Depression and Risk of Mortality in People with Diabetes Mellitus: A Systematic Review and Meta-Analysis

Objective To examine the association between depression and all-cause and cardiovascular mortality in people with diabetes by systematically reviewing the literature and carrying out a meta-analysis of relevant longitudinal studies. Research Design and Methods PUBMED and PSYCINFO were searched for articles assessing mortality risk associated with depression in diabetes up until August 16, 2012. The pooled hazard ratios were calculated using random-effects models. Results Sixteen studies met the inclusion criteria, which were pooled in an overall all-cause mortality estimate, and five in a cardiovascular mortality estimate. After adjustment for demographic variables and micro- and macrovascular complications, depression was associated with an increased risk of all-cause mortality (HR = 1.46, 95% CI = 1.29–1.66), and cardiovascular mortality (HR = 1.39, 95% CI = 1.11–1.73). Heterogeneity across studies was high for all-cause mortality and relatively low for cardiovascular mortality, with an I-squared of respectively 78.6% and 39.6%. Subgroup analyses showed that the association between depression and mortality not significantly change when excluding three articles presenting odds ratios, yet this decreased heterogeneity substantially (HR = 1.49, 95% CI = 1.39–1.61, I-squared = 15.1%). A comparison between type 1 and type 2 diabetes could not be undertaken, as only one study reported on type 1 diabetes specifically. Conclusions Depression is associated with an almost 1.5-fold increased risk of mortality in people with diabetes. Research should focus on both cardiovascular and non-cardiovascular causes of death associated with depression, and determine the underlying behavioral and physiological mechanisms that may explain this association.


Introduction
Depression is common in people with diabetes, affecting approximately 20% of the patients [1,2]. Two meta-analyses revealed that people with type 2 diabetes mellitus are 15-24% more likely to develop depression compared to people without diabetes [3,4]. Furthermore, individuals with previously diagnosed diabetes have an increased risk of depression relative to those with impaired glucose metabolism or undiagnosed diabetes [5]. Depressed individuals with diabetes report lower quality of life [6] have higher HbA 1c levels, indicating suboptimal glycemic control [7] and are characterized by poor self-care behavior that may contribute to suboptimal glycemic control [8]. They demonstrate lower levels of physical activity [9], have more negative appraisals of insulin therapy [10], are likely to be less adherent to the prescribed treatment regimen and have less healthy eating behaviors [8].
Several longitudinal epidemiological studies concluded that the combination of diabetes mellitus and depression is associated with higher mortality rates [11,12,13]. For example, Black et al. [11] demonstrated that people with comorbid depression and diabetes have a higher risk of developing diabetes-related complications and also mortality, than those with depression or diabetes alone or without either condition.
Lin and colleagues [14] showed that people with type 2 diabetes and comorbid depression had a 36% increased risk of developing microvascular complications such as end-stage renal disease, low vision or blindness, retinopathy, foot ulcers or amputations, compared to individuals with diabetes without depression. Furthermore, a 25% higher risk of developing macrovascular complications, such as myocardial infarction or stroke, was established.
However, the research question whether depression in diabetes is associated with an increased mortality risk has not been the subject of a systematic review and meta-analysis. Therefore, the objective of the present paper was to examine whether depression increases the risk for all-cause and cardiovascular mortality in people with diabetes, both by (a) reviewing the literature in a systematic way and (b) carrying out a meta-analysis of longitudinal studies on this subject.

Study Selection
Two authors (FvD, GN) independently evaluated the articles for eligibility. Studies that met the following criteria were included: 1) the study design was longitudinal, including both retrospective as prospective studies, 2) the study population included people diagnosed with diabetes (clinical or self-report) either as total sample or subgroup, 3) the outcome variable was mortality, 4) the association between baseline depression (yes/no, clinical diagnosis or self-report) and mortality during follow up was analysed. Studies presenting clinical trials meeting these inclusion criteria were not included in the meta-analysis, considering these studies describe the effect of a treatment in people with diabetes and depression on mortality and not the effect of depression on mortality in people with diabetes.
All discrepancies were resolved after rechecking the source papers and further discussion among both authors, and consultation of a third co-author (FP) where needed, with full consensus before inclusion. Regarding multiple reports on the same dataset, only one report was included in the meta-analysis, based upon population size (largest sample size), aim of the article (with mortality as the main end point) and primary analysis (as compared to secondary analysis). Three corresponding authors were contacted for additional information, but we did not receive further information. No restriction on type of language was placed. Figure 1 depicts the process of article selection by means of a flow diagram.

