Among the 360 participants randomised to receive oral insulin or oral placebo [19], rates of diabetes progression did not differ significantly between the two groups (HR 0.75 [95% CI 0.50, 1.12], p=0.16). Likewise, among the 298 participants randomised to receive i.v. injection of insulin or only observation [18], there was no difference between active therapy and no therapy (HR 0.95 [95% CI 0.68, 1.34], p=0.78) (electronic supplementary material [ESM] Fig. 1). Therefore, data from the placebo and treatment arms were combined for subsequent analysis. The duration of follow-up was 2.90 (IQR 2.82) years. Among the 658 study participants, 227 (34.4%, or ~12% per year) developed type 1 diabetes. Progressors developed diabetes at a median of 2.24 (IQR 2.13) years, while non-progressors were still free of diabetes at a median 3.41 (IQR 2.80) years of follow-up.

Impact of age on progression to type 1 diabetes

The impact of age on progression to stage 3 type 1 diabetes was analysed using baseline age as a continuous variable in a univariate Cox regression model. Progression to type 1 diabetes was inversely related to age (HR 0.96 [95% CI 0.94, 0.97] per 1 year, p<0.0001). Next, we assessed the relationship between age as a continuous variable and changes in fasting glucose, 2 h glucose and glucose sensitivity. Baseline age was not related to these variables among the entire cohort or in the progressor/non-progressor subgroups (all p>0.05). We used χ2 serial survival analyses to scan the entire age range of the cohort to identify the highest discriminant age for risk of progression to type 1 diabetes. In this analysis, age 14 years, which divided the cohort roughly into pre- and postpubertal children vs teenagers and adults, doubled the risk of type 1 diabetes development (Fig. 1). This age cut-off was used in subsequent analyses for the following purposes: (1) to compare baseline metabolic measures between progressors and non-progressors aged <14 years and ≥14 years; (2) to test the ability of fasting and model-derived metabolic variables to aid in diabetes risk prediction; and (3) to compare the trajectory of changes in insulin and glucose sensitivity between progressors and non-progressors and between older and younger progressor subgroups.

Fig. 1

Diabetes-free survival function of study participants by the discriminant age of 14 years (p=2.0×10−7). Age <14 years, n=448; age ≥14 years, n=210; 295 prepubertal, 221 peripubertal and 142 adults

Impact of age on baseline metabolic variables

Stratifying by an age cut-off of <14 and ≥14 years, progressors were still younger as compared with non-progressors and had higher baseline prevalence of IGT (Table 1). Sex distribution was similar and autoantibody patterns were not significantly different between any of the groups. As expected, anthropometric measures were higher in both progressors and non-progressors aged ≥14 years compared with progressors and non-progressors aged <14 years.

Table 1 Baseline anthropometric and clinical variables of the cohort by outcome and age

Fasting glucose was similar between all the groups, whereas 2 h plasma glucose at study entry was 15% higher in progressors (Table 2). Insulin sensitivity was lower in progressors and non-progressors ≥14 years compared with younger progressors and non-progressors. Similarly, fasting C-peptide, fasting insulin, insulin secretory rates and total insulin output were all higher in progressors and non-progressors aged ≥14 years. Glucose sensitivity, which is the mean slope of the dose–response function relating ISRs to glucose concentrations during the OGTT, was reduced by ~46% both in progressors aged <14 years and progressors aged ≥14 years compared with non-progressor groups. Beta cell rate sensitivity, potentiation and insulin sensitivity were similar when comparing progressors with non-progressors.

Table 2 Baseline metabolic and functional variables by outcome and age Diabetes risk prediction using age in combination with either fasting or OGTT/IVGTT-derived metabolic variables

To identify metabolic variables that were predictive for the development of diabetes, we ran a multivariate Cox model including sex, BMI, fasting plasma glucose, insulin, C-peptide concentrations and fasting ISR. In this model, only age <14 years was a significant predictor of progression to diabetes (HR 3.12 [95% CI 2.04, 4.86]). In contrast, in univariate analysis, sex-specific quartiles of beta cell glucose sensitivity were strongly associated with progression, predicting that at 3 years of follow-up, 50% of the participants in the lowest quartile (median glucose sensitivity 30 [IQR 13] pmol min−1 m−2 [mmol/l]−1) would develop diabetes vs 5% of those in the top quartile (median glucose sensitivity 148 [IQR 63] pmol min−1 m−2 [mmol/l]−1). Notably, the survival curves were evenly graded both in participants aged <14 years and in participants aged ≥14 years (Fig. 2). At follow-up, the insulin secretion/plasma glucose dose–response function was markedly shifted downward and to the right in progressors but not in non-progressors (ESM Fig. 2). A multivariate Cox model including both fasting and OGTT/IVGTT-based variables (Fig. 3) showed that in addition to beta cell glucose sensitivity, age <14 years, fasting C-peptide and 2 h glucose were significant predictors of progression, totalling an AUC of the receiver operating characteristic curve (AUCROC) of 0.81 (p=2.5×10−34). This was a large improvement over a model using only sex, age, BMI and fasting plasma glucose (AUCROC=0.63, p=2.7×10−6).

