Introduction

Depression is a frequent co-morbid condition in diabetes. Approximately 30 % of diabetes patients report significantly elevated depressive symptoms and one-third of these patients fulfil the diagnostic criteria of a major depressive disorder (Ali et al., 2006; Anderson et al., 2001; Barnard et al., 2006). Negative impact of the comorbidity on health-related quality of life was found to be striking (Goldney et al., 2004; Moussavi et al., 2007). Furthermore, depression in diabetes has been consistently associated with impairments of diabetes treatment (Gonzalez et al., 2008) and elevated risks of subsequent long-term complications (Black et al., 2003; de Groot et al., 2001) and mortality (Egede et al., 2005; Katon et al., 2008; Zhang et al., 2005).

Numerous studies have investigated the association between depression and glycaemic control in diabetes, commonly assuming that depressive conditions would interfere with adequate self-care and thus cause hyperglycaemia (Aikens et al., 2009; Ciechanowski et al., 2003; Georgiades et al., 2007; Lustman & Clouse, 2005). However, a meta-analysis of 24 studies reported a relatively small overall association between depression and glycaemic control (Lustman et al., 2000), and there is also evidence which conflicts with the hypothesis of a direct behavioural link (Lustman et al., 2005). Beyond that, results from treatment research reveal that the reduction of the depressive mood in diabetes commonly fails to yield significant improvement of glycated haemoglobin (Petrak & Herpertz, 2009; van der Feltz-Cornelis et al., 2010; Wang et al., 2008), which conflicts with the assumption of a direct interference between depression and glycaemic control. In sum, it can be stated that despite the important role of glycaemic control for the long-term prognosis of people with diabetes (Hoogwerf & Brouhard, 1994; UK Prospective Diabetes Study (UKPDS) Group, 1998) the suggested association between depression and glycaemic control is uncertain.

Recent studies frequently focussed on the impact of elevated diabetes-related distress on glycaemic control. Diabetes-related distress is a condition of inadequate emotional adjustment to the illness, which is frequently found in diabetes patients with elevated depressive symptoms (Kokoszka et al., 2009; Pouwer et al., 2005). Diabetes-related distress has been consistently associated with a poorer glycaemic control (Polonsky et al., 1995; Reddy et al., 2013; Snoek et al., 2000; Weinger & Jacobson, 2001) and some studies even suggest that diabetes-specific emotional problems may affect metabolic control rather than general depressive mood (Fisher et al., 2010; Pibernik-Okanovic et al., 2008).

Since evidence of a negative impact of depression on glycaemic control is equivocal, whereas there seems to be a negative impact of diabetes-related distress, the role of diabetes distress in people with diabetes and depression deserves closer attention. If diabetes-related distress played a significant role in mediating the effect of depression on glycaemic control—as suggested by some evidence (Pibernik-Okanovic et al., 2008)—this could explain the lack of improvement of glycaemic control through generic depression treatments (e.g., antidepressants). In this case, the efficacy of treatments in improving glycaemic control could possibly be enhanced by shifting the focus from depressive feelings alone to coping with diabetes.

We used multiple regression analyses to determine the impact of depression and diabetes-related distress on glycaemic control separately. In view of the latest findings (Fisher et al., 2010; van Bastelaar et al., 2010), we hypothesised that the association between depression and glycated haemoglobin would be mediated by diabetes-related distress.

Methods

At the German Diabetes Center Mergentheim, a tertiary referral centre for diabetes, 466 patients with diabetes were cross-sectionally assessed for affective disturbances using standardised self-report scales. Data collection comprised self-report assessments of depressive mood and diabetes-related distress and analysis of glycated haemoglobin. Medical conditions (diabetes types and long-term complications) were diagnosed according to the ICD-10 diagnostic criteria by the clinic’s physicians, body mass index (BMI) was computed from the current body weight measured at the clinic, and demographic (sex, age, level of education, partner relationship status) as well as diabetes-specific characteristics (diabetes duration, type of medical treatment, and mean number of blood glucose tests per day) were assessed by our study personnel in a structured interview.

