Introduction

Depression and diabetes are common coexisting conditions that have a debilitating impact on each other [1]. As diabetes not only negatively affects patients’ physical health but also their mental health, a substantive number of patients suffers from depressive symptoms [2,3]. Cognitive Behavior Therapy (CBT) is the most commonly used evidence-based psychological intervention for treating depressive symptoms in individuals with and without a chronic somatic disease [4–6]. In the past decade, Mindfulness-Based Cognitive Therapy (MBCT) has evolved as an upcoming therapy for treating depressive symptoms [7]. Although the field of research on MBCT for current depressive symptoms is still relatively young, existing studies have provided consistent empirical support for the effectiveness of MBCT in reducing depressive symptoms and increasing quality of life in healthy as well as in chronically ill people [8–10].

CBT and MBCT can be characterized as variants of cognitive therapy, but they do include fundamentally different treatment techniques. CBT is a structured program that focuses on behavioral activation and on helping patients to understand connections among thoughts, feelings, and behaviors in order to consequently alter underlying cognitions that maintain a depressed mood [11]. MBCT, on the contrary, involves practicing awareness and acceptance of dysfunctional thoughts, feelings, and bodily sensations by focusing on the present moment during mindfulness meditation and yoga exercises [7]. Because of the different focus of CBT and MBCT, it could be that these treatments are beneficial for different groups of patients.

A few studies have directly compared the effects of CBT and mindfulness-based interventions [12–15]. Those studies provided evidence for the effectiveness of both interventions with respect to improving a variety of psychological and medical outcomes. In a recent randomized controlled trial, we compared the effects of individual CBT and MBCT in reducing depressive symptoms in comparison to a waiting list control condition in patients with diabetes [16]. Both interventions were found to be effective in reducing depressive symptoms, diabetes-related distress, and anxiety as well as in increasing well-being, with no intervention being superior over the other. Positive effects sustained at long-term follow-up at nine months after treatment [17].

This evidence, suggesting comparable effectiveness of both CBT and MBCT, does not necessarily imply that every individual patient benefits equally from these interventions. Characteristics of the individuals presenting for treatment are assumed to impact the level of treatment efficacy [18,19]. The current study investigates factors that predict differential response to CBT and MBCT.

Much of the literature on predictors of treatment efficacy has focused on investigating general prognostic predictors of CBT in order to identify subpopulations of patients that are most likely to benefit from CBT [20,21]. Typically, these studies have investigated predictors grouped in four different domains, namely demographics, clinical characteristics, personality dimensions, and disease related characteristics in somatic populations. Although demographics as predictors for CBT outcome have widely been studied, results are inconclusive as to which characteristics are of influence. Whereas some studies identified age [22], gender [23,24], education [22,23], marital status [22,25] as predictors, other studies failed to identify any demographic predictors [26–28]. In contrast, clinical characteristics such as high levels of pretreatment depressive symptoms, chronicity, and history of depression have consistently been associated with poorer CBT treatment outcome [20,27,29]. Studies that investigated personality dimensions as possible predictors of outcome mainly focused on neuroticism [30]. Again, no clear pattern emerged in the literature. Although high levels of neuroticism were associated with poor CBT outcome in some empirical studies [30–32], an earlier review by Mulder et al. [30] concluded that the influence of neuroticism firmly depended on the methodology of the study as the strongest support for the association came from methodologically weaker studies. Finally, disease-related variables like type of treatment, comorbidity, and diabetes-related distress may also predict outcome to psychological treatment. Lustman et al. [33] found that CBT was overall effective in reducing depressive symptoms, but that patients scoring lower on compliance with self-monitoring of blood glucose (SMBG) and with presence of diabetes complications achieved less remission of depressive symptoms than did patients with higher compliance with SMBG and no diabetes complications. Accordingly, diabetes symptom burden or diabetes-related distress might interfere with depression treatment and thereby negatively influence its effectiveness.

