figure a

Introduction

Psychiatric disorders are common [1], can substantially impair quality of life [2], and are associated with elevated mortality rates [3, 4]. Excess mortality in individuals with a psychiatric disorder can be attributed not only to more frequent suicides and accidents but also to the high prevalence of comorbid somatic disorders in this population [5,6,7,8]. The prevalence of type 2 diabetes mellitus, for example, has been reported to be higher in people with bipolar disorder [9], schizophrenia [10] or major depression [11], compared with the general population. The global prevalence of type 2 diabetes has increased from 1990 onwards, current estimates ranging between 6.3% and 9.3%, and is expected to increase further within the next 20 years [12, 13]. During recent decades, systematic reviews have investigated the prevalence of type 2 diabetes in people with different psychiatric disorders. However, no systematic overview and critical appraisal of this literature is currently available.

To address this limitation, this umbrella review sought to summarise and critically assess the existing evidence on the prevalence of type 2 diabetes in people with a psychiatric disorder. We focused on systematic reviews of observational studies to generate a narrative synthesis of the prevalence estimates. Our review included a risk of bias assessment and meta-analyses of prevalence estimates from the primary studies included in the systematic reviews.

Methods

Protocol, registration and study design

We registered the protocol for this umbrella review at PROSPERO (registration no. CRD42020159870) and described it in an a priori published protocol [14]. We followed the Preferred Reporting Items for Systematics Reviews and Meta-Analyses (PRISMA) [15] and the Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines [16] (ESM Tables 1 and 2). The umbrella review covers systematic reviews describing the prevalence of type 2 diabetes in people with a psychiatric disorder.

Search strategy and eligibility criteria

We searched four electronic databases (PubMed, EMBASE, PsycINFO and Cochrane Database of Systematic Reviews) from their inception to 17 January 2021. We used a structured search strategy with searches in four blocks: (1) block I, psychiatric disorders; (2) block II, diabetes; (3) block III, prevalence; and (4) block IV, systematic reviews. Search words and Medical Subject Headings (MeSH) terms within each block of the search were combined with an ‘OR’. In the final search, the blocks (I–IV) were combined with an ‘AND’. We included only systematic reviews in English, Dutch, German or Scandinavian language, with no limitations in publication date. The complete block search is reported in ESM Table 3.

More specifically, we included systematic reviews of observational studies, investigating the prevalence of type 2 diabetes in people with a psychiatric disorder. We used a definition of psychiatric disorders as psychiatric diagnoses mentioned in the ICD (http://apps.who.int/classifications/icd10/browse/2016/en) or Diagnostic and Statistical Manual of Mental Disorders (DSM) classification systems, or elevated levels of clusters of psychiatric symptoms (DSM III or DSM IV, axis I or axis II; DSM5 section II; ICD-10, F00-F99) [17,18,19,20]. We did not include reviews focusing on either a single psychiatric symptom or distress not described as a condition in the ICD or DSM classification systems (e.g. work-related stress and sleep duration). We included systematic reviews where the psychiatric disorders were assessed by diagnosis, diagnostic interviews, hospital records, prescriptions of psychotropic medication, or self-reported, and where type 2 diabetes was assessed by diagnosis, medical records, prescription of glucose-lowering medication, or self-reports. Only systematic reviews focusing on adult samples (age ≥18 years) were included. If no information was available regarding type of diabetes, we expected that most of the individuals had type 2 diabetes and included the study. Studies with a design other than a systematic review were excluded. Similarly, reviews without an explicit literature search strategy were regarded as unsystematic and were thus excluded.

We removed duplicates in merged searches from different databases with the reference managing package EndNote X8 (Clarivate Analytics, USA). Two independent reviewers (NL and SHS) selected studies to be read in full text by screening titles and abstracts using the software package Covidence (Melbourne, Australia), and subsequently selected studies for final inclusion based on the full text of these studies, based on the predefined eligibility criteria. If at least one of the reviewers regarded a record as potentially eligible during the title/abstract screening, the record was included and evaluated at the full-text level. Any disagreements at the full-text level were resolved after consulting a third author (FR). Finally, the reference lists of the included systematic reviews were manually screened for other potentially eligible systematic reviews.

