Combined Analysis of Autoantibodies Improves Prediction of IDDM in Islet Cell Antibody-Positive Relatives

Prediction of insulin-dependent diabetes mellitus (IDDM) is still largely based on islet cell antibodies (ICAs), but it may be improved by combined analysis with other humoral markers. We examined autoantibodies to insulin (IAAs), glutamic acid decarboxylase (GAD), and M_r 37,000 and M_r 40,000 fragments of islet antigens (37 and 40 kDa) together with ICA subtypes in 101 family members with ICAs ≥10 Juvenile Diabetes Foundation units (JDF U) followed for up to 14 years, of whom 18 have developed IDDM. Life-table analysis showed a 43% risk of IDDM within 10 years for those with ICAs ≥10 JDF U, rising to 53% for those with ICAs ≥20 JDF U. The risk for ICAs ≥10 JDF U was 62% in the family members in the youngest age quartile (<13.2 years) and fell with increasing age to 4% in those >40.7 years of age (P = 0.03). ICAs ≥10 JDF U combined with IAAs gave a risk of 84% (P = 0.03 compared with IAA⁻), and ICAs ≥10 JDF U combined with GAD antibodies gave a risk of 61% (P = 0.018). The risk for ICAs >10 JDF U with antibodies to 37-kDa antigen was 76% (P < 0.0001). Risk increased with the number of autoantibodies, from 8% for ICAs alone to 88% with >3 autoantibodies (14 cases detected) (P < 0.0001). The increased risk associated with multiple antibodies was observed independent of age. The median time to diagnosis in those with antibodies to 37- and/or 40-kDa antigen was 1.5 years, compared with 7.2 years in those with IAAs and GAD antibodies in the absence of antibodies to 37/40 kDa. The intensity and range of the autoantibody response offers better overall prediction of diabetes than any single autoantibody specificity, although antibodies to 37-/40-kDa antigens may prove to be useful markers of early clinical onset. We found that 78% of future cases of IDDM in ICA⁺ relatives came from the 27% with multiple autoantibodies and estimate that 88% of individuals within this category will need insulin treatment within 10 years. We propose a simple predictive strategy based on these observations.