Search

While associations between the type and number of islet autoantibodies and progression to type 1 diabetes (T1D) have been reported, the effect of titer values is less well understood. We aim to quantify the ability of autoantibody titers at seroconversion to improve T1D onset prediction. Prospective cohorts in Finland, Germany, Sweden, and the US have followed 24662 children at increased genetic risk for development of islet autoantibodies and T1D. For the 1400 who seroconverted (523 developed T1D), the titers of insulin autoantibodies (IAA), glutamic acid decarboxylase autoantibodies (GADA), and insulinoma antigen-2 autoantibodies (IA2A) at time of initial and confirmed seroconversion, i.e., the respective first and consecutive second autoantibody-positive serum sample, were normalized (to log multiples of upper limit of normal) and analyzed. Prediction models using multivariate logistic regression with inverse probability censored weighting (IPCW) were trained using 10-fold cross validation. Discriminative power for disease was estimated using the IPCW concordance index (c-index) with 95% confidence intervals estimated via bootstrap. Multivariate Cox proportional hazards models were used to quantify the impact of the autoantibody titers. A baseline model with covariates for data source, sex, HLA-DR/DQ genotype, and age at initial and confirmed seroconversion had a performance of 64 c-index; 95% CI 62-65. Significant improvement was observed after adding IAA, GADA, IA2A positivity indicators at initial and confirmed seroconversion (74; 72-75). Adding the corresponding autoantibody titers resulted in significant additional gains (76; 75-76). Adjusted hazard ratios (HR) from the Cox model showed that autoantibody titers at confirmed seroconversion were predictive of diabetes (HR 1.29; 95% CI 1.20-1.38 for IAA, 1.12; 1.07-1.18 for GADA, and 1.19; 1.15-1.25 for IA2A, all p Islet autoantibody titers at seroconversion improve T1D prediction. Disclosure K. Ng: Employee; Self; IBM. V. Anand: None. R. Veijola: None. M. Maziarz: None. K. Waugh: None. W. Hagopian: None. P. Achenbach: None. T1di study group: n/a. Funding JDRF (1-IND-2019-717-I-X)

Four birth-cohort studies in the U.S. (DAISY, DEWIT), Sweden (DiPiS) and Finland (DIPP) were initiated in 1990s to early 2000. General population newborns and young first-degree relatives (FDR) of T1D patients, preselected on increased HLA risk, were followed for islet autoantibodies (AAb)and T1D. We harmonized their common variables, including AAb to insulin, IA-2, GAD, or ZnT8, HLA-DR-DQ risk genotypes, socio-demographics, diet, family history, glucose/HbA1c and anthropometric measures. The combined cohort included 22,312 subjects with AAb measured in 256,243 visits (mean 11±10 per subject) at 3-12 month intervals. During follow-up, 2279 (10%) subjects developed at least one AAb, 914 (4%) multiple AAb and 603 (3%) T1D. The cumulative risk of T1D differed by the number of AAb at seroconversion (Figure). T1D developed in 26% (95% CI 24-32%) of 1AAb, 62% (55-68%) of 2AAb and 84% (75-88%) of >=3AAb positive subjects in the 15y after seroconversion. T1D risk varied by HLA genotype: 40% (35-48%) in DR3/4-DQ2/8, 22% (18-31%) in DR4/x-DQ8/x and 21% (18-35%) in DR3/x-DQ2/x of 1AAb subjects and, respectively, in 82% (68-88%), 61%(50-65%)and 87% (32-98%)of 2AAb subjects. Once ≥2 AAb positive, HLA genotype was less predictive of progression to T1D. This harmonized dataset offers an unparalleled opportunity to validate and improve upon reported risk estimates for T1D development via novel analytical methods. Disclosure V. Anand: None. B. Liu: None. M. Ghalwash: None. E. Koski: None. K. Ng: None. J.L. Dunne: None. W. Hagopian: Research Support; Self; Novo Nordisk A/S. M. Lundgren: None. Å. Lernmark: None. J. Ilonen: None. R. Veijola: None. M. Rewers: None. Funding JDRF

The development of multiple islet autoantibodies (IA) predicts progression to type 1 diabetes (T1D), and forms the basis of current staging of T1D. However, age at seroconversion, persistence of IA, and frequency of sampling influence the determination of multiple IA status. Individual IA may appear at different timepoints and fluctuate. This analysis aims to understand how differences in the stringency of multiple IA status definition impact risk estimates of progression to stage 3 T1D. We combined data from three birth-cohort studies (N=18,537): DAISY (U.S.), DiPiS (Sweden) and DIPP (Finland), defining four groups by IA positivity pattern and accounting for variability in persistence and timing in concurrent and sequential patterns of IAA, IA2A, and GADA. Subjects were placed exclusively in their most stringent IA group. Cumulative incidence of progression to T1D was estimated by survival analysis and log-rank test, which showed group differences (p Disclosure B. Liu: None. M. Ghalwash: None. V. Anand: None. E. Koski: None. K. Ng: None. J.L. Dunne: None. M. Lundgren: None. H. Elding Larsson: None. W. Hagopian: Research Support; Self; Novo Nordisk A/S. R. Veijola: None. B.I. Frohnert: None. Funding JDRF

