Endogenous and Antipsychotic-Related Risks for Diabetes Mellitus in Young People With Schizophrenia: A Danish Population-Based Cohort Study

The prevalence of diabetes mellitus is four to five times higher among people with schizophrenia than in the general population (1). Those who suffer from both diabetes and schizophrenia have a three- to fourfold higher overall mortality rate than the general population, and at least one-third of those deaths can be attributed to diabetes (2). Diabetes contributes to excessive cardiovascular deaths and more than 20 years of reduced life expectancy in people with schizophrenia (3, 4). Young people with schizophrenia, age less than 40 years, are at high risk (5) for early-onset type 2 diabetes, which is a rapidly progressing severe disease subtype with increased risks for microvascular and macrovascular complications (4, 6–8).

Studies from the preneuroleptic era (9) have reported an increased prevalence of diabetes, impaired glucose tolerance, and increased insulin resistance in people with schizophrenia (10–12). Abnormal glucose tolerance is more prevalent in antipsychotic-naive people with schizophrenia than among healthy controls, and this association is independent of the effects of body mass index and of poor health habits (13–15). Moreover, increased prevalence of diabetes among the first-degree relatives of people with schizophrenia indicates a potential genetic link between schizophrenia and diabetes (16, 17). TCF7L2 and IGF2BP2, two of the most replicated susceptibility genes for type 2 diabetes, have been associated with schizophrenia (18–20). Furthermore, linkage studies, genetic association studies, and pathway analyses have supported shared genetic risks in these two disorders (21–23).

As soon as chlorpromazine was introduced in 1952, there were several reports of its association with diabetes and impaired glucose tolerance (10). Since the introduction of second-generation antipsychotic drugs, their association with diabetes has been investigated more systematically (24–28). A meta-analysis, not including data on aripiprazole, ziprasidone, and amisulpride, indicated that second-generation antipsychotics confer more risk for diabetes in schizophrenia than do first-generation antipsychotic drugs (29). However, a longitudinal study showed that both second-generation and first-generation antipsychotics increase the risk for diabetes and that second-generation antipsychotics, as a class, confer significantly less risk for incident diabetes than first-generation antipsychotics (30). Individual first-generation and second-generation antipsychotics may differ widely in their risks for diabetes (30–32). Clozapine may confer higher risk for diabetes than other second-generation antipsychotics (31, 33). A recent meta-analysis reported the incidence of type 2 diabetes in 2–24-year-olds exposed to antipsychotics as 3.09 (95% confidence interval [CI], 2.35–3.82) per 1,000 person-years, and it also showed that the risk for early-onset type 2 diabetes was significantly higher in those who received prescriptions for second-generation antipsychotics (34). Dose-response effects (30) and several plausible biological mechanisms (21) indicate a causal association between antipsychotics and diabetes. Antagonism of central and peripheral M₃, H₁, 5-HT_2c, other serotonergic, and adrenergic receptors may have direct diabetogenic effects, or they can increase the risk of diabetes by indirect mechanisms, such as weight gain and reduced insulin sensitivity (21, 31, 35).

There are several caveats in our current understanding of endogenous and antipsychotic-related risks of diabetes in schizophrenia. First, to our knowledge there has not been any large population-based cohort study investigating endogenous risk for diabetes in antipsychotic-naive people with schizophrenia compared with the general population. Second, population-based cohort studies focusing on the risks for early-onset diabetes in young people with schizophrenia are sparse (34). Third, despite a large volume of literature, the impact of the first-line therapeutic decision of choosing between first-generation and second-generation antipsychotics on the antipsychotic-related risk for diabetes in schizophrenia remains uncertain (29, 30). Fourth, previous studies investigating the antipsychotic-related risks for diabetes in schizophrenia have seldom been adjusted for the effects of family history of diabetes and for the dosage of other potentially diabetogenic psychotropic medications (36–38). Hence, this population-based cohort study included all young people (more than 2.7 million) born on Jan. 1, 1977, or later in Denmark, investigated endogenous risk for early-onset diabetes in antipsychotic-naive schizophrenia, and evaluated the antipsychotic-related risks in people with schizophrenia, after adjusting for the effects of potential confounders.