Data extraction

Two independent reviewers (NL and SHS) extracted the following information from the included systematic reviews, using an extraction form we developed before extraction of the information: name of first author; year of publication; country where the study was conducted; type of study (systematic review with or without meta-analysis); type of psychiatric disorder(s); assessment of psychiatric disorder(s); assessment of type 2 diabetes; number of primary studies of the prevalence of type 2 diabetes; total number of study participants; and primary findings including information on the prevalence of type 2 diabetes, ORs and RRs, when possible. The first author (NL) extracted the following information from all primary studies included in the systematic reviews: name of first author; year of publication; total number of study participants; number of study participants with type 2 diabetes; and/or prevalence estimates. We divided the included systematic reviews into categories of psychiatric disorders and conducted a narrative data synthesis for each of the identified categories of psychiatric disorders.

Statistical analysis

If more than one systematic review described the prevalence of type 2 diabetes for each of the identified categories of psychiatric disorders, we conducted separate meta-analyses to summarise the results of a large number of prevalence estimates quantitatively. This was a deviation from the original PROSPERO protocol, although the amendment was described in our published protocol [14]. We conducted meta-analyses based on data extracted from the primary studies that were included in the systematic reviews. We defined the numerator as number of people with a psychiatric disorder and type 2 diabetes, and the denominator as total number of people with a psychiatric disorder in the primary study. We included each primary study only once in the meta-analysis. If several primary studies were based on the same sample, we included the study with the largest sample size. If data regarding the numerator and denominator were not available in the primary study, we extracted data from the original systematic review, when possible, as the authors of the systematic reviews in some cases had contacted the author of the primary study to obtain the relevant information. Where no data was available, we excluded the primary study from the meta-analyses.

We used a random-effects model for all meta-analyses. In order to stabilise the variance, we used double arcsine transformations and the inverse variance method [21]. We calculated the I2 metrics to evaluate between-study heterogeneity [22, 23]. In our published protocol [14], we planned to use Egger’s test to examine publication bias [24]. However, concerns have been raised about the sensitivity of Egger’s test to detect asymmetry when the number of included studies is small [25]. Therefore, we decided to use two more recent methods, the Doi plot and the Luis Furuya-Kanamori (LFK) index, to identify publication bias. In the LFK index a value beyond ±1 is deemed consistent with publication bias and this index has been shown to be more sensitive than Egger’s test [25]. All meta-analyses were conducted in MetaXL version 5.3, a software package for meta-analysis in Microsoft Excel [21, 26].

Risk of bias assessment

Two independent reviewers (NL and either SHS or LK) assessed the risk of bias for each of the included systematic reviews, using the Risk of Bias in Systematic Reviews (ROBIS) tool [27]. A third author (LK or SHS) resolved disagreements. We assessed the following four types of risk of bias: (1) study eligibility criteria; (2) identification and selection of studies; (3) data collection and study appraisal; and (4) synthesis and findings. Based on how we assessed potential risk of bias within these four types, we assessed an overall risk of bias in the systematic review (high, low or unclear risk of bias).

Results

Study selection and characteristics

In total, we identified 32 systematic reviews including a total of 245 unique primary studies focusing on prevalence of type 2 diabetes in psychiatric disorders from 431 full-text screenings and 5155 records found in the dataset searches (Fig. 1). The included systematic reviews were focused on 11 categories of psychiatric disorders: (1) schizophrenia (n = 7) [10, 28,29,30,31,32,33]; (2) bipolar disorder (n = 4) [9, 31, 34, 35]; (3) depression (n = 3) [11, 31, 36]; (4) substance use disorder (n = 3) [31, 37, 38]; (5) anxiety disorder (n = 2) [31, 39]; (6) eating disorder (n = 2) [40, 41]; (7) intellectual disability (n = 2) [42, 43]; (8) psychosis (n = 2) [44, 45]; (9) sleep disorder (n = 2) [46, 47]; (10) dementia (n = 1) [48]; and (11) a ‘mixed’ group that comprised different types of psychiatric disorders (n = 9) [31, 49,50,51,52,53,54,55,56].

