Associations Between Changes in Depressive Symptoms and Social Support and Diabetes Management Among Low-Income, Predominantly Hispanic Patients in Patient-Centered Care

Hispanics with diabetes are less likely to follow recommended self-management instructions from health professionals, and this lower adherence to self-care behaviors is believed to explain racial disparities in glycemic control (1). Hispanic older adults with diabetes have severe health consequences, such as increased risk of kidney disorder and mortality, as noted in a 7-year study (2). Observed disparities in diabetes management appear to be related to social stressors, such as financial strain, and lack of resources for self-care behaviors (3). Studies have well documented the effects of comorbid depression on diabetes management among Hispanics with diabetes (4,5) as well as diverse populations (6). Hispanic patients with higher depressive symptoms are 3.27 times more likely to have poor glycemic control than Hispanics with lesser depressive symptoms (4). A study with 1,293 patients showed that diabetes-related complications, symptoms, and distress were correlated with emotional burden (5). Many scholars regard depression as having a bidirectional relationship with diabetes management and distress (6,7).

The social context in which patients with diabetes are embedded is a significant factor for self-care behaviors (8). Social support has been mostly studied using constructs measuring the quality of the social context in association with diabetes management (9,10). Social support can affect why and how individuals manage illnesses, help patients believe that they have the efficacy or capacity to implement recommended self-care behaviors, and reveal solutions to coping with barriers impeding successful diabetes management (11). A meta-analysis concluded that social support is effective in modifying behavior for better diabetes management (12). Older adults with diabetes who reported that they received support with taking medicine, engaging in enough physical activity, and going to health care providers were more likely to have improved health after 7 years (9).

The patient-centered medical home (PCMH) model is focused on treating comorbid depression and diabetes (13). Psychosocial interventions have recognized the significance of social support for low-income Hispanic patients with comorbid depression (14,15). Social support was recognized early as a promising target for psychosocial services developed to treat depressive symptoms among patients with chronic illnesses (16). In some models, promotoras and community health workers aim to improve diabetes management by enhancing social support and social resources, and strengthening social support is one of the main therapeutic activities shown to increase perceived social support (15). A study focusing on Hispanic patients with cancer with comorbid depression found that social support was a significant mediator that explained differences in daily functioning among patients with cancer who received collaborative depression care at safety-net clinics (17).

Yet, few empirical findings exist regarding relationships between changes in depressive symptoms and diabetes management. Although one study examined correlations between changes in depressive symptoms and frequency of four self-care behaviors—healthy diet, exercise, blood glucose monitoring, and foot care—findings were mixed (18). Because adherence to diabetes management can be measured in many ways (19–21), further studies are needed. To our knowledge, no study has examined whether improved social support can explain changes in diabetes management among individuals following a patient-centered care regimen, which is becoming the norm due to the implementation of the Affordable Care Act (22). Although social support has been identified as a significant predictor of diabetes management in observational studies (9,11,12), this correlation was not found in a study involving less acculturated Hispanic patients with diabetes (23). Specifically, perceived social support was not correlated with adherence to self-care behaviors among low-income Hispanics living in Texas, even when perceived support was very specifically measured in terms of behaviors associated with diabetes management (23). Thus, studies are needed to empirically examine whether changes in social support observed in primary care can explain the variance in diabetes management in this population. Last, depressive symptoms and perceived social support are often correlated, and their associations with diabetes management may not be clearly observed unless they are examined in the same statistical model. Patients with higher depressive symptoms also tend to present with increased interpersonal conflicts and lower social support (24,25). A prospective study found patients with type 1 diabetes with higher depressive symptoms had significantly lower social support (26). By examining changes in depressive symptoms and social support, we can document the effect sizes of those two variables in explaining the variance in diabetes management.

To fill these research gaps, we examined whether changes in depressive symptoms and social support during 6 months postbaseline can explain self-efficacy and adherence to recommended self-care behaviors for patients with diabetes at the 6-month follow-up. We conducted secondary analyses of data collected in a randomized clinical trial (RCT) that compared the effectiveness of a promotora-led self-management program and a newly launched PCMH model by a local government regarding diabetes management among Hispanic patients with diabetes relying on safety-net clinics in a low-income community. A promotora is a community health worker for Hispanic patients who continues to practice indigenous customs and conform to cultural norms (15). By providing culturally tailored services boosting diabetes management, this intervention intends to alleviate emotional distress as well as enhance diabetes management (15).