Method

Study Population

This population-based cohort study included people from the Danish Civil Registration System (39), which has recorded data on all Danish citizens since 1968. Each Danish citizen is assigned a unique 10-digit personal identification number at birth. This number is used in all Danish registries, enabling unambiguous linkage among them. The registration system provides data on gender, date of birth, date of death, date of migration, and parental identity. Our cohort included all people who were born in Denmark on or after Jan. 1, 1977. They were followed up until Jan. 1, 2013.

People With Diabetes

Information on diabetes was derived from the Danish National Patient Register (40) and the Danish National Prescription Registry (41). The Danish National Patient Register, established in 1977, collects data on all hospitalizations from nonpsychiatric hospitals, including dates of admission and discharge and discharge diagnoses, coded by the treating physicians according to the 8th revision of the International Classification of Diseases (ICD-8) (42) until the end of 1993 and the 10th revision (ICD-10) (43) thereafter. Since 1995, outpatient visits have also been recorded. We identified all hospital discharge diagnoses of diabetes. In ICD-8 these included codes 249 (insulin-dependent diabetes [type 1]) and 250 (non-insulin-dependent diabetes [type 2]). In ICD-10 these included codes E10 (brittle, juvenile, and ketoacidosis prone diabetes), E11 (adult-onset, maturity-onset, nonketotic, and stable diabetes), E12–14 (malnutrition, other specified and unspecified diabetes), H36.0 (diabetic retinopathy), and O24 (diabetes in pregnancy, childbirth, and puerperium) excluding O24.4 (gestational diabetes). Neither the ICD-8 nor ICD-10 provides specific codes for late-onset type 1 diabetes in adults or drug-induced diabetes. Because people with schizophrenia may develop type 2 diabetes at younger ages than the general population, and because the Danish registers include more frequent use of unspecified diabetes codes in people with schizophrenia, we examined the complete spectrum of ICD diabetes codes. Although using the Danish National Patient Register and Danish National Prescription Registry to identify people with diabetes has proven to yield high-quality and almost complete (44, 45) data, we cannot reliably distinguish type 1 and type 2 diabetes by using the registers (46, 47).

In addition to hospital contacts involving diabetes, we identified antidiabetic prescriptions from the Danish National Prescription Registry, because many people with diabetes are treated only in primary care. The prescription registry contains data from 1995 on all prescription drugs dispensed at all Danish pharmacies, and the information includes the Anatomical Therapeutic Chemical (ATC) classification system codes, dispensing dates, drug names, dose units, number of dose units in package, number of daily defined doses sold (48), and dispensing pharmacy codes. Antidiabetic drugs are available only by prescription in Denmark, so we identified all prescription redemptions for antidiabetic drugs (ATC code: A10). However, women with polycystic ovary syndrome may be treated with metformin (ATC code: A10BA02). Thus, to avoid inclusion of these nondiabetic people, we excluded all metformin prescriptions redeemed by women age 20 or more years in this cohort. Onset of diabetes was defined as the date of first admission with diabetes since birth or as the date of the first prescription for an antidiabetic drug since 1995 or later, whichever came first.

People With Schizophrenia

The Danish Psychiatric Central Research Register (49) was established as an electronic database in 1969 and has been recording data on all psychiatric admissions since 1969. Diagnoses in the Danish Psychiatric Central Research Register were coded in accordance with the ICD-8 (42) until the end of 1993 and ICD-10 (43) thereafter. This register includes data on all outpatient visits since 1995. We identified all hospital discharge diagnoses of schizophrenia (ICD-8: 295 [excluding 295.79] and ICD-10: F20). These diagnoses have been validated (50) and have contributed substantially to epidemiological research (49). As the data in the Danish National Prescription Registry have been complete only since 1995, we excluded all people (N=97) who developed schizophrenia before Jan. 1, 1996, to ensure that we had complete psychopharmacological data and at least 1 year of prescription use history for all people with schizophrenia after they had been diagnosed. Hence, we included all incident schizophrenia cases since Jan. 1, 1996.