Comparatively, fewer studies have been conducted examining potential predictors of MBCT outcome and of these studies, the majority focused on recurrently depressed patients in remission. Kuyken et al. [34], investigated several possible general prognostic predictors like demographics, severity of depression, and depression recurrence and found that only gender was a significant predictor. Female participants showed a greater reduction in depressive symptoms at 15-months follow-up as compared to males. Yet, as the number of males in the study was low, the result should be replicated before drawing firm conclusions. In addition, another study by Nyklicek et al. [35] also found that females showed larger decreases in depression after MBCT as compared to males. Prior research by Teasdale and colleagues [36,37] reported that for patients with three or more previous depressive episodes, MBCT significantly decreased the risk of relapse, whereas this was not the case for participants with fewer episodes. Another study focusing on personality factors as predictors found that students had more benefit from Mindfulness-Based Stress Reduction when scoring high on neuroticism [38]. Hence, given the scarcity of previous research, firm conclusions about prognostic indicators of response to MBCT cannot be drawn.

Identifying general prognostic predictors within a given treatment may provide indications about who benefits most from that treatment, yet it offers no information on which type of treatment is more effective for a patient with certain characteristics. In this context, it is important to consider the interaction between patient characteristics and treatment modality by comparing active treatments with each other. The identification of those so-called prescriptive predictors, or moderators, can indicate subpopulations that fare better with one treatment than the other. Gaining insight into which treatment works best for whom can enhance optimal treatment selection, resulting in optimized clinical decision making and a more personalized care.

To date, only three RCTs investigated prescriptive predictors of CBT and mindfulness-based interventions. Manicavasagar et al. [14], investigating the effectiveness of MBCT and CBT for individuals with major depressive disorder, revealed differential effects for history of depression. CBT was more effective for individuals with four or more episodes, while MBCT was equally effective irrespective of the number of previous episodes. Arch et al. [13], although not focusing on depression but on individuals with anxiety, found that severity of current depressive symptoms was a moderator of effectiveness: individuals with no to mild depression had more benefit from CBT, whereas people with moderate to severe depression fared better with the mindfulness intervention. In another study, Zautra et al. [15] compared the effects of CBT and a mindfulness intervention to an education group with respect to several psychological and pain-related outcomes in a non-clinical sample of patients with rheumatoid arthritis. It was revealed that patients with recurrent depressive symptoms had most benefit from the mindfulness intervention. Unfortunately, these studies focused on clinical characteristics only, while other characteristics such as demographics, disease-related characteristics, and personality may also differentially predict change after treatment [18].

To advance the knowledge of prescriptive predictors (i.e., moderators) of the effects of CBT and MBCT, the current study investigated a broad range of patient characteristics to identify prognostic and prescriptive predictors in CBT and MBCT for depressive symptoms in patients with diabetes. We examined clinical meaningful patient characteristics that can be easily assessed by clinicians, namely demographics, clinical characteristics, neuroticism, and diabetes related characteristics. First, we examined which patient characteristics differentially predicted response to CBT and MBCT. Second, we investigated potential prognostic predictors separately for CBT and MBCT. While no clear hypotheses were formulated with respect to demographic, clinical, and disease specific characteristics, we hypothesized that patients scoring high on neuroticism will benefit more from MBCT as compared to CBT.

Materials and methods

Procedure

Patient recruitment took place from June 2011 to February 2013 and the last follow-up was concluded in April 2014. The trial was conducted in the Netherlands and included 94 participants. The patient flow is shown in Fig 1. Patients were recruited through standard screening at somatic outpatient clinics, referral by physician, and self-referral. In case of elevated levels of depressive symptoms, patients were invited for an intake during which they were assessed for eligibility. When criteria were met and patients gave written informed consent for participation, they were included in the study. Computerized randomization was carried out stratified by gender, use of antidepressant medication, and baseline BDI-II score. Before randomization, patients were blinded for the treatment condition. Consequently, patients were only told that they were to be randomized to a psychological treatment for reducing depression, starting within three months after randomization. Researchers performed the randomization and informed the participants. Following randomization, participants received individual CBT (n = 32), individual MBCT (n = 31), or were allocated to a waiting list control condition (n = 31). Patients allocated to the control group were randomized for the second time at the end of the three-months waiting period and received either CBT (n = 15) or MBCT (n = 16). For the purpose of the present study, participants in the waiting list condition were included in the analyses on condition that patients still had elevated levels of depressive symptoms after the waiting period (i.e., BDI-II ≥ 14). Three of the 31 patients did not score above the cut-off and were excluded, resulting in a total sample of 91 participants.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flow of the study.