Hypothesis 1: changes in depressive symptoms and social support during 6 months postbaseline will independently predict differences in self-efficacy and adherence to recommended self-care behaviors for patients with diabetes at the 6- and 12-month follow-ups.

The study analyzed data collected from Hispanic patients with diabetes screened with comorbid depression at baseline in the A Helping Hand (AHH) study. AHH recruited Hispanic patients diagnosed with diabetes, cardiovascular disease, or heart failure from April 2014 to May 2015. In total, 348 patients from three safety-net clinics in East Los Angeles, CA, met eligibility criteria and completed informed consent and a baseline survey. We analyzed data from patients previously diagnosed with diabetes, and 333 patients (95.7%) met this eligibility. Initial analyses determined whether changes in depressive symptoms and social support predicted diabetes management at the 6-month follow-up. We found that 82 patients (24.6%) did not complete the 6-month survey, creating a sample of 251 patients (75.4%) for analysis. We conducted further analyses to determine whether changes in depressive symptoms and social support predicted diabetes management at the 12-month follow-up. Of 333 patients with diabetes, 113 patients (33.9%) did not complete either the 6- or 12-month follow-up, leaving 220 patients (66.1%) for analysis. AHH was an RCT that examined the effectiveness of a promotora-led self-management support program compared with a standard PCMH team featuring a physician, nurse, and medical assistant (15). Promotoras provided psychosocial services focused on six areas: engagement for initial rapport building, problem formulation to list targeted problems, education for teaching self-care management strategies and health information, action planning to develop action steps and implementation, community resource navigation and referrals to providers, and evaluation through feedback (15). Promotoras primarily provided services during in-person meetings, although services were delivered by phone when a patient was not available for an in-person meeting. In addition to six main sessions, three monthly booster sessions were offered to activate self-care management (15). PCMH usual care included depression screening by a primary care provider, referral to community health workers who were newly hired clinic promotoras, referral to behavioral health specialty care, and activation of patient care management (15). The main outcome study found no significant differences in terms of depression outcomes, psychological health, self-care management, physical condition, stress and social support, and glycemic control between the promotora-led intervention and PCMH usual care (15). The AHH study was reviewed and approved by the institutional review board of an affiliated university.

Table 1 features descriptive statistics of demographic and clinical factors in the total sample and among patients who were analyzed or dropped out at the 6-month follow-up. Among the baseline and 6- and 12-month follow-up surveys gathered from the AHH, our study analyzed data from patients who completed baseline and 6-month follow-up surveys (n = 251; 75.4%); 82 patients (24.6%) did not complete the 6-month survey. We compared the means of demographic and clinical factors between those whose data were analyzed and those who had dropped out. No demographic and clinical factors at baseline differed significantly between these two groups other than marital status (P < 0.05).

Depressive symptoms were assessed via the nine-item Patient Health Questionnaire (PHQ-9). Depressive symptoms are a continuous variable ranging between 0 and 27. The PHQ-9 has adequate internal consistency and test-retest reliability (27). It is widely used as a screening tool to discriminate between racially diverse primary care patients with and without depression (28). The internal consistency of the PHQ-9 measured at baseline and 6-month follow-up was 0.53 and 0.88, respectively (N = 251). To find a reason for lower internal consistency at baseline, we conducted a descriptive analysis on distribution of responses to individual items of the PHQ-9. There was more variability of responses to each item at baseline when compared with the 6-month follow-up, and this variability seemed to be translated into lower internal consistency at baseline (Supplementary Data). Overall, more patients endorsed questions on “little interest,” “feeling down,” “trouble falling or staying asleep,” and “feeling tired.” Changes in depressive symptoms were calculated by comparing depressive symptom scores at baseline and 6-month follow-up. Higher values represent higher depressive symptoms at 6-month follow-up compared with baseline. In other words, patients with positive values presented higher depressive symptoms at the 6-month follow-up compared with baseline.

Perceived social support was measured with the eight-item Modified Medical Outcomes Study (MOS) Social Support Survey (29). This abbreviated version was extracted from a 19-item survey that was part of the MOS in the early 1990s (30). A psychometric study of the modified version found comparable internal consistency, convergent validity, and divergent validity when compared with the longer version (29). The same study presented evidence for factorial validity; namely, that the instrument has two factors: instrumental social support (items 1–4) and emotional social support (items 5–8). Items ask how often someone is available when a respondent is in a particular situation, such as being confined to bed and needing to visit a doctor. Response options include: 1 (none of the time), 2 (a little of the time), 3 (some of the time), 4 (most of the time), and 5 (all of the time). For scoring, the average of responses to the eight items was calculated. Our study used total social support, which averaged responses to all eight items, instrumental social support (items 1–4), and emotional social support (items 5–8). The internal consistency measured at baseline and 6-month follow-up was 0.93 and 0.91, respectively, for total social support; 0.95 and 0.94, respectively, for instrumental support; and 0.91 and 0.91, respectively, for emotional support (N = 251).

