Risk of Developing Type 2 Diabetes in Individuals With a Psychiatric Disorder: A Nationwide Register-Based Cohort Study

Psychiatric disorders are common: the lifetime prevalence of any psychiatric disorder is estimated to be 29% for the world’s population (1). Individuals with a psychiatric disorder often not only have impaired quality of life (2), but are also at increased risk for multimorbid medical conditions (3,4) and have higher mortality rates equal to a reduced life expectancy of up to 15 years (5,6).

Several systematic reviews based on longitudinal studies have concluded that individuals with a psychiatric disorder, such as depression, insomnia, or an anxiety disorder, are at increased risk of developing type 2 diabetes (7). However, most often, the studies focus on a single psychiatric disorder or investigate a selected sample. New population-based longitudinal studies with a focus on different psychiatric disorders are needed to further increase the understanding of the risk of developing type 2 diabetes in individuals with a psychiatric disorder (7). In a recent nationwide, register-based cohort, Momen et al. (3) studied the associations between a broad range of psychiatric disorders and multiple medical conditions, including diabetes. Across all investigated psychiatric disorders, Momen et al. (3) detected increased incidence of diabetes in individuals with a psychiatric disorder compared with individuals without the psychiatric disorder of interest. However, the reported analyses did not differ between type 1 and type 2 diabetes, and the analyses were not adjusted for socioeconomic status or addressed potential age-specific differences.

In addition to focusing on incident type 2 diabetes in a broad range of psychiatric disorders, attention for sex- and age-specific differences is warranted. Sex differences in the prevalence of psychiatric disorders have been a stable finding in research (8). Internalizing psychiatric disorders, such as depression and anxiety disorders, are more common among women, whereas externalizing disorders, such as substance use disorders and behavioral disorders, are more common among men (9). These differences emphasize the need to take sex into account as a potential effect modifier when exploring the incidence of type 2 diabetes in individuals with a psychiatric disorder. Furthermore, age at onset of different psychiatric disorders as well as age at onset of type 2 diabetes differs (10,11). In a review from 2015, it is highlighted that individuals with a psychiatric disorder have earlier onset of type 2 compared with individuals without a psychiatric disorder (12), but the finding is based solely on studies focusing on individuals with schizophrenia. Therefore, more knowledge on the age-specific incidence of type 2 diabetes in people with different psychiatric disorders is required to ensure optimal timed preventative approaches as well as early detection. To the best of our knowledge, more detailed information from large, population-based studies focusing on a broad range of psychiatric disorders and incident type 2 diabetes is currently lacking.

Therefore, the aim of this study is to fill this gap in the scientific literature by investigating the risk of incident type 2 diabetes with respect to different psychiatric disorders in the entire adult Danish population. Specifically, we aim to determine the overall, sex-specific incidence rate ratios (IRR) and sex- and age-specific incidence rates (IR) for individuals with a diagnosed psychiatric disorder compared with individuals without a diagnosed psychiatric disorder of interest.

Denmark has population-wide registers covering many aspects of life, including socioeconomic conditions, migration, and health care–related aspects (13). All Danish residents are assigned a personal registration number (CPR-number) at birth or immigration (14); this number is used to link data from different Danish national registers at an individual level. In this nationwide, register-based dynamic cohort study, we included all individuals (≥18 years old) living in Denmark during the study period (1 January 1995 to 31 December 2018). The study used a dynamic cohort design, which means that new adults were continuously included in the study cohort. Thus, for all individuals, study entry was defined at 1 January 1995 or at the date when they turned 18 years old, whichever came later.

The study was approved by the Danish Data Protection Agency. According to Danish law, informed consent is not required for register-based studies. Authorization to data access is given by the Danish National Health Data Authority and requires permission from the Danish Data Protection Agency.

Information on type 2 diabetes was obtained from a nationwide diabetes register (11) constructed using data from all available Danish health care registers containing diabetes-related information (the National Patient Register [15], the National Health Services Register [16], the National Prescription Register [17], the Danish Adult Diabetes Database [18], and the Danish Registry of Diabetic Retinopathy [19]). The date of incident type 2 diabetes was defined as the first record of type 2 diabetes in the nationwide diabetes register during the study period. As the focus of the study was on incident type 2 diabetes, we excluded all individuals with type 2 diabetes registered before entry to study or within the first year of the study period (n = 65,638).