Exposure to Antipsychotics

We traced all oral and depot antipsychotic prescriptions (ATC code: N05A, excluding N05AN [lithium]) since 1995 using information from the Danish National Prescription Registry (48). The prescription registry is a powerful pharmacoepidemiological tool that provides complete high-quality data on all filled prescriptions for antipsychotics prescribed by various psychiatric departments, general practitioners, or other specialists after 1995. It does not record the drugs that are dispensed within hospitals. All Danish pharmacies are obliged to register all dispensed prescriptions, and they have a high economic incentive for completeness of prescription registration because of the universal reimbursement system for prescription drugs in Denmark. Antipsychotics were subdivided into first-generation antipsychotics, second-generation antipsychotics (ATC codes: N05AD03, N05AE03–4, N05AH04, N05AL01, N05AL05, N05AX08, and N05AX13), olanzapine, aripiprazole, and clozapine (see Appendix S1 in the data supplement accompanying the online version of this article).

Covariates of Interest

We chose our covariates a priori on the basis of their availability, previous research associating them with diabetes or schizophrenia, and their influence on the prescribing pattern of antipsychotics. From the Danish National Prescription Registry, we identified other potentially diabetogenic psychotropic medications, such as tricyclic (ATC code: N06AA) and tetracyclic (ATC codes: N06AA21, N06AX03, N06AX11) antidepressants (36) and valproate (ATC code: N03AG01) (37, 38). Data on gender, urbanicity, and calendar period were derived from the Danish Civil Registration System and the Danish National Patient Register (40). Using the parental identity from the civil registration system, we identified siblings and half-siblings, and we linked them to the national patient register and prescription register to establish family history of diabetes.

Statistical Analyses

We calculated age-specific and gender-specific incidence rates of diabetes and their 95% CIs. In order to investigate the endogenous risk for diabetes, we used Cox proportional hazards regression models to compute the adjusted hazard ratio for diabetes in antipsychotic-naive people with schizophrenia, compared with antipsychotic-naive people without schizophrenia. Follow-up began on the date of birth and ended on the date of diabetes diagnosis, first prescription of antipsychotics, emigration, or death or on Jan. 1, 2013. The first registered diagnosis of schizophrenia during follow-up was included as a time-dependent variable. Gender, family history of diabetes, urbanicity, and exposure to valproate and tricyclic or tetracyclic antidepressants were included as covariates.

We assessed antipsychotic-related risk for diabetes by including only people with schizophrenia who were antipsychotic naive at the time of their diagnosis. We employed Cox proportional hazard regression models estimating the adjusted hazard ratio by comparing the rates of diabetes in people with schizophrenia with and without antipsychotics. Follow-up began on the date of first diagnosis of schizophrenia and ended on the date of diabetes diagnosis, emigration, or death or on Jan. 1, 2013. Age was chosen as the time scale. Our Cox regression model included gender, family history of diabetes, and urbanicity as time-independent covariates and calendar period and daily defined doses of antipsychotics, valproate, and tricyclic or tetracyclic antidepressants as time-varying covariates. Moreover, we estimated adjusted hazard ratios and their 95% CIs that compared the risks for diabetes before and after starting selected antipsychotics in people with schizophrenia. Furthermore, we performed sensitivity analyses that followed the patients until their first type 2 diabetes contact (ICD-8: 250; ICD-10: E11, O24.1) and/or oral antidiabetic drug (ATC code: A10B), ignoring any previous contacts due to type 1 diabetes or insulin prescriptions. We tested the proportional hazards assumption by repeating the Cox regression models and including the primary independent variable as one of the time-varying covariates, and we did not find any violations. All analyses were performed by using the statistical software STATA 13.1 (StataCorp, College Station, Tex.).

Results

Population Characteristics

We followed 2,736,510 young people for a total of 49,582,279 person-years. The median duration of follow-up was 18.80 years (interquartile range, 16.99 years). The maximum period of follow-up was 35.96 years; thus, the oldest person in our cohort was less than 36 years old. A total of 8,945 (0.33%) people developed schizophrenia during follow-up. The mean duration of follow-up after the diagnosis of schizophrenia was 5.07 (SD=4.24) years, and the mean duration of follow-up after the first antipsychotic drug prescription was 6.40 (SD=4.34) years. Table 1 presents the sociodemographic and clinical characteristics of our cohort. People with schizophrenia were significantly more likely to have a family history of diabetes than other people (χ²>1300, df=1, p<0.001).

