Prospective Test Performance of Nonfasting Biomarkers to Identify Dysglycemia in Children and Adolescents

Abstract

Introduction:

Test performance screening measures for dysglycemia have not been evaluated prospectively in youth. This study evaluated the prospective test performance of random glucose (RG), 1-hour nonfasting glucose challenge test (1-h GCT), Hemoglobin A1c (HbA1c), fructosamine (FA), and 1,5-Anhydroglucitol (1,5-AG) for identifying dysglycemia.

Methods:

Youth ages 8–17 years with overweight or obesity (body mass index, BMI, ≥85^th percentile) without known diabetes completed nonfasting tests at baseline (n=176) and returned an average of 1.1 years later for two formal fasting 2-hour oral glucose tolerance tests. Outcomes included glucose-defined dysglycemia (fasting plasma glucose ≥100 mg/dL or 2-hour plasma glucose ≥140 mg/dL) or elevated HbA1c (≥5.7%). Longitudinal test performance was evaluated using receiver operating characteristic (ROC) curves and calculation of area under the curve (AUC).

Results:

Glucose-defined dysglycemia, elevated HbA1c, and either dysglycemia or elevated HbA1c were present in 15 (8.5%), 11 (6.3%), and 23 (13.1%) participants at baseline, and 16 (9.1%), 18 (10.3%), and 28 (15.9%) participants at follow-up. For prediction of glucose-defined dysglycemia at follow-up, RG, 1-h GCT, and HbA1c had similar performance (0.68 (95% CI 0.55–0.80), 0.76 (95% CI 0.64–0.89), and 0.70 (95% CI 0.56–0.84)), while FA and 1,5-AG performed poorly. For prediction of HbA1c at follow-up, baseline HbA1c had strong performance (AUC 0.93 (95% CI 0.88–0.98)), RG had moderate performance (AUC 0.67 (0.54–0.79)), while 1-h GCT, FA, and 1,5-AG performed poorly.

Conclusion:

HbA1c and nonfasting glucose tests had reasonable longitudinal discrimination identifying adolescents at risk for dysglycemia, but performance depended on outcome definition.

Introduction

Rising rates of obesity have fueled increases in the incidence of prediabetes and type 2 diabetes (T2D) in the pediatric population [1–3]. Although youth-onset T2D is increasing in incidence [4], the overall prevalence remains low, affecting approximately 0.7 per 1000 American youth [5]. In contrast, an estimated one-quarter of youth with obesity in the United States have prediabetes [6], defined as impaired fasting glucose (100–125 mg/dL), impaired glucose tolerance (140–199 mg/dL 2-hours after a 75 gm oral glucose load), or elevated hemoglobin A1c (HbA1c) (5.7–6.4%) [7]. The increasing prevalence of youth-onset obesity [8–10] and prediabetes will likely translate to an increasing lifetime burden of T2D due to high risk among youth with insulin resistance [11] or high-normal fasting glucose [12,13]. Early identification of prediabetes in youth may facilitate targeting of intensive lifestyle modification efforts, which may help reverse glucose abnormalities [14]. Screening tests currently endorsed by the American Diabetes Association (ADA) include HbA1c, fasting plasma glucose (FPG), and the 2-hour oral glucose tolerance test (OGTT) [7]. Each test has notable limitations in the pediatric population. Specifically, HbA1c has low sensitivity [15,16], low specificity [17], and limitations in the setting of rapid red blood cell turnover [15]; contributing to its poor cost-effectiveness [18]. Fasting glucose has low sensitivity [19] and OGTT has poor reproducibility [20] and may be challenging to obtain in practice. Random plasma glucose (RG) and 1-hour nonfasting glucose challenge tests (1-h GCT) offer greater discrimination between normal and dysglycemia measured by OGTT than HbA1c in cross-sectional studies of overweight or obese youth [21]. Similarly, alternative nonfasting measures of glycemia, including 1,5-Anhydroglucitol (1,5-AG) and fructosamine (FA), have been shown cross-sectionally to be good predictors of T2D in youth with obesity [22,23]. Studies in adults have linked higher nonfasting levels of HbA1c [24] and random glucose with high risk for future diabetes in adults [25]; however, the test performance of these measures in youth have not been evaluated prospectively. Due in part to a physiologic decrease in insulin sensitivity during puberty, glycemic abnormalities in youth may be transient [26,27], making stable longitudinal prediction of dysglycemia more challenging than in adults. In this study, we evaluated the prospective test performance of nonfasting biomarkers of glycemia to identify overweight/obese youth at risk for prediabetes/T2D.

