A Comparative Analysis of the Sensitivity, Specificity, Concordance, and the 5-year Predictive Power of Diabetes-related Autoantibody Assays

This study compares novel type 1 diabetes-related autoantibody assays developed to improve upon the standard radiobinding assay (RBA). Samples from 1505 individuals, followed for 5 years or to clinical type 1 diabetes, originally tested by RBA were aliquoted and sent blindly to 5 laboratories (BDC, IDR, DRI, MSD, Enable) to be tested by electrochemiluminescence (ECL) assays, Luciferase Immuno Precipitation System (LIPS) assays, multiplex antibody detection by agglutination-PCR (ADAP) assays, and N-terminally truncated GAD65 or IA2β autoantibody RBAs (tGADA/IA2βA). Findings: The fraction of samples that were concordant for negative/positive interpretations across all assays were 79.7% (GADA), 65.2% (IA-2A), 36.2% (IAA), and 67.5% (ZnT8A). The assays with the highest Youden index for predicting the previous RBA results differed by autoantibody: 0.65 LIPS(IDR) for IAA, 0.91 ECL(BDC) for ZnT8A, 0.82 tGADA RBA(IDR) for GADA, 0.91 ECL(MSD and BDC) for IA-2A. The Youden index for predicting 5-year type 1 diabetes varied significantly across assays and was highest for LIPS(DRI) for all autoantibody combinations, with little variation in the respective maximum Youden index. The discordance between assays makes it problematic to interpret positivity when comparing results from different assays. Longitudinal autoantibody assessments should be tested with the same assay.