Incidence of Diabetes

A total of 14,118 (0.52%) people developed diabetes in this cohort (see Figure S1 in the online data supplement). Age-specific incidence rates of diabetes in people with and without schizophrenia are presented in Table 2 (see also Figure S2 in the supplemental data). Gender-specific incidence rates of diabetes for men and women without schizophrenia were 0.26 (95% CI, 0.26–0.27) and 0.30 (95% CI, 0.29–0.31) per 1,000 person-years, respectively. Gender-specific incidence rates of diabetes for men and women with schizophrenia were 4.02 (95% CI, 3.35–4.83) and 4.69 (95% CI, 3.81–5.77) per 1,000 person-years, respectively.

Endogenous Risk for Diabetes

Among the people who were not exposed to any antipsychotics, 12,976 (0.5%) without schizophrenia and 11 (0.9%) with schizophrenia developed incident diabetes during our follow-up. Table 3 presents the incidence rates and the adjusted hazard ratios for diabetes in antipsychotic-naive people with schizophrenia. Family history of diabetes significantly increased the risk of diabetes (adjusted hazard ratio, 3.96; 95% CI, 3.82–4.11). People with schizophrenia had a higher incidence rate of diabetes, and they had an approximately threefold higher risk of diabetes (adjusted hazard ratio, 3.07; 95% CI, 1.71–5.41), compared with people without schizophrenia, after adjustment for the effects of potential confounders, including family history of diabetes.

Antipsychotic-Related Risks

A total of 4,322 (48.3%) people with schizophrenia were antipsychotic naive at the time of their diagnosis, and the remaining 4,623 (51.7%) had received an antipsychotic drug before schizophrenia was diagnosed. A total of 83 (1.9%) people with schizophrenia who were antipsychotic naive at diagnosis developed diabetes during follow-up. Gender-specific incidence rates of diabetes for antipsychotic-treated men and women with schizophrenia were 4.22 (95% CI, 3.17–5.62), and 3.33 (95% CI, 2.23–4.96) per 1,000 person-years, respectively. Table 4 presents the incidence rates and the adjusted hazard ratios for diabetes in people with schizophrenia, comparing their risks before and after they started treatment with selected antipsychotics. The rate for diabetes significantly increased by more than threefold (adjusted hazard ratio, 3.64; 95% CI, 1.95–6.82) in people with schizophrenia after starting any antipsychotic, compared with their rate before receiving antipsychotics, when the analysis accounted for the effects of potential confounders, including family history of diabetes.

First-Line Treatment With First-Generation or Second-Generation Antipsychotics

Compared with the people with schizophrenia who remained antipsychotic naive until the end of follow-up, those with schizophrenia who received monotherapy with either first-generation antipsychotics (adjusted hazard ratio, 3.06; 95% CI, 1.32–7.05) or second-generation antipsychotics (except clozapine) (adjusted hazard ratio, 3.44; 95% CI, 1.73–6.83) as their first-line treatment had significantly higher antipsychotic-related rates for diabetes after adjustment for the effects of potential confounders, including family history of diabetes. The first-line therapeutic decision of choosing between first-generation and second-generation antipsychotics did not make a statistically significant difference (χ²=0.13, df=1, p=0.72) in the rate for diabetes in people with schizophrenia. We repeated these analyses after excluding olanzapine from the class of second-generation antipsychotics and confirmed the earlier findings.

Risks Associated With Selected Antipsychotics

Starting either olanzapine (adjusted hazard ratio, 1.88; 95% CI, 1.36–2.59) or aripiprazole (adjusted hazard ratio, 2.35; 95% CI, 1.70–3.26) significantly increased the rate of diabetes by almost twofold in people with schizophrenia, after adjustment for the effects of potential confounders. We repeated these analyses including the duration of antipsychotic treatment before the start of either olanzapine (adjusted hazard ratio, 1.91; 95% CI, 1.37–2.65) or aripiprazole (adjusted hazard ratio, 1.69; 95% CI, 1.19–2.39) in the models and confirmed the earlier findings. Among 861 (9.6%) people with schizophrenia who received clozapine after their diagnosis, 39 (0.4%) developed incident diabetes after starting clozapine. Treatment with clozapine significantly increased the rate of diabetes by nearly fourfold (adjusted hazard ratio, 3.98; 95% CI, 2.77–5.73) in people with schizophrenia, compared with people with schizophrenia not receiving clozapine, when the analysis accounted for the potential confounders, including the family history of diabetes (see Appendix S5 in the online data supplement).

