Previous findings suggest that there are age-related endotypes of type 1 diabetes with different underlying etiopathological mechanisms in those diagnosed at age <7 years compared with those diagnosed at age ≥13 years. We set out to explore whether variation in demographic, clinical, autoimmune, and genetic characteristics of children and adolescents with newly diagnosed type 1 diabetes support the existence of these proposed endotypes.
We used data from the Finnish Pediatric Diabetes Register to analyze characteristics of 6,015 children and adolescents diagnosed with type 1 diabetes between 2003 and 2019. We described and compared demographic data, clinical characteristics at diagnosis, autoantibody profiles, and HLA class II–associated disease risk between three groups formed based on age at diagnosis: <7, 7–12, and ≥13 years.
We found significant age-related differences in most of the characteristics analyzed. Children diagnosed at age <7 years were characterized by a higher prevalence of affected first-degree relatives, stronger HLA-conferred disease susceptibility, and higher number of autoantibodies at diagnosis, in particular a higher frequency of insulin autoantibodies, when compared with older children. Those diagnosed at age ≥13 years had a considerably higher male preponderance, higher frequency of glutamic acid decarboxylase autoantibodies, longer duration of symptoms before diagnosis, and more severe metabolic decompensation, reflected, for example, by a higher frequency of diabetic ketoacidosis.
Our findings suggest that the heterogeneity of type 1 diabetes is associated with the underlying disease process and support the existence of distinct endotypes of type 1 diabetes related to age at diagnosis.
Introduction
Type 1 diabetes is a chronic disease characterized by autoimmune destruction of the insulin-producing β-cells of the pancreas, leading to insufficient insulin production and a subsequent diagnostic rise in blood glucose. Genetic predisposition and environmental factors appear to be involved, but despite extensive research, the disease process remains poorly understood. Recently, awareness of the heterogeneity of type 1 diabetes has increased. It has, for example, been reported based on birth cohort studies in at-risk children that seroconversion to positivity for insulin autoantibodies (IAA) shows a sharp peak during the second year of life, while seroconversion to positivity for antibodies to GAD antibody (GADA) stretches over several years (1,2). The heterogeneity poses a substantial challenge for diabetes research and may explain the high response variability observed in intervention and prevention trials. These challenges have encouraged attempts to find disease endotypes (i.e., subtypes defined by distinct pathophysiological mechanisms) (3). One interesting observation was the discovery of distinct immunohistological profiles correlating with age at diagnosis (4). Those diagnosed before age 7 years had a hyperimmune CD20Hi pattern of insulitis, a lower proportion of residual insulin-containing islets, signs of aberrant proinsulin processing, lower circulating C-peptide levels, and higher proinsulin-to–C-peptide ratios (5,6). In contrast, those diagnosed at age ≥13 years had a pauci-immune CD20Lo immune phenotype, more residual insulin-producing β-cells with less evidence of aberrant proinsulin processing, higher circulating C-peptide levels, and lower proinsulin-to–C-peptide ratios. Those diagnosed in the interim (age 7–12 years) had either one or the other of the above distinct immunohistological profiles.
The aim of our study was to explore the heterogeneity of type 1 diabetes in view of these proposed age-related endotypes. We set out to describe demographic, clinical, autoimmune, and genetic characteristics of a large population of Finnish children and adolescents with newly diagnosed type 1 diabetes and to assess possible differences with respect to age at diagnosis.
Research Design and Methods
Data Source
We based our study on data from the Finnish Pediatric Diabetes Register (FPDR) (7). The Ethics Committee of the Hospital District of Helsinki and Uusimaa (Helsinki, Finland) approved the study protocol for this register. It is a national register, which has been collecting data on children and adolescents diagnosed with diabetes since 2002. Upon diagnosis, all children and adolescents treated in pediatric units are invited to participate in this register, and coverage has been estimated to be >90% since 2003 (8).
Study Population
We identified all those registered in the FPDR for diagnosis of type 1 diabetes between 1 January 2003 and 31 December 2019 (n = 8,661). For the purposes of the study, we included only those with data on islet autoantibodies within 30 days from diagnosis and on HLA class II–associated disease risk (n = 6,243). When families had several children fitting the inclusion criteria, we included the child who had been diagnosed and registered first of the siblings and excluded the other sibling(s) (n = 226). We examined those diagnosed at age <1 year (n = 44) with the intention of excluding those with neutral or protective class II HLA genotype and no detectable autoantibodies or only one positive low-titer antibody at diagnosis, based on the likelihood that such individuals have monogenic diabetes rather than type 1 (9). There were two cases matching these exclusion criteria. One had a strong protective genotype and no autoantibodies; the other had a neutral genotype and a low positive islet cell antibody titer. The total number of cases included in our analyses was 6,015.
Data Set
The data set analyzed included demographic data, clinical characteristics at diagnosis, presence and titers of diabetes-associated islet autoantibodies, and HLA class II–associated disease risk.
Demographics
Age at diagnosis is a central factor in this study, used to stratify the data and to explore differences between the age groups. We used a division into three age groups: <7, 7–12, and ≥13 years.
In order to explore the possible differences in the season of birth and the season of diagnosis, we divided the calendar year into four seasons according to common practice: spring (March to May), summer (June to August), autumn (September to November), and winter (December to February).
We examined the family history of type 1 diabetes (i.e., whether cases had a first-degree relative [parent or sibling], a father, a mother, or a sibling with type 1 diabetes [yes or no for each category] and the number of first-degree relatives affected by type 1 diabetes). Those with a first-degree relative with type 1 diabetes were classified as having familial type 1 diabetes.
The FPDR does not record data on race, because the Finnish population is more homogenous than populations of most other developed countries, so this was not taken into account in the analyses.
