Subclasses of Glucose Trajectories in Early Childhood Stratified the Risk of Abnormal Glucose Tolerance in Adolescence and Young Adulthood

Abstract

Early-life exposures may shape long-term effects on glucose regulation. This study aimed to stratify long-term abnormal glucose tolerance (AGT) risk from early childhood. A total of 906 children were enrolled at baseline and reevaluated in adolescence and young adulthood. By using the latent class trajectory analysis, glucose trajectories of children were measured via five–time point oral glucose tolerance tests and then grouped into three latent subclasses: mild excursion–normal reversion (MN), moderate excursion–delayed reversion (MD), and severe excursion–delayed reversion (SD). Logistic regression was performed to estimate the risk of AGT and associations between cardiometabolic factors and subclasses. In adolescence, compared with the MN subclass, the risk of AGT was 1.7-fold in the MD subclass and 5.5-fold in the SD subclass, after adjusting for age, sex, BMI, and Tanner stage. In young adulthood, the adjusted risk of AGT was 3.6-fold and 11.6-fold in the MD and SD subclasses, respectively. During the full natural history of glucose tolerance, the risk of AGT was 3.6-fold in the MD subclass and 18.1-fold in the SD subclass, after adjusting for childhood covariates. MD and SD subclass membership was strongly associated with childhood hypertension, maternal gestational diabetes, and maternal hypertension during pregnancy. Glucose trajectory subclasses in early childhood effectively stratified the long-term risk of AGT. The association between maternal cardiometabolic health and childhood subclass membership highlighted that prenatal exposures may influence metabolic outcomes in offspring.

Article Highlights

• Abnormal glucose tolerance (AGT) in youth has become an alarming global public health issue; however, approaches to identify high-risk population among young people have not been well-established.

• Can the long-term risk of AGT be stratified by the subclasses of glucose trajectories defined in childhood?

• Subclasses defined in childhood can efficiently stratify the risk of AGT in adolescence and young adulthood. The subclass membership was strongly associated with cardiometabolic disorders in childhood and maternal cardiometabolic disorders during pregnancy.

• This subclass method provides a potential strategy to identify those at risk of later cardiometabolic disorders from childhood for more intensive evaluation of intervention. The close relationship between maternal cardiometabolic disorders and subclass membership of children highlighted the potential influence of gestational cardiometabolic health on the development of cardiometabolic disorders in offspring.

Graphical Abstract

Introduction

Abnormal glucose tolerance (AGT), including both prediabetes and diabetes, significantly contributes to the burden of noncommunicable diseases in young people (1,2). The prevalence of prediabetes in adolescents tripled from 11.5% in 1999–2002 to 36.3% in 2015–2020 based on the National Health and Nutrition Examination Survey in the U.S. (3). According to a large population-based cohort study in Hong Kong, in people <20 years old, the age-standardized incidence of type 2 diabetes increased from 4.6 to 7.5 per 100,000 person-years in boys and from 5.9 to 8.5 per 100,000 person-years in girls from 2005 to 2015 (4). Beyond the rapid increase in cases, young-onset diabetes (YOD) is considered to have more aggressive disease course and severe diabetes complications than adult-onset diabetes (5–7).

Despite the growing burden of YOD, effective methods for early identification of children at long-term risk for AGT during adolescence and young adulthood remain insufficiently investigated. Current screening approaches for AGT are primarily tailored for adults (8), which may overlook changes of glucose dysregulation occurring in early childhood. Investigating distinct glucose trajectory patterns in childhood can offer a promising approach to stratify long-term AGT risk and provide potential suggestions for earlier targeted interventions.

To address this critical question, we measured the glucose records of participants, from childhood to young adulthood, in a prospective cohort study. We classified participants into three subclasses by their glucose trajectory in childhood and aimed to test the hypothesis that long-term risk of AGT in adolescence and young adulthood could be stratified by the subclasses defined in childhood.

