Diabetes as a Risk Factor for Development of Elevated Depressive Symptoms in a Population-Based Cohort of Older Mexican-Americans

Study Design and Population

We analyzed data from the Sacramento Area Latino Study on Aging (SALSA), a population-based prospective cohort of Latino individuals aged 60 years and older. A full description of the design and methods for this study has been published elsewhere (20). Briefly, 1789 individuals were recruited from within the Census tracts of Sacramento, Yolo, Sutter, Solano, Yuba, and Placer counties with proportional densities of Latinos greater than 10% based on updated 1990 Census data. Recruitment occurred from March 1998 through June 1999, with an overall response rate of 82.2%. Interviews and clinical assessments were conducted in participants’ homes by bilingual/bicultural technicians and information was obtained regarding demographic, health, and functional status every 12 to 15 months. Participants chose English or Spanish for survey administration. There were a total of seven examinations in the home and six semi-annual phone interviews. Annual loss-to-follow-up was about 5%, primarily due to mortality.

Measures

Type 2 diabetes status and depressive symptoms were measured during seven annual home visits. Individuals were classified as having diabetes if they fulfilled any of five criteria: 1) were on anti-hyperglycemic medications (insulin or oral medications, as ascertained by medicine cabinet inventory performed during annual home visits), 2) self-reported a physician’s diagnosis, 3) had fasting blood glucose ≥126 milligrams/deciliter, 4) had HgbA1c ≥6.5% (which was measured starting at the third follow-up visit), or 5) death certificate listed diabetes as a cause of death (21).

The Center for Epidemiological Studies-Depression (CES-D) scale was administered to all SALSA participants at baseline and every 12–15 months. This scale has been widely used and validated in Latino and geriatric populations (22). A CES-D score of ≥16 was used to define depressive symptoms (23). This definition is commonly used in clinical studies, and a study of primary care outpatients found that this cut-off was associated with a 95% sensitivity and 70% specificity (23). Individuals with CES-D score < 16 and who were not using antidepressants were classified as “Normal. ” Individuals were classified as being “Depressed” if they: 1) had depressive symptoms with a CES-D score ≥16, or 2) used an antidepressant drug as derived from a medicine cabinet inventory (24). Antidepressant drugs included predominantly selective serotonin reuptake inhibitors (SSRIs) and tricyclic antidepressants (TCAs), with smaller numbers of individuals on atypical antipsychotics, noradrenergic and specific serotonergic antidepressants (NaSSAs), norepinephrine-dopamine reuptake inhibitors (NDRIS), and serotonin-norepinephrine reuptake inhibitors (SNRIs).

Socio-demographic variables included participant age, gender, education, marital status, place of birth, preferred language, and monthly household income level. Participants were also asked to self-report medical history. Hypertension was classified as self-report of a doctor’s diagnosis, an elevated systolic and diastolic blood pressure (>140/90) or use of an anti-hypertensive medication. Blood pressure, height and weight were measured at home visit and body mass index (BMI) was calculated as (kg/m²).

Statistical Analysis

To capture the variable nature of depressive symptoms over time, we used Multi-state Markov models (25) to model subject transitions between three possible states: 1) “Normal”: CES-D score < 16 and no antidepressant use), 2) “Depressed”: CES-D score ≥16 or antidepressant use, and 3) “Death. ” Multi-state Markov models describe the associations of covariates with probabilities of state transitions using hazard rates, assuming that hazard rates depend only on the currently occupied state and not on previously occupied states (25). We checked this assumption by fitting a multinomial logistic model for the current state that included the states occupied at one and two visits prior as covariates and the results indicated that the assumption was reasonable. The magnitudes of the regression coefficients associated with the states lagged by one visit were noticeably larger than the magnitudes of the regression coefficients of the states lagged by two visits, indicating that the multi-state Markov models described the data reasonably well. Our models allowed transitions from “Normal” to “Depressed”, “Normal” to Death, “Depressed” to “Normal”, and “Depressed” to Death (Figure). The model also allowed for no transitions, for example remaining “Normal. ” We included only participants who had at least 2 visits with a depressive symptoms score. Bivariate analyses identified covariates that were significantly associated with at least one transition. These covariates included gender and baseline measures of age, education, BMI, hypertension, and stroke. Age, education, and body mass index were categorized into 3 groups respectively for the purposes of this analysis (Table 2). We then fit a series of Multi-state Markov models to the data. Model 1 included only the main predictor of interest, time-dependent diabetes. Model 2 included adjustment for baseline age and gender. Model 3 added baseline covariates that were significantly associated with at least one transition: years of education, BMI, hypertension, and history of stroke. Nativity, income, having a primary care provider, and having insurance were not included in the final model as they were not statistically significantly associated with any of the transitions. Hazard ratios for each possible transition were estimated for each covariate effect. Because of concerns of misclassification of depression by including those on antidepressants who may have been on them for other reasons, we conducted a sensitivity analysis in which we examined the CES-D scores among those taking antidepressants and repeated bivariate analysis of the association of diabetes with depressive symptoms (CES-D score ≥16) with no inclusion of antidepressants in the outcomes. The number of observations with an antidepressant treatment was not sufficient to permit separate analyses due to power considerations.

All analyses were performed using the msm package in R version 3.1.1 (2011). This uses a continuous time model, while properly accounting for the fact that states other than death are only observed at intermittent follow-up times.