Endpoint
The endpoint was all-cause mortality and, where available, cardiovascular mortality.

Data Extraction and Statistical Analysis
The results of the data extraction were summarized in a systematic manner including the following information: first author name, publication year, country of study, length of follow-up, study design and sample size (mean age, % female, type of diabetes), number of depression cases, assessment method of diabetes/depression/mortality (Table 1), hazard ratios (HRs) with corresponding 95% CI and multivariable adjustment ( Table 2).
Data from all studies were pooled using the program Comprehensive Meta-Analysis version 2 (Biostat, Englewood NJ, 2005). If multiple hazard ratios were presented in a given article, the estimate that most closely adjusted for demographic variables and micro-and/or macrovascular complications (e.g. retinopathy, neuropathy) was selected, to expose the independent effect of depression.
When appropriate, the meta-analysis will be performed using the fixed effects or the random effects model: in case of homogeneity (or low heterogeneity) the fixed effects model will be used. If heterogeneity is substantial (above 50%) the random effects model will be used. In two studies [11,15] more than one measure of depression was used. Black and colleagues [11] measured depressive symptoms at baseline with the CES-D, and also used a modified version of the CIDI at two years follow up to determine whether patients suffered from a depressive disorder. Depressive symptom-scores were included, as they were measured at baseline. Sullivan et al. [15] reported minor and major depression based on the PHQ-9, the dichotomisation of PHQ-9$10 and the continuous PHQ-9 score. As combining major and minor depression into one estimate would require an additional step, the readily available dichotomised PHQ-9 score was used.
For three studies [11,14,16] hazard ratios needed to be combined in order to obtain the estimate of interest (example of procedure explained in Figure S1 in Supporting Information S1). In one paper [16] the hazard ratios for men and women were separately reported. In the case of Lin et al. [17] we assembled the hazard ratio for minor and major depression into one depression hazard ratio. For Black et al. [11], we combined the group with minimal depression (CES-D = 1-15) and without any depressive symptoms (CES-D = 0) into one group (CES-D , 16).
Five studies [11,18,19,20,21] used the no diabetes, nondepressed group as the reference category, while we were interested in the comparison of the two diabetes groups only (depressed versus non-depressed). Two of these studies [18,21] performed post hoc analyses in people with diabetes only, producing the desired estimate.
For the other three studies, we used the information from the four group scenario to calculate the HR for the comparison of the depressed and non-depressed diabetes groups (example of procedure explained in Figure S2 in Supporting Information S1).
In two papers [13,22] no hazard ratios or odds ratios were presented, so the odds ratios and confidence intervals for these studies were calculated based on the number of patients who died in each group.
Statistical heterogeneity was assessed using the I-squared statistic, which quantifies the percentage of total variation across studies due to heterogeneity rather than chance, with values of 50% or more indicating a substantial level of heterogeneity [23]. When study outcomes were heterogeneous based on this statistic, the potential influence of follow-up length, age, method of depression assessment, method of diabetes assessment, number of participants and the percentage of females included were examined. Differences in effect estimates between the subgroups were assessed by comparing the pooled effect estimates using chisquared analysis, comparing logarithms of these estimates. Additionally, a sensitivity analysis where each study is removed one by one was done.
Potential publication bias was assessed with Egger's test of the intercept [24]. A funnel plot was constructed by plotting the effect measure against the inverse of its standard error (see Figure S3 in Supporting Information S1. An asymmetric plot indicates a likely publication bias, and p,0.05 is considered representative of statistically significant publication bias.