Clinical Characteristics at Diagnosis
Clinical characteristics at diagnosis included BMI SD score (SDS), duration of symptoms prior to diagnosis (weeks), plasma glucose (mmol/L), glycated hemoglobin (HbA1c) (% [mmol/mol] ), blood pH, and plasma β-hydroxybutyrate (mmol/L). Diabetic ketoacidosis (DKA) was defined based on a pH value <7.30 and severe DKA based on a pH <7.10 (10). BMI SDS was calculated using the World Health Organization AnthroPlus software (11).
Autoantibodies
Diabetes-related autoantibodies analyzed in this study included autoantibodies to islet cells (ICA), insulin (IAA), GADA, islet antigen 2 (IA-2A), and zinc transporter 8 (ZnT8A). The specific radiobinding assays used have been described previously, and the cutoff limits used to define positivity were 2.5 JDFU for ICA, 1.57 relative units (RU) for IAA, 5.36 RU for GADA, 0.77 RU for IA-2A, and 0.50 RU for ZnT8A (12,13). Starting from the beginning of 2017, we analyzed antibodies to truncated GAD (amino acids 96–585) instead of full-length GAD, with the cutoff limit remaining the same. Results from the Diabetes Autoantibody Standardization Program and the Islet Autoantibody Standardization Program in 2010–2016 revealed sensitivities and specificities of 36–62% and 94–98% for IAA, 64–88% and 94–99% for GADA, 62–72% and 93–100% for IA-2A, and 62–70% and 99–100% for ZnT8A, respectively (14). As mentioned above, only cases with autoantibody samples taken within 30 days (median 5 days) pre- or postdiagnosis were included to ensure that the samples represented autoantibodies. This is particularly important regarding IAA, because with a delay of sampling, positive IAA titers might reflect the use of exogenous insulin rather than an autoimmune response (7). Based on the biochemical autoantibody profile at diagnosis, we classified cases into profile groups in which either IAA (IAA positivity and GADA negativity at diagnosis) or GADA (GADA positivity and IAA negativity at diagnosis) was the likely primary autoantibody at seroconversion (15). The GADA-dominant profile included those with GADA alone, GADA + IA-2A, GADA + ZnT8A, or GADA + IA-2A + ZnT8A, whereas the IAA-dominant profile comprised those with IAA alone, IAA + IA-2A, IAA + ZnT8A, or IAA + IA-2A + ZnT8A.
HLA Class II Genotype
The procedure used for genotyping has previously been described in detail (16,17). We based our analyses on HLA-DR/DQ genotype-based risk classification, in which cases are defined as carriers of genotypes conferring high risk, moderately increased risk, slightly increased risk, neutral risk, slightly decreased risk, and strongly decreased risk (16). For additional analysis, we combined the risk groups to form two groups: those with HLA class II–conferred increased risk and those with protective or neutral HLA class II genotypes.
Statistical Methods
We reported categorical data using frequencies and proportions, normally distributed continuous data using mean and SD, and skewed continuous data using median and interquartile range. To test for independence of age groups and categorical variables of interest, we used the Pearson χ2 test. For continuous variables of interest, we used, when appropriate, the Welch ANOVA to test for the differences in the means between the age groups. For skewed continuous data, we applied the Kruskal-Wallis test to analyze differences in the mean ranks between the age groups. In the subsequent pairwise comparisons, we used the independent samples Student t test or Mann-Whitney U test, as appropriate. All tests were performed as two tailed, and the significance level was set to 5%. Bonferroni correction for multiple comparisons was not applied because of its overly conservative nature. Multiplicity issues were taken into account by cautious interpretation of the results.
To further study the relationship between the age at diagnosis and the heterogeneity of type 1 diabetes, we performed additional analyses. By using age as a continuous variable, we studied the relationship between age at diagnosis and the likelihood of having IAA positivity and/or familial type 1 diabetes rather than having none of these. We chose IAA and familial type 1 diabetes for these additional analyses based on the fact that they showed highly significant differences between the age groups in the initial analysis. To examine the functional form of the relationship, we applied binary logistic regression models with the restricted cubic spline function of age as the explanatory variable. We tested whether the linearity assumption holds. In addition, we repeated similar analyses for the following subgroups: IAA positivity and familial type 1 diabetes, IAA positivity and nonfamilial type 1 diabetes, and IAA negativity and familial type 1 diabetes. In the presence of a linear relationship, we reported odds ratio and 95% CI for age; otherwise, we presented the results graphically.
We performed data analyses using IBM SPSS Statistics 25 and the Epi package from R software.
Data and Resource Availability
Data may be available upon request from the corresponding author (M.K.) but will be subject to ethical and legal considerations.
Results
Demographics
Our study population included 6,015 children ranging in age from 0.3 to 17.3 years at diagnosis of type 1 diabetes. There was a male predominance in each of the three age groups, but this was most conspicuous in the oldest age group, with 68.2% males (Table 1). Overall, 13.2% had one or more first-degree relatives with type 1 diabetes; 5.7% a sibling, 5.4% a father, and 3.1% a mother. Familial type 1 diabetes was more common in the youngest age group, in which a significantly higher proportion had at least one first-degree relative with type 1 diabetes and a mother or sibling with type 1 diabetes. The distribution of seasons of birth or diagnosis was not statistically different between the age groups. The most common season of birth was spring (26.3%), and the least common was winter (23.3%), whereas the most common season of diagnosis was autumn (26.9%), and the least common was spring (22.5%).
