Treatment for comorbid depressive disorder or subthreshold depression in diabetes mellitus: Systematic review and meta‐analysis

Abstract Objective To provide an estimate of the effect of interventions on comorbid depressive disorder (MDD) or subthreshold depression in type 1 and type 2 diabetes. Methods Systematic review and meta‐analysis. We searched PubMed, PsycINFO, Embase, and the Cochrane Library for randomized controlled trials evaluating the outcome of depression treatments in diabetes and comorbid MDD or subthreshold symptoms published before August 2019 compared to care as usual (CAU), placebo, waiting list (WL), or active comparator treatment as in a comparative effectiveness trial (CET). Primary outcomes were depressive symptom severity and glycemic control. Cohen's d is reported. Results Forty‐three randomized controlled trials (RCTs) were selected, and 32 RCTs comprising 3,543 patients were included in the meta‐analysis. Our meta‐analysis showed that, compared to CAU, placebo or WL, all interventions showed a significant effect on combined outcome 0,485 (95% CI 0.360; 0.609). All interventions showed a significant effect on depression. Pharmacological treatment, group therapy, psychotherapy, and collaborative care had a significant effect on glycemic control. High baseline depression score was associated with a greater reduction in HbA1c and depressive outcome. High baseline HbA1c was associated with a greater reduction in HbA1c. Conclusion All treatments are effective for comorbid depression in type 1 diabetes and type 2 diabetes. Over the last decade, new interventions with large effect sizes have been introduced, such as group‐based therapy, online treatment, and exercise. Although all interventions were effective for depression, not all treatments were effective for glycemic control. Effective interventions in comorbid depressive disorder may not be as effective in comorbid subthreshold depression. Baseline depression and HbA1c scores modify the treatment effect. Based on the findings, we provide guidance for treatment depending on patient profile and desired outcome, and discuss possible avenues for further research.


| SUMMATIONS
This systematic review and meta-analysis exploring psychotherapeutic, pharmacologic, and other interventions shows beneficial treatment effects for comorbid depression in type 1 and type 2 diabetes mellitus with moderate-to-large effect sizes for most intervention types.
Although all interventions were effective for depression, not all treatments were effective for glycemic control.
Effective interventions in comorbid depressive disorder may not be as effective in comorbid subthreshold depression.

| LI M ITATI O N S
Most of the selected studies did not meet all criteria to reduce the risk of bias and not all provided sufficient data to be included in the meta-analysis.
Further, some treatments were only evaluated in a single RCT.
There is a scarcity of data from many low-and middle-income countries.

