The study participants included were all enrolled TEDDY children who reported a persistent confirmed positivity for islet autoantibody as of May 31, 2021. These subjects were divided into four subcohorts depending on their islet autoantibody combination (IAA, GADA, IA-2A, and ZnT8A) of the first, second, third, or fourth appearing islet autoantibody. The progression from the first autoantibody to the second or type 1 diabetes, the second to the third or type 1 diabetes, and the third to the fourth or type 1 diabetes are referred to as transition states in this study. The given starting state islet autoantibody combination at first visit with positivity for the respective autoantibodies in each subcohort is presented in Table 1, together with the possible types of islet autoantibodies that could subsequently appear.

Cord-blood or heel stick capillary blood samples taken in the first months of life were used to identify high risk HLA DR-DQ genotypes meeting the eligibility criteria of the TEDDY protocol. Typing utilized PCR amplification, Sanger sequencing, oligonucleotide probe hybridization and/or denaturing gel electrophoresis (9). The HLA genotypes were then confirmed at 9–12 months of age using reverse line blot hybridization at a central HLA Reference Laboratory at Roche Molecular Systems, Oakland, CA (10).

Islet autoantibody surveillance for IAA, GADA and IA-2A started at three to four months of age. It was then repeated every third month until 4 years of age, and thereafter every 3 to 6 months until 15 years of age. In order to confirm the islet autoantibody positivity, IAA, GADA, and IA-2A were analyzed in two different reference laboratories, one in the United States at Barbara Davis Center for Childhood Diabetes at the University of Colorado Denver and the other in Europe at the University of Bristol in the U.K, by radiobinding assays as previously described (11–13). Both reference laboratories have high sensitivity and high specificity as well as concordance for the islet autoantibody assays. A detected autoantibody was considered as persistent when confirmed by both reference laboratories in a sample drawn at a consecutive follow-up study visit. Persistent autoimmunity was defined by the presence of one or several persistent confirmed autoantibodies. The ZnT8A surveillance started once a child was positive but not confirmed for another primary islet autoantibody (IAA, GADA, IA-2A). ZnT8A were also analyzed at one of the two reference laboratories and considered persistent upon two consecutive positive samples in one of the reference labs (14).

The HbA1c test was performed from the nine months TEDDY visit in islet autoantibody positive children and every third month thereafter until 15 years of age. This requirement for HbA1c measurement was added to the TEDDY protocol 4 years after the start of the study. HbA1c samples were analyzed using an ion-exchange HPLC method on a Tosoh G8 instrument at the Diabetes Diagnostic Laboratory (DDL) at the University of Missouri, standardized using the Diabetes Control and Complications Trial reference method (imprecision coefficient of variation < 1.3%) (15; 16).

The association between HbA1c and the transition from one islet beta cell autoantibody state to the subsequent autoantibody transition state was assessed using joint models of competing risk and longitudinal data. The joint models defined time as the time in years from the start of the initial state. The competing risks models studied the relationships between the covariates and the risks of transitioning from a given autoantibody state to the next autoantibody state or type 1 diabetes diagnosis. Four separate sets corresponding to the four different starting states, IAA first, GADA first, IAA + GADA, and IAA + GADA + ZnT8A were analysed. The corresponding sub-cohorts were defined as the subjects going through the four starting states under consideration. HLA (DR3/DR4 yes or no), gender, country, BMI-z score, and HbA1c were included as covariates in the competing risks model and proportional hazards for competing events are assumed. Longitudinal models are used to model the change of HbA1c and BMI-z scores over time. The trajectories of HbA1c and BMI-z scores were modeled by longitudinal mixed effects models with constant, linear, and quadratic orthogonal polynomials. All HbA1c data from the initial autoantibody visit were included and HbA1c data after the time of type 1 diabetes diagnosis were excluded. Due to the multiplicity of subcohorts and events, a consecutive p<0.01 was considered statistically significant. Technical details of the statistical models are described in the supplemented Technical Appendix (Figure 1A–D).

