Depression and Glycemic Control in Elderly Ethnically Diverse Patients With Diabetes: The IDEATel Project

Subjects were participants in the Informatics for Diabetes Education and Telemedicine (IDEATel) project (28,29), a demonstration project funded by the Center for Medicare and Medicaid Services (CMMS). Medicare beneficiaries with diabetes living in medically underserved areas were enrolled to evaluate the feasibility, effectiveness, and cost-effectiveness of telemedicine with this population. Subjects were recruited through primary care providers (PCPs) in urban and rural medically underserved areas and were included if they were ≥55 years of age, receiving Medicare, and had diabetes (defined by physician’s diagnosis and being treated with diet, oral hypoglycemic agents, or insulin). Excluded were those who had moderate/severe cognitive, visual, or physical impairments or severe comorbid disease. Subjects were randomized to a telemedicine case management intervention or usual care. Intervention subjects received a home telemedicine unit, i.e., a web-enabled computer used to upload blood pressure and blood glucose measurements, to videoconference with a nurse case manager and dietitian, and to access individualized graphic data displays and educational materials. The nurse case managers provided diabetes education and, under the supervision of an endocrinologist, treatment planning and consultation to PCPs who maintained treatment decision authority for their patients. A separate team of trained research nurses conducted physical and psychosocial assessments at baseline and 1-year follow-up. Over the 1st year of involvement, the mean number of home televisits (∼30 min, every 2–6 weeks depending on glycemic control) was 28.3 ± 15.2 (median 28), with a mean number of blood glucose uploads (560.2) and blood pressure uploads (184.6) indicating intensive program involvement. The data analyzed for the present study includes baseline (n = 1,665) subject demographics (age, sex, race/ethnicity, marital status, insulin use, education, years of diabetes, and pack-years smoked) and measures of glycemic control, comorbidity, diabetes symptom severity, functional disability and depression, and 1-year (n = 1,578) measure of glycemic control. Although depression was measured at baseline, it was not used as an exclusion criterion.

The SHORT-Comprehensive Assessment and Referral Evaluation (CARE) Depression Scale is a brief version of the CARE and measures depression in the elderly and is chosen for IDEATel because it has been used with ethnically diverse populations. Internal consistency reliability estimates were 0.87 for development samples, and interrater reliability was 0.94 (32). Cronbach’s α for the current sample was 0.86. The evidence for concurrent validity was high (e.g., correlation of 0.75 with diagnosis) (33). The SHORT-CARE includes diagnostic items that correspond to DSM criteria, and cut scores have been developed for classification of depression (31). A cutoff of seven yields the highest sensitivity and specificity for clinical depression (34).

The Charlson Comorbidity Index is a 17-item self-report of medical conditions. Weights (0–6) are assigned to various conditions reported. The weights are then added with an algorithm to ensure that the same condition is represented only once. It correlates significantly with short-term outcomes (36,37).

The Type 2 Diabetes Symptom Checklist assesses six dimensions of diabetes-related symptoms: hyperglycemic, hypoglycemic, ophthamologic, psychologic, cardiovascular, and neuropathic (38). Test-retest reliability ranged from 0.79 to 0.94. Internal consistency estimates ranged from 0.76 to 0.95. Cronbach’s α for the current sample was 0.91.

The Activities of Daily Living Scale of the CARE is a measure of functional disabilities used to assess elderly individuals of different races, ethnicities, and community settings (30,39,40). Concurrent and predictive validity is good (32,33). Cronbach α for the development sample were 0.84–0.95, and for the current sample it was 0.93.

Glycemic control was measured using HbA_1c (A1C) analyzed by boronate affinity chromatography with the Primus CLC 383 (Primus, Kansas City, MO).

IDEATel participants were enrolled through 700 PCPs (325 downstate and 375 upstate) and randomized within the practice. Therefore, all statistical analyses were adjusted for clustering within the PCP practices.

Variables included as covariates were age, sex, race/ethnicity, marital status, years of education, years of diabetes, insulin use, pack-years smoked, comorbidity, functional disability, and diabetes symptom severity. Variables were included based on clinical importance and preliminary analysis. Pearson’s correlation and t tests were used to determine significant associations between the possible covariates and baseline measures of depression and glycemic control. Variables were included as covariates if they had a P value of ≤0.10 with the baseline data. Variables included in the baseline models were also included in the 1-year outcome models.

The primary analysis was of the relationship between baseline subject variables and outcome measures, with depression as the main independent variable and change in A1C (1-year A1C controlling for baseline A1C) as the outcome variable. To perform prospective analyses, a mixed-model approach with random effect to adjust for clustering within the PCP practice was implemented, using the SAS PROC MIXED. Variance components covariance structure was used in all analyses. Each analysis controlled for baseline A1C and all subject variables, including functional disability and comorbidity, two factors that are critical to outcomes of the elderly.

Of 9,597 potential subjects assessed for eligibility, 929 were excluded by CMMS (e.g., not in CMMS system, died), and 6,467 were excluded for other reasons (e.g., refused, too sick, did not have diabetes, PCP refused). Telephone screens were accomplished for 2,201 individuals, 235 were excluded (e.g., vision, cognitive impairment), and 301 were not randomized (e.g., changed minds), leaving 1,665 randomized (844 intervention, 821 usual care). For prospective analyses, 105 usual care and 201 intervention subjects dropped out or were lost to follow-up. Subjects who completed the 1-year exam did not differ significantly from those who did not on age, race/ethnicity, sex, or baseline medical data. The baseline demographic and clinical data did not differ for the intervention and control groups (Table 1).