Materials and Methods

Research Design and Methods

Our longitudinal cohort was a convenience sample of children with overweight/obesity (BMI percentiles ≥85^th for sex/age) who were 8–17 years old at study entry and recruited from primary care clinics and pediatric specialty clinics in southeast Michigan. We excluded individuals with known diabetes, who were using medications known to affect glucose metabolism, or who were pregnant. This study was approved by the University of Michigan Institutional Review Board. Written informed consent was obtained from the parent/guardian for all participants, and participants 10 years or older provided written assent.

Participants attended study visits at the Michigan Clinical Research Unit, where a medical history, vital signs, anthropometry, and laboratory evaluation were performed (see Supplement Fig. 1 of research design). At visit 1, participants were requested to fast for a minimum of 12 hours prior to a 2-hour OGTT, in which venous blood samples were drawn at baseline and every 30 minutes up to 2 hours after glucose load (1.75 g/kg up to maximum of 75g). Additional measurements at visit 1 included HbA1c, FA, and 1,5-AG. At visit 2, which occurred 1–3 weeks after visit 1, a random (nonfasting) glucose was measured, and participants underwent a 1-hour GCT (glucose measured 1 hour after ingestion of 50g Glucola in nonfasting state). Two follow-up visits (visits 3 and 4) were requested to be completed approximately 1 year later. At both visits 3 and 4, participants were requested to fast for a minimum of 12 hours prior to a 2-hour OGTT. At study visit 3, HbA1c was collected. Pubertal status was measured by child participant or parent (if child <13 years) self-report at visits 1 and 3. Participants completed visits 1 and 2 between 2007–2019.

Laboratory analyses were performed by the Michigan Diabetes Research Center (MDRC) Core Laboratories. Glucose was measured using the glucose hexokinase method. Glucose, FA and 1,5-AG assays were run on a Randox rX Daytona chemistry analyzer (Randox Laboratories Limited, United Kingdom). HbA1c was determined using a Tosoh G7 HPLC Analyzer (Tosoh Biosciences Inc, South San Francisco, CA).

Study Definitions

Measured height and weight were converted to BMI percentiles for age and sex according to the 2000 Centers for Disease Control and Prevention growth curves. Impaired fasting glucose (IFG) was defined as FPG 100–125 mg/dL and impaired glucose tolerance (IGT) was defined as 140–199 mg/dL for the 2-h PG during the OGTT; prediabetes-range HbA1c was defined as 5.7–6.4% [7]. Diabetes was defined as FPG ≥126 mg/dL, 2-h PG ≥200 mg/dL, or HbA1c ≥6.5%, as these are the thresholds recommended by the ADA guidelines [7]. The primary outcomes of interest were glucose-defined dysglycemia (FPG ≥100 or 2hr-PG ≥140) at either follow-up visit 3 or 4. The secondary outcomes included elevated HbA1c (HbA1c ≥5.7%) at follow-up or either glucose-defined dysglycemia or elevated HbA1c (FPG ≥100 or 2hr-PG ≥140 or HbA1c ≥5.7%) at follow-up.

Analysis

Normally-distributed continuous variables were summarized by mean and standard deviation (SD) and variables with skewed distributions were summarized by median and interquartile range (IQR). Categorical variables were summarized using frequencies and chi-squared analysis were used to compare groups.

The objective was to assess the predictive ability of non-fasting biomarkers (visit 2) on glycemia outcomes (visit 3 and 4) using receiver operating characteristic (ROC) analyses and measuring sensitivity, specificity, and positive and negative predictive values. Non-fasting biomarkers included RG, 1-h GCT, 1,5,-AG, and FA. Sensitivity analyses were performed for individuals who had normal HbA1c and FPG and 2-h PG at baseline, excluding any participants with elevated FG or 2-h PG or HbA1c at baseline. Stratified analyses were performed by sex and race, although the number of dysglycemia outcomes was small, limiting these subgroup analyses. Two-sided P < 0.05 was considered to be statistically significant.

