding due to measurement error of these variables, or failure to control for other confounders. However, with regard to BMI, a previous validation study found that self-report of height and weight in the BWHS is highly accurate (32). In addition, the persistent association we find even among obese women suggests that at least for women with the highest number of depressive symptoms, there may be pathways independent from BMI. In addition, our group has previously reported significant associations of BMI (43) and behavior variables (44) with diabetes, suggesting that our covariate measures capture biologically relevant information.

In summary, we found a positive relationship between depressive symptoms and incident diabetes in a large population of African American women. Much, but not all, of this relationship was explained by lifestyle behaviors and obesity. In addition, we found an increase in risk in association with antidepressant use regardless of the severity of depressive symptoms. This may represent a direct effect of the medications. Alternately, there may be depression-associated biological factors among women with the severest symptoms, representing important paths for future studies.