Reproducibility of Glycemic Measures Among Dysglycemic Youth and Adults in the RISE Study

Abstract

Aims

Previous work found poor reproducibility for measures of glycemia in individuals at risk for dysglycemia. Differences between youth and adults have not been assessed. Using youth and adults in the Restoring Insulin Secretion Study, we tested variability and classification concordance for hemoglobin A1C (HbA1c), fasting and 2-hour glucose from oral glucose tolerance tests (OGTTs).

Methods

HbA1c and glucose on repeated samples obtained ∼6 weeks apart were compared in 66 youth (mean age 14.2 years) and 354 adults (52.7 years). Changes, coefficient of variation (CV), and concordance of diagnostic categories between the 2 visits were compared.

Results

Mean difference between the 2 visits in HbA1c was higher in youth than adults (P < .001), while fasting glucose was similar and 2-hour glucose was lower in youth (P = .051). CV was smallest for HbA1c compared to fasting and 2-hour glucose. For HbA1c, youth had higher CV (P < .001); whereas CV for 2-hour glucose was lower for youth (P = .041). Classification concordance by HbA1c was lower in youth (P = .004). Using OGTT or HbA1c for classification, intervisit variability produced discordant classification in 20% of youth and 28% of adults. Using both fasting glucose and HbA1c, intervisit variability reduced discordant classification to 16% of adults while not improving classification in youth.

Conclusions

Poor reproducibility and lack of classification concordance highlight the limitations of one-time testing, with important implications for assessing eligibility in clinical trials. Consideration should be given to using more than a single parameter for screening and diagnosis, especially when classification category is important.

Highlights:

• Tested the variability and classification of duplicate measures of dysglycemia in youth and adults in the Restoring Insulin Secretion (RISE) Study

• Disease status concordance was highest with HbA1c and lowest with 2-hour glucose

• Concordance by HbA1c was lower in youth than adults

• Use of OGTT or HbA1c produced discordant categorical classification in 20% of youth and 28% of adults

• Use of more than a single parameter should be considered when assessing eligibility or classifying individuals within the context of clinical trials of prediabetes and diabetes

Fasting and 2-hour glucose measurements from oral glucose tolerance tests (OGTTs) and glycated hemoglobin (HbA1c) are routinely used in clinical practice for the diagnosis of impaired glucose tolerance (IGT) and type 2 diabetes. They are also commonly used in the research setting for assessing eligibility and outcomes for clinical trials of prevention and treatment of type 2 diabetes. While guidelines recommend that in most circumstances clinical diagnosis requires a confirmatory test, this is not always feasible (1, 2). Thus, a single test may frequently be used for glycemic classification, particularly in clinical trials and epidemiologic studies.

Previous studies reported poor reproducibility of HbA1c and OGTT glucose measures in adult populations at risk for diabetes (3-7). In the Diabetes Prevention Program, only half of adults with a fasting and/or 2-hour glucose level that met criteria for diabetes on an initial OGTT were confirmed on a second OGTT (8). Similarly, poor reproducibility of the OGTT measures has been documented in overweight youth, particularly the 2-hour glucose (9). However, as there has not been a direct comparison of these 2 age groups using the same methods and central laboratory, it is uncertain whether reproducibility and glucose tolerance classification to confirm prediabetes status will be similar or different in obese youth and adults. Knowledge of such would be helpful when designing future clinical studies.

The Restoring Insulin Secretion (RISE) Study examined youth and adults in parallel, using standardized testing with high-quality analyte measurements performed in a central laboratory (10). These data therefore provide the unique opportunity to compare the 2 age groups for within-person variability, ie, reproducibility, of glycemic measures that are routinely used for clinical and research purposes, namely fasting glucose, 2-hour glucose, and HbA1c. Using this cohort, we examined the test-retest variability and glucose tolerance reclassification concordance for these measures and whether these features of repeatability differ between comparably dysglycemic youth and adults. The findings of these analyses have implications for the design of future clinical trials and epidemiological studies that enroll both youth and adults with prediabetes simultaneously.