Psychiatric disorders were defined based on information from the Danish Psychiatric Central Research Register (PCRR) during the study period (20). The PCRR contains information on all admissions to psychiatric hospitals since 1969 and all outpatients and emergency contacts since 1 January 1995 (20). Psychiatric disorders were defined based on subcategories from ICD-10, the Diagnostic Research Criteria subchapter F (see Table 1 for a full overview of definitions for each psychiatric disorder) (21). These definitions have previously been used in several nationwide register-based studies (3,5,10,22). In total, 10 categories of psychiatric disorders were defined. Additionally, we included a composite category comprising any psychiatric disorder. The date of diagnosis for each psychiatric disorder of interest was defined as the earliest date of diagnosis for the specific disorder in the PCRR during the study period.

Sociodemographic characteristics of the study population included sex, age, migration status, and highest level of education. Migration status and level of education were included as covariables due to their association with both psychiatric disorders and type 2 diabetes (23–26). All covariables were measured at entry to study: sex, age, and migration status were extracted from The Danish Civil Registration System (14). Age and calendar time were counted in years. Migration status was defined based on country of origin, and all individuals born outside Denmark were classified as immigrants. Data on highest level of education were extracted from the Educational Attainment Register (27,28) and divided into three categories (lower secondary or below, upper secondary, and tertiary or above) according to the International Standard Classification of Education (29). Individuals with missing data regarding education level were excluded from the analyses (n = 504,914).

We followed individuals from study entry until: 1) date of a diagnosis with type 2 diabetes (outcome), 2) date of emigration, 3) date of death, or 4) end of study period (31 December 2018), whichever came first. For each psychiatric disorder, we estimated incidence of type 2 diabetes during the follow-up period for individuals with and without the psychiatric disorder, respectively. Psychiatric disorders were considered as time-varying factors (i.e., during the study period, individuals were considered as unexposed to the specific psychiatric disorder until the first diagnosis of the specific disorder and then considered as exposed to the psychiatric disorder afterward). If an individual had several psychiatric disorders at the same time, the individual contributed with risk time for each of the psychiatric disorders simultaneously.

We calculated mean and SD for continuous variables and frequencies and percentages for categorial and binary variables. χ² tests for categorical and binary covariables and independent t tests for continuous covariables were performed to assess differences in the distribution between individuals with and without a psychiatric disorder. Furthermore, we calculated median (interquartile range) of age of type 2 diabetes for individuals with and without a psychiatric disorder.

We applied different multistate models (one for each category of psychiatric disorders) stratified by sex to provide overview of the individuals and their transitions during the study period (see examples in Supplementary Appendix 1). IRR for the occurrence of type 2 diabetes were estimated with multivariable adjusted Poisson regression models using log-person-time as the offset variable. Risk time (follow-up time) was split into 1-year band intervals by the timescale age. The Poisson regression models were stratified by sex. We conducted two versions of Poisson regression models: a model adjusted for age and calendar year (model A) and a model adjusted for age, calendar year, migration status, and level of education (model B). The continuous covariables (age and calendar year) were considered as time-varying factors, the effects of which were modeled with natural splines with five knots. Knots were placed so that the type 2 diabetes events were evenly distributed between the knots.

We modeled the age-specific IR of type 2 diabetes for individuals with and without each psychiatric disorder stratified by sex. The age-specific IRs were adjusted for calendar year, migration status, and highest level of education. In all statistical analyses, the level of significance was set at 0.05. All statistical analyses were performed using R version 4.0.2 (30). The Epi package, version 2.44, was used to handle the multistate models (31–33).

We followed 5,005,612 individuals for 89,732,494 person-years. As shown in Table 2, 50% of the individuals were men. At entry to the study, the mean age was 36 years (SD 17.18), 5% were immigrants, and 57% had a lower secondary education or below. During the study period, 334,739 individuals (7%) developed type 2 diabetes.