Adherence to self-care behaviors was measured with the MOS Specific Adherence Recommendations (MOS-SAR) (20,21). Response options range between 1 (none of the time) and 5 (most of the time). The MOS-SAR was developed to measure chronic illnesses, including diabetes, hypertension, and heart disease, and 11 self-care behaviors are included for final score calculation. They include taking prescribed medication, following a low-fat or weight-loss diabetes diet, exercising regularly, eliminating or reducing smoking and alcohol use, limiting stress, checking feet regularly, checking blood glucose, carrying supplies needed for self-care, and carrying something with sugar. An average was calculated based on responses to a second set of questions regarding frequency of implementing self-care behaviors recommended by physicians. The internal consistency of the MOS-SAR at baseline and 6-month follow-up was 0.83 and 0.65, respectively (N = 251).

Self-efficacy related to diabetes management was measured with the Self-Efficacy for Managing Chronic Disease scale (31). This measurement has six questions that ask about the extent to which respondents are confident about keeping fatigue, pain or physical discomfort, emotional distress, or other symptoms from interfering with what they want to do, being able to do things that can reduce the need to see a doctor, or how much illness affects their everyday life. Response options range between 1 (not at all confident) and 10 (totally confident). Average scores for these six items were calculated. Psychometric properties were empirically examined in a study featuring eight samples from five countries, including studies conducted in Spanish and English (32). In our study, internal consistency of this measure at baseline and 6-month follow-up was 0.94 and 0.93, respectively.

We examined longitudinal patterns of the variables of interest and conducted a series of paired t tests comparing the means of data at baseline and 6- and 12-month follow-ups. Then we developed hierarchical regression models to examine correlations between changes in depressive symptoms and social support variables during 6 months postbaseline and self-efficacy and adherence to recommended self-care behaviors at 6- and 12-month follow-ups. Each hierarchical regression model contained four submodels. Model 1 only examined correlations between the outcome measured at the 6- and 12-month follow-ups and the baseline value of the outcome while controlling for potential confounding factors, including health status when the outcome was measured, study arm, diabetes types, age, sex, education, language, and dysthymia (15). By controlling for the baseline value of the outcome, further analyses can rule out innately different levels of self-efficacy and adherence to diabetes management at baseline. By adding changes in depressive symptoms, model 2 estimated coefficients without the social support variable. Model 3 estimated coefficients for change in total social support while considering changes in depressive symptoms, together with other covariates included in model 1. Due to multicollinearity, we had to create model 4, which included changes in instrumental social support and emotional social support, changes in depressive symptoms, and other potential confounding variables. As the post hoc analyses, we controlled for the self-efficacy and adherence to diabetes management at the 6-month follow-up when examining diabetes management at the 12-month follow-up. These post hoc analyses were intended to find the time at which changes in depressive symptoms and social support affected diabetes management. SPSS 24.0 (SPSS Inc.) was used for statistical analyses. For hypothesis testing, we used P < 0.05 to signify statistical significance.

Table 2 displays longitudinal patterns of the means and SDs of depressive symptoms, perceived social support (including total, instrumental, and emotional support), self-efficacy and adherence to diabetes management, and the eight individual self-care behaviors in the MOS-SAR. Mean change in depressive symptoms indicates change between baseline and 6-month follow-up. Participants had an average decline of 7.21 in depressive symptom scores between baseline and 6-month follow-up. Social support generally improved; for instance, total social support increased by 21.43%. Changes in self-efficacy and adherence to diabetes management during the same period were relatively limited compared with changes in depressive symptoms and social support. With regard to the eight individual self-care behaviors, which are recommended for patients with diabetes, patients reported lower adherence to regular exercise, as reflected by a lower mean compared with other behaviors across the three waves.