Increased levels of HbA1c were associated with a higher risk of developing type 1 diabetes (HR 1.27, 95% CI [1.16, 1.39], p<0.001). The landmark plot shows the increase of HbA1c in IAA positive children from five years back from type 1 diabetes onset (Figure 1A). No statistically significant association between HbA1c and the transition from IAA as the single autoantibody to any subsequent second autoantibody (GADA, IA-2A, or ZnT8A) or multiple subsequent autoantibodies was found (Supplementary Table 1. A). In IAA positive children, HLA DR3/DR4 heterozygosity was associated with GADA as the second autoantibody (HR 1.89, 95% CI [1.27,2.800], p=0.002). Being an FDR with IAA was associated with two or more autoantibodies (HR 3.707, 95% CI [1.754,7.834], p<0.001). The HbA1c trajectory analysis in this subcohort showed a roughly linear increase of HbA1c over time (estimated covariate 0.34 SE 0.06, p<0.001).

Comparable to the IAA only subcohort, increased HbA1c was only significantly associated with type 1 diabetes as the subsequent transition event after GADA as a single autoantibody (HR 1.82, 95% CI [1.59,2.07], p<0.001). The landmark plot (Figure 1B) illustrates the linear increase of HbA1c during the five years prior to the type 1 diabetes event in GADA only positive children. HbA1c was not associated with the risk of any second islet autoantibody in GADA only positive children. However, lower HbA1c levels were significantly associated with IA-2A (HR 0.85, 95% CI [0.75,0.97], p=0.017) as a second autoantibody following GADA (Supplementary Table 1B). HLA DR3/DR4 heterozygosity was associated with IAA as the second autoantibody following GADA as the first (HR 2.16, 95% CI [1.43,3.26], p=0.001). The trajectory analysis of HbA1c present also for this GADA only subcohort a roughly linear increase of HbA1c over time (estimated covariate 0.64, SE 0.05, p<0.001).

Increased HbA1c levels were associated with type 1 diabetes (HR 1.82, 95% CI [1.58,2.10], p<0.001) in children positive for both IAA and GADA. The linear increase of HbA1c five years before type 1 diabetes clinical onset is presented in the landmark plot in Figure 1C. Increased HbA1c was not associated with any third autoantibody (Supplementary Table 1C). Similar to the two previously mentioned single autoantibody subcohorts, trajectories of HbA1c increased over time (estimated covariate 0.65, SE (0.06), p<0.001) in this subcohort of children with two autoantibodies. Female gender was associated with IA-2A as the third autoantibody preceded by GADA and IAA (HR 1.81, 95% CI [1.17,2.79], p=0.007).

In this subcohort with three autoantibodies, increased HbA1c levels were associated with type 1 diabetes clinical onset (HR 2.12, [1.79,2.51], p<0.001). The increase of HbA1c was linear (slope estimate 1.37, SE (0.148), p<0.001) with increasing rate (quadratic estimate 0.47, SE (0.092), p <0.001) over time as proximity to clinical onset of type 1 diabetes increases. The increased trajectories of HbA1c five years back from the development of type 1 diabetes in the subgroup with three autoantibodies is illustrated in Figure 1D. IA-2A as the fourth autoantibody was not associated with higher levels of HbA1c, but possibly suggested lower HbA1c levels (HR 0.90, 95% CI [0.82,0.99], p=0.036) (Supplementary Table 1D).

The main result of this study is the association of increasing HbA1c levels over time with significantly higher hazard ratios for type 1 diabetes, indicating a higher risk for the type 1 diabetes event, regardless of prior islet beta cell autoantibody number or combination. There was no association between increasing HbA1c and the transition to positivity for the second, third, or fourth islet beta cell autoantibody. However, the HbA1c trajectory analysis revealed a linear increase of HbA1c, in progression to type 1 diabetes, irrespective of the number and combinations of autoantibodies, larger HbA1c rate of increase with increasing autoantibody number from one to three autoantibodies, and finally increasing rate of HbA1c over time as proximity to type 1 diabetes diagnosis increases. The landmark plots presented a rise of HbA1c starting as early as five years prior to type 1 diabetes clinical onset. Nevertheless, the autoantibody transition from GADA or IAA, GADA and ZnT8A to IA-2A as the second or fourth autoantibody associated with lower levels of HbA1c is a novel finding emphasizing further investigation of autoantibodies and HbA1c together as biomarkers in the prediction of type 1 diabetes. To our knowledge, this is the first study evaluating the association between HbA1c and the progression to an additional autoantibody of specific combination in general population children who carried increased HLA-conferred risk of type 1 diabetes, had seroconverted positive for at least one islet autoantibody and were younger than 15 years of age.