| INTRODUC TI ON
No international consensus exists to guide treatment of comorbid depression in diabetes. Nonetheless, over the last three decades, clinicians have been seeing increasing numbers of patients with comorbid depression of various severity in diabetes (Khaledi et al., 2019;Zheng et al., 2018) due to the exploding prevalence of both diabetes and depression (GBD Disease & Injury Incidence & Prevalence Collaborators, 2018). This can amount to up to 30% depending on severity of symptoms and it occurs especially where the person with diabetes has elevated HbA 1 c despite treatment, or frequent episodes of hypoglycemia and increased glucose variability, diabetesrelated complications, and disengagement from treatments (Groot et al., 2001;Lustman, Anderson, et al., 2000;O'Connor et al., 2009).
Depression is a serious psychiatric disorder characterized by loss of interest or pleasure, depressed mood, and suicidal behavior (Ruengorn et al., 2012). Diabetes and depression can both seriously affect an individual's quality of life, and lead to functional disability, increased distress, and social burden (Renn et al., 2011). Depressive symptoms in people with diabetes can have a detrimental impact on engagement with diabetes management (Ciechanowski et al., 2000;Gonzalez, Peyrot, et al., 2008) and on glycemic control (Lustman, Anderson, et al., 2000) as well as on health-related outcomes (e.g., weight gain and diabetes-related complications) and associated healthcare costs (Black et al., 2003) As such, the high prevalence of this comorbidity is accompanied by high rates of morbidity and mortality worldwide (Hofmann et al., 2013;Lloyd et al., 2018;Nouwen et al., 2019). Epidemiological studies indicate there is a bidirectional relationship between diabetes and depression (Golden et al., 2008;Katon, 2008;Katon et al., 2007), in which individuals with diabetes have an increased risk of depression and vice versa; the presence of a depressive disorder can increase the risk of metabolic diseases such as diabetes (Renn et al., 2011) and there is an association between depression and diabetes complications (Groot et al., 2001;Van Steenbergen-Weijenburg et al., 2011).
Evidence is growing to suggest that depression may play a role in the pathogenesis of diabetes in a number of ways. Depression may be a consequence of similar environmental factors that govern glucose metabolism, and can also independently influence nutrition and lifestyle choices which can predispose individuals to the development of diabetes (Beydoun & Wang, 2010). Biological mechanisms have also been proposed through a dysregulated and overactive HPA axis, a shift in sympathetic nervous system tone toward enhanced sympathetic activity, and a pro-inflammatory state (Champaneri et al., 2010;Joseph & Golden, 2017). The role of inflammation is particularly pertinent. Laake et al. (2014) found that increased inflammation may be involved in the pathogenesis of depression in people with type 2 diabetes, which in turn could contribute to the increased risk of complications and mortality in this clinical population (Geraets et al., 2020).
The relationship between depressive symptoms and poorer diabetes self-care  applies also to subclinical or subthreshold depressive symptoms (Pibernik-Okanović et al., 2011) and not only to major depressive disorder. Subthreshold refers to those with two or more depressive symptoms who do not meet the diagnostic criteria for depression (Rodríguez et al., 2012).
Subthreshold depressive symptoms in people with diabetes have been found to be persistent but also associated with an increased risk of worsening over time (Bot et al., 2010;Nefs et al., 2012;Pibernik-Okanovic et al., 2008). Furthermore, an increased incidence of adverse health outcomes and suboptimal metabolic control has been observed not only in patients with the established diagnosis of depression but also in those suffering subthreshold depressive symptoms (Johnson et al., 2014). This indicates that even mild depression is clinically relevant, and implies that combined treatments could also be efficacious for people with diabetes and subthreshold depressive symptoms.
A lack of a clear understanding of the shared origins of depression and diabetes means that finding the most appropriate treatment guidance for treatment depending on patient profile and desired outcome, and discuss possible avenues for further research.

K E Y W O R D S
depression, diabetes mellitus, glycemic control, meta-analysis, systematic review, treatments for this comorbidity in this vulnerable patient group is difficult. In order to optimize health outcomes, feasible and effective interventions aiming to provide benefits to both physical and mental health are recommended (Baumeister & Bengel, 2012;Baumeister et al., 2014;Harkness et al., 2010). The focus of treatment strategies should be on the remission or improvement of depression, in addition to improvement in glycemic control as a marker of diabetes outcome .
Evidence shows that comorbid depression in diabetes can be treated with moderate success by psychological and pharmacological interventions, often implemented by using collaborative care (Katon, Von Korf, et al., 2004) and stepped care approaches (Stoop et al., 2015). However, there is conflicting evidence for the efficacy of antidepressants and psychological therapy in the improvement of glycemic control (Lustman, Anderson, et al., 2000;Lustman et al., 1997Lustman et al., , 1998aLustman et al., , 2000bLustman et al., , 2007. Petrak, Herpertz, et al. (2015)) claim that more research is needed to evaluate treatment of different subtypes of depression in people with diabetes as well as the effectiveness of new approaches to treatment.

| Rationale and objective
A previous systematic review of treatments for comorbid depression in diabetes indicated favorable effects on depressive outcome according to rating scales (Van der Feltz-Cornelis et al., 2010), but did not include data for subthreshold depression, which has been found to be related to poorer diabetes outcomes similar to DSM-5 depressive disorder Pibernik-Okanović et al., 2011). We updated and expanded this systematic review and meta-analysis of randomized controlled trials to provide an estimate of the effect of interventions for comorbid depressive disorder or subthreshold depression in type 1 diabetes and type 2 diabetes.
The interventions were compared with care as usual (CAU), waiting list (WL), placebo or another active comparator (e.g., another antidepressant or psychotherapy) on depression outcome and glycemic control, and, if possible, to provide treatment guidance for this condition.