Data were derived from the baseline assessments of a randomised trial (identifier NCT01812291) and supplementary cross-sectional collection. The study protocol was approved by the Ethics Committee of the State Medical Chamber of Baden-Wuerttemberg. All participants provided written informed consent prior to enrolment.

Study participation required adult age up to 70 years, sufficient German language skills, and written informed consent. Inclusion in the randomised trial additionally required the self-report of depressive symptoms (Center for Epidemiologic Studies Depression Scale score ≥16) in the clinic’s standard screening. Patients fulfilling these inclusion criteria were informed about the study and offered participation. Exclusion criteria were severe depression (F32.2), assessed in a structured clinical interview performed by trained psychologists, antidepressive treatment, suicidal ideation, acute severe mental and/or physical illness (e.g. schizophrenia, cancer), known from the current anamnesis, being bedridden, and guardianship.

According to our estimates, about 75 % of the clinic’s patients were eligible for study participation following to the study’s inclusion and exclusion criteria. Among those invited to participate, about 50 % indeed participated in the study, 30 % refused to participate, and 20 % were not able to participate for technical problems, e.g. appointment management or timing.

Instruments and measures

Assessment of depression

Depressive mood was assessed using the German version of the Center for Epidemiologic Studies Depression Scale (CES-D), a 20-item questionnaire with good psychometric properties (Radloff, 1977). The observed internal consistency (α) in this study was 0.92. Respondents report the frequency of common symptoms of depression on a four-point Likert scale (ranging from 0—‘rarely or never’ to 3—‘most of the time’). The total score ranges between 0 and 60 with higher values representing higher levels of depressive mood. A clinically significant depression level is indicated by the cut-off score of 23 in the German version (Hautzinger & Bailer, 1993).

Assessment of diabetes-related distress

Diabetes-related distress was measured using the German version of the Diabetes Distress Scale (DDS), a well evaluated psychometric questionnaire which has shown good test characteristics (Polonsky et al., 2005; Hermanns et al., 2009). In this study a consistency (α) of 0.89 was observed. The scale consists of 17 questions on diabetes-specific problems and emotional distress, for example ‘Feeling angry, scared and/or depressed when I think about living with diabetes’. Respondents indicate if they experience the described problems and evaluate their severity on the 6-point Likert scale (ranging from 1—‘no problem’ to 6—‘serious problem’). The specifications are added and averaged to a mean score between 0 and 6, whereby higher values indicate greater distress. A DDS score of 3 or higher is considered to represent a clinically meaningful level of diabetes-related distress (Fisher et al., 2012).

Assessment of glycaemic control

As measure of glycaemic control HbA1c was assessed. All blood samples were analysed in the diabetes centre’s laboratory using high performance liquid chromatography, performed with the Bio-Rad Variant II Turbo analyser. The period between blood sampling and questionnaire assessment was averagely less than a week.

Statistical analyses

The statistical analyses were performed using and SPSS 22.0.0 (IBM SPSS Statistics, New York, USA). The statistical methods comprised one-way analysis of variance, Student’s t test, Pearson’s χ2 test, bivariate correlation, multiple regression, and Sobel test. A P value of <0.05 was considered as criterion of statistical significance in all analyses.

Bivariate correlation analyses were used to determine significant predictors of glycaemic control. For this purpose, levels of depression and diabetes-related distress as well as demographic and clinical patient characteristics were analysed for possible associations with HbA1c. Subsequently, variables which were identified as putative predictors of glycaemic control as well as all potential confounders were used for the hierarchical multiple regression analysis, which served to separate the associations of depression and diabetes distress levels with glycaemic control, and control the effects of putative covariates.