Materials and Methods

Participants

Participants were youth in the RISE Pediatric Medication Study and adults in the RISE Adult Medication Study and RISE Adult Surgery Study who had either IGT or recently diagnosed type 2 diabetes. For the Pediatric Medication Study, individuals with a fasting plasma glucose ≥5 mmol/L (≥90 mg/dL) plus 2-hour glucose ≥7.8 mmol/L (≥140 mg/dL) and HbA1c ≤ 8.0% (≤64 mmol/mol), if treatment naïve, were eligible. Youth with type 2 diabetes meeting the same OGTT criteria who were on metformin for <3 months with a HbA1c ≤ 7.5% (≤58 mmol/mol) or metformin for 3 to 6 months and HbA1c ≤ 7.0% (≤53 mmol/mol) were also eligible. For the Adult Medication Study, those with a fasting plasma glucose 5.3 to 6.9 mmol/L (95–125 mg/dL) plus 2-hour glucose ≥7.8 mmol/L (≥140 mg/dL) and HbA1c ≤ 7.0% (≤53 mmol/mol) were eligible. For the Adult Surgery Study, individuals with a fasting glucose >5 mmol/L (>90 mg/dL) plus 2-hour glucose ≥7.8 mmol/L (≥140 mg/dL) and HbA1c < 7% (≤53 mmol/mol) were eligible. Complete details on participant recruitment and eligibility criteria have been published previously (10) and are available on the RISE website (https://rise.bsc.gwu.edu/web/rise/collaborators). Associated ClinicalTrials.gov identifier numbers are NCT01779362, NCT01779375, NCT01763346. The study was approved by the Institutional Review Boards of all participating centers, and written informed consent and/or child assent where age-appropriate was obtained prior to initiation of any study-related activities.

Procedures

Over a 4-year period (2013–2017), 1290 individuals were determined based on their history to be at high risk for IGT and type 2 diabetes and eligible to be screened for participation in one of the 3 RISE protocols. They were screened with a 75-gram OGTT and HbA1c following an overnight fast of at least 10 hours. Not all individuals who screened met the required glucose criteria and therefore did not proceed further. Youth participants with positive anti-glutamic acid decarboxylase or IA-2 antibodies were excluded.

Volunteers determined to be eligible for the youth and adult medication studies based on glucose criteria from the screening OGTT participated in a 3-week, placebo run-in period prior to randomization. The 267 adult and 91 youth (10-19 years with Tanner stage pubertal development II or greater) participants who successfully completed this run-in attended a baseline visit at which they had a repeat 75-gram OGTT and HbA1c following a 10-hour overnight fast (11). No medication or lifestyle intervention was implemented between the 2 OGTT visits. As mentioned, unlike adults the inclusion criteria for the Pediatric Medication Study allowed prior exposure to metformin; to avoid confounding due to metformin exposure, those youth currently or previously having taking metformin (n = 25) were excluded from the present analyses.

The Adult Surgery Study had no required run-in period after volunteers had been identified as eligible with the screening OGTT, and therefore these participants underwent a repeat 75-gram OGTT test at baseline about a month after the screening visit. Eighty-seven screen-eligible surgical study participants provided repeated data at baseline. One additional surgical study participant had a fasting glucose value at baseline that indicated he was not fasting; his results are excluded from the present analyses. The 267 Adult Medication Study and 87 Adult Surgery Study participants are presented together as “adults” for these analyses.

Weight, height, and blood pressure were collected on both occasions in a standardized manner (11, 12). Sex and race/ethnicity were self-reported at the screening visit.

Assays

All OGTT blood samples were immediately placed on ice, separated by centrifugation, and frozen at −80°C prior to shipment to the central biochemistry laboratory at the University of Washington (Northwest Lipid Metabolism and Diabetes Research Laboratories, University of Washington, Seattle, WA). Plasma glucose concentrations were measured by the glucose hexokinase method using Roche reagent on a Roche c501 autoanalyzer (Roche Diagnostics Inc., Indianapolis, IN). The method has interassay coefficient of variations (CVs) on quality control samples with low, medium, and high glucose of 2.0%, 1.7%, and 1.3%, respectively. HbA1fc was measured by ion-exchange high performance chromatography on a TOSOH G8 analyser (TOSOH Biosciences, Inc., South San Francisco, CA). The interassay CVs on low- and high-quality control samples were 1.9% and 1.0%, respectively.