Research Design and Methods

Data Source and Participants

In this prospective cohort study, offspring of mothers who participated in the Hyperglycemia and Adverse Pregnancy Outcomes (HAPO) study (9), from the Hong Kong center, were invited to participate in the baseline assessment during their childhood between 2009 and 2013. They were reevaluated in adolescence and young adulthood between 2013 and 2016 and between 2020 and 2023 (around 4 and 11 years after the baseline childhood assessment) (10,11). Participants who did not have five–time point glucose records to depict glucose trajectory were excluded from the childhood assessment. Details of the childhood assessment and follow-up assessments are shown in Supplementary Fig. 1. Informed written consent was obtained from the parents of children involved in the study, except for offspring who reached age 18 years at the time of follow-up assessment, in which case the consent was obtained from the offspring. Ethical approval for this study was obtained from the Joint Chinese University of Hong Kong-New Territories East Cluster Clinical Research Ethics Committee (Hong Kong, China). The study was reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) for cohort studies.

Definitions and Equations

Children and adolescents underwent a 2-h oral glucose tolerance test (OGTT) after at least 8 h of fasting, with a glucose load of 1.75 g/kg body weight (maximum 75 g). A 75-g OGTT was performed in participants who underwent reevaluation in young adulthood. Blood samples drawn during OGTT were tested for plasma glucose, plasma insulin, and lipid profile measurements. We defined glucose tolerance according to recommendations of the American Diabetes Association guideline (12): 1) normal glucose tolerance (NGT): fasting plasma blood glucose (FPG) <5.6 mmol/L and 2-h plasma glucose (PG) <7.8 mmol/L; 2) isolated impaired fasting glucose (i-IFG): FPG 5.6–6.9 mmol/L and 2-h PG <7.8 mmol/L; 3) isolated impaired glucose tolerance (i-IGT): FPG <5.6 mmol/L and 2-h PG 7.8–11.0 mmol/L; 4) combination of IFG and IGT (IFG + IGT): FPG 5.6–6.9 mmol/L and 2-h PG 7.8–11.0 mmol/L; and 5) diabetes: FPG ≥7.0 mmol/L or 2-h PG ≥11.1 mmol/L. AGT was defined as one of the following conditions: 1) i-IFG; 2) i-IGT; 3) IFG + IGT; or 4) diabetes.

BMI was calculated by dividing weight in kilograms by the square of height in meters. To define overweight or obesity in children and adolescents under 18 years old, we used the age- and sex-specific BMI cutoffs, aligned with an international standard definition survey, which included Hong Kong participants in the sample (13). For participants who underwent reevaluation in young adulthood (aged ≥18 years), a BMI cutoff ≥23 kg/m² was considered as overweight or obesity. Sum of skinfold was calculated as the total skinfold thickness at biceps, triceps, subscapular, and suprailiac sites. High total cholesterol, high triglycerides, low HDL cholesterol, high LDL cholesterol, and dyslipidemia were defined by lipid profiles according to the American College of Cardiology/American Heart Association guideline, including specific cutoffs for children, adolescents, and adults (14). Hypertension was defined according to the 95th percentile of age-, sex-, and height-specific cutoffs of blood pressure (BP) in children and adolescents (7). The cutoff of systolic BP ≥140 mm Hg and/or diastolic BP ≥90 mm Hg was used in young adults of age ≥18 years to define hypertension (15). Tanner staging was performed by trained individuals using breast and areolar development for girls and testicular volume (Prader orchidometer) for boys (11).

Gestational diabetes was diagnosed according to the recommendations of International Association of Diabetes and Pregnancy Study Groups (16). In brief, after fasting for at least 8 h, women underwent a 75-g OGTT at 24–32 weeks of gestation. FPG ≥5.1 mmol/L and/or 1-h PG ≥10.0 mmol/L and/or 2-h PG ≥8.5 mmol/L was defined as gestational diabetes. Women were classified as gaining weight “below,” “within,” or “excessive” in relation to recommendations, according to their prepregnancy BMI category, based on the 2009 Institute of Medicine guidelines for healthy pregnancy weight gain (17). Details of standardized gestational weight gain calculation in this study are provided in our previous publication (18). Hypertension during pregnancy included chronic hypertension, preeclampsia or eclampsia, and gestational hypertension (19,20).

Latent Class Trajectory Models

A latent class trajectory model is a statistical approach used to identify distinct subgroups (or latent classes) of individuals who follow similar trajectories over time in longitudinal data (21). Using the “lcmm” package in R (21), we applied the latent class trajectory analysis to classify participants into subclasses according to their glucose trajectories in early childhood, based on five–time point glucose measurements (0, 15, 30, 60, and 120 min) during the OGTT.

