Cognitive Behavioral Therapy Versus Sertraline in Patients With Depression and Poorly Controlled Diabetes: The Diabetes and Depression (DAD) Study: A Randomized Controlled Multicenter Trial

Depression is a common and potentially life-threatening comorbidity in diabetes. The known adverse effects include poor treatment adherence (1), hyperglycemia (2), increased health care costs (3), increased complications (4), cognitive decline (5), and increased mortality (6). Hence, the treatment of comorbid depression is essential for the clinical care of patients with diabetes (7). Treatment goals focus not only on remission or the improvement of depression but also on the improvement of poor glycemic control (8) to prevent or delay long-term complications (9).

Interventions were evaluated in randomized controlled trials and summarized in two recent meta-analyses (10,11). Both meta-analyses concluded that depression could be treated with moderate success in patients with diabetes using psychological, pharmacological, or combined interventions. The overall result for glycemic control is controversial, mostly due to severe methodological limitations and the heterogeneity of the examined populations and interventions (11). In addition, collaborative care studies with sound methodology (e.g., studies by Ell et al. [12] and Katon et al. [13]), tested the efficacy of this approach as a whole. However, these studies were not able to identify single effective components of compound treatments or evaluate the superiority of one single treatment over another because of a design that allowed a switch between pharmacological and psychological interventions. Most recently, beneficial long-term effects of CBT to improved treatment adherence (compared with treatment as usual) in patients with diabetes with depression were observed for adherence and glycemic control; however, the hypothesized superior long-term treatment effects for depression were not confirmed (14). In summary, the evidence for a single treatment that could significantly improve depression outcomes and glycemic control at the same time remains so far inconclusive (10,11).

One potential reason for contradictory long term-results regarding depression outcomes is a lack of distinction between the initial treatment response and maintenance of the treatment effect. Whereas pharmacological and psychological treatments of depression have similar efficacy for the reduction of depression (10,11), whether the same strategies have a similar efficacy on the maintenance of reduced depression is unclear. Randomized trials comparing pharmacological and psychological treatments for maintenance of treatment response on depression are not yet published in people with diabetes. This would require including only people with diabetes and depression who initially yield a clinically significant depression reduction after treatment and to test different treatment strategies for maintenance.

Against this background, we conducted the Diabetes and Depression (DAD) Study, a randomized controlled multicenter trial in which the efficacy of psychotherapy (diabetes-specific cognitive behavioral therapy [CBT]) was compared for the first time with a pharmacological treatment for depression (sertraline) in patients with poorly controlled diabetes and major depression who responded initially to these treatments with a clinically significant reduction of depression.

The study protocol was published in detail previously (15).

Four coordinating trial centers (located in Bochum/Dortmund, Mainz, Düsseldorf/Köln, and Bad Mergentheim) organized the recruitment and treatment in seventy trial centers of outpatient secondary care (specialized diabetes practices and ambulatory care health services in clinics) located in different parts of Germany (predominantly in the Rhine-Main area, Ruhr area, and Düsseldorf/Köln) from April 2006 through May 2009. Eligibility criteria included insulin-treated diabetes (type 1 or 2), 21–69 years of age, major depression according to DSM-IV criteria, and HbA_1c >7.5% (58 mmol/mol) within the 9 preceding months and again in the screening measurement. Key exclusion criteria were suicidal ideations, psychotic symptoms, bipolar disorder, substance abuse or dependence in the past 6 months, psychotherapy in the preceding 3 months, current use of mood stabilizers, neuroleptics, antidepressants, or benzodiazepines, and liver enzyme elevations to exclude severe liver dysfunction (for a detailed list, see [15]).

The following eligibility and exclusion criteria were revised and amended to the protocol in August 2006 (when 60 patients were randomized) and April 2007 (when 198 patients were randomized): the age range was changed from 21–65 years to 21–69 years to increase the number of eligible subjects. The inclusion criterion of HbA_1c >8% (64 mmol/mol) was changed to >7.5% (58 mmol/mol) because we faced serious difficulties finding poorly controlled patients in the secondary care recruitment centers. To enhance comparability with international studies and to prevent a selection bias, our independent scientific advisory board (15) recommended excluding patients only in case of clinically significant suicide risk or history of attempted suicide in the past 12 months. Therefore, patients with values above 1 in the Hamilton Depression Rating Scale (HAMD-17) “suicide” item (item 3, range 0–4) were not included in the trial. The exclusion criterion regarding the current use of psychotropic drugs was modified to allow the inclusion of patients treated with low-potency neuroleptic drugs in low doses (<300 mg chlorpromazine dose equivalents per day) because these drugs are often used to treat sleeping disorders and restlessness in patients with diabetes. Post hoc analysis revealed that none of the patients received low-potency neuroleptic drugs at the beginning or during the study.