Table 3 presents results from hierarchical regression models examining diabetes management outcomes at the 6-month follow-up. This hierarchical regression examined four models in a consecutive manner. Two sets of hierarchical regression analyses were conducted with self-efficacy and adherence to diabetes management as the outcome. Changes in depressive symptoms during 6 months postbaseline were significantly correlated with self-efficacy at the 6-month follow-up (P < 0.001). This suggests that patients with more decline in depressive symptoms over 6 months reported higher self-efficacy to manage diabetes. Effect sizes of changes in depressive symptoms were ∼0.40, meaning that when patients with diabetes experienced a decline in depressive symptoms of 1 SD, they increased self-efficacy by 0.40 SD. Adding changes in depressive symptoms (model 2) accounted for 15.0% of the adjusted explained variance. In model 3, changes in total social support were significantly correlated with self-efficacy (P < 0.01), suggesting that patients whose total social support increased over 6 months reported higher self-efficacy at the 6-month follow-up. Adding total social support change increased the adjusted explained variance compared with model 2 by 2.5%. Results from model 4 show that changes in emotional social support were correlated with the outcome (P < 0.05), yet instrumental social support was not (P > 0.05). With respect to adherence to diabetes management, changes in depressive symptoms over 6 months significantly predicted differences in adherence to diabetes management in models 2, 3, and 4 (P < 0.001). The direction of the association indicates that patients reporting lower depressive symptoms at the 6-month follow-up engaged in significantly more frequent self-care behaviors at the 6-month follow-up when the baseline value of the outcome was controlled for. Adding changes in depressive symptoms (model 2) increased the adjusted explained variance by 7.1%. Yet, no social support variables were significantly correlated with adherence to diabetes management (P > 0.05).

Table 4 presents results from a hierarchical regression model examining diabetes management outcomes at the 12-month follow-up. Changes in depressive symptoms explained different levels of self-efficacy (P < 0.01). Standardized coefficients were reduced by almost half compared with analyses focusing on the 6-month follow-up. Changes in depressive symptoms (model 2) added 2.7% to the explained variance compared with model 1. No social support variables were significantly correlated with self-efficacy related to diabetes management at 12-month follow-up. When adherence to diabetes management at 12-month follow-up was examined, changes in depressive symptoms were still significantly correlated with the outcome in model 2 (P < 0.05), model 3 (P < 0.05), and model 4 (P < 0.01). Standard coefficients reflecting changes in depressive symptoms declined to a lesser extent compared with those for self-efficacy related to diabetes management. Social support variables were not significantly predictive of diabetes management at the 12-month follow-up (P > 0.05). Post hoc analysis was conducted controlling for the self-efficacy and adherence to diabetes management at the 6-month follow-up when examining diabetes management at the 12-month follow-up. We found that changes in depressive symptoms were not a significant factor in this post hoc analysis (P > 0.05). This post hoc finding suggests that the influence of changes in depressive symptoms on self-efficacy and adherence to diabetes management lasts for at least 6 months since depressive symptoms were measured. (Results are available upon request.)

The AHH study offered a suitable setting to examine this research question because one group received PCMH usual care and the other group received promotora-led self-management services. Due to services offered in the PCMH model, as designed and implemented by the Los Angeles County Department of Health Services, the primary outcome study found no difference between the two groups in depression, diabetes management, physical health, and glycemic control (15). Changes in depressive symptoms during 6 months postbaseline consistently predicted self-efficacy and adherence to diabetes management. Results regarding relationships between social support and diabetes management were mixed. Patients reporting an increase in total social support and emotional social support during 6 months postbaseline reported higher self-efficacy at the 6-month follow-up. Yet, social support variables were not correlated with self-efficacy at the 12-month follow-up or adherence to diabetes management at 6- or 12-month follow-ups.

Consistent empirical findings have supported the significance of mental health care for patients with chronic illnesses, including diabetes (33,34). A study found that medication adherence and glucose monitoring were clearly enhanced when participants received cognitive behavioral therapy (35). However, few studies have attempted to assess whether changes in depressive symptoms observed in primary care explain the variance in diabetes and chronic illnesses. A systematic review concluded that no intervention studies focused on comorbid depression had presented evidence of a statistically significant relationship between depression and self-care behaviors (33). Also, some studies showed that adherence to self-care behaviors was not significantly higher in the intervention group when compared with the control group, although the intervention group demonstrated significantly lower depressive symptoms on average in a clinical trial (34,36). An exceptional study that analyzed data collected from an RCT testing the effectiveness of culturally tailored collaborative depression care for low-income Hispanic patients found that only exercise was prospectively predicted by depression remission observed at 12 months postbaseline (18). Yet, this particular study measured adherence to diabetes management with the Summary of Diabetes Self-Care Activities, which does not suggest aggregating scores for four individual self-care behaviors (19). In other words, this study was not able to test whether a decline in depressive symptoms observed in primary care predicts enhanced overall adherence to diabetes management (18). Our study measured adherence to diabetes management with the MOS-SAR, allowing us to examine the effect of changes in depressive symptoms on adherence to overall self-care behaviors. Despite our findings supporting the significance of treating comorbid depression to self-efficacy and adherence to diabetes management, our study is unable to provide evidence that promotora-led intervention can enhance diabetes management by virtue of treated depressive symptoms. Trief et al. (37) conducted mediation analysis testing significance of self-care behaviors in associations between receiving an intervention and HbA_1c level. The same analytic approach would offer more convincing evidence for the argument that alleviation of depressive symptoms is necessary for improved diabetes management, yet we were not able to use the same analysis because the parent RCT failed to find different levels of diabetes management in the promotora-led intervention group when compared with the PCMH model control group (15). Further studies should examine significance of depressive symptoms as a mediator that explains different diabetes management levels between an intervention group and a control group in an RCT.

