A total of 3304 families (441 trios) were included in the dataset analysed (available on 1st July 2009). Inclusion criteria have been described previously [5
]. Briefly, an eligible family consisted of at least 2 siblings with T1D diagnosed before the age of 35 and treated with insulin within 6 months of diagnosis without an interruption longer than 6 months thereafter. Occasional exceptions were made to these criteria (through assessment by an eligibility committee) if other clinical data supported the diagnosis of type 1 diabetes. The parents of the affected sibpair, all affected and up to 2 unaffected siblings were invited to participate if available. In population groups with a low prevalence of type 1 diabetes, trio families, consisting of one affected patient and his/her parents, were also included. In order to avoid duplication, each family member was asked if they had participated in this study before a new inclusion was started. All of the participating centers were approved by the Office for Human Research Protection (Department of Health and Human Services, US). The local Ethics Committees approved the study and all participants signed a written informed consent before inclusion.
Clinical information was obtained using questionnaires delivered at each of the participating centers. Information was obtained directly from the participating family members and/or from their clinical records. Clinical data obtained included gender, ethnic background, age of onset, family history of diabetes, estimated body size at diagnosis (categorised as heavy, normal, thin) and self-reported AAID. AAID was considered to be present if any of the following was reported: thyroid, celiac or Addison’s disease, rheumatoid arthritis, multiple sclerosis, systemic lupus erythematosus, vitiligo, psoriasis, inflammatory bowel disease, pernicious anemia or myasthenia gravis.
Glutamic acid decarboxylase (GAD65 [GADA]) and the intracellular portion of protein tyrosine phosphatase (IA-2ic
[IA-2A]) antibodies were measured in three central laboratories (Bingley et al, Bristol, UK (for the European and UK networks); Colman et al, Southbank, AUS (for the Asia-Pacific Network) and Eisenbarth et al, Aurora, CO, US (for the North American Network)) in serum from all the affected siblings. All laboratories used similar radio-binding assays, with different local standards, with values assigned by calibration against the World Health Organization (WHO) reference reagent. Results reported on samples included in the Diabetes Autoantibody Standardization Program (DASP) were compared, and standard methods for reporting in WHO units/mL were developed [8
]. When antibody titres were above values in the standard curve, index estimations were chosen, since they gave a higher concordance among labs than extrapolation [8
]. High antibody titres were defined as values above the respective standard curve.
HLA-genotyping was also performed centrally (Noble, Erlich et al, Oakland, CA, US, for the North American Network; Tait et al, Melbourne, Australia, for the Asia-Pacific Network and Carlson et al, Malmö, Sweden, for the European and UK Networks) in all participating family members, using a PCR-based, sequence-specific oligonucleotide probe system, as previously described [9
]. Images of the results were scanned, and probe intensities were measured as pixel values and then imported into Sequence Compilation and Rearrangement Evaluation (SCORE) software for final genotype calling. T1D-related high-risk and protective haplotypes were defined according to previous data from the T1DGC [9
], i.e. DRB1
*0302 (DR4), DRB1
*0302 (DR4), DRB1
*0201 (DR3) and DRB1
*0302 (DR4) as those associated with highest susceptibility and DRB1
*0303 (DR7), DRB1
*0503 (DR6), DRB1
*0602 (DR2) and DRB1
*0301 (DR11), as those associated with strongest protection.
Statistical analysis was performed using SPSS for Windows (SPSS Inc, Chicago, IL, US) and R. Continuous variables are described as median (range) and qualitative variables, as percentages. Differences between siblings with and without AAID were analysed using Wilcoxon-Mann-Whitney’s test and chi-squared. Yates’ correction was applied to the latter, except in 2×2 tables where the expected frequency for a cell was below 5, in which case Fisher’s exact test was used. Differences in the distribution of high-risk, protective and single HLA-DRB1-DQA1-DQB1 haplotypes were also analysed (Chi-squared with Yates’ correction, or Fisher’s exact test, as described above) in the subjects with unequivocal haplotypes.
To identify factors associated with AAID, a multivariate regression analysis (general additive model) was performed. In order to avoid interference by family size (i.e, bias in favour of factors present in larger families), only 2 affected siblings per family (the first 2 diagnosed, present in most families) were included in the analysis. Gender, age of onset, time since diagnosis, antibody positivity and presence of AAID in first degree participating relatives were included in the model as independent variables and analysed in all the families. In addition, the number of HLA haplotypes, associated with high risk of, or protection from T1D, were added to the model. Furthermore, the specific HLA haplotypes associated with higher risk of AAID in the descriptive analysis were included in a model together with the clinical predictors. For the latter analyses, we only included the families in whom DRB1-DQA1-DQB1 haplotypes could be unequivocally inferred. In order to identify factors specifically associated with single disorders, the multivariate analysis was repeated using the most common AAID, i.e. thyroid and celiac disease, as dependent variables.