The medical records of the patients were reviewed to assess age, type of diabetes, insulin treatment, and HbA_1c levels. Patients were contacted by telephone to confirm basic eligibility criteria and to invite them to a baseline screening. Depression was measured with a questionnaire-based screening (Center for Epidemiologic Studies Depression Scale [CES-D]) (16,17), followed by the Structured Clinical Interview for DSM-IV (SCID) (18) for patients with a CES-D value >22. The severity of depression was measured with the HAMD-17 (19). Health-related quality of life was assessed with the Short Form Health Survey (SF-36) (20), and diabetes-specific stress was assessed with the Problem Areas in Diabetes Questionnaire (PAID) (21). Medical diagnostics included documentation of current medication, concomitant diseases, onset and treatment of diabetes, liver enzymes, and HbA_1c (using high-performance liquid chromatography). All blood samples were analyzed in a central laboratory (Bioscientia Laboratory, Mainz, Germany). Moreover, patient had to participate in a short diabetes education program to be randomized (see 9interventions).

Patients were randomized using block randomization. For each coordinating institution, a separate randomization procedure using permuted blocks stratified by type of diabetes was performed. Randomization lists were computer generated and maintained by the Interdisciplinary Centre for Clinical Trials Mainz (IZKS) and conducted by fax.

The study was a single-blind trial; that is, the treatment evaluation was conducted by research psychologists unaware of the treatment allocation and not involved in the recruitment or treatment of the patients.

Eligible patients were assigned to CBT or sertraline treatments (Fig. 1). Diabetes treatment was not part of the trial protocol and was continued “as usual.” After 12 weeks, those of both groups who responded to the short-term therapy phase (≥50% reduction of the HAMD-17 baseline score or HAMD-17 posttreatment score ≤7) were included in a 12-month, long-term phase. These patients constituted the intention-to-treat (ITT) population. Nonresponders of the short-term phase were excluded from the treatment protocol. All patients entering the long-term phase received diabetes treatment as usual in the trial center at 3-month intervals during the following 12 months. Sertraline responders received continuous treatment as relapse prevention. CBT responders did not receive further treatment but were encouraged to work with a patients’ manual in the sense of a bibliotherapy (22) during the 1-year follow-up phase. The difference in the active treatment duration between both interventions corresponds to usual clinical practices and thus ensures external validity. Generally, group CBT is offered for a limited time, assuming that “carry-over” effects will stabilize the results (23,24), whereas sertraline is given for a longer period as relapse prevention in patients responding to the initial treatment (25). Patients of both groups underwent the same number of visits to control for the amount of physician contact.

At the 12-month follow-up, both treatment groups were reexamined regarding the primary and secondary outcome variables. The primary outcome was change of glycemic control, defined as the difference in the HbA_1c value from baseline to the end of the long-term phase. Secondary outcomes were the reduction of the HAMD-17 score, remission of depression (not fulfilling the DSM-IV-TR criteria for depression according to the SCID and HAMD scores ≤7), HbA_1c decrease of ≥1%, and changes in SF-36 and PAID scores (all analyses were compared with the baseline values to the end of the trial).

Initially, the primary outcome was improvement of glycemic control, defined as a decrease of at least 1% in HbA_1c value (yes/no) from baseline to the end of the long-term phase. Owing to advice from the advisory board, the analysis of the primary outcome parameter was changed in November 2008 to “change of glycemic control,” defined as the difference in the HbA_1c value from baseline to the end of the long-term phase. By changing the dichotomous into a continuous end point, the statistical power of the trial could be enhanced and the planned number of trial subjects had to be reduced. All changes in eligibility criteria or outcome were amended to the protocol during the trial and before breaking of the blinding after approval by the ethic committee.

CBT leads to a better improvement of glycemic control compared with sertraline treatment at the 1-year follow-up in patients who initially responded to short-term therapy (CBT or sertraline) with regards to improvement in depression. This superiority of the CBT regarding glycemic control was expected because the CBT focused not only on depression but also on diabetes-related aspects (e.g., problems with self-management and adherence to treatment). In contrast, the sertraline group received a purely psychopharmacological treatment, without further focus on their diabetes-related problems.

CBT and sertraline are equally effective in remission of depression a) after 12 weeks of treatment and b) at the 1-year follow-up.