| ME THOD
This systematic review and meta-analysis was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (Liberati et al., 2009). We searched MEDLINE, Embase, the Cochrane Central Register of Controlled Trials, and Web of Science using Ovid software. The full search strategy and keywords used have been published elsewhere  and are shown in the appendix (pp 1-2). The reference lists of selected RCTs and reviews were checked for relevant studies that were not included in the databases. The search was supported by the Centre for Reviews and Dissemination at the University of York.
The protocol for this review is registered on PROSPERO and can be found here: https://www.crd.york.ac.uk/prosp ero/displ ay_record.php?ID=CRD42 01914 7910 The final search results were restricted to studies completed before 28th August 2019. Inclusion criteria for studies were that they should be randomized clinical trials, provide a treatment intended to have an effect on both comorbid depressive symptoms and glycemic control in type 1 diabetes and/or type 2 diabetes, and have a control arm (e.g., CAU, placebo, WL or active comparator). The intervention had to be described sufficiently in order to be classified as a psychotherapeutic, medical, pharmacological, collaborative care or other type of intervention. A glossary providing an explanation about the interventions and a list of acronyms are provided in the appendix (pp 3-4).
Participants were adult patients with diabetes and comorbid depressive or subthreshold depression, which was defined as the presence of two or more core depressive symptoms, but not meeting the DSM-5 diagnostic criteria for depressive disorder (Rodríguez et al., 2012). No restriction was placed on type of intervention or publication language. Studies were not included if depressive disorder or depressive symptoms were not established in a systematic manner such as by semistructured interview or questionnaire at baseline. Studies were selected in a two-stage process. First, titles and abstracts from the electronic searches were scrutinized by two independent reviewers (SA and CFC). Second, if the abstract met inclusion criteria, we obtained full texts and final decisions were made about study inclusion. Disagreement regarding inclusion status was discussed. Consensus was reached in all cases.
Two reviewers (SA and CFC) independently extracted data for participants' characteristics, interventions, and study outcomes. A proforma as used in the original systematic review (Van der Feltz-Cornelis et al., 2010) was used to extract data from the included studies, now also including subthreshold depression from the search hits. The extracted data included: author and year; country; study type; sample size; age; baseline depression measure/ diagnostic tool; baseline depression score, baseline glycemic control score, intervention details; control group, length of follow-up; diabetes and depression outcomes with regard to: i) the change in depression score from baseline to last follow-up using any validated self-report measure of depressive symptomatology and ii) the change in levels of biological marker of glycemic control from baseline to last follow-up. Assessment of glycemic control could be using HbA 1 c, which provides an integrated measure of mean blood glucose levels over the last 6-8 weeks, or FBG, which gives an indication of the blood glucose concentration at the moment of assessment. If both were reported, we used the HbA 1 c to calculate a standardized mean difference. The difference in means of each outcome was the primary measure within each study.
Additional outcomes on adherence to recommendations of healthcare providers with regard to self-care behaviors were extracted if reported. Authors were approached for additional data when questions arose.

| Risk-of-bias assessment
The Cochrane risk-of-bias tool (McGuire et al., 1998) was used to assess random sequence generation (selection bias), allocation concealment (selection bias), blinding of participants and personnel (performance bias), blinding of outcome assessment (detection bias), incomplete outcome data (attrition bias), selective reporting (reporting bias), and other biases. Risk of bias was assessed by SA and CFC independently.
Initial disagreements were resolved by consensus (Appendix pp 6-9). As psychotherapy trials often have limitations in the possibility for blinding (Van der Feltz-Cornelis & Ader, 2000), studies with limited blinding procedures were not excluded from the analysis. GRADE assessments were made (Guyatt et al., 2008) to give the confidence in each reported effect size. They are reported in the appendix (Appendix page 6-9).