Ethical considerations

The SALSA study was approved by the Institutional Review Boards at the University of Michigan, the University of California, Davis, and the University of California, San Francisco.

Conclusion

In this large population-based cohort of older Latinos, diabetics had higher rates of developing elevated depressive symptoms or initiating antidepressant therapies that were more likely to persist over ten years of follow-up. We also found a high prevalence of diabetes with almost half (47%) of the participants meeting diagnostic criteria for diabetes by the end of the study. The associations we found between diabetes and elevated depressive symptoms and antidepressant therapy use were not explained by other important medical conditions or socioeconomic factors. This study adds to prior research that individuals with diabetes are at higher risk for depressive symptoms, and provides more information about older Latinos, a rapidly growing segment of the U. S. population.

Most prior research among Mexican-origin individuals has focused on diabetes outcomes among participants with concomitant diabetes and depressive symptoms (18, 19). Few have examined the risk of depressive symptoms in individuals with diabetes compared to those without diabetes (16, 17). A national longitudinal study of older Mexican-Americans with diabetes demonstrated greater incidence of microvascular and macrovascular complications, disability and mortality among individuals with concomitant depressive symptoms (measured at baseline using the CES-D, and once during 7 years of follow-up using the Composite International Diagnostic Interview, CIDI) (18). In a randomized controlled trial of a diabetes management intervention among Mexican-Americans in San Diego County, California, lower depression scores on the Patient Health Questionnaire-9 (PHQ-9) were associated with lower hemoglobin A1c levels, greater social and environmental support for disease management, better diabetes self-management, and lower BMI and triglyceride levels in two years of follow-up (19).

Previous studies reporting on the incidence of depression among diabetics in the general population have questioned whether increased contact with the medical system (and the diabetes diagnosis itself) led to increased diagnosis and medical labeling of depression (14, 15). This population-based study has the advantage of systematic and direct ascertainment of depressive symptoms and medication information, regardless of frequency of contact with the medical system and whether the individual was being treated for diabetes or not. Further, the long length of follow-up allows us to examine dynamic changes in depressive symptoms over time using the Markov Transitions Model. The increased rate of transition from “Depressed” to Death further suggests a synergistic effect between diabetes and depressive symptoms on mortality.

Several limitations should be considered in interpreting our study findings. While it has been postulated that there may be a bidirectional association between diabetes and depressive symptoms (17, 26), our analysis approach was only designed to evaluate the association between diabetes and depressive symptoms. Other studies have used different measures to assess for depressive symptoms (such as the PHQ-9 or Beck Depression Inventory) rather than the CES-D, limiting the ability to compare results. Furthermore, the CES-D is a commonly used and well-validated measure of depressive symptoms, with high sensitivity and specificity for depression (23), but given it was not accompanied by a full diagnostic interview for depression we cannot speak to the incidence of clinical depression among our participants. A psychiatric consult may not be routinely feasible in primary care. Increasingly, primary care physicians are relying on validated depression screening tools, such as the PHQ-9, to diagnose and prompt further evaluation and monitoring of patients. We did not collect information on self-reported depression diagnosis, so it is possible that individuals were taking antidepressant medications for other conditions (such as for neuropathic pain, insomnia or other mood disorders). However, similar to prior longitudinal studies, we decided to include antidepressant use as part of the outcome measure to avoid misclassification of individuals with adequately treated depression and thus low CES-D scores while taking an antidepressant medication (17, 27). Sensitivity analyses demonstrated that participants on antidepressants had higher mean CES-D scores compared to those who were not taking those medications. This suggests that inclusion of antidepressant use in our outcome definition does not lead to differential misclassification of the outcome.

Our study has several implications for primary and mental health care providers serving older Mexican-origin Latinos. There is evidence that in routine clinical practice there is under-diagnosis of depression among individuals with diabetes, and that this is particularly true in older adults, despite recognition that depressive symptoms worsen diabetes outcomes and lead to increased dementia and all-cause mortality (6, 9, 10, 28). Our study supports the need to more actively screen older Latino adults with diabetes for depressive symptoms and to evaluate them for treatment. Additionally, given the association between diabetes and obstructive sleep apnea (OSA), and the increased incidence of depression among individuals with OSA, it is important to screen for treatable sleep disorders among individuals with diabetes (29). Treatment of depression has been shown to improve depressive symptoms, and studies on collaborative care suggest that such models for treatment of depression can have favorable effects on cardiovascular risk factors and be cost-effective (6). Furthermore, individuals with diabetes and sub-threshold symptoms of depression may benefit from early culturally-appropriate interventions to prevent depression (30). Culturally-sensitive models of care, with a reliance on team-based models, need to be further explored in the treatment and prevention of depression among older Mexican-Americans and other ethnic minorities. A study of Latinos in a safety net setting also emphasized the high rates of relapse for depressive symptoms, so individuals previously treated for depression must undergo continued monitoring even after remission (6). Future research is needed to identify factors that influence vulnerability or resiliency to depression and diabetes, as well as into programs aimed at preventing the development of depression in individuals with diabetes.

In older Latinos, we found that diabetes and depressive symptoms interact in clinically significant ways over 10 years of follow-up. Specifically, diabetes increases risk of depressive symptoms and the transition to death is increased among individuals with diabetes and depressive symptoms. Older Latinos with diabetes should be actively screened for depressive symptoms and prioritized with closer follow-up and a team-based approach. Given increasing concurrent mental health conditions and chronic diseases, new models to integrate culturally appropriate mental health services into primary care are needed.