To determine the influence of demographic covariates on ROC analysis, univariable logistic regression was performed using individual demographic variables as predictors (e.g., age, sex, race, ethnicity, family history, baseline BMI percentile, and change in BMI percentile between visits 1 and 3) with outcomes of glucose-defined dysglycemia (FPG ≥100 or 2hr-PG ≥140), elevated HbA1c (HbA1c ≥5.7%), and either dysglycemia or elevated HbA1c. Variables with significant associations (P <0.05) were included in multivariable logistic regression models with one predictor each (e.g., baseline HbA1c or baseline FG) to compare areas under the ROC curve of the fitted values with and without the demographic variable of interest. In the case of overlapping or significantly lower AUC, models with the additional demographic variable were rejected in favor of the simpler model without it.

Statistical analyses were performed using Stata 14.0 (Stata-Corp, College Station, TX).

Results

Of 358 participants who enrolled in the study, 176 (49.2%) completed all 4 visits. The duration between visits 1 and 3 averaged 1.1 (SD 0.4) years and ranged from 4.1 months to 2.0 years. Visit 4 occurred at a median of 14 days (interquartile range 7–28) after visit 3. One participant had a missing HbA1c at visit 3 and two had missing 2-hour glucose values during Visit 4. The cohort included more females than males and one-quarter of participants were Black. All participants were overweight or obese at Visit 1, including 57% with obesity (Table 1). Most participants were pubertal or post-pubertal at both visits 1 and 3. Approximately 10% of participants had both glucose-defined dysglycemia and elevated HbA1c at visit 3 or visit 4. Approximately 15% of participants had either dysglycemia or elevated HbA1c at visit 3 or visit 4. There was poor reproducibility of IFG between visits 3 and 4, with 83.3% (5/6) of participants with IFG at visit 3 having normal FPG at visit 4 and 1.1% (2/176) of participants with normal FPG at visit 3 having IFG at visit 4. There was poor reproducibility of IGT between visits 3 and 4, with 63% (5/8) of participants with IGT at visit 3 had reverted to normal 2-h PG at visit 4 and 1.2% (2/166) with normal 2-h PG at visit 3 having IGT at visit 4. In response, we grouped visit 3 and 4 together when defining dysglycemia.

Table 1.

Characteristics of study participants

	Visit 1	Visit 3	Visit 4	Visits 3 or 4
Participants, n	176	176	176
Age in years, mean (SD)	13.4 (2.4)	14.5 (2.5)	14.6 (2.5)
Sex, n (%)
Female	99 (56.2)
Male	77 (43.8)
Race, n (%)
Black	44 (25.0)
White	101 (57.4)
Other	31 (17.6)
Ethnicity, n (%)
Hispanic	12 (6.8)
Non-Hispanic	164 (93.2)
Family History of type 2 diabetes (self-reported), n (%)
Yes	75 (42.6)
No or Unknown	101 (57.4)
Weight status at study enrollment, n (%)
Overweight (BMI ≥85th percentile and <95th percentile)	75 (42.6)
Obese (BMI ≥95th percentile)	101 (57.4)
Female Tanner stage (self-reported), n (%)
Tanner 1	1 (1.0)	1 (1.0)	Not collected
Tanner 2–5	90 (86.5)	83 (79.8)	Not collected
Unknown	13 (12.5)	20 (19.2)	Not collected
Male Tanner stage (self-reported), n (%)
Tanner 1	9 (11.1)	6 (7.4)	Not collected
Tanner 2–5	44 (54.3)	51 (63.0)	Not collected
Unknown	28 (34.6)	24 (29.6)	Not collected
Glycemia Status, n (%)
Abnormal Fasting Plasma Glucose (>100 mg/dl)	5 (2.8)	6 (3.4)	3 (1.7)
Abnormal 2-h Plasma Glucose (≥140 mg/dl)	11 (6.3)	9 (5.1)a	5 (2.8)
Abnormal Fasting Plasma Glucose or 2-h Plasma Glucose	15 (8.5)	12 (6.8)a	7 (4.0)	16 (9.1)
Abnormal HbA1c (5.7–6.4%)	11 (6.3)	18 (10.3)b	Not collected	18 (10.3)
Abnormal Fasting Plasma Glucose or 2-h Plasma Glucose or HbA1c	23 (13.1)	24 (13.7)b	7 (4.0)	28 (15.9)