| Statistical analysis
As a first step, overall meta-analysis was performed for all RCTs comparing all treatments with CAU, WL, or placebo for the combined effect on depressive outcome and glycemic control (illness burden). Then, we performed an analysis of illness burden in the studies reporting on depression versus the studies reporting on subthreshold depression. Then, studies were grouped according to the mode of treatment (pharmacotherapy, psychotherapy, collaborative care, online, phone and group interventions, exercise), depression severity (both as depression scores at baseline, and as classification of major depressive disorder or subthreshold depression), and depressive or diabetes outcome. Effect sizes were calculated. Outcomes from individual studies were pooled using a random-effects model (DerSimonian & Laird, 1986), as this approach assumes that there could be clinical and methodological heterogeneity that might affect the findings. All pooled analyses were reported with 95% confidence intervals (CIs). The effects were presented in terms of standardized effect sizes (Cohen's d). An effect size of 0.5 indicates that the mean of the experimental group is half a standard unit larger than the mean of the control group. It is generally assumed that an effect size of 0.56-1.2 represents a large clinical effect, while effect sizes of 0.33-0.55 are moderate, and effect sizes of 0-0.32 are small (Lipsey & Wilson, 1993). A meta-regression was conducted to assess whether baseline levels of depressive severity (scores on depression questionnaires) (Appendix pp.15) or glycemic control (HbA 1 c) influenced the effect of the intervention. Between-study heterogeneity was assessed using the I (Khaledi et al., 2019) statistic (Higgins, 2003). Publication bias was examined by constructing a Begg funnel plot (Begg, 1994) and Duvalls trim and fill (Rothstein et al., 2005). We adhered to published guidance of the Cochrane handbook (Higgins et al., 2019) throughout. We used the statistical program Comprehensive Meta-Analysis, version 2 (Biostat, 2005) to conduct random-effects meta-analyses.

Results are shown in
Overall meta-analysis in the RCTs comparing all treatments with CAU, WL, or placebo for the combined effect on depressive outcome and glycemic control showed an effect size of 0.485; 95% CI 0.360; 0.609, p < .0001 (Appendix pp 10-12).

Yes No
Depression: Both groups improved significantly in HDRS scores (mean difference 0.62; p=.003) Diabetes: No difference in HbA1c (mean diff 0.11; n.s.) No significant difference between both conditions. This study is not pooled in the meta-analysis.

n/a
Depressive symptoms but not glycemic control improved in both MgCl2 and imipramine groups. No control group so study not included in meta-analysis.   n/a Escitalopram appears to be better than Agomelatine for improving both depression and glycemic control. No control group so study not included in meta-analysis.

Diabetes: No significant improvement
in HbA 1 c levels.
n/a No differences in improvements in depressive symptoms between IPT and sertraline. No significant effect on glycemic control was shown for either intervention.

n/a
Sertraline and CBT both improve depression after 12 weeks. Significant advantage of sertraline over diabetes specific CBT for improving depressive symptoms over one year, but not glycemic control.

No No
Depression: BDI total score, mean difference 5.6; p=.03 Diabetes: HbA 1 c, no significant difference, no outcome reported.
Depression: Δ − 0.868 Diabetes: Δ 0 Poorly controlled (HbA1c ≥ 9%) as inclusion criterion. Improvement in depression but not in glycemic control in nortriptyline versus. control. Nortriptyline may have negative impact on glycemic control.

No No
Depression: Treatment x time interaction effect on CES-D scores (p<.001) was significant. Diabetes: No significant treatment effect found for HbA 1 C levels (p >.05).

n/a
Significant improvement in depressive symptoms but not glycemic control in webbased-CBT group versus active control. Study not entered in meta-analysis due to lack of reported data.

Depression: Δ0.735
Diabetes: Δ0.133 Significantly greater improvement in depressive symptoms in internet guided self-help versus active control. No effect on glycemic control.

Sub No No
Depression: PHQ9 scores improved overall and the group x time interaction was significant (p<.001). 51% in iCBT versus 18% in TAU improved reliably. Diabetes: No significant interaction effect for HbA1c levels (p=.750).

Depression: Δ0.782
Diabetes: Δ 0.142 Significantly greater improvement in depressive symptoms but not glycemic control in Web-based CBT group versus care as usual.
No follow-up data for care as usual group limits conclusions.