The autoantibody positive TEDDY cohort represents the strength of this study with a relatively large number of autoantibody positive children from the general population, followed from birth until 15 years of age in an accurate islet beta cell autoantibody surveillance program for various numbers and combinations of islet beta cell autoantibodies. The heterogeneity in the TEDDY cohort and the relatively large number of autoantibody positive children made it possible to distribute the children in different subcohorts with different combinations and numbers of autoantibodies.

The benefit of the statistical joint model used in this study that combined longitudinal and survival models is that the estimates of factors such as HbA1c were comparable across the different autoantibody categories since the underlying hazard function was the same (17).

One limitation of this study was the inability to analyze all combinations of islet beta cell autoantibodies (IA-2A first and ZnT8A first or both without any of IAA or GADA) due to limited number of children or progression to type 1 diabetes diagnosis in less than three months. Another limitation was the HbA1c not analyzed in TEDDY until four years after the study had started, therefore some children had limited HbA1c information.

Consistent with our results, it was reported in the population-based prospective Finnish Diabetes Prediction and Prevention (DIPP) study that a 10% increase of HbA1c during 3–12 months in children with multiple islet autoantibodies predicted type 1 diabetes diagnosis after a median time of 1.1 years. Moreover, mean HbA1c levels remained stable in autoantibody positive children who did not progress to type 1 diabetes (4). Similar results were recently reported in an international study showing that an increase of HbA1c of 20% to 30% from a previous sample predicted type 1 diabetes onset and appearance of first autoantibody but not any multiple autoantibodies (18). Our study adds to these two findings by evaluating whether increased HbA1c is associated with a specific autoantibody type or combination as well as the risk of developing a subsequent autoantibody or type 1 diabetes in seroconverted children with high HLA risk.

The disease pathway to type 1 diabetes is heterogenous and associated with many factors including HLA genotype, age, age at the first appearing islet autoantibody, type of first appearing autoantibody, gender, and BMI giving rise to different endotypes (2; 19). Considering this, the TEDDY cohort was analyzed in two subcohorts with IAA or GADA as the first appearing autoantibody. The present analysis showed, however, that lower levels of HbA1c in GADA single positive TEDDY children was associated with the risk to develop IA-2A as the second autoantibody.

Irrespective of blood glucose levels reflecting beta cell function, a decrease in hemoglobin or iron can increase the HbA1c level (20–22). The HbA1c trajectories in the current study show an increase in HbA1c over time in autoantibody positive children driven by those who progress to type 1 diabetes. This HbA1c increase is still within normoglycemic ranges detected up to years before onset. We have previously shown in autoantibody positive TEDDY children an inverse association of mean corpuscular hemoglobin (MCH) and mean corpuscular volume (MCV) with HbA1c (23), indicating a decrease in iron levels generally caused by an aberrant erythropoiesis or a reduction in iron intake or absorption (24; 25). Thus, further research is required to clarify whether there are common factors associated with erythropoiesis and development of type 1 diabetes.

Unexpectedly, increased risk for developing IA-2A in GADA positive children and those with IAA, GADA, and ZnT8A was associated with lower levels of HbA1c. This may be explained by a more aggressive autoimmune attack on the beta cells leading to autoantibody spreading or insulin leakage into the bloodstream, as IA-2A positivity is known to confer a rapid progression risk of type 1 diabetes (14; 26; 27).

Type 1 diabetes disease process is extremely heterogenous and varies with age, genetics, BMI, and sex (2; 28). The prediction of disease progression in pre-symptomatic type 1 diabetes children at stage 1 or 2 is currently made by autoantibody surveillance programs together with regular OGTTs and HbA1c monitoring. Within these follow-up programs, diabetic ketoacidosis can effectively be prevented, but close follow-up is costly and limits public health implementation (29; 30). However, biomarkers predicting the progression from one stage of type 1 diabetes to the next are limited, and more accurate predictive biomarkers are needed to complement the autoantibody screening. Given that the risk of multiple autoantibodies is age-related and declines exponentially by age (31), there is a need for additional studies evaluating age-related biomarkers. An ability to predict time more accurately to type 1 diabetes progression would improve clinical trial designs and move us closer towards personalized medicine. This study shows the high impact of HbA1c as a time predictive biomarker for type 1 diabetes onset. Thus, the joint model analysis designed in this study could be further developed with HbA1c as a tool predicting time to type 1 diabetes diagnosis.