^aTwo patients (1.1%) had a glucose level in the diabetes range (2-h Plasma Glucose ≥ 200 mg/dl)

^bOne participant did not have a HbA1c result for visit 3.

Supplement Table 1 shows the demographic breakdown by these outcomes and the mean and SD of the baseline nonfasting tests. We evaluated associations with age, sex, race, and ethnicity and found significant associations between age and HbA1c (ß = −0.02% (95% CI −0.04–−0.001) per year of age, P = 0.03), age and 1-h GCT (ß = 1.6 mg/dL (95% CI 0.2–2.9) per year of age, P = 0.02), and sex and 1,5-AG (22.4, SD 5.8 μg/mL in females versus 24.7, SD 6.9 μg/mL in males; P = 0.02).

Table 1 shows the demographic characteristics and relationship of baseline weight status and glycemia at visit 1 with glycemic outcomes for visits 3 and 4. Only 1.7% (n=3) had glucose-defined dysglycemia at both visits 3 and 4. Regarding secondary outcomes, 10.3% (n=18) had elevated HbA1c at visit 3 and 16% (n=28) had at least one abnormal glucose or HbA1c at follow-up.

Table 2 shows the relationship of glycemic category at baseline with follow-up for glucose and HbA1c based outcomes. Depending on the outcome variable, 5–13% of patients with normal baseline glycemia became abnormal by visits 3 and 4 and 40–78% of patients with abnormal baseline glycemia remained abnormal.

Table 2.

Longitudinal progression of glycemic category at follow-up by glycemic category at Visit 1

		Glycemic category at follow-up a
		Values for FPG or 2-h PG at Visits 3 or 4		HbA1c at Visit 3b		Values for FPG and 2-h PG, HbA1c at Visits 3 or 4b
Glycemic category at Visit 1	Total, n	Normal	Abnormal	<5.7%	≥5.7%	Normal	Abnormal
FPG or 2-h Plasma Glucose, n (% of baseline category)
Normal	161	150 (93.2%)	11 (6.8%)	147 (91.9%)	13 (8.1%)	140 (87.0%)	21 (13.1%)
Abnormal	15	9 (60.0%)	6 (40.0%)	10 (66.7%)	5 (33.3%)	8 (53.3%)	7 (46.7%)
HbA1c, n (% of baseline category)
Normal	165	152 (92.1%)	13 (7.9%)	154 (93.9%)	10 (6.1%)	145 (87.9%)	20 (12.2%)
Abnormal	11	7 (63.6%)	4 (36.4%)	3 (27.3%)	8 (72.7%)	3 (27.3%)	8 (72.7%)
FPG, 2-h Plasma Glucose, or HbA1c, n (% of baseline category)
Normal	153	143 (93.5%)	10 (6.5%)	144 (94.7%)	8 (5.3%)	137 (89.5%)	16 (10.5%)
Abnormal	23	16 (69.6%)	7 (30.4%)	13 (56.5%)	10 (43.5%)	11 (47.8%)	12 (52.2%)

^aAbnormal Fasting Plasma Glucose = Fasting plasma glucose ≥100 mg/dl; Abnormal 2-h Plasma Glucose = 2-h PG ≥ 140 mg/dl; Abnormal HbA1c = HbA1c 5.7–6.4%

^bOne participant did not have a HbA1c result for visit 3. Two participants did not have 2-hour glucose at visit 4.