Demographic characteristics of the whole study cohort of 6,015 children diagnosed with T1D and by three groups according to age at diagnosis
. | Overall . | Age at diagnosis, years . | P . | ||
---|---|---|---|---|---|
0–6 . | 7–12 . | ≥13 . | |||
n | 6,015 | 2,361 (39.3) | 2,706 (45.0) | 948 (15.8) | |
Sex | <0.001 | ||||
Male | 3,413 (56.7) | 1,269 (53.7) | 1,497 (55.3) | 647 (68.2) | |
Female | 2,602 (43.3) | 1,092 (46.3) | 1,209 (44.7) | 301 (31.8) | |
1 vs. 3 <0.001 | |||||
2 vs. 3 <0.001 | |||||
Season of birth | 0.337 | ||||
Spring | 1,581 (26.3) | 625 (26.5) | 704 (26.0) | 252 (26.6) | |
Summer | 1,522 (25.3) | 592 (25.1) | 675 (24.9) | 255 (26.9) | |
Autumn | 1,511 (25.1) | 570 (24.1) | 717 (26.5) | 224 (23.6) | |
Winter | 1,401 (23.3) | 574 (24.3) | 610 (22.5) | 217 (22.9) | |
Season of diagnosis | 0.236 | ||||
Spring | 1,353 (22.5) | 542 (23.9) | 588 (21.7) | 223 (23.5) | |
Summer | 1,478 (24.6) | 586 (24.8) | 674 (24.9) | 218 (23.0) | |
Autumn | 1,619 (26.9) | 658 (27.9) | 709 (26.2) | 252 (26.6) | |
Winter | 1,565 (26.0) | 575 (24.4) | 735 (27.2) | 255 (26.9) | |
Familial T1D | |||||
One or more first-degree relatives with T1D | 793 (13.2) | 374 (15.8) | 325 (12.0) | 94 (9.9) | <0.001 |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
N of first-degree relatives with T1D | <0.001* | ||||
0 | 5,222 (86.8) | 1,987 (84.2) | 2,381 (88.0) | 854 (90.1) | |
1 | 723 (12.0) | 345 (14.6) | 294 (10.9) | 84 (8.9) | |
2 | 64 (1.1) | 26 (1.1) | 30 (1.1) | 8 (0.8) | |
3 | 5 (0.1) | 3 (0.1) | 1 (0.0) | 1 (0.1) | |
4 | 1 (0.0) | 0 (0) | 0 (0) | 1 (0.1) | |
1 vs. 3 <0.001† | |||||
1 vs. 2 <0.001† | |||||
Paternal T1D | 322 (5.4) | 141 (6.0) | 137 (5.1) | 44 (4.6) | 0.204 |
Maternal T1D | 185 (3.1) | 93 (3.9) | 76 (2.8) | 16 (1.7) | 0.002 |
1 vs. 3 0.002 | |||||
1 vs. 2 0.025 | |||||
Sibling with T1D | 340 (5.7) | 160 (6.8) | 137 (5.1) | 43 (4.5) | 0.008 |
1 vs. 3 0.015 | |||||
1 vs. 2 0.010 |
. | Overall . | Age at diagnosis, years . | P . | ||
---|---|---|---|---|---|
0–6 . | 7–12 . | ≥13 . | |||
n | 6,015 | 2,361 (39.3) | 2,706 (45.0) | 948 (15.8) | |
Sex | <0.001 | ||||
Male | 3,413 (56.7) | 1,269 (53.7) | 1,497 (55.3) | 647 (68.2) | |
Female | 2,602 (43.3) | 1,092 (46.3) | 1,209 (44.7) | 301 (31.8) | |
1 vs. 3 <0.001 | |||||
2 vs. 3 <0.001 | |||||
Season of birth | 0.337 | ||||
Spring | 1,581 (26.3) | 625 (26.5) | 704 (26.0) | 252 (26.6) | |
Summer | 1,522 (25.3) | 592 (25.1) | 675 (24.9) | 255 (26.9) | |
Autumn | 1,511 (25.1) | 570 (24.1) | 717 (26.5) | 224 (23.6) | |
Winter | 1,401 (23.3) | 574 (24.3) | 610 (22.5) | 217 (22.9) | |
Season of diagnosis | 0.236 | ||||
Spring | 1,353 (22.5) | 542 (23.9) | 588 (21.7) | 223 (23.5) | |
Summer | 1,478 (24.6) | 586 (24.8) | 674 (24.9) | 218 (23.0) | |
Autumn | 1,619 (26.9) | 658 (27.9) | 709 (26.2) | 252 (26.6) | |
Winter | 1,565 (26.0) | 575 (24.4) | 735 (27.2) | 255 (26.9) | |
Familial T1D | |||||
One or more first-degree relatives with T1D | 793 (13.2) | 374 (15.8) | 325 (12.0) | 94 (9.9) | <0.001 |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
N of first-degree relatives with T1D | <0.001* | ||||
0 | 5,222 (86.8) | 1,987 (84.2) | 2,381 (88.0) | 854 (90.1) | |
1 | 723 (12.0) | 345 (14.6) | 294 (10.9) | 84 (8.9) | |
2 | 64 (1.1) | 26 (1.1) | 30 (1.1) | 8 (0.8) | |
3 | 5 (0.1) | 3 (0.1) | 1 (0.0) | 1 (0.1) | |
4 | 1 (0.0) | 0 (0) | 0 (0) | 1 (0.1) | |
1 vs. 3 <0.001† | |||||
1 vs. 2 <0.001† | |||||
Paternal T1D | 322 (5.4) | 141 (6.0) | 137 (5.1) | 44 (4.6) | 0.204 |
Maternal T1D | 185 (3.1) | 93 (3.9) | 76 (2.8) | 16 (1.7) | 0.002 |
1 vs. 3 0.002 | |||||
1 vs. 2 0.025 | |||||
Sibling with T1D | 340 (5.7) | 160 (6.8) | 137 (5.1) | 43 (4.5) | 0.008 |
1 vs. 3 0.015 | |||||
1 vs. 2 0.010 |
Data are reported as n (%). Spring, March to May; summer, June to August; autumn, September to November; winter, December to February. Analyses were performed using Pearson χ2 test unless otherwise indicated. T1D, type 1 diabetes.