No No
Depression: CES-D scores mean difference = -6.8 (p<.01). At 6 months 35% of intervention versus 80% of control remained depressed. Diabetes: No significant improvements for FBG or HbA1c levels.
Depression: Δ0.964 Diabetes: Δ0.272 Significant improvement in depressive symptoms but not glycemic control in SWEEP psychoeducation group compared to control group.  Note: The first column indicates the first author, year of publication and country study was conducted. The second column shows the sample size, % type 1 diabetes and type 2 diabetes and the Mean[SD] age of participants. The third column indicates how depressive disorder/presence of clinically significant symptoms or subthreshold disorder was diagnosed or defined. The fourth column describes the intervention, including the follow-up (FU) time periods. Column 5 shows the Baseline data for both diabetes (e.g., HbA 1 c) and depression (e.g., depression questionnaire) outcomes. Column 6 shows the outcome data for both the diabetes and depression outcomes. Column 7 shows the effect size of the intervention on both the diabetes and depression outcomes. Column 8 describes the conclusions drawn from the study. Column 9 indicates whether the study focused on participants with depressive disorder or clinically significant symptoms (as noted by MDD) or subthreshold disorder (sub). Columns 10 and 11 show whether the intervention included an intervention component or focus on adherence, respectively. The number of trials and participants for each intervention is shown in the row indicating intervention type.  Higgins et al., 2019;Johnson et al., 2014;Nefs et al., 2012;Pibernik-Okanović et al., 2011;Pibernik-Okanovic et al., 2008;Rodríguez et al., 2012;Simson et al., 2008)

| D ISCUSS I ON
This systematic review and meta-analysis shows beneficial treatment effects for comorbid depression in type 1 and type 2 diabetes Significantly greater improvement in depressive symptoms and glycemic control in boxing intervention group versus control group.

Depression: Δ0.342
Diabetes: Significantly greater improvement in depressive symptoms but not glycemic control in HOPE telehealth intervention versus care as usual control group.
Depression: Δ 0.722 Diabetes: Δ −0.032 Light therapy was not significantly better at reducing depressive symptoms in comparison to placebo, and had no effect on glycemic control.  (Koopmans et al., 2009), and this may be worse in case of comorbid depression Lysy et al., 2008 This review also shows that interventions that are effective in depressive disorder may not be as effective in subthreshold depression. In this group, psychotherapy and online treatment had large, significant effect sizes on depressive symptoms, but group therapy and psychoeducation were not effective. Looking at glycemic control as an outcome, psychotherapy had a large, significant effect and group-based therapy had a small, significant effect, while online treatment and psychoeducation had no significant effect at all. Consequently, the preferred treatment for both depression and glycemic control in comorbid subthreshold depression would be psychotherapy.

No No
The finding that psychoeducation is not more effective than CAU in subthreshold depression, both for depression outcome and glycemic control, is an important finding as in stepped care models, psychoeducation has been suggested as a first step in diabetes-related distress or subthreshold depressive symptoms (Huang et al., 2013). Furthermore, psychoeducation was supposed to be a good start for improving self-management and in that way improving glycemic control. This line of thought is not supported by our results. Also, the finding that group therapy is highly effective in depressive disorder, but not in subthreshold depression, might suggest that patients with subthreshold depression might benefit more from individual treatment tailored to their specific needs rather than from group participation, something that has been suggested earlier (Huang et al., 2013). Treatment of comorbid subthreshold depressive disorder could be psychotherapy both in patients with elevated or normal HbA 1 c. The latter group might also benefit from online treatment. If glycemic control is a target, our analysis shows that it makes sense to target patients with high baseline levels of depression and of HbA 1 c, as they are likely to benefit most from treatment on both symptom levels.
In our flowchart, we recommend collaborative care in comorbid MDD and multimorbidity or problems requiring complex case management. Although effect sizes for some other treatment modes are found to be larger in our meta-analysis, none of those were evaluated in patients with such a complex and multimorbid profile, whereas several systematic reviews show that outcomes in such patient groups improve by collaborative care (Faridhosseini et al., 2014;Tully & Baumeister, 2015).
One RCT (Guo et al., 2014) found that metformin improved glycemic control but also depressive outcomes, compared to placebo, in patients with type 2 diabetes. Although a small study with only 58 participants, this finding is of interest and may contribute to the expanding field of evaluation of medicines that are normally prescribed for physical conditions for their effect in treatment of depression (Arteaga-Henríquez et al., 2019;Che et al., 2018;Köhler et al., 2014).
Further research could explore the mechanism for metformin in im-

