Discussion
In an 8-year nationwide outpatient cohort study examining 294 642 individuals with type 2 diabetes and 1 271 537 individuals with depression, we observed bidirectional associations between type 2 diabetes and depression. Incident type 2 diabetes in 2015 was associated with a 22.7% increased risk of subsequent depression, and a 2015 incident depression diagnosis was associated with a 15.3% increased risk of subsequent type 2 diabetes, both after adjusting for chronic medical conditions, region of residence, and area-level deprivation.
Between 2012 and 2014, a substantial number of individuals were diagnosed with a form of diabetes, while approximately double that number had existing depression, highlighting the considerable public health impact of these diseases. After excluding these prevalent cases, our findings regarding the 2015 incidence of depression across age and sex align with, yet are not identical to, previous research.13 14 Notably, we observed a linear increase until the 30–34 years age group without a distinct peak during adolescence, which is likely attributable to our youngest age group being 18–19 years. A shift toward a midlife peak incidence of depression in the German adult population might be influenced by societal factors and reporting biases among younger individuals. This notion is supported by one of our prior studies, which highlighted an elevated risk of comorbid depression and obesity for middle-aged men and age-specific differences in reporting of depression diagnoses in a representative sample for the German adult population.15 Additionally, while Pedersen et al identified a second peak in the elderly,13 in our analysis, this peak was discernible only among men, suggesting potential differences in causes of depression in later life.16 The gender disparity in incidence across several studies necessitates more nuanced, gender-sensitive approaches in depression epidemiology and intervention strategies.
A previous study using Danish registers has shown that individuals with newly diagnosed type 2 diabetes are at increased risk of initiating pharmacological treatment with an antidepressant with an HR of 1.51, while the risk of having a psychiatric hospital contact was not increased to the same extent (HR=1.14).4 We add to this evidence by showing a 22.7% increased risk of being diagnosed with depression after type 2 diabetes diagnosis in outpatient care, using longitudinal data from approximately 86% of the German population.10 Both type 2 diabetes and depression, being common in outpatient settings, do not always necessitate medication or inpatient treatment, emphasizing the significance of our findings in an outpatient context. Biological mechanisms, including the disruption of insulin signaling17 and stress pathway activation,18 may contribute to higher depression risk following type 2 diabetes. Additionally, the psychological ramifications of chronic disease diagnoses, such as type 2 diabetes, can trigger social and emotional distress, potentially increasing the risk of depression. While factors such as medical comorbidities19 and environmental influences, such as low socioeconomic status,20 might confound this relationship, the higher risk of depression remained significant after accounting for these factors.
In contrast to previous research conducted using Danish hospital diagnosis and prescription data,9 which indicated a stronger association between depression and subsequent type 2 diabetes compared with the association between type 2 diabetes and subsequent depression, our analyses based on German outpatient data suggest a weaker association between type 2 diabetes following depression. Notably, when an alternative definition for type 2 diabetes relying solely on hospital diagnosis was employed, Wimberley et al reported an even stronger association between depression and subsequent type 2 diabetes.9 Since hospital diagnoses tend to capture more severe cases, it raises the possibility that the link between depression and type 2 diabetes may be influenced by disease severity. Indeed, our subgroup analyses also reveal a stronger association between the highest level of depression severity and subsequent type 2 diabetes. These discrepancies, driven by variations in study settings, warrant careful consideration, especially in the context of future targeted interventions aimed at mitigating the burden of this comorbidity. Our findings underscore the importance of prioritizing interventions to prevent depression following the diagnosis of type 2 diabetes, given the multiple distress factors associated with this scenario, including challenges in adhering to medical recommendations, social stigma, and reduced quality of life.21 22 One potential approach involves reducing barriers in both primary and specialty care settings to facilitate screening for depression and diabetes-related distress, which could positively influence the risk of developing depression and also improve diabetes care.23 The proposed mechanisms through which depression precipitates type 2 diabetes involve both direct biological effects and indirect influences via health behaviors. Another consideration is the potential diabetogenic effects of certain antidepressant medications.24 Moreover, individuals newly diagnosed with depression may already exhibit preclinical alterations, such as altered insulin resistance, potential precursors to type 2 diabetes, a phenomenon similarly observed in bipolar disorder.25 Recent Mendelian randomization studies also pointed to a causal relationship between depression and type 2 diabetes, with approximately 36.5% of this effect being channeled through body mass index.26 This suggests that obesity could be a shared underlying factor linking depression and type 2 diabetes.
In our analysis of both directions of association, age stratification highlighted differential risks across age groups. Younger individuals were more susceptible to developing the comorbidity than their middle-aged or elderly counterparts. However, the ≥90 years age group stood out as an exception. This deviation may be influenced by survivorship bias, indicating those reaching such advanced ages might possess unique health or genetic profiles not typical of the wider population, altering perceived risk dynamics. Moreover, there is an increased risk of misclassification among the elderly due to potential non-capture of depressive episodes predating the pre-observational period. Intriguingly, our analyses consistently highlight a heightened risk in younger individuals, which may be potentially attributed to unique challenges they face, particularly when considering health disparities in relation to their same-aged peers. Early onset of type 2 diabetes or depression in younger ages might also flag a group with an unusually high medical burden, especially because younger individuals tend to seek medical consultations less frequently, resulting in fewer diagnoses.15 27 Lastly, the age trends we observed may be shaped by our methodological choices. Specifically, we excluded those with existing depression or diabetes between 2012 and 2014. Consequently, elderly individuals newly diagnosed with depression or type 2 diabetes at a more advanced age may not carry a heightened risk of the other condition compared with their controls, representing a selectively ‘healthier’ cohort, having reached advanced ages without either condition, suggesting an inherent resilience.