During the study period, 512,466 individuals (10%) were in- or outpatients at a public psychiatric hospital. In Supplementary Appendix 2, a descriptive overview of individuals with and without the different psychiatric disorders is presented. Of all individuals with a psychiatric disorder, 43,953 individuals (9%) developed type 2 diabetes during the study period. Overall, individuals with psychiatric disorders were most often women compared with individuals without psychiatric disorders (56% vs. 49%). However, individuals with substance use disorders, schizophrenia, intellectual disabilities, developmental disorders, and behavioral disorders were more often men compared with individuals without the specific psychiatric disorder. Individuals with a psychiatric disorder, except individuals with organic disorders, were younger at study entry compared with individuals without the specific psychiatric disorder. Individuals with any psychiatric disorder, schizophrenia, mood disorders, and neurotic disorders were more often immigrants compared with individuals without the specific psychiatric disorder (for any psychiatric disorder: 6% vs. 5%). Regardless of the specific psychiatric disorder, individuals with a psychiatric disorder more often had a low level of education at study entry compared with individuals without a psychiatric disorder (for any psychiatric disorder: 64% vs. 56%).

Table 3 shows the IRR for type 2 diabetes when individuals with a psychiatric disorder were compared with individuals without the specific psychiatric disorder. Across all psychiatric disorders, IR were higher in individuals with a psychiatric disorder for both men and women, when adjusted for age, calendar year, migration status, and highest level of education. Men and women with any psychiatric disorder had, respectively, 47% and 65% increased IR when compared with individuals without any psychiatric disorder (IRR: men, 1.47 [95% CI 1.45–1.50]; women, 1.65 [95% CI 1.62–1.68]). The highest IRRs for men were found in individuals with schizophrenia, eating disorders, or behavioral disorders, whereas the highest IRR for women were found in individuals with schizophrenia, personality disorders, intellectual disorders, or behavioral disorders.

Figure 1 shows the age-specific IR of type 2 diabetes for individuals with versus without a psychiatric disorder for men and women adjusted for calendar time, migration status, and level of education. Overall, we found that the IR of type 2 diabetes increased exponentially with higher age for both men and women with and without a psychiatric disorder. For the oldest ages, the IR stabilized and, in some occasions, showed a tendency to decrease. Furthermore, when observing the IR of type 2 diabetes in Fig. 1 for individuals with a psychiatric disorder compared with individuals without a psychiatric disorder, we observed that the difference was particularly high in the younger age-groups (<50 years), after which the difference attenuated. This pattern was found for the majority of the psychiatric disorders. In four psychiatric disorders (eating disorders, intellectual disabilities, developmental disorders, and behavioral disorders), we found wide CIs when trying to estimate the IR of type 2 diabetes in the older population (Fig. 1), leading to inconclusive results for this age group in these psychiatric disorders. When we calculated median (interquartile range) of age of type 2 diabetes onset for individuals with and without a psychiatric disorder, we found similar results supporting the findings that individuals with a psychiatric disorder develop type 2 diabetes earlier compared with individuals without a psychiatric disorder (see Supplementary Appendix 3 for an overview).

In this nationwide longitudinal study, we have investigated the sex- and age-specific incidence of type 2 diabetes in individuals with a psychiatric disorder. Across all categories of psychiatric disorders that were studied, individuals with a psychiatric disorder had increased risk of incident type 2 diabetes compared with individuals without a psychiatric disorder. These findings were comparable for men and women.

In line with previous studies, we found that individuals with internalizing psychiatric disorders were more often women, whereas externalizing psychiatric disorders were more common among men. Across all psychiatric disorders, except eating disorders, we found higher IRR of type 2 diabetes in women than in men. The number of men diagnosed with an eating disorder was very low, and only cautious interpretation of these findings should be made.

Our findings were in line with previous findings from Momen et al. (3), although we in general detected higher estimates of type 2 diabetes incidence in individuals with a psychiatric disorder compared with individuals without a psychiatric disorder. These differences could probably be explained by a more nuanced measurement of diabetes, while we solely focused on type 2 diabetes, and a more comprehensive adjustment for socioeconomic status in our study. As migration status and level of education are associated with psychiatric disorders as well as type 2 diabetes (23–26), we expect that our comprehensive adjustment has also slightly affected the incidence estimates.