https://doi.org/10.1371/journal.pone.0179941.g001

Results

Patient characteristics

The sample consisted of 91 Caucasian participants. Patient characteristics are presented in Table 1. Overall, participants had a mean age of 53, 50% was female, and most patients were in a relationship. There was chance baseline imbalance between groups on several variables. Participants randomized to the CBT condition were older, more likely to have one or more comorbidities, and scored lower on neuroticism as compared to the MBCT condition. These variables were controlled for in relevant analyses by including them as a main effect.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Patient characteristics.

https://doi.org/10.1371/journal.pone.0179941.t001

Prescriptive predictors of change

Data were assessed on multicollinearity. All VIF scores were below 10, indicating no multicollinearity for continuous variables. Type of diabetes and diabetes treatment and type of diabetes and diabetes comorbidity had a phi-coefficient slightly above 0.3 (0.308 and 0.310). Considering that type of diabetes and insulin are also theoretically closely related constructs, yet type of diabetes and diabetes comorbidity not necessarily, we decided to remove diabetes treatment from the analyses.

Fifteen different putative predictors were investigated in three different domains. In the final model combining all significant effects of the analyses within the three domains, a main effect was found for history of care (b = 4.11, SE = 1.64, p = 0.013) and one interaction effect for condition x education (dummy high levels of education; b = 8.17, SE = 3.54, p = 0.021) at post-treatment. For interpretation of the results, means for depressive symptoms and effect sizes of significant predictors are shown in Table 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Mean depression scores for levels of education and history of care, separately for CBT and MBCT on all measurements.

https://doi.org/10.1371/journal.pone.0179941.t002

The interaction effect showed that individuals with high levels of education as compared to individuals with medium or lower levels benefitted less from treatment in the CBT condition than in the MBCT condition (see also Fig 2). Thus, directly after treatment, MBCT might be more advantageous than CBT for individuals with higher levels of educational attainment. Investigating the effect of education on post-treatment depressive symptoms separately for CBT and MBCT revealed that it remained a significant predictor within the CBT condition (b = 7.51, SE = 2.98, p = 0.012), but not within the MBCT condition (b = -1.54, SE = 2.56, p = 0.55). At 9-months follow-up, no significant effects were found, neither main effects, nor interaction effects.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Education as a moderator of depression (BDI-II).

CBT = cognitive behavior therapy; MBCT = mindfulness-based cognitive therapy; T1 = pre- treatment, T2 = post-treatment, T3 = at 9-months follow-up.

https://doi.org/10.1371/journal.pone.0179941.g002

Discussion

The main purpose of the present study was to investigate whether subgroups of patients with diabetes and comorbid depressive symptoms could be identified that benefitted more from CBT than from MBCT or vice versa. Educational attainment was the only factor that differentially predicted treatment outcomes. The subgroup of individuals with higher educational attainment (relative to individuals with medium or lower levels) reported more favorable outcomes in the MBCT condition than in the CBT condition at post-treatment. This effect was found directly after treatment and did not sustain at 9-months follow-up. For all other demographics, disease-related characteristics, clinical and personality factors, no moderating effects were found. With respect to prognostic predictors (i.e. predictors within treatment conditions), individuals with a history of psychological or psychiatric treatment (relative to those without a history of care) had poorer treatment outcomes in CBT and MBCT. Again, this effect was only evident directly after the intervention and did not remain at 9-months follow-up.

One finding is that level of education was found as a prescriptive predictor of treatment outcome in CBT and MBCT directly after the intervention. Yet, education could not be identified as prognostic predictor within the MCBT condition, but only within the CBT condition. As noted in the introduction, earlier studies that directly compared CBT and MBCT investigated only clinical factors and overlooked other potential prescriptive predictors such as demographics and personality. Previous investigations comparing CBT with Interpersonal Psychotherapy (IPT) or examining CBT and MCBT separately, have neither found that education was a prescriptive nor a prognostic predictor of treatment outcome [26,48,49]. As our findings are inconsistent with findings of earlier studies and as the effect was only temporary, it might be due to chance. Replication of this result with a larger sample size is needed to estimate the magnitude of our finding.