As recommended in the literature (22), various latent class trajectory models were fitted and compared before the final model was determined. We considered three possible polynomial specifications—linear, quadratic and cubic—to allow for curvilinear developmental patterns of the glucose trajectory. Each of these polynomial models was modeled as a class solution, starting with a standard one-class heterogenous mixed model. The Bayesian Information Criterion (BIC) values were used to make comparisons between three polynomial models. Once we identified the best fitted quadratic polynomial model (lowest BIC values compared with the linear and cubic polynomial specifications), we progressed to increase the number of latent classes to explore the optimal number of subclasses.

We retrieved the following indices to evaluate the goodness-of-fit and subclass discrimination: BIC, distribution of subclass membership probabilities, and interpretability of the identified patterns (23). We increased the number of latent subclasses from one to five to test the possible low BIC methods. Both the three-subclass method and four-subclass method had low but similar BIC values (15,931.41 vs. 15,954.71). We further tested the cumulative distribution of posterior probability in both methods (Supplementary Fig. 2), with a lower value observed in the three-subclass method. We then tested the mean of posterior probabilities of each class in the three-subclass method (Supplementary Table 1). The mean value of posterior probabilities of three subclasses ranged from 0.779 to 0.844 (mean probability between 0.8 and 1.0 indicates a very good classification). For further validation, we performed fivefold cross-validation tests to evaluate the posterior probabilities of the three-subclass method. The mean of posterior probabilities of each subclass was higher than 0.8 in both the training data set and testing data set (Supplementary Table 2).

We finally used a three-class latent trajectory model with a quadratic polynomial specification according to the criteria above. In detail, participants were classified into mild excursion-normal reversion (MN) subclass, moderate excursion-delayed reversion (MD) subclass, and severe excursion-delayed reversion (SD) subclass (Fig. 1).

Figure 1.

Glucose trajectories of three subclasses among 906 children. The solid lines represented the glucose trajectories estimated by the latent class trajectory analyses. The shaded areas represented the 95% CIs of glucose trajectories.

Cardiometabolic Risk Factors and the Subclass Membership

After identifying the three glucose trajectories, we further explored the cardiometabolic risk factors that affected the subclass membership of children. These risk factors included the nonglucose cardiometabolic disorders in childhood (overweight or obesity, hypertension, and dyslipidemia) and maternal cardiometabolic disorders during pregnancy (gestational diabetes, excessive gestational weight gain, and hypertension). We combined the MD and SD subclasses as the at-risk group, because of their higher risk of AGT, and considered the MN subclass as the reference group. Logistic regression was used to evaluate the association between each cardiometabolic risk factor and the dichotomized subclass membership.

Statistical Analyses

Baseline characteristics were presented as mean ± SD, median and interquartile range, or n and percent, as appropriate. One-way ANOVA tests (normally distributed data) or Kruskal-Wallis tests (nonparametric analyses) were used to compare differences among subclasses, as appropriate. χ² tests or Fisher exact tests were used for comparisons between categorical variables. Logistic regression was used to estimate the odds ratios (ORs) and 95% CIs for dichotomous outcomes to evaluate the risk of AGT. Statistical significance was defined as a two-sided P value <0.05, unless otherwise specified. All statistical analyses were performed using R version 4.2.2 software (https://www.r-project.org/).

Data and Resource Availability

Data are available from the corresponding author upon reasonable request.

Results

Characteristics of Participants in Childhood

A total of 906 participants in childhood, aged 7.0 ± 0.4 years, underwent the baseline childhood assessment. Children were classified into three subclasses according to their glucose trajectories during the OGTTs. The majority of children (60.3%) were classified into the MN subclass, and 35.9% of the children were stratified into the MD subclass, with 3.9% of the children classified into the SD subclass (Table 1). Children in the MN subclass had both the lowest fasting glucose and lowest glucose excursion during the whole OGTT period, while children in the SD subclass had both the highest fasting glucose and highest glucose excursion (Fig. 1). Children in the MD subclass had a modest level of fasting glucose and glucose excursion and a relative delayed reversion compared with the MN subclass. Children in the SD subclass had a delayed glucose reversion compared with those in the MN and MD subclasses.

Table 1.