Kruskal-Wallis test.
Mann-Whitney U test.
Clinical Characteristics
The degree of metabolic decompensation at the time of diagnosis was more severe in older children; blood pH was lower, DKA and severe DKA were more common, and β-hydroxybutyrate levels were higher (Table 2). The duration of symptoms prior to diagnosis was also longer in the oldest children, HbA1c levels were higher, and BMI SDS was lower. Blood glucose levels at diagnosis did not differ significantly between the age groups.
Clinical characteristics of the whole study cohort of 6,015 children diagnosed with T1D and by three groups according to age at diagnosis
. | Overall . | Age at diagnosis, years . | P . | ||
---|---|---|---|---|---|
0–6 . | 7–12 . | ≥13 . | |||
BMI SDS | −0.22 ± 1.37 | −0.09 ± 1.23 | −0.20 ± 1.43 | −0.62 ± 1.45 | <0.001* |
1 vs. 3 <0.001† | |||||
1 vs. 2 0.003† | |||||
2 vs. 3 <0.001† | |||||
pH, median (IQR) | 7.38 (0.08) | 7.39 (0.06) | 7.37 (0.10) | 7.36 (0.10) | <0.001‡ |
1 vs. 3 <0.001§ | |||||
1 vs. 2 <0.001§ | |||||
2 vs. 3 <0.001§ | |||||
DKA (pH <7.30) | 1,127 (19.4) | 318 (14.0) | 568 (21.7) | 241 (26.4) | <0.001‡ |
1 vs. 3 <0.001§ | |||||
1 vs. 2 <0.001§ | |||||
2 vs. 3 0.004§ | |||||
Severe DKA (pH <7.10) | 298 (5.1) | 69 (3.0) | 152 (5.8) | 77 (8.4) | <0.001‡ |
1 vs. 3 <0.001§ | |||||
1 vs. 2 <0.001§ | |||||
2 vs. 3 0.006§ | |||||
Plasma glucose, mmol/L, median (IQR) | 23.9 (12.5) | 23.8 (12.7) | 23.8 (12.5) | 24.1 (12.2) | 0.106‡ |
HbA1c, mmol/mol | 94.6 ± 27.6 | 82.3 ± 20.6 | 101.4 ± 27.8 | 104.7 ± 30.2 | <0.001* |
1 vs. 3 <0.001† | |||||
1 vs. 2 <0.001† | |||||
HbA1c, % | 10.8 ± 2.5 | 9.7 ± 1.9 | 11.4 ± 2.5 | 11.7 ± 2.8 | <0.001* |
1 vs. 3 <0.001† | |||||
1 vs. 2 <0.001† | |||||
β-Hydroxybutyrate, mmol/L, median (IQR) | 1.7 (4.4) | 1.3 (3.5) | 2.1 (4.8) | 2.3 (4.9) | <0.001‡ |
1 vs. 3 <0.001§ | |||||
1 vs. 2 <0.001§ | |||||
Duration of symptoms, weeks | <0.001‡ | ||||
No symptoms | 99 (1.6) | 39 (1.7) | 45 (1.7) | 15 (1.6) | |
<1 | 1,184 (19.7) | 550 (23.3) | 481 (17.8) | 153 (16.1) | |
1–4 | 3,140 (52.2) | 1,340 (56.8) | 1,352 (50.0) | 448 (47.3) | |
>4 | 1,079 (17.9) | 250 (10.6) | 581 (21.5) | 248 (26.2) | |
Unknown | 513 (8.5) | 182 (7.7) | 247 (9.1) | 84 (8.9) | |
1 vs. 3 <0.001§ | |||||
1 vs. 2 <0.001§ | |||||
2 vs. 3 0.036§ |
. | Overall . | Age at diagnosis, years . | P . | ||
---|---|---|---|---|---|
0–6 . | 7–12 . | ≥13 . | |||
BMI SDS | −0.22 ± 1.37 | −0.09 ± 1.23 | −0.20 ± 1.43 | −0.62 ± 1.45 | <0.001* |
1 vs. 3 <0.001† | |||||
1 vs. 2 0.003† | |||||
2 vs. 3 <0.001† | |||||
pH, median (IQR) | 7.38 (0.08) | 7.39 (0.06) | 7.37 (0.10) | 7.36 (0.10) | <0.001‡ |
1 vs. 3 <0.001§ | |||||
1 vs. 2 <0.001§ | |||||
2 vs. 3 <0.001§ | |||||
DKA (pH <7.30) | 1,127 (19.4) | 318 (14.0) | 568 (21.7) | 241 (26.4) | <0.001‡ |
1 vs. 3 <0.001§ | |||||
1 vs. 2 <0.001§ | |||||
2 vs. 3 0.004§ | |||||
Severe DKA (pH <7.10) | 298 (5.1) | 69 (3.0) | 152 (5.8) | 77 (8.4) | <0.001‡ |
1 vs. 3 <0.001§ | |||||
1 vs. 2 <0.001§ | |||||
2 vs. 3 0.006§ | |||||
Plasma glucose, mmol/L, median (IQR) | 23.9 (12.5) | 23.8 (12.7) | 23.8 (12.5) | 24.1 (12.2) | 0.106‡ |
HbA1c, mmol/mol | 94.6 ± 27.6 | 82.3 ± 20.6 | 101.4 ± 27.8 | 104.7 ± 30.2 | <0.001* |
1 vs. 3 <0.001† | |||||
1 vs. 2 <0.001† | |||||
HbA1c, % | 10.8 ± 2.5 | 9.7 ± 1.9 | 11.4 ± 2.5 | 11.7 ± 2.8 | <0.001* |
1 vs. 3 <0.001† | |||||
1 vs. 2 <0.001† | |||||
β-Hydroxybutyrate, mmol/L, median (IQR) | 1.7 (4.4) | 1.3 (3.5) | 2.1 (4.8) | 2.3 (4.9) | <0.001‡ |
1 vs. 3 <0.001§ | |||||
1 vs. 2 <0.001§ | |||||
Duration of symptoms, weeks | <0.001‡ | ||||
No symptoms | 99 (1.6) | 39 (1.7) | 45 (1.7) | 15 (1.6) | |
<1 | 1,184 (19.7) | 550 (23.3) | 481 (17.8) | 153 (16.1) | |
1–4 | 3,140 (52.2) | 1,340 (56.8) | 1,352 (50.0) | 448 (47.3) | |
>4 | 1,079 (17.9) | 250 (10.6) | 581 (21.5) | 248 (26.2) | |
Unknown | 513 (8.5) | 182 (7.7) | 247 (9.1) | 84 (8.9) | |
1 vs. 3 <0.001§ | |||||
1 vs. 2 <0.001§ | |||||
2 vs. 3 0.036§ |
Data are reported as mean ± SD or n (%) unless otherwise indicated.