Results
A total of 400 potentially relevant articles were retrieved by the database searches ( Figure 1). From this set, 34 full-text articles were assessed for eligibility. Of these, 15 articles met our inclusion criteria and were included in the systematic review. Three of these articles [13,22,25] used a different measure of effect size (odds ratio) which cannot be pooled in a meta-analysis with hazard ratios [26]. However, odds ratios can be converted to risk ratios, which then can be combined with hazard ratios in a meta-analysis as forms of relative risks [27].
One additional relevant article that was not indexed for MEDLINE yet, was included in both the review and metaanalysis [15]. The characteristics and extracted data of the 16 articles are presented in Table 1, and the hazard ratios with corresponding 95% confidence intervals and covariates used for analysis in Table 2.

All-cause Mortality
Sixteen studies were included in the meta-analysis, comprising 109046 individuals with diabetes and including 21443 (19.7%) with comorbid depression. The follow-up periods of included studies ranged from 2-10 years, with a mean follow-up of 6 years. The mean age at baseline ranged from 62 to 76 years, with exception of the only article focusing on type 1 diabetes, with a mean age of 39 years at baseline [16]. Twelve analyses (75%) showed a statistically significant association between depression and mortality in individuals with diabetes.
The pooled hazard ratio for mortality was significantly increased in patients with diabetes and depression compared with those without depression (HR 1.46, 95% CI 1.29-1.66, p,0.0001) ( Figure 2). The I 2 value was 78.6%, demonstrating high heterogeneity in the study results. After conducting subgroup analyses, there were no significant differences for follow-up length, age, method of depression assessment, method of diabetes

Cardiovascular Mortality
Five of the 16 studies, comprising 11375 individuals with diabetes and including 2619 (23%) with comorbid depression, also specifically reported on cardiovascular mortality as separate endpoint. Two examined cardiac mortality, two cardiovascular disease mortality and one coronary disease mortality. The followup periods ranged from 5-8 years, with a mean follow-up of 6.6 years. The mean age at baseline ranged from 63 to 68 years. Two out of 5 studies found a significant association between depression and cardiovascular mortality in people with diabetes [19,21], and there was a trend in a third study [17]. After pooling the HRs for cardiovascular mortality, the HR was significant (HR = 1.39, 95% CI 1.11-1.73, p,0.0001) (Figure 2).

