Principal findings
In a large prospective study in the UK, we confirmed the higher mortality risk associated with each of depression and diabetes, and also identified synergistic effects of depression and diabetes on different causes of mortality beyond that expected from their individual effects. This pattern remained even after adjusting for a wide range of potential confounding factors.
Strengths and limitations of this study
Our final model adjusted for a number of covariates that could be potential mediators of the observed associations. An important consideration when distinguishing between confounders and mediators is whether factors change over time and do not or do lie on the causal pathway, respectively [30]. While genetically determined factors such as sex and ethnicity clearly preceded the onset of our exposure and outcome and may confound their association, lifestyle factors and physical measures such as smoking status and BMI could be potential mediators of the observed associations. By adjusting for these covariates we can identify the extent to which any association is independent of these factors. The pattern of results was similar in models before and after the inclusion of potential mediators (ESM Figs 1, 2). However, our final model may have underestimated the strength of the association between depression and/or diabetes and risk of mortality and the magnitude of the interaction effect between depression and diabetes on risk of circulatory diseases. Further research is required to perform a formal mediation analysis, ideally using a dataset with time-varying information on covariates, if criteria for establishing causality are met [31].
Our study has a number of strengths. It is one of just a few cohort studies investigating the relative importance of depression and diabetes and their synergistic effects on risk of all-cause and circulatory mortality in the general population and within a universal healthcare setting. Furthermore, to our knowledge, this is the first study to describe the effects of comorbid depression and diabetes on risk of mortality from cancer and causes of death other than circulatory disease and cancer. We used a large prospective cohort that contained detailed information on a range of potential confounding factors. The large sample size and subsequent large number of deaths meant we had sufficient power to investigate the individual and synergistic effects of our exposure groups, to study cause-specific mortality risks and to stratify our analysis by sex. A further advantage of our study is that, in contrast to many other prospective cohort studies, there was limited attrition, because we relied on administrative health records to ascertain outcomes.
Our study has some limitations. The UK Biobank had a low response rate (5.5%), which resulted in a relatively healthy cohort from a higher socioeconomic background than that of the general population [32]. However, it has been argued that this is unlikely to influence estimates of associations between diseases, given that there are large numbers of participants with different levels of risk factors in the sample [32]. Nonetheless, selection bias might have influenced some of the results of this analysis. As previously described, selection into a cohort can introduce collider bias that can work in any direction [33]. However, without further information on the population from which the cohort is drawn or from unselected cohorts, it is not possible to determine the presence of bias or the direction of bias in the strength of the association.
In addition, there is potential for misclassification because participants may have under-reported depression, diabetes and comorbid depression and diabetes at baseline. Although our measurement at baseline used hospital records and self-report, it is possible that we misclassified some participants’ exposure status and have underestimated the mortality risks associated with depression and diabetes. We may have further inaccurately estimated the mortality risks associated with the presence of diabetes, as we did not identify individuals with undiagnosed diabetes. Furthermore, individuals may have been misclassified as depressed if they took antidepressants for treatment of chronic pain, such as neuropathic pain. This could lead to overestimation of mortality risks associated with depression if chronic pain is associated with higher mortality risks than depression (and underestimation if chronic pain is associated with lower mortality risks). We were not able to update exposure status during follow-up. In addition, we may have missed a small number of deaths occurring outside the UK, but this is likely to have occurred non-differentially across the four exposure groups. This may, however, have further biased our findings towards the null. Although key confounding factors were adjusted for in this analysis, residual confounding might explain some of the observed effect, for example if the measurement error of lifestyle factors and comorbidities was systematically different among the four exposure groups.
Strengths and weaknesses in relation to other studies
Our findings are in keeping with previous studies reporting a high risk of all-cause and circulatory mortality risk among people with comorbid depression and diabetes that exceeds the risk due to having either diabetes or depression alone [13, 16,17,18,19,20,21]. While the strengths of the associations between comorbid depression and diabetes and risk of all-cause and circulatory mortality were similar in some previous studies [17,18,19, 21], others observed much higher HRs of 3.64 [20], 3.71 [13] and 4.56 [16] for risk of all-cause mortality and 3.27 for circulatory mortality risk [17]. Potential explanations for the observed differences are the use of very selected reference groups, such as people with a score of 0 on the Centre for Epidemiologic Studies of Depression Scale [13, 16], and differences in the study populations [17, 20].