T1D, type 1 diagnosis.
Welch ANOVA.
Independent samples Student t test.
Kruskall-Wallis test.
Mann-Whitney U test.
β-Cell Autoimmunity
In each age group, the most common number of positive autoantibodies at diagnosis was four, detected in approximately one-third of the participants. Only 2.1% of the total study population tested negative for all five autoantibodies at diagnosis, but this was most common in the oldest age group (4.1%) and rarest in the youngest age group (0.9%) (Table 3). Positivity for one and two autoantibodies was also more frequent in the oldest age group, whereas positivity for all five autoantibodies was most common in the youngest age group.
Autoantibodies and HLA-DR/DQ genotype–based risk classification of the whole study cohort of 6,015 children diagnosed with type 1 diabetes and by three groups according to age at diagnosis
. | Overall . | Age at diagnosis, years . | P . | ||
---|---|---|---|---|---|
0–6 . | 7–12 . | ≥13 . | |||
N of positive autoantibodies | <0.001 | ||||
0 | 125 (2.1) | 21 (0.9) | 65 (2.4) | 39 (4.1) | |
1 | 291 (4.8) | 100 (4.2) | 126 (4.7) | 65 (6.9) | |
2 | 671 (11.2) | 249 (10.5) | 304 (11.2) | 118 (12.4) | |
3 | 1,444 (24.0) | 559 (23.7) | 662 (24.5) | 223 (23.5) | |
4 | 1,984 (33.0) | 765 (32.4) | 910 (33.6) | 309 (32.6) | |
5 | 1,500 (24.9) | 667 (28.3) | 639 (23.6) | 194 (20.5) | |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 0.001 | |||||
IAA positive | 3,282 (54.6) | 1,741 (73.7) | 1,209 (44.7) | 332 (35.0) | <0.001 |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 <0.001 | |||||
IAA, RU, median (IQR) | 2.0 (7.3) | 6.1 (17.96) | 1.2 (3.68) | 0.8 (2.3) | <0.001 |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 <0.001 | |||||
GADA positive | 4,002 (66.5) | 1,467 (62.1) | 1,850 (68.4) | 685 (72.3) | <0.001 |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 0.025 | |||||
GADA, RU, median (IQR) | 14.5 (56.2) | 11.4 (49.5) | 16.2 (58.4) | 19.7 (72.8) | <0.001 |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 0.014 | |||||
IA-2A positive | 4,473 (74.4) | 1,726 (73.1) | 2,058 (76.1) | 689 (72.7) | 0.024 |
1 vs. 2 0.024 | |||||
2 vs. 3 0.039 | |||||
IA-2A, RU, median (IQR) | 69.0 (120.7) | 50.8 (116.0) | 81.7 (122.4) | 69.5 (124.4) | <0.001 |
1 vs. 3 0.038 | |||||
1 vs. 2 <0.001 | |||||
ZnT8A positive | 4,167 (69.3) | 1,526 (64.6) | 1,987 (73.4) | 654 (69.0) | <0.001 |
1 vs. 3 0.017 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 0.009 | |||||
ZnT8A, RU, median (IQR) | 3.5 (24.6) | 2.4 (17.2) | 5.0 (30.5) | 3.4 (26.3) | <0.001 |
1 vs. 3 0.002 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 0.006 | |||||
ICA positive | 5,477 (91.1) | 2,210 (93.6) | 2,451 (90.6) | 816 (86.1) | <0.001 |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 <0.001 | |||||
ICA, JDFU, median (IQR) | 49.0 (152.0) | 49.0 (143.0) | 49.0 (152.0) | 49.0 (177.0) | 0.097 |
IAA/GADA-dominant profile* | <0.001 | ||||
IAA | 936 (36.1) | 574 (65.7) | 302 (24.3) | 60 (12.7) | |
GADA | 1,656 (63.9) | 300 (34.3) | 943 (75.7) | 413 (87.3) | |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 <0.001 | |||||
HLA-DR/DQ risk group | <0.001 | ||||
Increased | 4,887 (81.2) | 1,972 (83.5) | 2,189 (80.9) | 726 (76.6) | |
Neutral or protective | 1,128 (18.8) | 389 (16.5) | 517 (19.1) | 222 (23.4) | |
1 vs. 3 <0.001 | |||||
1 vs. 2 0.015 | |||||
2 vs. 3 0.004 |
. | Overall . | Age at diagnosis, years . | P . | ||
---|---|---|---|---|---|
0–6 . | 7–12 . | ≥13 . | |||
N of positive autoantibodies | <0.001 | ||||
0 | 125 (2.1) | 21 (0.9) | 65 (2.4) | 39 (4.1) | |
1 | 291 (4.8) | 100 (4.2) | 126 (4.7) | 65 (6.9) | |
2 | 671 (11.2) | 249 (10.5) | 304 (11.2) | 118 (12.4) | |
3 | 1,444 (24.0) | 559 (23.7) | 662 (24.5) | 223 (23.5) | |