Figure 1 shows the ROC curves and Table 3 shows the AUC of each of the nonfasting tests for each of the outcome measures. All ROC analyses were without adjustment for demographic variables, based on lack of significant improvement in model fit as described above. Figure 1a shows the ROC curves for the outcome of glucose-defined dysglycemia at visits 3 or 4. The outcome of glucose-defined dysglycemia at both visits 3 and 4 was not evaluated using ROC analysis due to the very small number of participants with persistent dysglycemia (n=3). Baseline RG (AUC 0.68 (95% CI 0.55–0.80)), 1-h GCT (AUC 0.76 (95% CI 0.64–0.89)), and HbA1c (0.70 (95% CI 0.56–0.84)) had similar levels of discrimination. Figure 1b shows the ROC curves for the outcome of elevated HbA1c (≥5.7%) at visit 3. HbA1c had the highest discrimination compared with all other tests, with AUC 0.93 (95% CI 0.88–0.98). Figure 1c shows the combined outcome of either glucose-defined dysglycemia or elevated HbA1c. Baseline HbA1c (AUC 0.81, 95% CI 0.71–0.90) had the highest AUC, followed by RG (AUC 0.65, 95% CI 0.55–0.75) and 1-h GCT (AUC 0.63, 95% CI 0.50–0.76). FA and 1,5-AG had poor discrimination for all outcomes.

Fig. 1.

Receiver Operating Characteristic (ROC) curves by glycemic outcome (a) glucose-defined dysglycemia (Fasting Plasma Glucose ≥100 mg/dL or 2-h Plasma Glucose ≥140 mg/dL); (b) elevated HbA1c (HbA1c ≥5.7%); (c) any dysglycemia (Fasting Plasma Glucose ≥100 mg/dL, 2-h Plasma Glucose ≥140 mg/dL, or HbA1c ≥5.7%)

Table 3.

Longitudinal progression of glycemic category at follow-up by glycemic category at Visit 1

	Fasting Plasma Glucose ≥100 mg/dL, 2-h Plasma Glucose ≥140 mg/dL (n = 176)	HbAlc ≥ 5.7% (n = 175)	Fasting Plasma Glucose ≥ 100 mg/dL, 2-h Plasma Glucose ≥ 140 mg/dL, or HbA1c ≥ 5.7% (n = 176)
HbAlc	0.70 (0.56–0.84)	0.93 (0.88–0.98)	0.81 (0.71–0.90)
Random glucose	0.68 (0.55–0.80)	0.67 (0.54–0.79)	0.65 (0.55–0.75)
1-h Glucose Challenge Test	0.76 (0.64–0.89)	0.57 (0.40–0.74)	0.63 (0.50–0.76)
Fructosamine	0.48 (0.33–0.63)	0.47 (0.35–0.59)	0.49 (0.38–0.60)
1,5-anhydroglucitol	0.50 (0.35–0.64)	0.45 (0.31–0.59)	0.49 (0.37–0.61)

Supplement Table 2 shows the test characteristics of baseline HbA1c, random glucose, and 1-h GCT for predicting glucose-defined dysglycemia, elevated HbA1c, or either at visits 3 or 4 longitudinally for each of the outcomes. For the outcome of glucose-defined dysglycemia, HbA1c had low sensitivity (25%) but high specificity (96%) at a threshold of 5.7%; at a threshold of 5.4%, sensitivity more than doubled (56%), while specificity remained high (77%). For the combined outcome of either abnormal glucose or HbA1c, an HbA1c threshold of 5.7% had a sensitivity of 29% and a specificity of 98%, while a threshold of 5.4% had a sensitivity of 64% and specificity of 81%. For all outcomes, RG had low sensitivity even below a threshold of 100 mg/dL: sensitivity ranged from 53–61% at a threshold of 90 mg/dL, while specificity was 58–66%. Similarly, 1h-GCT had low sensitivity across all outcomes: at a threshold of 110 mg/dL, sensitivity was 53%, 39%, and 43% and specificity was 85%, 84%, and 86% for glucose-defined dysglycemia, elevated HbA1c, and any abnormal glucose or HbA1c, respectively. In Supplement Table 3, we conducted sensitivity analyses. First, we evaluated test performance among the 153 adolescents who had normal FPG, 2-h PG, and HbA1c at visit 1, excluding any with dysglycemia. Although 1h-GCT had reasonable test performance for the outcome of glucose-defined dysglycemia, it performed poorly for the outcomes of elevated HbA1c or any dysglycemia or HbA1c elevation. HbA1c remained the strongest predictor of the combined outcome of glucose-defined dysglycemia or elevated HbA1c. There were no significant differences in test performance by sex or race, although these subgroup analyses were limited by sample size. As only 2 participants had diabetes-range values at follow-up, results were largely unchanged when the outcome of IGT or IFG rather than dysglycemia was used (results not shown).