Contrary to predictions based on the existing literature (9,11,12), changes in social support had a negligible association with both self-efficacy and adherence to diabetes management. We believe that three possible explanations deserve further consideration. First, as our study found, the strength of the association between changes in social support and diabetes management could be too negligible to detect in a low-powered study. A review study noted different findings from observational studies depending on sample size (10). Studies with larger samples tend to find significant relationships between social support and diabetes management (10). Yet, this reason does not seem convincing because effect sizes of social support–related variables were minimal in our study. Second, it is possible that social support only works as a buffer to the negativities from social and diabetes-related stressors (38). For instance, Baek et al. (39) found that social support was associated with diabetes-related distress only if patients had higher diabetes-related burden. This finding suggests that social support might become potent only if patients experience more diabetes-related burden, and further studies are recommended for empirical scrutiny for this hypothesis. However, additional analyses testing moderating effects of social support did not offer evidence for buffering effects when functional disability was used as a proxy for diabetes burden (40). Lastly, it is possible that measuring social support may not capture nuanced mechanisms by which social support facilitates self-efficacy and behavioral modification regarding chronic illness management. For instance, a study analyzed data from 463 adults and found that significant correlations between social support from family, friends, and community and individual self-care behaviors were observed only if social support was specific to behaviors (41). In other words, rather than perceptions of general social support associated with chronic illnesses as measured by the eight-item Modified MOS Social Support Survey, supportive behaviors specifically targeting individual self-care behaviors, such as healthy diet or exercise, are likely more significant predictors for diabetes. Last, social support may not be the sole function of social networks (41–43). Social support is one of many psychosocial mechanisms by which diabetes management is affected by input from social networks (41). This effect is often more pronounced among immigrants or people living in ethnic enclaves, which was generally the case in our sample (44). The social networks in which low-income Hispanic individuals are embedded may influence diabetes management outcomes by virtue of different levels of social support, availability of accurate health information, and norms and rules shared among in-group members. Functions other than social support likely influence whether Hispanic patients with diabetes adhere to a primary care physician’s recommendations (23).

Our study has several limitations spanning different research activities. First, the AHH study conducted recruitment using convenience sampling at three safety-net clinics that were not randomly selected. Findings from the study should not be generalized to the population of interest (i.e., patients with diabetes and elevated depressive symptoms who use patient-centered primary care safety-net clinics). To our knowledge, no population-level research had previously studied this particular population. Second, ∼25 and 34% of patients did not complete 6- and 12-month follow-up surveys, respectively, and these patients were excluded from the study. We found no significant baseline differences in most demographic, clinical, and psychosocial attributes other than marital status between those who remained and dropped out. However, it is possible that patients dropped out because they might not have benefited from either promotora-led self-management services or PCMH usual care to the same extent as those who remained.

Our study presented empirical results explicating whether changes in depressive symptoms and social support can explain differences in self-efficacy and adherence to diabetes management among low-income Hispanic patients with diabetes and elevated depressive symptoms receiving patient-centered primary care from a safety-net clinic. We found evidence of the significant associations between changes in depressive symptoms and self-efficacy and adherence to diabetes management. Further studies using different methodological approaches to measure different functions of social networks (e.g., information dissemination and normative pressure) are highly recommended.

Acknowledgments. The authors thank Eric Lindberg for exceptional editorial support.

Funding. This study is supported by the Patient-Centered Outcomes Research Institute (grant AD-1304-7364). This study analyzed data collected in the AHH (trial registry no. NCT02147522, clinicaltrials.gov).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. H.O. conceived of a research question, wrote the manuscript, and analyzed data. K.E. collected the data and reviewed and edited the manuscript. H.O. is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.