Overall, the risk to develop type 2 diabetes was particularly elevated in the younger people with a psychiatric disorder (<50 years old). This finding was in line with previous studies reporting earlier onset of type 2 diabetes in individuals with psychiatric disorders compared with individuals without psychiatric disorders (12). An unhealthy lifestyle, such as unhealthy dietary pattern (increased consumption of processed meat and sugar-sweetened beverages, decreased intake of whole grains, coffee, and heme iron), physical inactivity, and smoking (34,35), as well as the use of psychotropic medication (36,37) are well-known risk factors for the development of type 2 diabetes It might be the case that an unhealthy lifestyle is more common in younger individuals with a psychiatric disorder or that an unhealthy lifestyle in young adulthood increases the risk of negative consequences in the future compared with individuals in other age groups. Furthermore, as the onset of a psychiatric disorder and therefore the use of psychotropic medication often are initiated in early adulthood (38), and many psychotropic medications cause substantial weight gain (39), it is plausible that this may increase the risk of type 2 diabetes in the younger populations. Finally, competing risk of death might have influenced the findings, as older individuals with a psychiatric disorder were selected by having survived. Therefore, only the healthiest individuals from the older population were left, and their risk of developing type 2 diabetes was not higher than older individuals without a psychiatric disorder. This assumption was supported by the fact that IR of type 2 diabetes for individuals with and without a psychiatric disorder did not differ as much in the older populations compared with the younger populations. However, more research is needed to understand potential mediating mechanisms and to illuminate the age-specific differences in the onset of type 2 diabetes.

In individuals with an eating disorder, intellectual disability, developmental disorder, or a behavioral disorder, we found wide CIs when estimating the IR of type 2 diabetes in the older population. These wide CIs may partially be explained by competing risk of death potentially introducing bias. However, the wide CIs could probably also be explained by the fact that these psychiatric disorders mainly have their onset in childhood and adolescence, and therefore, only a limited number of older individuals received these diagnoses in a psychiatric hospital setting. Therefore, we should be careful when interpreting findings from the older population for these psychiatric disorders. In case of a more specific interest into these psychiatric disorders, a focus only on the younger population and information on psychiatric diagnoses from the childhood could probably give more accurate information of the risk of type 2 diabetes.

The increased risk of type 2 diabetes detected across all investigated psychiatric disorders might indicate that common denominators play a role, such as genetic factors or chronic stress (12,40). Future research might benefit by a focus on factors such as level of distress caused by the psychiatric disorder rather than a focus on a specific type of psychiatric disorder. In a cross-sectional study from 2014, de Jonge et al. (41) found associations between 16 types of psychiatric disorders and diabetes. When de Jonge et al. (41) adjusted for psychiatric multimorbidity, only depression and impulse control disorder remained associated with diabetes, which highlights that future studies should also focus on psychiatric multimorbidities.

The findings from this cohort study highlight the importance to investigate whether we can find better ways to prevent type 2 diabetes in individuals with a psychiatric disorder. While most research has focused on the association between depression and incident type 2 diabetes (7), this nationwide study demonstrates that the increased risk of type 2 diabetes is not limited to individuals with depression. In contrary, we observed an increased risk of incident type 2 diabetes in individuals with any psychiatric disorder. Thus, a broader perspective on individuals with any psychiatric disorder as being at increased risk might be meaningful. We acknowledge the heterogeneity within each of the categories of psychiatric disorders and expect that different mechanisms may explain the increased risk of incident type 2 diabetes in individuals with a psychiatric disorder. Despite this, our findings highlight the importance to not simply focus on a specific psychiatric disorder when assessing type 2 diabetes in vulnerable populations.

Optimal treatment of type 2 diabetes requires early detection to minimize diabetes complications and increase life expectancy (42). The findings from the current study clearly emphasize the need to prioritize prevention initiatives and early detection of type 2 diabetes among individuals with a psychiatric disorder to improve their prognoses. Previous intervention studies have shown that pharmacological and behavioral interventions have a positive impact on individuals with severe psychiatric disorders (43). However, little is known regarding prevention of type 2 diabetes in individuals with a psychiatric disorder (44), which is partially due to the fact that individuals with psychiatric disorders are often excluded from traditional intervention studies, and our knowledge regarding the effect of intervention and preventions studies in this population is limited (45). Our findings also highlight the importance of focusing on prevention of type 2 diabetes in the population of individuals with a psychiatric disorder.