Given the poor glycemic control of the patients and considering that most patients treated with insulin had likely undergone diabetes education previously (according to the German guidelines [26]), a short diabetes education program (2 × 3 h) was offered to all patients as an update by trained diabetes educators (education manual [15]).

CBT was administered by clinical psychologists who had undergone systematic training regarding the manual (CBT manual [18]). CBT was delivered in groups of 4 to 10 patients in an outpatient setting within a 12-week period. This treatment consisted of 10 sessions (20 h) using a manualized semistructured CBT for depression, including different diabetes-specific aspects, to improve adherence to diabetes treatment and coping with diabetes. Psychoeducation included information about the association of mood, activities, and the development and maintenance of depression. Participants learned about the link between diabetes and mood and ways to influence impaired mood with cognitive techniques.

Furthermore, participants were encouraged to discuss diabetes-specific goals, such as HbA_1c target values, with their diabetologists and to specify behavioral goals to improve their glycemic control (15). Individual goal achievement was assessed, and possible barriers to the goal attainment were identified and modified, if possible. Each participant received a patient workbook that included theoretical background, worksheets, and exercises for each session. Patients were encouraged to continue working with the book after the end of the short-term phase to stabilize and generalize the improvement (patients’ manual [15]).

Treatment was started at a dose of 50 mg/day and could gradually be raised to 200 mg/day, with changes not exceeding 50 mg/week. Dose changes according to response and side effects were based on the Clinical Global Impression Rating (27) and the Udvalg for Kliniske Undersøgelser side effects rating scale (28).

The power calculation was based on expected differences in HbA_1c levels of the comparison groups in the ITT sample that included all randomized patients who entered the 12-month follow-up phase. Considering randomized controlled trials evaluating CBT (29) or selective serotonin reuptake inhibitors (30), a treatment difference of 1.0 ± 1.6% HbA_1c could be assumed to be relevant. Therefore, with 2 × 46 evaluable subjects, the trial had an 85% power of detecting a treatment difference of 1.0% HbA_1c by means of a t test on a two-sided significance level of α = 0.05. Assuming a response rate of 40% after the short-term treatment, 230 (2 × 115) subjects had to be randomized (statistical analyses plan) (15).

The primary outcome of the trial was defined as the difference in HbA_1c values from the baseline to the end of the long-term phase comparing both treatment groups using an ANCOVA controlling for baseline HbA_1c value and a baseline HAMD-17 score. Categorical outcomes (e.g., remission of depression) were analyzed using logistic regression analysis controlling for baseline HbA_1c values and baseline HAMD-17 scores. Improvement in HAMD-17, SF-36, and PAID scores were evaluated using ANCOVA controlling for baseline HbA_1c values and respective baseline scores. Analyses were repeated controlling for potential confounders. A correlation analysis was performed to identify confounders (age, sex, coordinating institution, diabetes type, diabetes complications, education years, income, single/recurrent episode[s], and comorbidity) with other mental disorders. Baseline variables associated (P < 0.10) with long-term outcome variables (HbA_1c, HAMD-17 score, SF-36, and PAID score, respectively) were included as further control variables. All analyses were conducted for the ITT population. Subgroup analyses for the type of diabetes were also performed. Owing to the purely exploratory character of these analyses, no adjustment for multiplicity was done.

Clinical monitoring, data management, pharmacovigilance, regulatory affairs, and statistical analyses were conducted by the IZKS Mainz and are described in detail elsewhere (15).

• Data management: A data management plan describing data management procedures, data collection, data flow, and the data validation was established for the DAD study.

• Monitoring: Clinical on-site monitoring was performed by personal visits from a clinical monitor according to standard operating procedures of the IZKS. The monitor visited each site at regular intervals to ensure compliance with the study protocol, good clinical practice, and legal aspects.

• Independent scientific advisory board: An independent scientific advisory board reviewed the outcome every 12 months.

• Quality assurance of depression severity assessment: To assure the quality of the HAMD rating, we established a validated training procedure (for details [15]).

• Adherence to the CBT manual: CBT sessions were videotaped to ensure adherence to the manual and to give continuous supervision by one of the authors (F.P.), who is a CBT trainer and supervisor.

• Adverse events reported by the patients or detected by the investigator were monitored and recorded continuously according to good clinical practice (15).

The trial was approved by the Medical Ethics Committee Hessen and was conducted according to the Declaration of Helsinki (31). All patients gave written informed consent to participate in the trial.

The recruitment is described in Fig. 2. The baseline characteristics of the patients showed no significant differences between intervention groups as tested with t tests or χ² tests (Table 1). The results of the drop-out analyses are reported in Supplementary Appendix 1.