Characteristics of participants in childhood

	MN subclass	MD subclass	SD subclass	P value
n (%)	546 (60.3)	325 (35.9)	35 (3.9)	—
Age, years	7.0 ± 0.4	6.9 ± 0.5	7.0 ± 0.5	0.324
Male, n (%)	275 (50.4)	178 (54.8)	14 (40.0)	0.172
BMI, kg/m2	15.04 ± 2.14	15.11 ± 2.56	14.87 ± 1.86	0.799
Overweight/obesity, n (%)	69 (12.6)	45 (13.8)	3 (8.6)	0.646
Sum of the skin fold, mm	34.98 ± 15.29	37.51 ± 19.16	35.77 ± 15.39	0.106
Sum of the skin fold, ≥85th percentiles, n (%)	78 (14.4)	51 (16.2)	4 (11.4)	0.660
GLU 0 min, mmol/L	4.54 ± 0.30	4.62 ± 0.41	4.77 ± 0.76	<0.001
GLU 15 min, mmol/L	6.65 ± 0.97	7.49 ± 1.14	8.80 ± 1.19	<0.001
GLU 30 min, mmol/L	6.77 ± 1.08	8.67 ± 0.97	10.55 ± 0.91	<0.001
GLU 60 min, mmol/L	5.06 ± 0.93	6.99 ± 1.05	9.75 ± 1.12	<0.001
GLU 120 min, mmol/L	4.94 ± 0.84	5.80 ± 0.79	6.04 ± 1.16	<0.001
AGT, n (%)	1 (0.2)	12 (3.7)	5 (14.3)	<0.001
Mean SBP, mmHg	101 [96, 106]	103 [95, 109]	102 [97, 106]	0.370
Mean DBP, mmHg	61 [56, 66]	62 [57, 67]	59 [56, 66]	0.428
Hypertension, n (%)	30 (5.5)	36 (11.1)	2 (5.7)	0.009
TC, mmol/L	4.41 ± 0.76	4.58 ± 0.69	4.52 ± 0.71	0.004
High TC, n (%)	101 (18.6)	72 (22.3)	8 (22.9)	0.388
TG, mmol/L	0.70 [0.50, 0.90]	0.60 [0.50, 0.80]	0.70 [0.55, 0.85]	0.001
High TG, n (%)	87 (16.0)	36 (11.1)	5 (14.3)	0.139
HDL-c, mmol/L	1.64 ± 0.34	1.69 ± 0.35	1.69 ± 0.38	0.081
Low HDL-c, n (%)	6 (1.1)	0 (0.0)	1 (2.9)	0.072
LDL-c, mmol/L	2.42 ± 0.65	2.58 ± 0.62	2.52 ± 0.63	0.001
High LDL-c, n (%)	40 (7.4)	40 (12.4)	2 (5.7)	0.036
Dyslipidemia, n (%)	169 (31.1)	95 (29.4)	10 (28.6)	0.844

Descriptive statistics are presented as mean ± SD, median [interquartile range], or number (%). Cutoffs of dyslipidemia in children: high TC ≥5.1 mmol/L; high TG ≥1.1 mmol/L; low HDL ≤1.0 mmol/L; high LDL ≥3.4 mmol/L. DBP, diastolic BP; GLU, serum glucose; HDL-c, HDL cholesterol; LDL-c, LDL cholesterol; SBP, systolic BP; TC, total cholesterol; TG, triglycerides.

There was no statistically significant difference in age, sex, BMI, and sum of skinfold among children in the three subclasses (Table 1). The proportion of AGT was nearly four times higher among children in the SD subclass compared with children in the MD subclass (14.3% vs. 3.7%), but it was only 0.2% in the MN subclass. Compared with the MN subclass, children in the MD and SD subclasses had higher proportion of hypertension (5.5% vs. 11.1% vs. 5.7%, P = 0.009). In general, children in the MD and SD subclasses had higher lipid levels than those in the MN subclass (P < 0.05 in total cholesterol, triglycerides, and LDL cholesterol), although there was no statistical difference in the proportion of dyslipidemia among subclasses (P = 0.844).

Stratified Risk of AGT Among Subclasses in Adolescence and Early Adulthood Stages

We reevaluated the children when they were in adolescence and in young adulthood. A total of 519 children had follow-up glucose records in adolescence, and a total of 400 children had records in young adulthood (Supplementary Fig. 1A). There were no significant differences in childhood characteristics between participants with and without follow-up records (Supplementary Table 3). The longitudinal patterns of baseline characteristics remained largely consistent from childhood through adolescence and young adulthood (Table 1 and Supplementary Table 4). Notably, while hypertension and few components of lipid profiles were statistically different among subclasses in childhood (Table 1), these differences were not statistically significant in adolescence or young adulthood (Supplementary Table 4).