Sensitivity Analyses

We conducted sensitivity analyses limited to potential type 2 diabetes diagnoses in this cohort (N=5,488, 0.20%) and confirmed the earlier findings of a higher rate of diabetes in antipsychotic-naive people with schizophrenia compared with the general population (adjusted hazard ratio, 4.71; 95% CI, 2.60–8.51) and an increased rate of diabetes after the start of any antipsychotic drug in people with schizophrenia (adjusted hazard ratio, 3.79; 95% CI, 1.97–7.26) without statistically significant differences between first-generation and second-generation antipsychotics (χ²=0.04, df=1, p=0.85) (see Appendixes S2, S3, and S5 in the online data supplement). We performed more sensitivity analyses after excluding 19 people who developed diabetes more than 12 months after the last prescription for any antipsychotic, to distinguish direct antipsychotic effects from the potential effects of antipsychotic-related weight gain. The results of this sensitivity analysis also corroborated our earlier findings. The antipsychotic-related risk for diabetes after starting any antipsychotic was significantly higher (adjusted hazard ratio, 3.28; 95% CI, 1.74–6.17) than the risk in people with schizophrenia not treated with antipsychotics, and the first-line therapeutic decision of treatment with either first-generation or second-generation antipsychotic monotherapy did not make a statistically significant difference (χ²=0.78, df=1, p=0.38). Moreover, we performed sensitivity analyses by including women ages 20 or more years receiving treatment only with metformin (N=8,551). We confirmed our findings of endogenous risk of diabetes in antipsychotic-naive schizophrenia (adjusted hazard ratio, 2.10; 95% CI, 1.30–3.38) and increased rate of diabetes after the start of any antipsychotic drug in people with schizophrenia (adjusted hazard ratio, 3.56; 95% CI, 2.09–6.06) without a statistically significant difference (χ²=2.03, df=1, p=0.15) between the first-line therapeutic decision of treatment with either first-generation antipsychotics (adjusted hazard ratio, 4.27; 95% CI, 2.21–8.27) or second-generation antipsychotics (adjusted hazard ratio, 2.98; 95% CI, 1.67–5.32) monotherapy.

Discussion

This nation-wide population-based cohort study reveals an increased endogenous risk for diabetes in antipsychotic-naive people with schizophrenia compared with people without schizophrenia. It confirms the antipsychotic-related risks for diabetes in people with schizophrenia after adjusting for potential confounders, including family history of diabetes. Moreover, it adds evidence that first-line treatments with first-generation and second-generation antipsychotics do not significantly differ on their antipsychotic-related risks for diabetes. Furthermore, it documents that the antipsychotic-related rate of diabetes in people with schizophrenia who have received other antipsychotics further increases fourfold after they start taking clozapine. The strengths of this study include the largest sample size to date, the longest follow-up, the antipsychotic-naive status of the people with schizophrenia, the focus on early-onset diabetes in young people, and statistical adjustments for the effects of family history of diabetes and other potentially diabetogenic medications.