According to Baron and Kenny’s (1986) mediator model, the disappearance of a significant association between depressive mood and HbA1c—when diabetes-related distress is added to the regression model—would suggest that diabetes-related distress mediates the association. In this case, the Sobel test should be used to test the indirect effect of depression on glycaemic control, mediated by diabetes distress, for statistical significance. For this purpose, Preacher and Hayes’s (2004) Sobel test macro was utilised.

Results

Sample characteristics

The sample characteristics are presented in Table 1. 68 % of the patients were diagnosed with type 1, 31 % with type 2, and 1 % with another specific type of diabetes. Most patients used exclusive (75 %) or medication-combined (19 %) insulin treatments, while only 6 % used non-insulin medical treatments (tablets and/or incretin mimetics). The mean HbA1c value was 72 mmol/mol respectively 8.8 %, and almost half of the patients (45 %) were diagnosed with one or more long-term complications.

Table 1 Sample characteristics
Full size table

To assess potential sources of bias, the study sample was compared to a representative sample of 1,172 patients at the diabetes centre. Compared to these typical patients, the study participants were significantly younger (43 ± 15 vs. 50 ± 15 years, P < 0.001), more likely to be female (54 vs. 43 %, P < 0.001), had a higher prevalence of Type 1 diabetes (68 vs. 51 %, P < 0.001), a lower BMI (29 ± 7 vs. 30 ± 7 kg/m2, P < 0.001), and a lower rate of long-term complications (45 vs. 59 %, P < 0.001). There were no significant differences regarding diabetes duration, diabetes treatment, and HbA1c.

Prevalences of significant depression and diabetes distress levels

Prevalence data as well as clinical characteristics of the patient subgroups with elevated depression or diabetes-related distress levels are shown in Table 2. According to the CES-D’s cut-off score (≥23), 203 patients (44 %) reported a clinical level of depression. The mean depression score in patients classified as ‘depressed’ was 31 ± 6; in ‘non-depressed’ patients it was 12 ± 6. Comparing type 1 and type 2 diabetes patients, no significant differences regarding depression levels or percentages with elevated scores were observed.

Table 2 Clinical characteristics and HbA1c values of patients with and without significant levels of depression and/or diabetes-related distress
Full size table

150 patients (32 % of the sample) had DDS scores of 3 or higher, indicating a clinically meaningful level of diabetes-related distress. Patients classified as ‘meaningfully distressed’ had an average DDS score of 3.6 ± 0.5; in ‘not meaningfully distressed’ patients it was 2.0 ± 0.5. Comparing diabetes types, a slightly higher mean distress level was observed in people with Type 2 diabetes (2.7 ± 0.9 vs. 2.5 ± 0.9, P < 0.05). Additionally, the Type 2 subgroup showed a significantly higher prevalence of clinically meaningful diabetes distress levels (39 vs. 28 %, P = 0.026).

The coincidence of depression and diabetes distress and its association with poor glycaemic control

Depression levels and diabetes distress levels were strongly associated (r = 0.58, P < 0.001). However, according to the cut-off scores of the CES-D and the DDS, only half (53 %) of the patients with elevated depression concomitantly experienced clinically meaningful diabetes-related distress (as shown in Table 2). Indeed, the coincidence of depression and diabetes distress was prevalent in 23 % of the total sample, and it was only this subgroup which showed significantly reduced glycaemic control compared to patients with neither depression nor diabetes distress (HbA1c: 77 vs. 70 mmol/mol respectively 9.2 vs. 8.6 %, P = 0.011). Glycaemic control of patients with elevated depression only or elevated diabetes distress only in contrast did not significantly differ from that of controls (both P > 0.80).

Diabetes distress as mediator of the association between depressive mood and poor glycaemic control

The bivariate analyses of variables predicting HbA1c revealed significant positive associations (indicating poorer glycaemic control) with higher levels of depression (P = 0.007) as well as diabetes-related distress (P < 0.001), as to be expected from the literature. Furthermore, HbA1c was positively (indicating poorer glycaemic control) associated with the use of combined ‘insulin plus medication’ treatments and negatively (indicating better glycaemic control) with higher age, higher level of education, having a partner relationship, Type 1 diabetes, a longer diabetes duration, and a higher mean number of blood glucose tests per day (all P < 0.05). All results are shown in Table 3.