The immediate quarter following incident type 2 diabetes or depression diagnosis emerged as the most vulnerable period, possibly due to detection bias from temporarily increased medical attention. Distinctively, while depression risks decreased over time following type 2 diabetes, the risk of type 2 diabetes following depression remained consistently elevated, potentially reflecting the distinct nature of diabetes, which might take longer to manifest, whereas depression could be more immediately triggered.
Through sex-stratified analyses, we observed minimal divergence between sexes. Previous findings of higher comorbidity rates in women parallel our observation of elevated type 2 diabetes risk following depression in women.28 29 Notably, hormonal fluctuations during the perimenopausal phase have been linked to elevated depression rates in longitudinal studies involving premenopausal women.30 Additionally, the perimenopausal period is associated with weight gain and the onset of metabolic diseases.31 These biological factors could potentially contribute to the observed disparities. Furthermore, variations in mental healthcare utilization by gender, with women being more likely to access healthcare services, may also influence comorbidity risk differently between men and women.32 Meanwhile, psychosocial factors and medical factors, such as obesity and alcohol use,33 might differentially influence the risk of depression for men and women following type 2 diabetes diagnosis, thereby meriting further exploration.
One of the limitations of our study is the inherent assumption that individuals without a specified outcome or exposure diagnosis survive until the end of follow-up. Differential mortality rates between cases and controls, attributed to their distinct disease exposures, could pose competing risks. This becomes increasingly relevant given the heightened mortality often associated with type 2 diabetes, predominantly due to cardiovascular comorbidities.34 While age-matching addresses this to a degree, the lack of mortality data likely biases our results. However, this might lead to a more cautious interpretation of the association, rather than overestimating it. In other words, our study may underestimate the impact of type 2 diabetes on the risk of developing depression due to the unaccounted deaths among individuals with diabetes who might have developed depression had they survived. Additionally, factors such as individuals leaving SHI, migrating to other countries, or not using medical healthcare services would result in a loss to follow-up. Unfortunately, we cannot account for these reasons as the claims data lack such information. Consequently, our analyses may carry a bias and cannot be generalized to individuals who did not seek statutory healthcare during the observational period. Another limitation lies in the reliance on accurate diagnosis and precise data input by healthcare professionals and/or administrators.
A broader perspective, as depicted by Lindekilde et al,35 suggests an array of psychiatric disorders, beyond depression, as potential risk factors for incident type 2 diabetes. Furthermore, it has been reported that antidepressant medication may mediate the relationship between depression and type 2 diabetes.24 Our study, however, did not incorporate data on the relationship between type 2 diabetes and other psychiatric disorders or medication, limiting the scope of our findings. We did not have access to diabetes biomarkers like HbA1c. Nonetheless, we employed two distinct diagnostic lists based on ICD-10-GM codes to delineate cases and controls. This approach served a dual purpose—it not only ensured the exclusion of individuals with prior diabetes and depression diagnoses within our cohorts, allowing us to confidently identify all detected cases as incident cases to the highest degree possible, but also facilitated the differentiation between type 1 and type 2 diabetes cases. In the context of diabetes, the broader definition encompassed ICD-10-GM codes for all types of diabetes, while the narrower definition included only type 2 DM, malnutrition-related DM, other specified DM, and unspecified DM. Given that adult type 1 diabetes cases constituted a mere 0.3% of the total 9.7% diabetes prevalence in Germany in 2010, as indicated by data from both inpatient and outpatient sectors,36 the likelihood of misclassifying individuals with type 1 diabetes as type 2 diabetes cases using the narrow definition remains low. Additionally, although we made efforts to control for potential confounding factors, the possibility of unmeasured or residual confounding (e.g., individual lifestyle factors) cannot be entirely ruled out in observational studies of this nature.
While we acknowledge the potential for bias due to not adjusting for the matching factors age and sex in our Cox proportional hazards model, we mitigated this limitation through stratified analyses by age group and sex. These stratifications allowed us to explore risk within subgroups stratified by these key demographic variables, providing valuable insights into the associations under study. The third matching factor (residential area) was taken into account by including the district of residence as random effect.
Our study design aimed to estimate the risk of developing type 2 diabetes following the initial diagnosis of depression, irrespective of the course of the depressive illness, including whether it was episodic or chronic. Consequently, individuals without depression at baseline were censored from the control group of the depression cohort if they subsequently developed depression during the follow-up period, but individuals in the depression group were not similarly excluded if their depression resolved during the follow-up. This approach was chosen to focus on the association between incident depression and type 2 diabetes, but we acknowledge that it may be perceived as inconsistent with the treatment of individuals in the depression group. Moreover, our results might not be representative of those experiencing recurrent depressive episodes, particularly in the context of age-stratified analyses. Additionally, there is potential for misclassification, where individuals diagnosed with depression before our pre-observational period might inadvertently be labeled as incident cases. Given that our study used only outpatient claims data, the results may not generalize to inpatient treatments typically involving more severe manifestations.
While our use of a matched cohort study design and adjustments for covariates enhanced the internal validity of our findings, it is important to acknowledge that this approach may have reduced statistical power due to a smaller sample size. Nevertheless, given the enormity of the data set including almost the entire German adult population, practical feasibility and computational resources played a significant role in our methodology selection. Matching more controls or including the entire data set would have posed substantial computational challenges without necessarily substantially enhancing statistical power. Our primary limitation lies in the availability of additional data rather than the sample size itself. Lastly, we did not consider time-dependent variables and this may have influenced the trajectories of depression and type 2 diabetes over the observed span.