4 | 1,984 (33.0) | 765 (32.4) | 910 (33.6) | 309 (32.6) | |
5 | 1,500 (24.9) | 667 (28.3) | 639 (23.6) | 194 (20.5) | |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 0.001 | |||||
IAA positive | 3,282 (54.6) | 1,741 (73.7) | 1,209 (44.7) | 332 (35.0) | <0.001 |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 <0.001 | |||||
IAA, RU, median (IQR) | 2.0 (7.3) | 6.1 (17.96) | 1.2 (3.68) | 0.8 (2.3) | <0.001 |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 <0.001 | |||||
GADA positive | 4,002 (66.5) | 1,467 (62.1) | 1,850 (68.4) | 685 (72.3) | <0.001 |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 0.025 | |||||
GADA, RU, median (IQR) | 14.5 (56.2) | 11.4 (49.5) | 16.2 (58.4) | 19.7 (72.8) | <0.001 |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 0.014 | |||||
IA-2A positive | 4,473 (74.4) | 1,726 (73.1) | 2,058 (76.1) | 689 (72.7) | 0.024 |
1 vs. 2 0.024 | |||||
2 vs. 3 0.039 | |||||
IA-2A, RU, median (IQR) | 69.0 (120.7) | 50.8 (116.0) | 81.7 (122.4) | 69.5 (124.4) | <0.001 |
1 vs. 3 0.038 | |||||
1 vs. 2 <0.001 | |||||
ZnT8A positive | 4,167 (69.3) | 1,526 (64.6) | 1,987 (73.4) | 654 (69.0) | <0.001 |
1 vs. 3 0.017 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 0.009 | |||||
ZnT8A, RU, median (IQR) | 3.5 (24.6) | 2.4 (17.2) | 5.0 (30.5) | 3.4 (26.3) | <0.001 |
1 vs. 3 0.002 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 0.006 | |||||
ICA positive | 5,477 (91.1) | 2,210 (93.6) | 2,451 (90.6) | 816 (86.1) | <0.001 |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 <0.001 | |||||
ICA, JDFU, median (IQR) | 49.0 (152.0) | 49.0 (143.0) | 49.0 (152.0) | 49.0 (177.0) | 0.097 |
IAA/GADA-dominant profile* | <0.001 | ||||
IAA | 936 (36.1) | 574 (65.7) | 302 (24.3) | 60 (12.7) | |
GADA | 1,656 (63.9) | 300 (34.3) | 943 (75.7) | 413 (87.3) | |
1 vs. 3 <0.001 | |||||
1 vs. 2 <0.001 | |||||
2 vs. 3 <0.001 | |||||
HLA-DR/DQ risk group | <0.001 | ||||
Increased | 4,887 (81.2) | 1,972 (83.5) | 2,189 (80.9) | 726 (76.6) | |
Neutral or protective | 1,128 (18.8) | 389 (16.5) | 517 (19.1) | 222 (23.4) | |
1 vs. 3 <0.001 | |||||
1 vs. 2 0.015 | |||||
2 vs. 3 0.004 |
Data are reported as n (%) unless otherwise indicated. Analyses were performed using the Kruskal-Wallis test and, for pairwise comparisons, the Mann-Whitney U test.
IAA-dominant profile, IAA positivity and GADA negativity at diagnosis; GADA-dominant profile, GADA positivity and IAA negativity at diagnosis.
When looking at the individual autoantibodies, ICA was the most common, with an overall frequency of >90% positivity, with highest positivity (93.6%) in the youngest age group. The most obvious difference between age groups was observed for IAA, with 73.7% positivity in the youngest versus 35.0% positivity in the oldest age group. In contrast, GADA positivity was more common in the oldest age group. IA-2A and ZnT8A positivity and titers were highest in the middle age group.
A total of 2,592 cases could be classified based on their autoantibody profile at diagnosis as representing either the IAA-dominant profile or the GADA-dominant profile; 36.1% had the IAA-dominant profile, and 63.9% had the GADA-dominant profile. The IAA-profile was significantly more common in the youngest age group and the GADA-profile in the oldest age group, with the intermediate age group falling between these two.
HLA-DR/DQ Genotype-Based Risk Classification
A vast majority of cases carried HLA-DR/DQ–conferred susceptibility to type 1 diabetes (81.2%) (Table 3). Increased risk was seen most commonly in the youngest age group (83.5%), whereas neutral or protective HLA-DR/DQ genotypes were more frequent in the oldest age group (23.4%). The distribution of the five HLA risk categories in the three age groups is shown in Supplementary Fig. 1.