Fig. 1
figure 1

Study flowchart

Full size image

We present the characteristics of the 32 systematic reviews in Table 1. Nineteen (59%) of the 32 systematic reviews originated from Europe, six (19%) from the USA, four (13%) from Australia and one (3%) from each of Chile, China and Ethiopia. Different methods were used to identify psychiatric disorders and type 2 diabetes, including diagnostic criteria, prescription data, screening, admission status, medical records, diagnostic tests, surveys, questionnaires, interviews and self-reports. In most of the systematic reviews (91%, 29/32), multiple types of assessments were used. A considerable proportion (41%, 13/32) did not report the type of assessment for psychiatric disorders or type 2 diabetes. Several systematic reviews were based on an overlapping set of primary studies. We report an overview of the 245 unique primary studies of the prevalence of type 2 diabetes in ESM Table 4.

Table 1 Characteristics of 32 systematic reviews included in the umbrella review investigating prevalence of type 2 diabetes among people with psychiatric disorders
Full size table

Risk of bias assessment

We report an overview of the risk of bias assessment in Table 2 and ESM Fig. 1. Twenty of the included systematic reviews had a high risk of bias [28,29,30,31,32,33,34,35,36, 41, 42, 45, 47,48,49, 51, 52, 54,55,56] while 12 had a low risk of bias [9,10,11, 37,38,39,40, 43, 44, 46, 50, 53]. We found high concerns regarding the identification and selection of studies in 19 reviews (59%) and concerns regarding the data collection and study appraisal in 18 reviews (56%).

Table 2 Presentation of ROBIS assessment of the 32 systematic reviews included in the umbrella review
Full size table

Narrative synthesis

An overview of prevalence estimates for each category of psychiatric disorder is shown in Table 3. In the 32 systematic reviews, prevalence estimates for type 2 diabetes in people with identified psychiatric disorders ranged between 1.3% and 66.0%. When only including systematic reviews with low risk of bias (n = 12), the prevalence estimates of type 2 diabetes ranged from 5.1% to 22.3%. In the systematic reviews with low risk of bias, the highest prevalence estimates of type 2 diabetes was reported in people with a sleep disorder (22.3%) [46], psychosis (18.9%) [44] and a substance use disorder (11.9–15.6%) [37, 38]. Forty-four per cent (14/32) of the included systematic reviews made comparisons with control groups and reported increased risk of type 2 diabetes in people with psychiatric disorders. In one of the included systematic reviews investigating people with an intellectual disability, three primary studies reported less prevalence of diabetes in people with an intellectual disability compared with the general population [42].

Table 3 Overview of prevalence estimates of type 2 diabetes for 11 categories of psychiatric disorders identified in 32 systematic reviews included within the umbrella review
Full size table

Meta-analyses of the prevalence of type 2 diabetes

For ten of the 11 categories of psychiatric disorders, more than one systematic review reported the prevalence of people with type 2 diabetes. We therefore conducted new meta-analyses of type 2 diabetes prevalence estimates for each of these categories of psychiatric disorders. These ten meta-analyses included between six and 153 unique primary studies depending on the psychiatric disorder, and an overview of the estimated prevalence estimates for each of the categories is presented in Table 4. According to our meta-analyses, people with a sleep disorder have the highest prevalence of type 2 diabetes (39.7% [95% CI 34.9, 44.7%]), followed by those with a binge eating disorder (20.7% [95% CI 7.6, 37.4]), a substance use disorder (15.6% [95% CI 10.3, 21.7]), an anxiety disorder (13.7% [95% CI 7.7, 20.9]), bipolar disorder (11.4% [95% CI 7.8, 15.6]) and psychosis (11.1% [95% CI 7.3, 15.5]). We found the lowest prevalence of type 2 diabetes in people with an intellectual disability (8.1% [95% CI 6.5, 9.8]). We present forest plots for all meta-analyses in ESM Fig. 2.

