The contribution of autoantibodies, HLA markers and age to long-term estimates of risk of type 1 diabetes were examined after a median of 11 years (range 7.5-14) during the follow-up in a cohort of 234 siblings (aged 2-29 years) of French children with recent-onset type 1 diabetes, of whom 12 (5.1%) developed diabetes. We evaluated islet cell antibodies (ICA) by indirect immunofluorescence and autoantibodies to insulin (IAA), to the 65 kDa isoform of glutamic acid decarboxylase (GADA) and to the IA-2 protein (IA-2A) by radioligand assay in sequential serum samples. Among the 234 siblings of type 1 diabetic patients screened, 27 were positive for at least one antibody, 11 of whom progressed to develop type 1 diabetes during the follow-up (sensitivity, 92%, predictive value, 41%). Among the four antibodies tested individually, ICA had the highest sensitivity (83%) but a poor predictive value (59%) and IA-2A the highest predictive value (70%). IAA and GADA both exhibited poor sensitivity and predictive value. Combinations of antibodies achieved better predictive values than antibodies tested individually. Satisfactory predictive values were obtained for the combination of GADA with IA-2A (83%), for any combination of at least two antibodies other than ICA (70%) and for the combination of ICA with at least one other antibody (69%). The risk estimates were highest in the presence of three or four antibodies, whether comprising ICA or not, but with a concomitant loss of sensitivity. For most antibody combinations, cumulative risks showed progression from approximately 50% after 5 years to 100% after 13 years. HLA-DR3/4 was significantly more frequent in siblings developing type 1 diabetes than in non-diabetic siblings (9/12 vs. 39/217, relative risk (RR)=14, P</=0.0001). The predictive value of HLA-DR3/4 was low (19%); however, taking into account the presence of HLA-DR3/4 in subjects who were positive for more than one antibody resulted in a higher predictive value (67%, vs. 20% in non-DR3/4 subjects, P</=0.02). In addition, siblings developing diabetes were younger at entry than those who did not (mean =7.5 +/-1.23 vs. 12.5 +/-0.39 years, respectively; P</=0.01). Ten of 12 were aged less than 10 years compared with 106/222 non-diabetic siblings (RR =5.4, P</=0.03). Moreover, younger age was associated with a more rapid development of type 1 diabetes. In conclusion, our results show that the combination of IAA, GADA and IA-2A autoantibodies in sequential serum samples is satisfactory for the identification of subjects at risk of developing type 1 diabetes. Additional factors such as younger age and HLA-DR3/4 as markers of progression to disease may contribute to more efficient prediction in antibody positive subjects.
Copyright 1998 Academic Press