Discussion

To our knowledge, this is the one of the first studies to evaluate and compare prospective test performance of multiple nonfasting biomarker tests for identifying dysglycemia in a pediatric population. Although previous studies have evaluated test performance among adolescents, they focused solely on HbA1c and/or fasting tests (FPG), or used a cross-sectional study design [15,28]. For the combined outcome of glucose-defined or HbA1c-defined dysglycemia, we found that HbA1c, RG, and 1-h GCT all had reasonable test performance and discrimination for identifying children with abnormal glucose/HbA1c levels and highlight the internal consistency of HbA1c. According to the 2021 ADA guidelines [29], HbA1c is the only nonfasting recommended test to screen for T2D in children/adolescents. Our results suggest that random glucose and 1-h GCT, which avoid potential inaccuracies of HbA1c in the setting of abnormal red blood cell turnover, may be additional options for T2D screening given their accuracy and the lack of a need for fasting.

This study confirms for clinicians that HbA1c is an effective and convenient nonfasting screening test that can be used in clinical practice, and accordingly, it is increasingly used in primary care settings for assessing for diabetes risk in children [30]. Clinicians need to be aware of the limitations of HbA1c, including the fact that a number of non glycemic factors impact HbA1c, including disorders that affect blood cell turnover, hemoglobinopathies, medications, race, and age [15] but paired with glucose measures it can be a useful screening tool. For example, in one of our health systems (University of Michigan), elevated BMI percentiles in children older than 10 years of age triggers an alert in the electronic health record, which recommends to clinicians that an HbA1c and random glucose be drawn.

We found that HbA1c had the highest AUC for identifying prospective dysglycemia outcomes for the glucose- and HbA1c-defined outcome and for the HbA1c-defined outcome alone. This is likely due to the lower short-term variability and greater reproducibility of prediabetes outcomes for HbA1c compared with glucose measures [31], and because both outcome definitions included HbA1c. The test performance of HbA1c for the glucose-defined outcome measure alone was lower. Accordingly, the 1-h GCT, a stimulated nonfasting glucose test, had the highest AUC for the glucose-defined dysglycemia outcome, although we did not find statistical differences in AUC between the HbA1c, RG, and 1-h GCT for that outcome.

The convenience of T2D screening matters to providers, as a national survey conducted in 2011 found that pediatricians and family practitioners were 5.6 times more likely to order nonfasting tests as their initial test of choice if they valued test convenience over accuracy of the test, and 2.2 times more likely if they did not have an on-site blood draw station at their practice [32]. At that time, providers reported that in light of the 2010 ADA guidelines, they would be more likely to order HbA1c test alone or an HbA1c in conjunction with additional nonfasting tests to screen for T2D [32], and this has been borne out by studies in a number of health systems demonstrating increased uptake of HbA1c as a screening modality in pediatric primary care [27,30].

The previous pediatric cross-sectional literature has reported sensitivity ranges of 5–75% and specificity ranges of 58% to 98% for HbA1c as a biomarker for identifying dysglycemia/T2D [15,28]. The higher specificity levels found in this study and across the literature suggest that a normal HbA1c can be reassuring in ruling out the presence of dysglycemia [15,28]. Providers should however take caution when using HbA1c to diagnose T2D, as there are a number of nonglycemic factors that impact HbA1c results [15]. A recent study reported that 2% of normal weight adolescents have HbA1c in the prediabetes range (5.7–6.4%), calling into question the specificity of HbA1c to discriminate normal from abnormal glucose metabolism [17].