The response rate after 12 weeks of treatment was 45.8% (n = 115 of 251) for the total sample, with 42.1% (n = 53 of 126) for CBT and 49.6% for sertraline treatment (n = 62 of 125). This difference was not statistically significant (P = 0.231). The 115 treatment responders of both groups constituted the ITT analysis group and were included in the long-term phase of the study.

Primary and secondary outcomes are reported in Table 2. After 15 months, the mean HbA_1c remained nearly unchanged compared with the baseline measures in both intervention groups, with no significant difference between the groups. The estimated treatment difference from the ANCOVA model amounted to −0.27% (95% CI −0.62 to 0.08, P = 0.129), indicating no significant difference between the groups. Hence, the main hypothesis of the study, which expected an advantage for CBT treatment, could not be confirmed. The mean HbA_1c values did not change substantially during the course of the study in either group. Depressive symptoms assessed by the HAMD-17 strongly decreased in both groups, with a significant advantage for sertraline (P = 0.020) and a moderate effect size difference between groups (d = 0.48, 95% CI 0.11–0.86, P = 0.011). A nonsignificant trend (P = 0.083) was observed for remission of depression in favor of sertraline.

In contrast with the physical component of the SF-36, which remained nearly unchanged compared with the baseline score for both groups, the mental component of the SF-36 improved considerably in both treatment groups. This result was similar for diabetes-related stress, with a strong decrease in the PAID score in both intervention groups (SF-36 and PAID differences between groups were statistically not significant).

The results of the per-protocol analyses, which included all patients who completed the treatment, demonstrated nearly identical results for the primary outcome and the depression-related secondary outcome (Supplementary Appendix 2).

Although patients with type 1 diabetes showed a slight increase in HbA_1c after 15 months compared with the baseline, a decrease was observed in patients with type 2 diabetes (Supplementary Fig. 1). The unadjusted mean difference between groups was 0.63% (adjusted P = 0.0036). Patients with type 2 diabetes showed an improvement in HbA_1c when treated with CBT compared with the sertraline treatment, which was associated with an increase in mean HbA_1c values (unadjusted mean difference between treatments 0.66%, adjusted P = 0.0475; Supplementary Fig. 1A). Patients with type 1 diabetes treated with sertraline demonstrated a stronger improvement in depressive symptoms after 15 months compared with patients who received the CBT treatment (P = 0.0044). In contrast, the difference between the interventions for patients with type 2 diabetes was not statistically significant (Supplementary Fig. 1B). Accordingly, remission rates were better for patients with type 1 diabetes in the sertraline group compared with the CBT group (odds ratio 0.28, 95% CI 0.095-0.810; P = 0.017; Supplementary Fig. 1C).

Throughout the trial, 73 patients had at least one hospitalization (CBT: n = 44; sertraline: n = 29), and 2 patients died (CBT: n = 1; sertraline: n = 1).

Contrary to our primary hypothesis, CBT did not lead to a better improvement of glycemic control compared with sertraline treatment after 15 months. In fact, the very poor glycemic control remained nearly unchanged during the entire trial in both intervention groups.

The initial response rate regarding depressive symptoms after 12 weeks of treatment was 45.8% for all randomized patients (in agreement with the first part of our secondary hypothesis), with similar response rates for the CBT and sertraline groups.

For the patients who qualified for the long-term phase of the study by their initial response to treatment, we observed a significantly better depression outcome for patients treated with sertraline after 15 months, which was mainly due to better maintenance of reduced depression in the sertraline group. This overall difference in the HAMD score (0.48 effect size) can be considered as moderate (32).

For the remission rates of depression, the observed differences between groups failed to achieve statistical significance.

Subgroup analyses showed differential treatment effects depending on the type of diabetes. A moderate HbA_1c improvement was shown when patients with type 2 diabetes were treated with CBT and a slight deterioration was shown when treated with sertraline, which led to a moderate difference of 0.66% HbA_1c between treatments. This effect was not seen in patients with type 1 diabetes.

In contrast, patients with type 1 diabetes benefited more from sertraline treatment than the CBT group with regard to depressive symptoms. This was also the case for remission of depression in patients with type 1 diabetes, with nearly 74% of patients treated with sertraline achieving remission compared with only 44% of the patients treated with CBT. This differential effect was not observed in patients with type 2 diabetes.