Fig. 2

Diabetes-free survival function by sex-specific quartiles of beta cell glucose sensitivity in participants aged <14 years (a) (n=448, p=3.4×10−15) or ≥14 years (b) (n=210, p=2.4×10−12). Colours indicate glucose sensitivity quartiles (red, green, blue and orange, from the lowest to the highest quartile). Each quartile includes 112 participants aged <14 years and 53 participants aged ≥14 years; progressors, n=227; non-progressors n=431

Fig. 3

Multivariate Cox proportional hazard model of diabetes progression in the entire cohort (n=658). Plots show HR and 95% CI. Insulin secretion represents total insulin output over the 2 h of the OGTT; AIR represents the AIR to i.v. injection of glucose

Time trajectories of changes in insulin and glucose sensitivity

The time course of the most relevant metabolic variables was reconstructed using all available OGTTs (n=4152). For each variable, the time trajectory was split by outcome (progressors vs non-progressors). As depicted in Fig. 4, in progressors, both fasting and 2 h glucose levels rose rapidly within 0.5–1.0 years before diagnosis as compared with the stable glucose levels of non-progressors. This biphasic pattern was in phase with declines in glucose sensitivity and rate sensitivity, potentiation and insulin sensitivity. Using year −1 as the trajectory inflection point, the rate of decline in glucose sensitivity until year −1 was only slightly faster in progressors than in non-progressors (−9.4 [−27.0] vs −4.0 [−24.5] pmol min−1 m−2 [mmol/l]−1 per year, median [IQR]; p=0.03). In contrast, the ‘late’ (year −1 to year 0) rate of decline was approximately fourfold faster in progressors than in non-progressors (−25.8 [−57.2] vs −6.5 [−72.3], median [IQR], p<0.0001).

Fig. 4

Time course of plasma glucose levels, beta cell function variables and insulin sensitivity by outcome (progressors vs non-progressors). Time 0 is the time of diabetes diagnosis or study end. Numbers of participants are given at the bottom of the figure. NP, non-progressors; P, progressors

Finally, we aimed to compare the trajectories of the change in glucose sensitivity and insulin sensitivity in older and younger progressors. When stratifying by baseline age (Table 3), there were no significant differences in the yearly rate of decline in either glucose sensitivity (−13.7 [21.2] vs −11.9 [21.5] pmol min−1 m−2 [mmol/l]−1, median [IQR], p=0.52) or insulin sensitivity (−22 [37] vs −14 [40] ml min−1 m−2, median [IQR], p=0.07) when comparing the progressors aged <14 years with progressors aged ≥14 years. When focusing on early and late time segments, the rate of decline in glucose sensitivity until year −1 was similar in older vs younger progressors (−4.1 [−24.0] vs −6.7 [−26.3] pmol min−1 m−2 [mmol/l]−1 per year, median [IQR], p>0.05). In addition, the ‘late’ (year −1 to year 0) rate of decline did not significantly differ (−14.1 [−55.8] vs −19.8 [−54.2] pmol min−1 m−2 [mmol/l]−1 per year, median [IQR], p>0.05) between older and younger progressors. When the trajectories of the progressors in the two age groups were plotted against their respective baseline age (Fig. 5), the patterns of change of the physiological variables were remarkably similar between younger and older participants (confirming the data presented in Table 3). Overall, these data indicate that the impact of baseline age is to shorten the time at which the physiological variables start to change (i.e. the inflection point of the biphasic time course), with similar rates of change in both beta cell glucose sensitivity and insulin sensitivity observed during this period of active progression.

Table 3 Yearly rate of change in metabolic variables between baseline and follow-up in progressors and non-progressors by age groupFig. 5

Time course of plasma glucose levels, beta cell function variables and insulin sensitivity by baseline age in progressors aged <14 years (n=184) and ≥14 years (n=43). Within each age group, trajectories are first aligned to the age of diagnosis (as in Fig. 4). Then, trajectories are shifted by the median group-specific age at diagnosis (last time point is the median age at diagnosis in the group)

A graphical summary of the overall changes in glucose and insulin sensitivity is shown in Fig. 6, which includes data from a group of healthy children under 14 years of age and another group of participants over the age of 14 years, each sex- and BMI-matched to the corresponding age group of our DPT-1 participants. On this graph, younger participants lie to the right of older participants on the scale of insulin sensitivity whether they are progressors, non-progressors or healthy control individuals. Furthermore, healthy participants of either age group have far better glucose sensitivity and insulin sensitivity than either DPT-1 progressors or non-progressors. Progressors start with worse glucose sensitivity than non-progressors and lose more glucose and insulin sensitivity over time. However, among progressors aged <14 years and progressors aged ≥14 years, the slopes of decline in glucose sensitivity and insulin sensitivity were not different, suggesting that age does not influence patterns of metabolic decline in autoantibody-positive individuals followed longitudinally until stage 3 type 1 diabetes onset.

Fig. 6

Scattergram of glucose sensitivity against insulin sensitivity in progressors and non-progressors by age <14 or ≥14 years (progressors aged <14 years, n=184; progressors aged ≥14 years, n=43; non-progressors aged <14 years, n=264; non-progressors aged ≥14 years, n=167). Historical data from healthy control groups of participants aged ≥14 years (n=72; age 30 ± 2 years, BMI 24.2 ± 5.0 kg/m2, mean ± SD) and younger individuals (n=68; age 9 ± 2 years, BMI 23.6 ± 1.1 kg/m2, mean ± SD) are also plotted. Squares, baseline values; circles, follow-up values; dashed arrows connect non-progressors; solid arrows connect progressors. Plots are mean ± SEM