Relationship Between IAA, Familial Type 1 Diabetes, and Age at Diagnosis
Age at diagnosis was associated (P < 0.001) with the likelihood of having IAA positivity and/or familial type 1 diabetes rather than having IAA negativity and nonfamilial type 1 diabetes. The likelihood decreased with increasing age at diagnosis (Fig. 1), demonstrating three different patterns: a steep almost-linear decrease up to age 7 years, followed by a modest decrease between age 7 and 12 years, and remaining almost stable for the rest of the age range. These patterns appeared to correspond to the three age groups used in the main analyses. In fact, a majority (77%) of the youngest age group had IAA positivity and/or familial type 1 diabetes, with the proportion decreasing to 51% in the middle age group and further to 40% in the oldest age group. Moreover, when further splitting by the presence of IAA positivity and familial type 1 diabetes, we observed similar patterns for the likelihood of having both of these (P < 0.001 for nonlinearity; P < 0.001 for association) or having IAA positivity and nonfamilial type 1 diabetes (P < 0.001 for nonlinearity, P < 0.001 for association) rather than having IAA negativity and nonfamilial type 1 diabetes and a linear relationship (P = 0.529 for nonlinearity; P = 0.002 for linear association) with negative slope (odds ratio 0.94; 95% CI 0.92–0.98) for the likelihood of having IAA negativity but familial type 1 diabetes rather than having IAA negativity and nonfamilial type 1 diabetes (Supplementary Figs. 2–4).
The odds of having IAA and/or familial type 1 diabetes rather than having IAA negativity and nonfamilial type 1 diabetes in relation to the age at diagnosis, when estimated using binary logistic regression analysis with the restricted cubic spline function of age as the explanatory variable (P < 0.001 for nonlinearity; P < 0.001 for the overall association). The gray area corresponds to the 95% confidence band, and the two vertical dashed lines are set to 7 and 13 years to demonstrate the division into the three age groups used in the main analyses.
The odds of having IAA and/or familial type 1 diabetes rather than having IAA negativity and nonfamilial type 1 diabetes in relation to the age at diagnosis, when estimated using binary logistic regression analysis with the restricted cubic spline function of age as the explanatory variable (P < 0.001 for nonlinearity; P < 0.001 for the overall association). The gray area corresponds to the 95% confidence band, and the two vertical dashed lines are set to 7 and 13 years to demonstrate the division into the three age groups used in the main analyses.
Conclusions
Increased awareness of the heterogeneity of diseases and the ensuing need for personalized therapy has led to an increasing interest in the endotype concept. This concept has been developed particularly in asthma research, in which it is seen as closely related to the treatment response (18). Although treatment response was not considered as part of the definition of endotype in this study, an increased understanding of the different pathophysiological pathways of type 1 diabetes would provide the possibility of precision medicine approaches to therapy, in particular for prevention and intervention.
For the purposes of our study, we analyzed the data in predetermined distinct groups according to age at diagnosis, based on the evidence previously presented (4–6). As was expected and as can be seen from our results, the studied characteristics were not distributed between the age groups in a discrete fashion but rather existed on a somewhat continuous spectrum. The heterogeneity of type 1 diabetes is in itself a continuum, adding to the challenge of finding distinct disease endotypes. Our findings do, however, emphasize the association between the heterogeneity of type 1 diabetes and age at diagnosis, and the age groups used appear justified when looking at the relationship between the presence of IAA and/or familial diabetes and age.
Almost all of the characteristics studied in this population of 6,015 children and adolescents showed statistically significant differences in their distribution among the three age groups. The exceptions were seasonality, paternal type 1 diabetes, plasma glucose, and ICA titer. As hypothesized, for most of the characteristics, the differences were significant between the youngest and oldest age groups, in particular, while the middle age group had intermediate values. IA-2A and ZnT8A positivity and ZnT8A titer were the only exceptions, because they were observed to be most common/highest in the middle age group. Our findings support the existence of disease endotypes related to the age groups analyzed and the underlying differing immunopathological processes.
In this nationwide cohort from Finland, type 1 diabetes diagnosed before age 7 years was characterized by a stronger familial clustering and HLA-DR/DQ–conferred disease susceptibility, shorter duration of symptoms, and less severe metabolic decompensation at diagnosis. The higher frequency of IAA and higher IAA titers in the youngest age group are in line with the hyperimmune CD20Hi pattern of insulitis. Conversely, those developing the disease later on seemed to have a stronger male preponderance, weaker familial clustering and HLA class II–conferred genetic risk, longer duration of symptoms before diagnosis, more severe metabolic decompensation, and less autoimmune reactivity, characterized, however, by GADA predominance. The increased frequency of IAA in the youngest age group and the higher prevalence of GADA in the oldest age group are in line with earlier studies in children with preclinical disease (1,2). From this point of view, it is not surprising that the IAA-dominant profile was more than five times more common in the youngest age group when compared with the oldest age group, while the GADA-dominant profile was 2.5 times more frequent in the oldest age group in comparison with the youngest one.
Many of the characteristics clustering in either the youngest or the oldest age group can be considered to be logically interrelated. For example, genetic susceptibility and familial disease are intuitively related, and familial disease in turn increases disease awareness, which affects delay in seeking medical attention and, thus, clinical state at diagnosis. Clinical characteristics are also interconnected. Differences in BMI SDS can, for example, be seen to be related to the extent of dehydration associated with metabolic decompensation or that to weight loss associated with a longer duration of symptoms preceding diagnosis.
Age-related distributions of various characteristics of type 1 diabetes have been previously reported, with largely similar age trends, but some distinct differences. The male preponderance demonstrated in our study is in line with previous reports of male excess in high-incidence countries (19,20). However, we found this male preponderance to be surprisingly accentuated in the oldest age group, with a male-to-female ratio of ∼2:1. Male excess has previously been observed to be more common in those diagnosed beyond age 15 years, with a male-to-female ratio of ∼3:2 (20,21), but to our knowledge, the extent of male preponderance observed in the current study is stronger than previously reported. Earlier evidence has also suggested seasonal variation in the diagnosis of type 1 diabetes. Results vary, but many studies report higher incidence during the colder seasons of winter and autumn and lower incidence during spring and summer, with the seasonal variation being more pronounced in older children (22–24). We observed a similar pattern of seasonality to that previously reported, but no significant difference between the age groups. We did, however, observe an age-related difference in the frequency of familial type 1 diabetes, with a positive family history being more prevalent in the youngest age group. Prior reports on the relationship between age at diagnosis and the distribution of familial and sporadic disease are controversial, with some reporting a trend similar toward the one we observed (25,26) and others failing to find age-related differences (7,27,28).