Progression to type 1 diabetes (T1D) is strongly predicted by multiple risk factors including genetic and non-genetic variables such as islet autoantibodies (AAb). Little is known, however, about the impact of interactions among the known risk factors on T1D risk prediction. Through consideration of interactions among risk factors, we aimed to evaluate time-to-diagnosis for T1D from seroconversion for AAb using a novel time-to-event model (RankSvx). RankSvx simultaneously predicts subject-level event times and risk scores as measured by Mean Absolute Error (MAE) and Concordance Index (CI), respectively. We control for subject-level characteristics (4 HLA groups, sex, and their interactions with AAb, age at seroconversion and visit-level AAb count) and use LASSO for feature selection. We also explored up to 12 months of AAb dynamics (i.e., appearance or disappearance of AAb in 3-month intervals) post seroconversion using an interaction model and compared it to a non-interaction (baseline model at seroconversion) and a traditional Cox model. We used a harmonized dataset of the DAISY, DiPiS, and DIPP studies with 1,880 AAb positive subjects of whom 512 (27%) progressed to T1D. First, we found the interaction model was superior to the baseline model without interactions (Wilcoxon Rank test p-value: 0.016). Second, incorporating AAb dynamics further improved T1D onset time prediction. When including post seroconversion AAb dynamics at 12 months, the interaction model performed 15% better than the baseline model (CI: 0.83 vs. 0.68, MAE: 2.5 years). Finally, the top identified interactions included: 1) young age at seroconversion interaction with IA-2A, and 2) HLA Group that includes at least one DR3-DQ2.5 haplotype interaction with ZnT8A or GADA. Our results are consistent with recent findings in the TEDDY dataset of patterns of AAb appearance and genetic association and age-related T1D incidence. Further investigations on seroconversion and post seroconversion endotypes are necessary. Disclosure B. Liu: None. V. Anand: None. M. Ghalwash: None. K. Ng: None. J.L. Dunne: None. R. Veijola: None. M. Rewers: None. M. Lundgren: None. Funding JDRF

Clinical type 1 diabetes (T1D) is preceded by asymptomatic islet-autoimmunity (IA) with presence of islet autoantibodies (AAb). However, both AAb profiles (clusters) during IA phase and progression rate to T1D are heterogeneous. It is unclear if AAb profiles affect the progression to T1D similarly in different countries. We analyzed data from 3 prospective birth cohorts, DAISY (U.S.), DiPiS (Sweden), and DIPP (Finland), to address this question. We developed a similarity algorithm that considers subject-level temporal IA profiles (based on 4 AAbs - IAA, GADA, IA2A, and ZnT8A positivity or negativity), age at which AAb developed, and imbalance in AAb positivity by weighting positive AAb matches higher than negative ones. Using this subject-level similarity measure, we performed hierarchical clustering of AAb positive subjects having at least 3 visits. We used a log-rank test to identify the number of clusters in 1125 DIPP subjects (329 T1D cases), 309 DAISY subjects (70 cases), and 226 DiPiS subjects (33 cases). Each dataset was individually clustered. We identified three clusters in each dataset associated with different rate of progression from seroconversion to T1D - slow progressors (SP), moderate progressors (MP) and fast progressors (FP). These were distributed as follows: DIPP (654 SP with 3% 10-year T1D risk, 356 MP with 60% risk, 115 FP with 96% risk); DAISY (215 SP with 6% 10-year risk, 92 MP with 62% risk, 2 FP with 100% risk); and DiPiS (160 SP with 3% 10-year risk, 62 MP with 58% risk, 4 FP with 100% risk). Finally, we tested the generalizability of the clusters by 1) learning a clustering from one data set “learned clusters,” 2) classifying subjects from another (independent) data set with the “learned clusters,” and 3) measuring how often the subjects were assigned to the same cluster (SP, MP, FP) as the “internal” clustering. The average accuracy of cluster transfer was high (mean = 87.9%, sd = 1.1%) suggesting that the IA profile predicts progression rate to T1D independently of the population. Disclosure M. Ghalwash: None. V. Anand: None. B. Liu: None. E. Koski: None. K. Ng: None. J.L. Dunne: None. M. Lundgren: None. M. Rewers: None. R. Veijola: None. Funding JDRF