Glucose Tolerance Classification

Participants were classified into categories of glucose tolerance according to the American Diabetes Association's Standards of Care based on the HbA1c value and using fasting and 2-hour glucose concentrations (1). We categorized participants in the following ways: (1) using each measure alone, (2) using fasting glucose and HbA1c, (3) using OGTT criteria (fasting + 2-hour) alone, (4) using either OGTT criteria or HbA1c (current American Diabetes Association criteria), and (5) using both OGTT and HbA1c, as detailed in Supplementary Table S1 (13).

Calculations and Statistical Analysis

Reproducibility was assessed by evaluating the 3 glycemic measures—HbA1c, fasting glucose, and 2-hour glucose—as continuous variables as well as categories of glycemic status. For each individual, for comparison of the continuous values we calculated change from screening to baseline as (1) absolute change regardless of direction, (2) percent change, (3) absolute percent change regardless of direction, and (4) CV per person. CV was calculated for each individual as a measure of between-sample reliability by dividing the standard deviation of the screening and baseline visit values by the corresponding mean of the screening and baseline visits. For HbA1c, calculations of percent change and CV are presented for National Glycohemoglobin Standardization Program percent rather than International Federation of Clinical Chemistry mmol/mol. In addition, we examined the concordance of glycemic categorization defined separately for each glycemic measure as categories of glycemic status [normal glucose tolerance (NGT), impaired fasting glucose (IFG), IGT, type 2 diabetes] across visits.

Continuous variables were expressed as mean ± SD for normally distributed data; categorical variables were summarized as percentages (%). Unadjusted group comparisons were performed using analysis of variance for normally distributed continuous variables by age group (youth vs adult), race/ethnicity within each age group, and body mass index (BMI) category within each age group. Pearson's chi-squared test was used to compare categorical variables. Within-individual agreement variability for each measure was assessed using Pearson correlation coefficients and analyzed graphically using Bland-Altman plots for assessing patterns of disagreement between the 2 measurements (14). The Pearson correlation coefficients were compared for difference between youth and adults using a two-sided z-test for equality. P-values <.05 were considered nominally statistically significant, with no adjustments made for multiple tests. Analyses were performed using SAS 9.4 (SAS Institute, Cary, NC).

Results

Participant Characteristics

Table 1 summarizes characteristics at the time of the screening visit of the 66 youth and 354 adult participants with evaluable repeat data. Among youth, there were fewer males and White participants compared to adults. The 2 groups did not differ in their weights; however, BMI was slightly higher in the youth due to their shorter height. Among the youth participants, approximately 60% were Tanner stage V. As listed in Table 2, at the screening visit, youth and adults had a mean HbA1c value of 5.8% (40 mmol/mL) with an average 2-hour glucose in the IGT range. The fasting glucose concentration was lower among youth. Nearly 30% of the adults and 20% of the youth cohort were classified as having diabetes based on the screening visit OGTT values (fasting and/or 2-hour glucose) with this being 35% and 27%, respectively, at the baseline visit.

Table 1.

Demographics and measures of glycemia in RISE youth and adults at screening

Demographics	Youth N = 66	Adult N = 354	P-value
Age (years)	14.2 ± 2.0	52.7 ± 9.4	<.001
Sex (% female)	47 (71.2)	182 (51.4)	.003
Race/ethnicity			<.001
White (%)	19 (28.8)	166 (46.9)
Black (%)	14 (21.2)	97 (27.4)
Hispanic (any race) (%)	25 (37.9)	67 (18.9)
Other (%)	8 (12.1)	24 (6.8)
Weight (kg)	98.5 ± 22.4	101.0 ± 18.3	.332
Height (cm)	163.6 ± 9.1	169.3 ± 9.7	<.001
BMI (kg/m2)	36.5 ± 5.9	35.2 ± 5.1	.057
BMI percentile	98.7 ± 1.7	—	—
Tanner stages (II/III/IV/V) (n)	4/10/12/40	—	—

Abbreviations: BMI, body mass index; RISE, Restoring Insulin Secretion.