The risk of AGT in children of the MD subclass and SD subclass was higher than for those of the MN subclass in all models (Table 2). In adolescence, compared with counterparts in the MN subclass, ORs of AGT for children in the MD subclass was 1.71 (95% CI 1.00–2.92) and was up to 5.51 (95% CI 2.01–14.39) for those in the SD subclass, after adjusting for age, sex, BMI, and Tanner stage. In young adulthood, the ORs of AGT for children in the MD and SD subclasses were 3.63 (95% CI 1.63–8.51) and 11.55 (95% CI 3.04–41.42) after adjusting for age, sex, and BMI.

Table 2.

Stratified risk of AGT in adolescence and early adulthood

Models	MN subclass (OR, 95% CI)	MD subclass (OR, 95% CI)	SD subclass (OR, 95% CI)
Adolescence
n	313	184	22
Model 1	Ref (1.00)	1.78 (1.05, 3.02)	4.80 (1.78, 12.26)
Model 2	Ref (1.00)	1.75 (1.03, 2.98)	5.49 (2.01, 14.35)
Model 3	Ref (1.00)	1.71 (1.00, 2.92)	5.51 (2.01, 14.39)
Early adulthood
n	249	137	14
Model 1	Ref (1.00)	3.62 (1.65, 8.37)	13.28 (3.56, 46.65)
Model 2	Ref (1.00)	3.80 (1.72, 8.88)	11.78 (3.11, 42.00)
Model 3	Ref (1.00)	3.63 (1.63, 8.51)	11.55 (3.04, 41.42)

Model 1: Crude model with three subclasses only, participants in the adolescence stage were additionally adjusted for Tanner stage; model 2: model 1 + age + sex; model 3: model 2 + BMI. Age, sex, BMI, and Tanner stage measured in adolescence were adjusted for models of adolescence. Age, sex, and BMI measured in young adulthood were adjusted for models of young adulthood.

Natural History of Glucose Tolerance in Childhood, Adolescence, and Early Adulthood

A total of 308 participants had full glucose tolerance records throughout childhood, adolescence, and young adulthood (Supplementary Fig. 1B), among which 303 children were NGT and 5 children were AGT at baseline (98.4% vs. 1.6%).

We classified the 303 NGT children into two groups based on their glucose tolerance status: children who remained NGT in both adolescence and early adulthood (NAEA, n = 243) (Supplementary Table 5) and children who developed AGT in adolescence and/or early adulthood (AAEA, n = 60). Only one participant from the AAEA group incurred diabetes in young adulthood. Participants in the NAEA group were more likely to be classified into the MN subclass compared with the MD and SD subclasses when they were in childhood (68.7% vs. 29.6% vs. 1.6%, P < 0.001). Participants in the AAEA group were more likely to be classified in the MD subclass and SD subclass compared with the NAEA group (51.7% vs. 29.6% for the MD subclass, 13.3% vs. 1.6% for the SD subclass, P = 0.021).

Among children with NGT in childhood, the proportion of AAEA increased from the MN subclass to the MD subclass and further increased in the SD subclass (11.2% vs. 30.1% vs. 66.7%, P < 0.001) (Table 3). The risk of AAEA was higher among children classified into the MD subclass and SD subclass than for those classified in the MN subclass. Compared with children in the MN subclass, OR of AAEA was 3.56 (95% CI 1.90–6.81) for children in the MD subclass and further increased to 18.11 (95% CI 5.12–74.71) for those in the SD subclass, after adjusting for age, sex, BMI, and Tanner stage.

Table 3.

Risk of AAEA stratified by novel subclasses among children with NGT at baseline

Models	MN subclass	MD subclass	SD subclass
n	188	103	12
AAEA, n (%)	21 (11.2)	31 (30.1%)	8 (66.7)
Model 1, OR (95% CI)	Ref (1.00)	3.61 (1.94, 6.87)	17.66 (5.00, 72.94)
Model 2, OR (95% CI)	Ref (1.00)	3.61 (1.93, 6.89)	17.69 (5.00, 73.15)
Model 3, OR (95% CI)	Ref (1.00)	3.56 (1.90, 6.81)	18.11 (5.12, 74.71)

Model 1: three subclasses + Tanner stage; model 2: model 1 + age (childhood) + sex; model 3: model 2 + BMI (childhood). Subclasses were defined in childhood stage by estimated glucose trajectories using the latent class trajectory analyses.