provement of depression in diabetes.
Our study has several strengths. First, we included data without language restriction from studies identified by a comprehensive search of the published literature. We included studies exploring the effect of treatment in subthreshold depression. Our sensitivity analysis excluding high risk-of-bias studies confirmed the findings, the fixed model meta-analysis refuted the null hypothesis, and we found no indication for publication bias. Second, we provided relative effect sizes for several new treatment modalities compared to the treatments already explored in the first systematic review, we differentiated the treatment effect on depressive outcomes versus glycemic control, and by performing meta-regression we showed the influence of baseline depression severity on both depression outcome and glycemic control, whereas baseline HbA1c only influenced glycemic control as an outcome. This combination of findings enabled us to provide clinicians with innovative guidance about which interventions may suit best, depending on patient profile. These strengths make our study the most comprehensive systematic review and meta-analysis of treatment for comorbid depression in diabetes yet undertaken.
Our analysis has several limitations. First, most of the included studies did not meet all criteria to reduce risk of bias, mostly due to unclear reporting and to small samples. Despite our efforts to contact authors for missing data, we were unable to include such data in three studies due to lack of response (Ell et al., 2011;Petrak, Herpertz, et al., 2015;van der Sluijs et al., 2018), which may have to do with the long timeframe of this systematic review. The need for low risk-of-bias studies in this field remains, with proper reporting of methodology and of outcomes. Second, the planned moderator analyses on the effect of add-on exercise on treatment outcome and on adherence as an outcome of treatment could not be performed because of insufficient data (Appendix pp. 22). Third, some treatments were only evaluated in one RCT. This probably reflects that, although many of these "new treatments" have been used for some time and have been felt to be useful by patients and clinicians, at least in primary care, researchers had not actively examined these "new" treatments until recently. In view of their clinical relevance, we emphasize this limitation. We strongly suggest further research is needed especially in group-based treatment and exercise, that seem to have promising results. Another limitation concerns the provenance of the studies. Although this is a study with a global perspective in terms of included studies, it is clear that there is a scarcity of data from many low-and middle-income countries, as shown by the map in Figure 1. The imbalance is of growing importance because it is likely that the low-and middle-income countries will have the greatest increases of comorbidities of prevalence and incidence of diabetes and depression. In countries in which the attention to mental health problems is minimal or absent and the investment in the care for diabetes is appropriate, the guidance for treatment that we could deduce from this systematic review and meta-analysis is particularly relevant and may improve care for comorbid depression.
Furthermore, the studies in this meta-analysis do not present results for type 1 diabetes and type 2 diabetes separately despite the different types of diabetes affecting different groups of the population; for example, type 2 diabetes tends develop more commonly in older people compared with a peak incidence of type 1 diabetes in adolescence and young adulthood. The lack of studies in type 1 diabetes alone with comorbid depression or comorbid subthreshold disorder is striking and research is needed to fill this gap.
A clearer understanding of the mechanisms underpinning why some treatments are more effective for patients with depressive disorder than for subthreshold depression and vice versa would also greatly benefit this area of research and for this purpose studies might provide more detailed information about the contents of the intervention. In particular, the idea that interventions aiming to improve self-management lead to better adherence and better diabetes and depression outcomes should be challenged in research as studies reporting on adherence as an outcome are lacking. Studies are also needed to develop standardized techniques or tools to help better identify particular subtypes of patients taking into account their depression severity and glycemic control. These suggestions will further aid in the identification and personalization of appropriate treatment plans for patients with diabetes and depression as outlined above.

ACK N OWLED G M ENT
Not Applicable.

AUTH O R CO NTR I B UTI O N S
This systematic review was designed by CFC and co-authors.
Screening and data extraction were completed by SA and CFC.
The meta-analysis was performed by CFC. The initial version of the manuscript was written by SA and CFC. Following this, input was provided from all authors. All authors approved the final version of the manuscript.

FU N D I N G I N FO R M ATI O N
This study was financially supported by Hull York Medical School.
The funder had no role in study design, data collection, data analysis, data interpretation, writing of the report, and decision to submit the paper for publication. The corresponding author had full access to all data in the study and had final responsibility for the decision to submit for publication.

E TH I C A L A PPROVA L
Ethical approval was not required for the current study as the data entered in the meta-analysis were collated from previous clinical trials in which informed consent had already been obtained.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data available on request.