Table 4 Prevalence estimates of type 2 diabetes in psychiatric disorders from random-effects meta-analyses of 245 primary studies from 32 systematic reviews
Full size table

In all meta-analyses, a high level of between-study heterogeneity was detected (I2 0.89–0.99). We found substantial publication bias in the four meta-analyses focusing on schizophrenia, substance use disorder, binge eating disorder and mixed group of psychiatric disorders. We found only minor publication bias in the meta-analysis of type 2 diabetes prevalence in people with psychosis. We present Doi plots for all meta-analyses in ESM Fig. 3.

Discussion

In this umbrella review, we systematically summarised and critically assessed the existing evidence on the prevalence of type 2 diabetes in people with a psychiatric disorder.

We meta-analysed data from primary studies for ten categories of psychiatric disorders revealing prevalence estimates between 8% and 40%. Our meta-analyses were based on primary studies from several systematic reviews and therefore were built on more data than previous meta-analyses. Overall, we found that most of the meta-analyses revealed type 2 diabetes prevalence estimates that were comparable with existing meta-analyses within each of the categories of psychiatric disorders [9,10,11, 38,39,40, 44, 46, 50, 53]. However, meta-analysed data revealed notably high prevalence estimates for sleep disorder (40%) and binge eating disorder (20%). Sleep disorder, as defined in DSM5, constitute a subgroup of psychiatric disorders [19] and have high comorbidity with several other somatic diseases [57, 58]. In our umbrella review, most of the primary studies were conducted among people with additional somatic diseases such as chronic kidney disease. It is likely that this physical comorbidity contributes to the high type 2 diabetes prevalence estimates in people with a sleep disorder. A previous review suggested the association between sleep disorders and type 2 diabetes is bidirectional, with sleep disorders being a risk factor for type 2 diabetes and type 2 diabetes (especially when combined with poor metabolic control) being a risk factor for sleep disorders [59]. For people with a binge eating disorder, the type 2 diabetes prevalence estimates were based on data from four primary studies with limited sample sizes. The results should be interpretated with caution, due to large CIs. Additionally, there might be an effect of publication year, with more recent publications reporting higher prevalence of type 2 diabetes in people with a binge eating disorder (see ESM Fig. 2).

In 44% of the included systematic reviews, comparisons with control groups were made, revealing an increased prevalence of type 2 diabetes in people with a psychiatric disorder. This increased prevalence of type 2 diabetes was detected for most of the investigated psychiatric disorders, suggesting a shared vulnerability in people with psychiatric disorders to the development of type 2 diabetes. In line with these findings, our meta-analysed type 2 diabetes prevalence estimates typically exceeded those for global prevalence estimates of type 2 diabetes [12, 13]. However, better understanding of these differences and explaining factors are still needed. The global prevalence estimates of 6.3% to 9.3% for type 2 diabetes is as reported in 2017 and 2019, respectively [12, 13], whereas the primary studies included in the meta-analyses were conducted between 1980 and 2020. The comparison of the global prevalence estimates with the findings from our meta-analyses suggest that people grouped across all the investigated psychiatric disorders more often have type 2 diabetes. However, more refined comparisons should be made between prevalence estimates in the future to better account for differences in populations groups, study settings and the broad range of years as well as methods used to ascertain type 2 diabetes.