Table 3 Bivariate associations of demographic, medical and affective patient characteristics with HbA1c
Full size table

Subsequently, the hierarchical multivariate regression of glycaemic control was performed to separate the associations and test their independence (see Table 4). In the first model, the impact of depression was analysed, revealing a significant negative association between depression and glycaemic control (P = 0.007). In the second model, diabetes-related distress was added. Results indicated a significant association between diabetes distress and glycaemic control while controlling for depression (P = 0.001). Notably, the association between depression and glycaemic control while controlling for diabetes distress was insignificant (P = 0.76). In the final model, all potential covariates were added to control for confounding effects. Results indicated significant positive associations of glycaemic control with higher age, higher education, and higher frequency of blood glucose tests, and significant negative ones with the use of combined medical treatments (insulin plus medication) and the presence of long-term complications (all P < 0.05). However, the associations between depressive mood, diabetes distress and glycaemic control were not substantially altered (see Table 4). The pattern of findings suggested a mediation of the association between depressive mood and glycaemic control by diabetes-related distress. Consequently, the Sobel test was applied to confirm the putative mediation. Results indeed indicated a significant indirect association between depressive mood and glycaemic control, mediated by diabetes-related distress (Z = 3.20, P = 0.001).

Table 4 Hierarchical multiple regression of HbA1c on levels of depression and diabetes-related distress as well as potential covariates
Full size table

Discussion

We observed high rates of depressive mood and diabetes-related distress in our study (which is partly explained by the clinical nature of the sample) and—as reported in earlier studies (Pouwer et al., 2005; Kokoszka et al., 2009; Fisher et al., 2008)—found these affective conditions to be clearly correlated, indicating a significant interdependence. On the other hand, not more than half of the patients with clinically meaningful depression levels reported concomitant diabetes distress, which suggests that in diabetes care a distinction between ‘depression without’ versus ‘depression with diabetes distress’ may be reasonable.

Comparably to former findings (Aikens et al., 2009; Georgiades et al., 2007; Lustman et al., 2000), our study identified a significant bivariate association between depressive mood and glycaemic control. However, the association disappeared in the multivariate analysis with the consideration of diabetes-related distress. This finding indicates that the association between depressive mood and glycaemic control is mediated by diabetes distress, which was confirmed by the Sobel test. The observed mediator effect was not affected by demographic and diabetes-specific covariates.

In sum, these results suggest that diabetes-specific distress may be the decisive link between depression and reduced glycaemic control. Accordingly, depression alone seems to be less important in predicting hyperglycaemia than the often coincidental emotional maladjustment to diabetes. Thus, depression may interfere with glycaemic control particularly when accompanied by diabetes-related distress, and the inconsistent findings regarding depression-related hyperglycaemia may be explained by the presence or absence of this essential factor. This concept (that particularly the coincidence or interaction of depression and diabetes distress may impair metabolic control) is also supported by previous findings (Pibernik-Okanovic et al., 2008; van Bastelaar et al., 2010).

According to the meta-analytic literature, the treatment of depression in diabetes (with both behavioural and medical interventions) has indeed been successful with respect to the reduction of depressive symptoms and suffering. Yet, contrary to expectations, the reduction of depression alone has commonly failed to yield significant improvement of glycaemic control (Petrak & Herpertz, 2009; van der Feltz-Cornelis et al., 2010). A recent randomised controlled trial, which specifically analysed the effects of depression treatments (cognitive-behavioural therapy versus sertraline) on HbA1c in depressed diabetes patients, once more confirmed this lack of improvement (Petrak et al., 2011). However, behavioural interventions can indeed improve metabolic control (Ismail et al., 2004), suggesting that the adjustment to the patients’ specific needs may be decisive.