Cardiometabolic Risk Factors Contributed to the Membership of Subclasses

Among cardiometabolic risk factors in childhood, hypertension was the main one associated with the MD or SD membership (OR 2.03, 95% CI 1.24–3.36) (Supplementary Table 6). Overweight or obesity and dyslipidemia were not statistically associated with the subclass membership (Supplementary Table 6). Mothers of children in the MN and the MD or SD subclasses had comparable characteristics in terms of age at delivery, parity, and percentage of weight gain during pregnancy (Supplementary Table 7). However, a higher proportion of mothers in the MD or SD subclass experienced gestational diabetes (19.2% vs. 13.6%, P = 0.030) and hypertension (10.3% vs. 7.5%, P = 0.183). Compared with mothers of children in the MN subclass, those of children in the MD or SD subclass had a 51% higher risk of gestational diabetes (OR_{gestational diabetes} 1.51, 95% CI 1.06–2.17) (Supplementary Table 6) and a 68% higher risk of hypertension during pregnancy (OR_{hypertension during pregnancy} 1.68, 95% CI 1.02–2.77). Excessive gestational weight gain was not associated with the subclass membership.

Discussion

Our results highlighted that the subclasses, defined by glucose trajectories during OGTT in early childhood, effectively stratified the risk of AGT in adolescence and in young adulthood. Additionally, the subclass method accurately stratified the risk of AAEA over the life course trajectory from childhood to young adulthood. Our findings also demonstrated that membership in the MD or SD subclass was strongly associated with hypertension in childhood and maternal cardiometabolic risk factors during pregnancy.

Stratifying the long-term risk of AGT at an early stage could uniquely contribute to early detection, effective interventions, and potential reduction of complications in YOD (6,24–27). Individuals with YOD (aged 10–19 years) have been estimated to have an accelerated rate of β-cell function loss at 25–30% per year versus 7% in those with later-onset type 2 diabetes, and the loss of β-cell function in youth did not revert or differ by treatments (24). In addition, among youths diagnosed with YOD (average age 14.0 ± 2.0 years at enrollment), over 60% of the participants developed at least one diabetes-related complication by young adulthood (mean age 26.4 ± 2.8 years) (6,25). The heavy burden of more rapid β-cell dysfunction and related complications among individuals with YOD could be foreseeably prevented or delayed, provided that high-risk individuals are identified in a timely manner. Supporting this possibility, the subclass method in our study demonstrated the ability to identify high-risk individuals at both an early age and an earlier stage of dysglycemia. Notably, the majority of children in the AAEA group were classified as having prediabetes, with only a single participant progressing to a diagnosis of diabetes in young adulthood, which highlights the potential of this method to identify at-risk individuals in a timely manner. Furthermore, our efficient subclassification method could inform the frequency of screening for earlier detection of YOD. Children identified as low risk may not require immediate intervention, allowing for more cost-effective management. In contrast, those classified as high risk could benefit from early education and personalized follow-up strategies aimed at mitigating the progression of dysglycemia. By enabling early risk stratification, this subclass-based approach provides a proactive framework to identify individuals who could benefit from timely interventions to delay or prevent YOD.