A prognostic predictor irrespective of treatment condition was history of psychological treatment. Patients who received psychological or psychiatric therapy prior to the study had on average lower improvements in depression scores after treatment. Again, when investigating long-term effectiveness, history of treatment did not have an effect. To the best of our knowledge, no existing study investigated the predictive effects of history of psychological treatment. Strikingly, history of recurrent episodes of depression did not predict response to CBT or MBCT in the current study. Also, history of treatment was not related to history of depression, therefore it is unclear whether previous treatment was related to a depressive episode or other psychological problems. Accordingly, more research is needed to assess the predictive influence of history of psychological treatment and care on the effectiveness of CBT and MCBT.

An important observation from this study is that only one prognostic predictor of treatment outcome could be identified, which, in addition, was solely significant immediately after treatment and not at 9-months follow-up. Consistent with previous research [30,33], most demographics, clinical characteristics, or diabetes related characteristics did not moderate or predict outcome. Thus, there was no evidence that either CBT or MBCT was more effective for individuals with certain age, gender, marital status, type of diabetes, or diabetes-related distress. In addition, current results add to the body of research proposing that presence of major depression and history of depressive episode neither moderate nor predict outcome [34,50–53]. This finding seems to be in contradiction with previous MBCT literature stating that MBCT is only effective in preventing recurrence of depression in patients with three or more previous depressive episodes [36,37] and in line with a study finding that MCBT was also effective for individuals with one or two prior episodes [54]. Yet, those studies focused on recurrently depressed patients in remission, whereas the current study investigated patients with current depressive symptoms.

Another main finding was that, in contradiction with our hypothesis, neuroticism neither moderated nor predicted the effects of CBT and MBCT. Whereas some research suggests a tendency to attribute high levels of neuroticism to disadvantageous CBT outcomes [30–32] and beneficial MBCT outcomes [38], other studies cannot validate this trend [30,35]. Those studies were, however, not conducted in patients with diabetes. In a recent study in patients with diabetes, negative affect, a concept closely related to neuroticism, was not identified as a predictor [35]. It is therefore not surprising that our study found no effect.

A plausible explanation of current findings is that CBT and MBCT are accessible interventions that are effective for a broad population of patients with diabetes, including for example old and young people and individuals with severe or mild symptoms. Nevertheless, it could be possible that other factors that are less easy to assess for clinicians, such as attachment style, or coping style would differentially predict treatment outcome. Given the scarcity of previous research comparing CBT and MBCT, more studies are needed to explore other potential prognostic factors and draw firm conclusions.

As CBT and MBCT are general, not diabetes specific interventions to improve depressive symptoms, we believe that current findings can be generalized to other populations. No diabetes specific characteristic was identified as a prognostic predictor, which might indicate that disease related characteristics are less important in choosing the most effective treatment for a given person.

The main limitations of the randomized trial (as also discussed in our earlier paper) include an underpowered sample size and relatively high attrition rates [16]. The current study is therefore likely to be underpowered and analyses are seen as exploratory. Another limitation of this study is that history depression and treatment were assessed by means of self-report, which has possibly resulted in bias.

Conclusion

The current study is the first to directly compare a range of prognostic and prescriptive predictors for CBT and MBCT for the treatment of depressive symptoms in patients with diabetes. As both CBT and MBCT are effective interventions for reducing depressive symptoms, identifying for whom which treatment may be the most beneficial has implications for clinical decision making. Despite examining a large number of potential prognostic and prescriptive predictors, only few variables influenced treatment response and the effects did not sustain in the long term. These results suggest that patient characteristics only make a modest contribution to treatment outcome and that both treatments are effective for individuals with different characteristics. Future research could consider the contribution of other plausible prognostic factors, or map differences in the therapeutic process, to further advance our understanding of differential response to psychotherapy.

Supporting information

Acknowledgments

The authors thank all the patients who participated in the study, the psychologists who delivered the MBCT and CBT sessions and the medical specialists, secretaries as well as the research assistants for their effort. The authors are particularly grateful to Annemieke Roos and Klaas Hoogenberg from the Martini Hospital Groningen, Jet de Hoop from the Medical Center Leeuwarden and Nynke Rauwerda from the Hospital Rivierenland Tiel for facilitating the study.