In the Pathways Study, Katon et al. (2004) provided a depression-specific treatment for depressed diabetes patients, but depression care alone produced no effect on HbA1c. In a subsequent trial, they combined antidepressants with motivational coaching and problem-solving-support for illness-related problems and achieved a significant reduction of depression scores and improvement of glycaemic control (Katon et al., 2010).

Our findings correspond to these results as they indicate that diabetes-related emotional problems can mediate the negative impact of depression on glycaemic control. Accordingly, generic depression treatment may be inappropriate for specific groups of depressed patients. In order to improve of glycaemic outcomes, adequate care of diabetes-related aspects may be essential.

The results have to be qualified by several limitations. Firstly, the data base of our study was cross-sectional, thus our analysis of statistical mediation does not allow causal inference regarding the relations between depressive mood, diabetes-related distress, and glycaemic control. Longitudinal studies are clearly needed in this regard. Secondly, depression and diabetes-related distress levels were assessed with self-report scales, which are associated with common problems of accuracy and bias. Furthermore, the term ‘depression’ refers to a meaningfully elevated level of depressive mood which should not be confounded with the presence of a clinical depressive disorder. Thirdly, our sample comprised a majority of people with Type 1 diabetes contrasting the balanced ratio of Type 1 and 2 diabetes (each approximately 50 %) in our clinic. The higher rate of Type 1 might explained by a self-selection process, as persons of younger age appeared more willing to participate, favouring the inclusion of the younger Type 1 patients. Finally, our measure of glycaemic control, HbA1c, corresponds to blood glucose values of the previous 2 months (Nathan et al., 2007) while the psychometric measures relate to the last two (CES-D) or four (DDS) weeks. This variance of time frames might also be taken into account as limiting factor.

The strengths of our study lie in the standardised data acquisition, which enabled high reliability of the psychometric data, high accuracy of HbA1c analyses (standardised estimation in the centre’s laboratory), and adequate temporal coincidence of blood sampling and psychometric assessment. This approach ensures high internal validity of our results, supporting our conclusions.

The associations with HbA1c observed in this study were generally rather small, which corresponds to the suggestion by Johnson (1992) that the achievement of good metabolic control depends on a number of physical, medical, mental and environmental factors. The final regression model integrating all assessed aspects including medical variables explained a total of 18 % of the variance, which supports the assumption that the prediction of glycaemic control is indeed difficult. However, since the majority of patients clearly exhibited elevated HbA1c values, the relatively small effect sizes may be partly explained by the underrepresentation of good metabolic control in this study, and a more representative sample could probably provide a more reliable estimation of effects. Nevertheless, our effect sizes are well comparable to the previous findings (Aikens et al., 2009; Lustman et al., 2000; Richardson et al., 2008; van Bastelaar et al., 2010), supporting their reliability. We consider depression and diabetes distress as only one part of the puzzle of optimal glycaemic control. Nevertheless, it is one part that can and should be addressed as it may improve patients’ quality of life and their diabetes control.

In sum, the presented findings point out that depression may actually be associated with poor glycaemic control, but this may particularly apply when diabetes-related distress is present concomitantly—suggesting that the experienced emotional distress is closely related to aspects of diabetes (Fisher et al., 2014). Accordingly, in the care of depressed patients we should pay primal attention to the patients’ experiencing of the disease and their diabetes-related concerns. The assessment of diabetes-related distress, particularly in high-risk patients, may be beneficial in understanding individual causes of hyperglycaemia. For this purpose, the DDS or the Problem Areas in Diabetes Scale (PAID) are well established instruments (Polonsky et al., 1995, 2005; Snoek et al., 2000). If diabetes distress is an essential predictor of poor glycaemic control in depressed patients, the appropriate coordination of psychosocial treatments regarding this aspect may be decisive to warrant effective interventions and help diabetes patients achieve optimal diabetes outcomes (Hermanns et al., 2014).