An important limitation of our study is that we could not reliably distinguish between type 1 and type 2 diabetes. Misclassification of diabetes type is likely, and a complete classification is not possible on the basis of the information available in Danish national registers (47, 51). We can only distinguish between type 1 and type 2 diabetes on the basis of ICD codes and insulin-only therapy, which potentially indicates type 1 diabetes. Moreover, people with schizophrenia are often diagnosed with ICD-10 codes for unspecified diabetes. However, sensitivity analyses limited to potential type 2 diabetes corroborated our findings. A substantial proportion of incident diabetes in young people with schizophrenia is expected to be early-onset type 2 diabetes (52). Early-onset type 2 diabetes can be a distinct, severe, rapidly progressing disease phenotype (4, 7, 8). This may limit generalizability of our findings to older people with schizophrenia, but it may emphasize their clinical importance. Besides, the possibility of increased surveillance of people with schizophrenia by the health services, and consequent surveillance bias, may explain the increased incidence of diabetes in antipsychotic-naive schizophrenia. However, prior evidence suggests that the proportion of undiagnosed diabetes may be higher among people with schizophrenia than in the general population (53, 54). Moreover, studying a relatively homogeneous and predominantly Caucasian population from a single country may limit generalizability of our findings, but this should strengthen their internal validity (55). Other limitations of this study include the lack of data on other potential confounders, such as obesity, dyslipidemia, smoking, other substance abuse, diet, and sedentary lifestyle; the lack of data on drugs dispensed within hospitals in the Danish National Prescription Registry; and the grouping of diverse antipsychotics together during our analyses. Controlling for all potential confounders is not possible in register-based research (56), and such residual confounding could have influenced our results.

A high prevalence of endogenous diabetes in antipsychotic-naive people with schizophrenia was reported in a different Danish cohort (31, 33), and our findings confirm a higher incidence of endogenous diabetes in young people with schizophrenia. In our study, young people with schizophrenia were significantly more likely to have a family history of diabetes. This finding indicates familial aggregation and genetic overlap of these two disorders. A recent population-based cross-sectional study from Australia reported that people with psychosis are at increased risk of diabetes if they have a family history of diabetes or if they do not have a family history of diabetes but are taking antipsychotics (57). Our results, adjusted for family history of diabetes, clarify this misconception and establish that people with schizophrenia are at increased risk for diabetes regardless of their family history of diabetes and their antipsychotic drug exposure.

Previous studies have reported that second-generation antipsychotics as a class confer more risk for diabetes than first-generation antipsychotics (29). However, evidence to the contrary also exists (30). Our findings add important evidence to this debate. Similarly, aripiprazole has been reported to either lower (31) or increase (58) the risk for diabetes in schizophrenia. Our findings do not support a protective effect of aripiprazole but document its association with diabetes risk. There is a possibility that aripiprazole and first-generation antipsychotics might have been prescribed more often for diabetes-prone people with schizophrenia, and such channeling bias might have influenced pertinent findings. Similarly, olanzapine might have been prescribed more often for people perceived to be less prone for diabetes by their treating clinicians, and this could have underestimated the associated risks. To address the diabetogenic potential of olanzapine, a three-phase algorithm for antipsychotic treatment that limits use of olanzapine to only a second-line antipsychotic agent has been proposed (33). First-line treatment with second-generation antipsychotics including olanzapine did not differ significantly from first-generation antipsychotics on their risks for diabetes in this cohort. Olanzapine has not been a second-line antipsychotic in many countries, including Denmark at the time of this cohort, and it is currently recommended as a first-line antipsychotic for schizophrenia by many national (59) and international (60) guidelines. Our findings do not add evidence to change the prevailing guidelines regarding prescription of olanzapine in schizophrenia. However, they confirm the substantial increase in diabetes risk from starting clozapine treatment for people with schizophrenia.

Shared genetic factors, negative symptoms, substance abuse, and unhealthy lifestyle in schizophrenia may contribute to its endogenous risk for diabetes. Future studies should focus on underlying biological mechanisms connecting these two disorders and on potential preventive interventions. Further studies investigating antipsychotic-related risks of individual antipsychotics and of their polypharmacy (30, 61) for diabetes in people with schizophrenia are warranted. Antipsychotic-related risk for diabetes is often discussed as one of the many factors in the therapeutic decision-making process of choosing an antipsychotic agent in psychiatric clinics. Combined endogenous and antipsychotic-related risks for diabetes in schizophrenia mandate more attention. Psychiatric services should either develop specific protocols or closely collaborate with primary care facilities to screen for diabetes among people with schizophrenia in the community. Promoting a healthy lifestyle, early detection by regular, at least yearly, screening, and effective treatment of diabetes should be integral parts of multidisciplinary management of schizophrenia.