Interestingly, while most studies have previously demonstrated younger children to be at higher risk for DKA at diagnosis (29), we observed DKA to be most common in the oldest age group. This may seem somewhat contradictory, considering the general notion that early disease manifestation is a sign of a more aggressive disease process, and vice versa. However, the extent of metabolic decompensation at diagnosis is likely to be largely affected by social factors affecting diagnostic delay. The higher HbA1c and longer duration of symptoms observed in older children suggest a delay in seeking medical attention, which likely explains the increased risk of DKA. The extent of metabolic decompensation may thus not be an apt indicator of the underlying disease process. Disease manifestation at an early age can, however, in itself be seen to be indicative of aggressive disease progression, and a more aggressive autoimmune process in young children is also supported by the higher number of diabetes-related autoantibodies present at diagnosis in this age group. Aggressive early-onset disease type has also been linked to carriage of high-risk HLA class II alleles. The high-risk group within the HLA-DR/DQ genotype–based risk classification we used consisted of those with the heterogeneous DR4-DQ8/DR3-DQ2 diplotype (16). Our observation of the higher prevalence of this high-risk genotype in the youngest children is in agreement with previous studies (30,31), including a recent study by Inshaw et al. (32), which was based on the same rationale and ensuing age categories. Inshaw et al. also discovered that the strongly protective haplotypes DRB1*15:01-DQB1*06:02 and DRB1*07:01-DQB1*03:03 were less common in those diagnosed at age <7 years compared with ≥13 years (32). These same haplotypes are associated with strong protection in the risk classification we used, and therefore, the greater prevalence of protective genotypes we observed in the oldest children is in agreement with this finding.
Previous studies have explored various characteristics of type 1 diabetes in different age groups, but to our knowledge, no studies exist with such an extensive population or wide scope of characteristics. Finland has the world’s highest incidence of type 1 diabetes (33). Therefore, the nationwide FPDR with high coverage provides a large study population, allowing the generation of robust data with the possibility of clarifying some previously controversial issues. The considerable homogeneity of the Finnish population is nonetheless a factor to take into account when generalizing the results. In addition, because of the pediatric nature of the register, it does not include all adolescents diagnosed beyond age 15 years, resulting in the smaller size of the oldest age group. The register also limits us to analyzing data recorded at the time of diagnosis. Accordingly, we are unable to directly study certain characteristics that might be of interest because of their apparent age relatedness, such as type of autoantibody at initial seroconversion and subsequent delay in diagnosis or postdiagnosis remission, glycemic control, and comorbidity. However, the possibility of stratifying patients based on characteristics at diagnosis, an easily identifiable stage of disease, provides an undeniable advantage, supporting the relevance of this study, despite its cross-sectional design.
Despite including comprehensive data from the time of diagnosis, the FPDR does have some additional limitations. The analysis of genetic disease susceptibility available for all participating children is limited to the class II HLA-DR/DQ genotypes, and accordingly, other predisposing genetic factors cannot be considered. Moreover, the register does not include data on serum C-peptide or proinsulin levels. With our large study population and samples taken at diagnosis, it would have been interesting to test whether we could confirm the findings of Leete et al. (5) (i.e., lower C-peptide levels and higher proinsulin-to–C-peptide ratio in children diagnosed at age <7 years and vice versa in those diagnosed at age ≥13 years). Because the classification of types of diabetes is challenging, particularly in very young children and in adolescents (34), we cannot exclude misclassification. We did, however, attempt to minimize this possibility for those diagnosed at age <1 year by excluding those without HLA risk genotype or obvious autoimmune activity. Furthermore, there may be some inaccuracy in the clinical data recorded. These data should ideally represent the pretreatment state, but this may not always be the case. Laboratory values are in principle based on analyses performed in the hospital laboratory, but in some cases, values measured by bedside devices may have been recorded instead. Weight and height data may have sometimes be based on the last known measurement rather than measurement upon arrival to the hospital, or they may have been measured later during hospitalization. Records on the duration of symptoms prior to diagnosis are also open to error. However, in all likelihood, these potential errors would be equally distributed among the study population and would accordingly not affect the results of the comparisons between age groups.
Our study confirms that there are definite age-related differences in demographic, clinical, autoimmune, and genetic characteristics at the time of the diagnosis of type 1 diabetes in children and adolescents. Our observations support the existence of distinct age-related endotypes of type 1 diabetes characterized by differing immunopathological processes. We believe that the differences in disease mechanisms in those diagnosed before age 7 years compared with those diagnosed at age ≥13 years should be further explored. A better understanding of the mechanisms behind disease heterogeneity will help in the search for effective individualized therapies. Future intervention trials might consider these age groups for stratification of participants.
This article contains supplementary material online at https://doi.org/10.2337/figshare.18857753.
A list of the investigators of the Finnish Pediatric Diabetes Register can be found in the supplementary material online.
Article Information
Acknowledgments. The authors thank all the participants, hospitals, and personnel of the Finnish Pediatric Diabetes Register.
Funding. The study was supported by the Academy of Finland (decision 292538), the Sigrid Jusélius Foundation, Finska Läkaresällskapet, and the Liv and Hälsa Fund.
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. A.P. conceived the study idea and wrote the first manuscript version. A.P. and A.B. planned and performed the statistical analyses. A.P. and M.K. designed the study. T.H. was in charge of the autoantibody assays. J.I. was responsible for the HLA genotyping and the classification into HLA risk groups. All authors reviewed and approved the manuscript. M.K. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.