Our study uniquely extends these findings to risk of cancer mortality and causes of death other than circulatory disease and cancer. In patients with diabetes, a previous study reported an increased risk of non-CVD, non-cancer mortality in people with comorbid depression and diabetes, whereas there was no association with risk of CVD and cancer mortality [22]. However, with small number of deaths in some groups, this study may have been underpowered to detect statistically significant differences. As this study was based on patients with diabetes, the joint effect of depression and diabetes could not be examined.
Possible explanations for our findings
The underlying mechanisms for the synergistic effect of depression and diabetes on mortality risk remain to be established. As we found synergistic effects of depression and diabetes for risk of different causes of mortality, it is unlikely that the underlying mechanism is organ or disease specific [34]. A more general explanation for the excess mortality risk among those affected by both depression and diabetes is that depression might make adoption and maintenance of a healthy lifestyle, including smoking cessation and self-management, more difficult. For example, depression has been shown to be a risk factor for medical non-compliance among individuals with comorbidities [5, 6], which might lead to adverse effects such as poor glycaemic control among individuals with comorbid depression and diabetes [4]. Second, individuals with mental–physical comorbidity may receive suboptimal quality of care, which in turn may increase their risk of adverse events [35, 36]. As such, the negative consequences of depression and diabetes may be aggravated among those with comorbid depression and diabetes because of the lack of successful treatment or self-management strategies for both conditions. However, more research is needed to further explore this hypothesis.
Implications of this study and future research
Our findings highlight the scope for improved care and treatment of people with depression, particularly those with diabetes. Despite the availability of guidelines on encouraging psychological well-being in people with diabetes [37], depression continues to be overlooked in clinical practice [38]. Screening for depression in clinical practice, particularly among those with diabetes, may be a helpful first step to identify patients at high risk of adverse effects. However, a requirement of screening programmes is the provision of cost-effective interventions to individuals identified as being at high risk of adverse events. This is particularly challenging in this context due to the lack of cost-effective interventions to reduce adverse outcomes in this patient group [39,40,41]. An RCT found that allocating a trained depression care manager and offering an antidepressant and interpersonal psychotherapy to patients with comorbid depression and diabetes may reduce the 5 year mortality rate [42]. However, the statistical methods used by Bogner et al [42] were criticised as they may have resulted in model overfitting [43], and few health systems are likely to have the resources to provide such interventions to the large number of people who might be eligible. Thus, further RCTs are needed to identify cost-effective interventions that reduce the risk of mortality and improve quality of life in patients with one or both of depression and diabetes.
A particular focus of future studies should be the potential synergistic effect of depression and diabetes not only on risk of circulatory mortality but also on cancer mortality and mortality from other causes, as this is the first study to report this. It would be helpful to establish whether the synergistic effect of depression and diabetes on mortality risk is observed in other settings and for participants with type 1 and type 2 diabetes. Furthermore, future studies should attempt to identify mechanisms that may be responsible for the synergistic effect of depression and diabetes on risk of mortality in order to inform the development and testing of interventions. Finally, the temporality of depression and diabetes deserves further attention, with one recent study suggesting smaller joint effects of depression and diabetes when both disorders are ascertained at the same point in time than when depressive symptoms develop after diagnosis of diabetes [18].
Conclusions
In summary, we found that individuals with depression and diabetes were at high risk of all-cause mortality and mortality from cancer, circulatory disease and causes other than circulatory disease or cancer. In the fully adjusted model, the combined association between depression and diabetes was additive for risk of circulatory mortality and synergistic (i.e. supra-additive) for risk of cancer and mortality from causes other than circulatory disease and cancer (described in Table 2, ESM Fig. 1). Although some progress has been made in the past, our findings highlight the need for further research and the potential for the improved treatment of depression, particularly in people with diabetes.