In the general population, the prevalence of type 2 diabetes has increased considerably over the past decades [13, 60, 61] due to an increase in the population incidence of obesity and type 2 diabetes, a decrease in age-standardised mortality and a growing proportion of the ageing population [62]. However, with the exception of people with a binge eating disorder, the forest plots (ESM Fig. 2) show that when the primary studies are ranked according to publication year, there is no indication of increase in prevalence estimates of type 2 diabetes in people with a psychiatric disorder with publication year of the study over the past two or three decades. This might be explained by issues such as changes in prescription patterns of psychotropic drugs [63] and the high likelihood of undiagnosed type 2 diabetes in people with a psychiatric disorder [64]. These issues warrant further investigation.

Overall, this umbrella review highlights the importance of focusing on prevalent type 2 diabetes across different psychiatric disorders, as we see a general pattern of elevated type 2 diabetes prevalence rates in people with a psychiatric disorder. Much of previous research has focused on schizophrenia and depression [10, 65,66,67], but our findings suggest that the prevalence of type 2 diabetes is also high in several other psychiatric disorders, emphasising the need for future studies that cover the full range of psychiatric disorders. Future collaborations should follow the example of the initiative of the European Depression in Diabetes (EDID) Research Consortium [68], combining researchers from different fields and countries to collaborate with a focus on a psychiatric disorder and diabetes.

Strengths and limitations

To the best of our knowledge, this is the most comprehensive summary on type 2 diabetes in psychiatric disorders to date providing an overview across different psychiatric disorders in one and the same paper. We have summarised all existing systematic reviews in the field and included all primary studies to estimate the type 2 diabetes prevalence for each psychiatric disorder.

There are also several limitations that should be acknowledged. First, our umbrella review included studies published during an extended period of several decades, during which time the clinical context has changed considerably. Changes in treatment guidelines [69] and lifestyle factors such as physical activity [70], dietary habits [71] and smoking behaviours [72] will presumably have had an impact on prevalence estimates of type 2 diabetes in people with a psychiatric disorder. We have not explored effects of these developments in this umbrella review. Second, we have not included the most recent observational studies, which are not included in a systematic review. However, primary studies up to 2020 have been included in the umbrella review, and we therefore estimate that this gap is modest. Third, the prevalence of type 2 diabetes was not available for some common psychiatric disorders such as attention deficit hyperactivity disorder and autism. Fourth, most of the included systematic reviews had high risk of bias regarding the identification and selection of studies, and there were also concerns regarding data collection and study appraisal. Most of the systematic reviews rated as having high risk of bias were rated as such because no efforts were taken to minimise error during the conduction of the study (e.g. no description of two independent reviewers conducting the procedures). Last, we detected publication bias in several of the identified categories of psychiatric disorders. All conducted meta-analyses yielded wide CIs, and we found a high level of heterogeneity. This might be explained by different types of study populations due to different purposes of the included systematic reviews and primary studies. Some systematic reviews had no restrictions on the study population, whereas other reviews had a narrower focus on elderly populations or hospital inpatients. Furthermore, different assessment methods of psychiatric disorders and type 2 diabetes were used. More homogenous prevalence estimates could be expected if all primary studies used the gold standards for diagnosis of type 2 diabetes [73].

Conclusions

For more than a decade, the importance of screening and adequate treatment of type 2 diabetes in people with psychiatric disorders has been highlighted [74]. By providing the most comprehensive review on this issue to date, this umbrella review confirms that type 2 diabetes is a common comorbidity in people with a psychiatric disorder, particularly a sleep disorder, a binge eating disorder, a substance use disorder or an anxiety disorder. Our review identified a need for future research to identify contributors to this comorbidity. Future research must determine whether underdiagnosis of type 2 diabetes in people with psychiatric disorders means that the present figures are lower-bound estimates. Reliable information about prevalence and a better understanding of biological and behavioural factors driving increased prevalence of type 2 diabetes in people with psychiatric disorders will be crucial to developing cost-effective strategies for the management of type 2 diabetes in people with psychiatric disorders.