This register-based dynamic cohort study is based on a nationwide, largely unbiased data set with limited reporting bias and includes long-term follow-up. A major strength of our study design is that it allows us to analyze data from the entire adult Danish population from 1995 to 2018, giving us the opportunity to investigate the association between a broad range of diagnosed psychiatric disorders and incident type 2 diabetes. When using data from the Danish registries, the risk of selection bias is very low. However, there are important limitations to this study that should be acknowledged. Initially, we recognize the that our missing data might have impacted the findings. We excluded >500,000 individuals from the analyses as they lacked information regarding highest level of education. Despite a high validity and coverage of the Educational Attainment Register (28), it is likely that the exclusion has introduced selection bias, as individuals without information on education are more often migrants or from the older part of the population (28). This selection bias might have led to an underestimation of the results. We only investigated psychiatric disorders that were diagnosed and treated in hospitals, leading to records of only the most severe cases. So, the impact of undiagnosed psychiatric disorders and diagnoses from the general practitioner and general psychologist could not be investigated, which may have led to an underestimation of the associations if these individuals had type 2 diabetes to the same extent as individuals diagnosed with a psychiatric disorder from the hospital. Our reason for not including children is the duration of our follow-up period (maximum 23 years); given the typical age of onset of type 2 diabetes is 45–50 years, inclusion of individuals started at 18 years of age. Additionally, this study is only based on data from 1 January 1995 and onward, and it was therefore not possible for us to include diagnoses of psychiatric disorders prior to this date. Combined with the fact that we only included adults, this means that we have no information on psychiatric disorders from childhood, which probably will result in underdetection of individuals with onset of psychiatric disorders in childhood and thus a potential underdetection of exposed individuals. This limitation was particularly relevant when investigating the psychiatric disorders with onset in childhood (e.g., intellectual disabilities and development disorders). To provide a comprehensive overview of all psychiatric disorders, we analyzed categories of psychiatric disorders rather than specific psychiatric disorders. Furthermore, we have not investigated the full spectrum of coexisting psychiatric disorders nor did we focus on the potential impact of psychiatric multimorbidity, which falls beyond the scope of our study and should be addressed in future studies (22). We expect that key drivers of the increased risk of incident type 2 diabetes may be depression and impulse control disorders, in line with the results of a previous cross-sectional study (41). In future research, it will be meaningful to focus on the impact of somatic and psychiatric multimorbidity. As this study is based on register-based data, it was also not possible to identify individuals with undiagnosed type 2 diabetes. Previous research has suggested a higher proportion of undiagnosed type 2 diabetes in individuals with psychiatric disorders compared with the general population (12), and it is possible that the increased risk of incident type 2 diabetes is even higher than reported in this study.

Across all investigated psychiatric disorders, we found that the IR of type 2 diabetes was higher in individuals with a psychiatric disorder compared with individuals without a psychiatric disorder. When comparing the IR of type 2 diabetes for individuals with and without a psychiatric disorder, we found the biggest difference in the younger population, highlighting the need for both prevention and early detection of type 2 diabetes in individuals with a psychiatric disorder.

Funding. This study was funded by an unrestricted PhD grant from the Faculty of Health Sciences, University of Southern Denmark.

Duality of Interest. G.S.A. own shares in Novo Nordisk A/S. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. N.L. contributed to the design of the study and was responsible for data management, data analyses, interpretation of the results, as well as writing the first draft of the manuscript. S.H.S. contributed to the data management and the data analyses, oversaw the interpretation of the results, and reviewed and edited the manuscript. L.J.D. contributed to data management and data analyses, interpretation of the results, and reviewed and edited the manuscript. K.H.R., O.P.-R., J.E.H., M.L., G.S.A., and F.P. have contributed to the design of the study, interpretation of the results, and reviewed and edited the manuscript. F.P. is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.