Our findings also underscore the importance of early-life cardiometabolic risk factors and the utility of glucose trajectory subclasses. It was interesting that, despite the absence of statistically significant differences in the individuals with overweight or obesity among the subclasses from childhood through young adulthood (Table 1 and Supplementary Table 4), our subclassification method effectively stratified the risk of AGT. This finding is supported by previous research suggesting that the single phenotypic BMI may not be a strong predictor for the evaluation of the long-term risk of diabetes, particularly in cases where there is a genetic-phenotypic discordance of BMI (28). Even in people with normal body weight, those classified into the discordantly high group faced a higher risk of type 2 diabetes than those who were overweight (28). In addition, we also observed the differences in LDL cholesterol and triglyceride levels across subclasses in early childhood. Although the values of these lipid profiles did not exceed diagnostic thresholds, they may still contribute to the cardiometabolic risk stratification. Furthermore, our study demonstrated that hypertension was associated with MD or SD subclass membership in childhood (Supplementary Table 6), but no significant differences in hypertension prevalence were observed among subclasses during adolescence and young adulthood (Supplementary Table 4). This discrepancy may be attributed to several factors. First, physiological changes during growth, such as pubertal development and hormonal fluctuations, may obscure early cardiometabolic patterns, making subclass differences less detectable in later stages (29). Second, lifestyle modifications over time—such as improvements in diet and physical activity—may mitigate initial risk differentials observed in childhood (30). Third, loss to follow-up or reduced sample sizes in later assessments could decrease the statistical power to detect meaningful differences in hypertension across subclasses. These findings underscore the potential of early cardiometabolic disorders to be normalized or mitigated over time, highlighting both the plasticity of cardiometabolic patterns and the importance of early-life intervention. Instead of offspring metabolic risk factors, our results showed maternal gestational diabetes and hypertension during pregnancy were strongly associated with membership in the subclasses, suggesting a significant “glycemic memory” of maternal cardiometabolic risk factors (Supplementary Table 6). Our findings highlighted the potential influence of gestational cardiometabolic health on the development of cardiometabolic disorders in offspring, suggesting that maternal factors may play a more significant role than childhood adiposity in shaping the long-term risk of AGT in offspring.

Our study has several strengths. The first one is the long follow-up period. We described the natural history of the trajectory of glucose tolerance from childhood to adolescence to early adulthood. Second, our subclass method could inform a potential risk stratification approach to identify high-risk individuals that benefit from strategies to delay the loss of β-cell function and to prevent diabetes and diabetes-related complications in later life. Third, the close relationship between maternal cardiometabolic disorders and subclass membership in children highlighted the potential influence of gestational cardiometabolic health on the development of cardiometabolic disorders in offspring. Our study also has limitations. First, the subclass method is reliant on performing the OGTT, which has been acknowledged to have poor reproducibility. Nonetheless, we optimized our glycemic evaluation by several measures, such as ascertaining at least 8 h of fasting before the OGTT, and having blood samples taken by well-trained staff according to structured protocols. The consistently increased risk of AGT in both adolescence and young adulthood also reflected the high quality of glucose records measured in childhood. Second, we did not have sufficient insulin records for every participant, and hence the evaluation of β-cell function was limited. Third, our study only included HAPO children from the Hong Kong center (limited to Chinese ethnicity) such that variation of the glucose trajectories among different ethnicities could not be evaluated in this study. Fourth, children in our study exhibited a narrow BMI range, which limited our ability to assess its impact on AGT over time. We recommend interpreting the findings with caution, and further research is needed for validation. Fifth, we did not have access to records of antidiabetic medication usage. However, we assumed a very low proportion of antidiabetic medication usage among participants, as only one individual in the AAEA group had been diagnosed with diabetes. Finally, our study used a prospective design where glucose trajectory subclasses were defined exclusively during childhood, so we did not include modeling of glucose trajectories during adolescence and young adulthood.

Conclusion

We have identified subclasses defined in childhood that can efficiently stratify the risk of AGT in adolescence and young adulthood. This subclass method provides a potential strategy to identify those at risk for later glycemic disorders from childhood for more intensive evaluation of intervention. The close relationship between maternal cardiometabolic disorders and subclass membership of children highlighted the potential influence of gestational cardiometabolic health on the development of cardiometabolic disorders in offspring.

This article contains supplementary material online at https://doi.org/10.2337/figshare.29168333.

Funding Statement

Y.Z. acknowledges Vice-Chancellor’s PhD scholarship from the Chinese University of Hong Kong. Y.Z. acknowledges travel grants from the European Association for the Study of Diabetes and from the Research Grants Council of the Hong Kong SAR, China (grants CUHK 473408, 471713, 14118316, 14118718, and 14102719). The HAPO study was funded by the National Institute of Child Health and Human Development and National Institute of Diabetes and Digestive and Kidney Diseases (R01-HD34242 and R01-HD34243). The HAPO Follow-up Study was funded by grant 1U01DK094830 from the National Institute of Diabetes and Digestive and Kidney Diseases and the Eunice Kennedy Shriver National Institute of Child Health and Human Development. R.C.W.M. acknowledge support provided by the University Grants Committee Research Grants Matching Scheme.