The short-term treatment response regarding depression after CBT or sertraline treatment in the current study is in line with randomized controlled trials in major depression with and without comorbid diabetes (33–37). Although negative results regarding glycemic control were observed in most of the studies in the field (10,11), it was still unexpected for us. Our expectation regarding the assumed advantage of CBT was based on the specific intervention, targeting not only depression but also adhering to diabetes treatment recommendations, which was a unique approach in published studies when we started our trial in 2006. Meanwhile, results of a trial with some similarities to our study became available (14): CBT with a focus on increased treatment for adherence and depression was compared with enhanced treatment as usual for patients with uncontrolled type 2 diabetes and depression. After 4 months of treatment, CBT for adherence and depression led to better results for adherence, depression, and glycemic control. Those differences were no longer observed after 8 and 12 months for depression but remained stable for adherence and glycemic control. Hence, like in our study, these results demonstrate the difficulties identifying a specific treatment with simultaneous effect on depression and glycemic control. However, the observed long-term influence on glycemic control makes it promising to include the topic of adherence in depression treatments for this group of patients.

The inclusion of secondary care of patients with poor diabetes control with major depression and the unintended fact that we included only Caucasians restricts the generalizability of our results. Adherence to diabetes treatment was not directly assessed, which is a further limitation of our study.

A strength of our work is that we conducted the study with a sound methodology, including multiple quality assurance aspects. These included the measurement of depression with SCID and HAMD interviews by trained clinical psychologists (instead of questionnaires, as in some important trials on this topic [12,13]), the video-controlled monitoring of adherence of psychologists to the CBT manual, the establishment of an independent scientific advisory board, the data management, and data validation with intense validation strategies. Finally, we limited bias due to center or investigator effects by including eight clinical psychologists and 70 trial sites across Germany.

Several facts might explain the lack of effects regarding control in our study.

Previous trials in the research field of diabetes and depression to a large extent included primary care patients (e.g., [12,13]) and the rare positive effects on glycemic control were mainly seen in studies conducted in primary care patients (13,38,39). In contrast, we recruited participants exclusively from secondary care study sites. In Germany, it can be expected that most patients with type 1 diabetes are treated in secondary care (40), whereas almost all patients with type 2 diabetes are treated in primary care and are referred to secondary care only when treatment fails. Thus, recruiting patients with type 2 diabetes in secondary care is already selecting patients in whom usual treatment is failing. When considering the mean HbA_1c baseline value of 9.3% (78 mmol/mol) in our sample, we had to assume that we recruited a group of very “difficult-to-treat patients” even in the secondary care setting and at least with respect to their glycemic control.

The depression in most of our patients was moderate to severe. Considering that trials with positive results in glycemic control preferentially used questionnaires to assess depression (13,39), we likely included patients with a more severe depression and potentially poorer treatment adherence, which in turn might have contributed to our results (41).

It might be the case that treatment has to be individualized even more distinctly to achieve better glycemic control (i.e., not only with respect to the type of diabetes), as suggested by our subgroup results, but also regarding the group intervention approach.

Both treatments were effective to improve depression in poorly controlled patients with diabetes, but sertraline was slightly more effective to maintain these effects compared with CBT. Despite an intense treatment focus on the improvement of adherence to diabetes treatment in the CBT group, glycemic control did not improve in either intervention. The selection of very difficult-to-treat patients with poorly controlled diabetes and depression and currently unknown aspects of the comorbidity of diabetes and depression might have contributed to the results of our trial and the differences that we observed with regards to the results of previous studies.

Our results demonstrate that CBT and sertraline, two widely used interventions in clinical practice, cannot be considered as sufficiently effective in the treatment of secondary care patients with poor diabetes control and depression. “Mood repair” alone does not automatically result in improved diabetes outcomes. Further research needs to address potential additional mechanisms that mediate the effect of depression on glycemic control.

Acknowledgments. The authors thank all of the patients for their participation in the DAD trial.

Funding. This work was supported by the Kompetenznetz Diabetes mellitus (Competence Network for Diabetes Mellitus) funded by the German Federal Ministry of Education and Research (BMBFhttp://dx.doi.org/10.13039/501100002347) (No. 01KG0505). All authors received financial support from the BMBF for the submitted work. The BMBF had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript.

Duality of Interest. C.H. has served as a consultant for Servier (Paris, France) and Janssen-Cilag (Beerse, Belgium); has served on the speakers bureaus of Bristol-Myers Squibb, Eli Lilly, Janssen-Cilag, Pfizer, and Servier; and is the managing director of the psiac GmbH, Mainz, which provides an Internet-based drug interaction program for psychoactive drugs. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. All authors designed the study, were responsible for its conduct, and contributed to the writing and editing of the manuscript. F.P., K.K., and C.R. were responsible for study management and data collection. C.R. undertook data analysis. F.P. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.