Discussion
This meta-analysis of 16 longitudinal studies shows a positive association between depression and subsequent mortality rates in people with diabetes. Compared to those without depression, depressed individuals had a 46% increased risk for all-cause mortality. Although based on only 5 studies, our results also show a 39% increased risk for cardiovascular mortality associated with the presence of depression in diabetes.
Previous meta-analyses have also found positive associations between depression and mortality rates in the general population (RR = 1.81, 95% CI 1.58-2.07) [28] and in patients with  [29]. The triad of depression, diabetes and cardiovascular disease is closely interrelated. Premature cardiovascular disease is the most common cause of morbidity and mortality in people with diabetes [30] and co-morbid depression appears to increase the risk of developing vascular conditions in this group [11,15,17]. However, depression is also common in people with established cardiovascular disease [31]. Rather than being a (in)direct causal factor, depression in diabetes may be secondary to having cardiovascular complications. It may owe its association with mortality to the increased risk of new cardiovascular events in people with established cardiovascular conditions [32]. We took this issue into account by including the risk estimates that were adjusted for existing vascular disease, and still found an almost 1.5-fold increased risk of mortality in depressed people with diabetes. Further prospective studies are needed to examine whether depression exerts a negative influence on mortality through the development of new vascular complications. These studies may also explore whether people with comorbid diabetes and depression face increased mortality risks beyond cardiovascular deaths, as suggested by recent results from the Pathways Epidemiologic Follow-up Study [17].
There are several potential behavioral or physiological mechanisms that could explain the increase of mortality for people with diabetes and depression. Depression is correlated with a decline in health-maintenance behaviors (e.g. physical activity, smoking, diet) in general [33], which is also true for people with diabetes [34]. In addition, depression is associated with several biological alterations; activation of the hypothalamic-pituitary-adrenal axis and proinflammatory cytokines, sympathic nervous system dysregulation, decrease in heart rate variability and cardiac fibrillation threshold, which can contribute to an increased risk of cardiovascular mortality, but also mortality of other causes [35].
In line with previous studies examining post-myocardial infarction depression [29] or depression in community samples [28], we did not observe a difference in mortality risk between studies assessing depression using a self-report questionnaire versus a clinical psychiatric interview. Only a minority of 30-40% of people with diabetes with an increased level of depressive symptoms suffers from clinically relevant depressive disorder [36,37]. However, both major depression and self-reported depressive symptoms appear to be chronic/recurrent conditions in people with diabetes [38,39], and both are associated with the development of diabetes complications [11]. Furthermore, depressive symptoms have been shown to predict the development of major depression [40].
The results need to be considered in relation to the study limitations. One important limitation in carrying out a metaanalysis is the inevitability to combine data from studies that are not equally designed. This meta-analysis included studies with differing study design and characteristics, and the results demonstrated significant heterogeneity. After conducting subgroup-analyses on follow-up length, age, method of depression assessment, method of diabetes assessment, number of participants and the percentage of females included, heterogeneity remained. However, after excluding the three articles presenting odds ratios (converted to risk ratios) the heterogeneity decreased substantially. This may be due to the fact that odds ratios and hazard ratios are different risk estimates, even after converting odds to risk ratios and combining them with relative risks [41].
In addition, the included studies often reported multiple hazard ratios, each adjusted for different covariates. To reveal the independent effect of depression on mortality we selected the hazard ratio that was most closely adjusted for both demographic and micro-and macrovascular complications. Unfortunately, these estimates sometimes also include adjustment variables through which depression may influence mortality rates, e.g. smoking, physical activity, HbA 1c. By correcting for these potential mediators the hazard ratio can be an underestimation of the real effect of depression on mortality in people with diabetes. With respect to type of diabetes of study participants, five articles did not specify this information. Moreover, only one article reported on individuals with type 1 diabetes, and two articles reported on a combined study population of people with both type 1 and type 2 diabetes. Because type 2 diabetes is the most prevalent form of diabetes, cohort studies with patients with type 1 diabetes are scarce. Finally, we found an indication for publication bias: negative or insignificant result are often not submitted for publication by authors, or rejected by reviewers and editors. This form of bias generally results in an overestimation of the effect.
Despite these limitations several strengths should also be acknowledged. First, our meta-analysis comprises both all-cause and cardiovascular mortality. In addition, the independent effect of depression on mortality was assessed by adjusting for both demographic variables and micro-and macrovascular complications where possible.
It is still unclear whether adequate depression recognition and subsequent depression treatment can help to decrease mortality rates. Bogner et al. [42] have conducted the Prevention of Suicide in Primary Care Elderly: Collaborative Trial (PROSPECT) study that examined a care-management intervention for older primary care patients with depression. The study had a median follow-up of more than 4 year. The authors reported that depressed patients with diabetes in the intervention category were less likely to have died during the 5-year follow-up interval than depressed diabetic patients in usual care after accounting for baseline differences among patients (adjusted hazard ratio 0.49 [95% CI 0.24-0.98]). However, the statistical methods used by Bogner et al. [42] were criticized, as they may have resulted in model overfitting [43,44]. Screening for depression in clinical practice may be a helpful first step, and should be embedded in collaborative care approaches [45]. Effective intervention strategies include cognitive behavioral therapy and treatment with antidepressant medication [46]. Given the close association of depression with suboptimal self-care behaviors [34], interventions that target behavioral mechanisms directly (e.g. coping skills training) may be of value as well.
In conclusion, the results of this meta-analysis suggest that depression is associated with a 1.5-fold increased risk of all-cause mortality in people with diabetes. Although based on only five studies, similar results were found for cardiovascular mortality. In consideration of the study limitations and strengths, we believe that a 1.5-fold increased risk of all-cause (and cardiovascular) mortality in people with diabetes is not an over-or underestimation, but could be an accurate risk estimation.
Future studies are encouraged to explore whether the association between depression and mortality is similar for people with type 1 and type 2 diabetes, and to address the behavioral or physiological pathways that may explain this association.