Detailed analyses of the results of the IDEATel project have been reported (41). Limited data are provided here in order to provide a context for this study. Mean A1C decreased in the intervention group from 7.35 to 6.97% and in the control group from 7.42 to 7.17%. When comparing these group changes, the net adjusted reduction was 0.18% (P = 0.006). When the data from the subgroup of subjects with baseline A1C >7% were analyzed, these differences were greater, showing a net adjusted reduction of 0.32% (P = 0.002) of intervention versus control subjects.

See Table 2. There was a significant correlation between depression and A1C (r = 0.104), indicating that depression was associated with poorer glycemic control at baseline. Subjects reporting more depressive symptoms were younger (r = −0.159), female (t = 0.210), Caucasian or Hispanic (t = 0.284), never married (t = 0.050), less educated (r = −0.207), insulin users (t = 0.084), heavier smokers (r = 0.061), and reported more medical comorbidities (r = 0.218) and more diabetes-related symptoms (r = 0.284).

Table 3 provides data on the relationship between baseline depressive symptoms and A1C. In each analysis, all other covariates were controlled. There was a trend for depression to predict baseline A1C (estimate = 0.026). Other predictors of higher A1C were being Caucasian (estimate = 0.803), male (estimate = −0.172), using insulin (estimate = 0.476), having more years of education (estimate = 0.013), having more years of diabetes (estimate = 0.016), having more diabetes symptoms (estimate = 0.006), and having poorer activity of daily living function (estimate = 0.047).

Depression was examined using three different measurement approaches (Table 4). In each analysis, we also examined whether the two groups, intervention and control, differed in whether depression predicted A1C (depression × group interaction term), and we ran separate analyses for the two groups to further assess this issue.

The first analysis was of depressive symptoms, treated as a continuous variable, predicting change in A1C (controlling for baseline A1C and other subject variables). Baseline depressive symptoms did not predict change in A1C (estimate = 0.016, P > 0.350), neither for the control (estimate = 0.001, P > 0.911) nor intervention (estimate = −0.003, P > 0.769) group.

Next, depression was treated as a dichotomous variable, and the cutoff score of seven was used to define depressed versus not-depressed groups, as recommended (34). At baseline, 31.7% (n = 528) exceeded the cutoff, while at follow-up 27.8% (n = 393) did so. Baseline depression did not predict change in A1C (estimate = 0.164, P > 0.105), neither for the control (estimate = 0.166, P > 0.164) nor intervention (estimate = 0.141, P > 0.156) group.

Finally, depression was defined as use of antidepressant medication. A study of the Diabetes Prevention Program research group found that individuals who take antidepressant medications often have a diagnosis of depression but do not have higher depression measure scores, and they argue for using this factor as a marker of depression (42). At baseline, antidepressant use did not correlate with any of the medical variables. In prospective analyses, depression did not predict change in A1C (estimate = −0.101, P > 0.293), neither for the control (estimate = −0.040, P > 0.754) nor intervention (estimate = 0.053, P > 0.641) group.

In all analyses, racial/ethnic groups did not differ on whether depression was a predictor of change in A1C. There were three factors that did predict glycemic control: age (older subjects had lower 1-year A1C), race (Caucasian subjects had higher 1-year A1C), and insulin use (insulin users had higher 1-year A1C).

In this large sample of elderly diabetic patients, we did not find evidence that depression prospectively predicts change in glycemic control, neither for the group that participated in the IDEATel intervention nor for those who received usual care. Whether we defined depression as a continuous variable (i.e., number and severity of depressive symptoms), as a dichotomous variable (i.e., depressed/not depressed), or as use of antidepressants, baseline depression did not predict change in A1C. We note that IDEATel did not target depression as a primary outcome, nor were patients systematically excluded due to depression scores, although PCPs may have screened out their more depressed patients before telephone screen.

We hypothesized that depressed patients would derive less benefit from the intervention, as the hopelessness/helplessness of depression might interfere with taking an active health care role. We did not find this to be the case. We also hypothesized that depressed patients receiving usual care would demonstrate poorer glycemic control 1 year later; this hypothesis was not supported.

Participants reported more depressive symptoms if they were female, never married, less educated, heavier smokers, had more comorbid medical conditions, and poorer glycemic control, findings similar to other reports (9,12). This supports the findings of relationships between depression and varied negative outcomes, confirms the limited research on the impact of depression on the elderly, and extends the data to an ethnically diverse sample of elders, a group that is often underrepresented in research.

Studies that have explored the relationship between depression and glycemic control have yielded inconclusive findings. While we did find a significant baseline relationship, this became a trend when other variables were controlled; thus, the relationship was not strong. If a stronger relationship exists, it may have been masked by limited variability in glycemic control or depression. However, in light of a recent population-based study that found no relationship between depression and hyperglycemia when comorbid diseases were controlled (12), one might reasonably conclude that the relationship between depression and glycemic control is a weak one for this sample of elderly individuals. We do not know if results would have been different for other age-groups.

There are several limitations of the study. As noted earlier, the limited variability in A1C or depression might have masked a relationship between the two. We used a self-report measure of depression that may not have been sufficiently sensitive. Also, we do not have information about the individuals who were excluded before the telephone screen, other than knowing that they were significantly older (though still elderly) than the final sample (73.9 vs. 70.8 years mean age). Finally, ∼36% of the drop-outs (n = 306), as compared with 31.7% of the participants, scored above the cutoff for clinical depression. The possibility that PCPs may have excluded their most depressed patients and that a greater percentage of subjects lost to follow-up were depressed may have affected our ability to find a relationship between depression and follow-up A1C.

Without evidence to the contrary, depression should not be used to exclude elderly patients from participation in interventions that target medical outcomes. Future research should assess the impact of depression on outcomes other than glycemic control. Finally, further research should specifically focus on the elderly with comorbid depression and diabetes, given the paucity of research with this ever-growing group.