Risks of Psychiatric Disorders and Suicide Attempts in Children and Adolescents With Type 1 Diabetes: A Population-Based Cohort Study

Participants

We used data from the Swedish Pediatric Diabetes Quality Registry (SWEDIABKIDS), Swedish National Diabetes Register (NDR), and the Swedish National Patient Register (NPR) to identify patients (probands) born in Sweden from 1964 onward with onset of type 1 diabetes before 18 years of age. Since 2000, the SWEDIABKIDS has been recording data from all Swedish pediatric diabetes centers reaching 99% national coverage in 2007 (13). In 1996, the NDR was initiated as a quality measure in diabetes care at a national level (14). Annual reports on patients treated for diabetes are delivered nationwide by hospitals as well as outpatient and primary care health centers. NDR provides information on type of diabetes, date of onset, and insulin regimen. The NPR contains data on somatic and psychiatric inpatient health care since 1973 and outpatient care since 2001 (15).

For every patient with childhood-onset type 1 diabetes, we randomly selected 100 control subjects (unexposed individuals) from the Total Population Register. Probands and unexposed individuals were matched by sex, year, and county of birth. The unexposed individuals were required to be born in Sweden after 1972, not have diabetes of any type before the age of 18 years, and be alive and living in Sweden at the time of diagnosis of diabetes of the matched proband (description of study design in Supplementary Fig. 1). Probands were divided into cohorts born 1973–1986, 1987–1996, and 1997–2009 based on the time of diagnosis of type 1 diabetes. These cohorts were chosen because they coincide with changes in ICD classification in Sweden (introduction of ICD-9 was 1987 and ICD-10 was in 1997). To investigate the mechanism behind the association between type 1 diabetes and childhood psychiatric disorders, we also studied psychiatric outcomes in full siblings of the patients with type 1 diabetes (siblings of probands). Biological relationship with patients with type 1 diabetes was established through the Multi-Generation Register, which links children to their biological parents (16). Healthy siblings of probands were defined as children of the same parents without a diagnosis of diabetes of any type before the age of 18 years. Similar to the comparisons between probands and unexposed, 100 healthy control sibling pairs (unexposed siblings) were randomly selected and matched to each case-sibling pair by sex, year, and county of birth of both siblings. Both subjects in the relative control pair were required to be free of a diabetes diagnosis of any type prior to the age of 18 years (Supplementary Fig. 2).

Exposure

Type 1 diabetes was defined as having such a diagnosis in the SWEDIABKIDS or NDR or an ICD code in the NPR (ICD-8: 250.00–250.09; ICD-9: 250A–250X; ICD-10: E10). ICD-8 and -9 do not distinguish between type 1 and type 2 diabetes, but in our age range (≤18 years) these codes identify type 1 diabetes with high positive predictive value, as type 2 diabetes is rather infrequent in the Swedish population up to 18 years of age (17). We excluded 70 children who were later rediagnosed as having monogenic diabetes (not type 1 diabetes). Furthermore, 153 patients were diagnosed with type 1 diabetes outside Sweden and therefore excluded from further analysis owing to uncertain date of diagnosis and psychiatric morbidity prior to diabetes.

Outcome Variables

Outcome variables related to psychiatric disorders were obtained from the NPR. The variables were defined in accordance with ICD classifications: 1) any psychiatric disorders (ICD-8 codes 290–315, ICD-9 codes 290–319, and ICD-10 codes F00-F99), 2) suicide attempt (ICD-8/ICD-9 codes E950-E959 or E980-E989 and ICD-10 codes X60–X84 or Y10–Y34), 3) psychotic disorders (ICD-8 codes 295 and 297–299; ICD-9 codes 295, 297, and 298; and ICD-10 codes F20–F29), 4) mood disorders (ICD-8 codes 296 and 300.4; ICD-9 codes 296, 300E, and 311; and ICD-10 codes F30–F39), 5) anxiety, dissociative, stress-related, and somatoform disorders (ICD-8 code 300, except 300.4, and code 307; ICD-9 code 300, except 300.E, and 308–309; and ICD-10 codes F40–F45 and F48), 6) eating disorders (ICD-9 codes 307B and 307F and ICD-10 code F50), 7) psychoactive substance misuse (ICD-8 codes 291, 303, and 304; ICD-9 codes 291, 303, 304, 305A, and 305X; and ICD-10 codes F10–19F), 8) attention-deficit hyperactivity disorder (ICD-9 code 314 and ICD-10 code F90), 9) autism spectrum disorder (ICD-9 code 299 and ICD-10 code F84), 10) intellectual disability (ICD-8 codes 310–315, ICD-9 codes 317–319, and ICD-10 codes F70–F79), and 11) other behavioral disorders (ICD-9 codes 312–313 and ICD-10 codes F91–F98).

Covariates

Information on sociodemographic characteristics was obtained by linkage to the biological parents through the Multi-Generation Register (16). Parental psychiatric morbidity was defined as at least one event of the following: psychiatric diagnosis (ICD-8 codes 290–315, ICD-9 codes 290–319, and ICD-10 codes F00–F99), suicide attempt (ICD-8/ICD-9 codes E950–E959 and ICD-10 codes X60–X84 in the NPR), or death by suicide (obtained from the Cause of Death Register) (18). Data on parental country of birth came from the Migration Register. Parental level of education was extracted from the Education Register, the LISA database (the longitudinal integration database for health insurance and labor market studies [with Swedish acronym]), and the population censuses from the years of 1970, 1975, and 1985 (19,20). In all patients, the highest level of education obtained by either parent was used in the multivariate analyses. The Medical Birth Register provided data on perinatal factors (21).

Statistical Analysis

Baseline characteristics were compared between probands and controls by a χ² test or Student t test for mean values (Supplementary Table 1). Because of baseline differences between groups, not fully controlled by matching, subsequent analyses were additionally adjusted for covariates. The risk of psychiatric disorders (outcome) related to diagnosis of type 1 diabetes (exposure) was estimated with the Cox proportional hazards model conditioned on the matching variables (sex and year and county of birth). Follow-up began on diagnosis of diabetes (the date of first dose of insulin) and ended with first reported event of the diagnosis of psychiatric disorder, emigration from Sweden, death, or end of the study period (31 January 2010). Censoring dates of death or emigration were extracted from the Cause of Death Register and the Migration Register, respectively (18,22). To control for censored observations, we calculated prevalence of psychiatric disorder by the Kaplan-Meier method. The prevalence of psychiatric disorders was estimated for the 10-year period after the onset of diabetes. This enables further comparison of estimates between cohorts and previously reported clinical samples.

Likewise, using the sibling data, we analyzed the relation between exposure (having a sibling with type 1 diabetes) and outcome of psychiatric disorders with the Cox model. The sibling cohorts were observed since their birth to the end of observation defined as above. Similarly, siblings of control subjects were observed from birth. As a side note, this will automatically result in a slightly higher prevalence among the control subjects to probands compared with control subjects of siblings. A robust sandwich estimator function was used to adjust for family clusters. Missing data were not replaced by any method but categorized as “unknown” for analyses. Statistical analyses were undertaken with SAS software (version 9.3; SAS, Cary, NC). The study was approved by the Regional Ethics Committee at Karolinska Institutet, Stockholm, Sweden.