We can compare our results for glucose-defined only outcomes with our previous cross-sectional study [21]; AUC for RG was 0.66 (0.60–0.73)(cross-sectional) vs. 0.68 (0.55–0.80) (prospective), AUC for 1h-GCT was 0.68 (0.61–0.74) (cross-sectional) vs. 0.76 (0.64–0.89) (prospective); AUC for HbA1c was 0.54 (0.47–0.61) (cross-sectional) vs. 0.70 (0.56–0.84) (prospective). Although our findings demonstrated a higher AUC point estimate for HbA1c when used for longitudinal prediction, the confidence intervals did overlap with the cross-sectional study.

We observe that test performance of FA and 1,5-AG was poor for both glucose- and HbA1c-defined dysglycemia at follow-up. Fructosamine, a glycated amine, offers a measure of intermediate glycemia due to its half-life of 2–3 weeks, but the fact that it represents a shorter time frame may account for its poor test performance. Although FA has been found to correlate well with HbA1c in pediatric patients with type 1 and type 2 diabetes [33], its use for discrimination of lower-range glycemic abnormalities appears to be more limited [21,22]. Because 1,5-AG competes with glucose for renal reabsorption, its poor test performance for prediabetes diagnosis is likely due to the very minimal time individuals with prediabetes experience with plasma glucose > 180 mg/dL, the average renal threshold. Similarly, although it appears to perform well for discrimination of T2D in youth with obesity, 1,5-AG is a weaker predictor of prediabetes-range abnormalities in youth [22,23]. Thus, both FA and 1,5-AG are likely to be less useful in the pediatric population, in which undiagnosed T2D is infrequent [28]. In addition, due to variation with age [34,35] and sex [36], use of 1,5-AG as a screening tool in practice would likely require use of age- and sex-specific thresholds.

Strengths of our study include the longitudinal study design, the use of repeated 2-h OGTTs at follow-up, and evaluation of nonfasting tests in a pediatric population but we also acknowledge limitations. First, we combined the outcomes of impaired fasting glucose and impaired glucose tolerance, which may not be ideal given that they differ in their pathophysiology. Second, 2-h OGTT, a gold standard test, has a demonstrated lack of reproducibility [20], which was confirmed in our study. The fact that we did not control for diet composition in the days leading to an OGTT could have contributed to this. We thus chose to define dysglycemia/T2D as the presence of abnormalities on at least one occasion, either visit 3 or 4; had we chosen to require persistent dysglycemia, the proportion with abnormal results would decrease substantially.

Third, the incidence of T2D was very low, and the rate of prediabetes-range abnormalities was also lower than that found in a recent nationally-representative sample [6], so our findings may not be representative of higher-risk populations. Fourth, there was variation in the followup time, with a median time between initial and follow-up visits of approximately 1 year, so our findings cannot necessarily be extrapolated to a longer duration. Fifth, there was minimal change in BMI percentile between visits; thus, it is possible that our findings overestimate the longitudinal performance of the evaluated tests, as BMI change is associated with longitudinal changes in dysglycemia [27]. Sixth, we acknowledge that a subset of patients elected to complete all the study visits and seventh, we did not have a significant population of Hispanic children, both of which could affect generalizability. Finally, the majority of children were pubertal at the time of visit 1 based on self-report rather than investigator-confirmed, but the age range was relatively wide, and we did not perform physical exams to assess progression of puberty during the study. An estimated one-quarter of American youth are now eligible for risk-based prediabetes/T2D screening criteria from the ADA [28]. As even high-normal fasting glucose in childhood is associated with a more than 2-fold increased risk of developing adult prediabetes and T2D [12], it is possible that individuals with even transiently abnormal test results fall on the continuum of diabetes risk would benefit from lifestyle interventions, although further evidence of the benefit of lifestyle interventions is needed. Cost-effective, feasible, and efficient screening methods are needed to appropriately screen this higher-risk group. Although the cost-effectiveness of HbA1c as a screening tool for T2D in youth is less favorable than the RG and 1h-GCT, its convenience and test performance highlight the importance of HbA1c in the arsenal of screening strategies [18]. Future studies are needed to assess these nonfasting markers in a higher-risk population over a longer time period to define the optimal frequency of T2D screening.

Data Availability Statement:

Datasets generated during and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.