Data shown are mean ± SD or n (%). P-value for youth vs adult comparison.

Table 2.

Comparison of screening vs baseline measures of glycemia in RISE youth and adults

	Youth(N = 66)			Adults(N = 354)			P-value*
	Screening	Baseline	P-value§	Screening	Baseline	P-value§
Glucose values
HbA1c (%)	5.78 ± 0.64	5.68 ± 0.56	0.002	5.79 ± 0.38	5.78 ± 0.40	.084	.878
HbA1c (mmol/mol)	39.68 ± 7.00	38.54 ± 6.11	0.002	39.78 ± 4.19	39.63 ± 4.35	.084	.878
Fasting glucose (mmol/L)	5.87 ± 0.85	5.93 ± 0.93	0.417	6.20 ± 0.55	6.15 ± 0.65	.085	<.001
2-hour glucose (mmol/L)	10.01 ± 2.41	9.89 ± 2.46	0.516	10.18 ± 2.01	10.15 ± 2.37	.823	.553
Glycemic status (fasting + 2 hr)			<0.001			<.001	.117
NGT (%)	—	4 (6.1)		—	18 (5.1)
IFG only (%)	—	3 (4.5)		—	39 (11.0)
IGT (with and without IFG) (%)	53 (80.3)	41 (62.1)		251 (70.9)	172 (48.6)
Diabetes (%)	13 (19.7)	18 (27.3)		103 (29.1)	125 (35.3)
Difference between visits	Youth (n = 66)			Adults (n = 354)			P-value †
HbA1c (%)	0.10 ± 0.26			0.01 ± 0.15			<.001
HbA1c (mmol/mol)	1.14 ± 2.83			0.15 ± 1.64			<.001
Fasting glucose (mmol/L)	−0.06 ± 0.64			0.05 ± 0.55			.131
2-hour glucose (mmol/L)	0.12 ± 1.55			0.02 ± 1.92			.685
Absolute difference
HbA1c (%)	0.21 ± 0.19			0.09 ± 0.12			<.001
HbA1c (mmol/mol)	2.27 ± 2.03			0.97 ± 1.33			<.001
Fasting glucose (mmol/L)	0.45 ± 0.45			0.41 ± 0.37			.383
2-hour glucose (mmol/L)	1.19 ± 0.99			1.50 ± 1.20			.051
Percent difference (%)
HbA1c	1.60 ± 4.24			0.22 ± 2.58			<.001
Fasting glucose	−1.38 ± 10.02			0.60 ± 8.93			.105
2-hour glucose	0.37 ± 14.46			−0.75 ± 19.41			.655
Absolute percent difference (%)
HbA1c	3.49 ± 2.87			1.53 ± 2.08			<.001
Fasting glucose	7.47 ± 6.76			6.57 ± 6.06			.280
2-hour glucose	11.63 ± 8.48			15.02 ± 12.30			.033
Coefficient of variation (%)
HbA1c (%)	2.18 ± 1.84			0.94 ± 1.29			<.001
Fasting glucose	4.52 ± 4.10			4.02 ± 3.50			.300
2-hour glucose	7.21 ± 5.54			9.20 ± 7.51			.041

Abbreviations: HbA1C, hemoglobin A1C; IFG, impaired fasting glucose; IGT, impaired glucose tolerance; NGT, normal glucose tolerance; RISE, Restoring Insulin Secretion.

Data shown are mean ± SD or n (%).

P-value for paired t-test screening vs baseline comparison (within age group).

P-value for youth vs adult comparison at screening visit.

P-values for youth vs adult comparison. Difference is the value from visit 1 (screening) minus the value from visit 2 (baseline). Absolute difference is the absolute value of the difference (regardless of direction). Percent difference is 100*(difference/visit 1 value). Absolute percent difference is 100*(absolute difference/visit 1 value).

Variation of Glycemic Measures in Youth and Adults

Table 2 compares data from the screening and baseline visits. The baseline visits took place 14 to 86 days after the screening visit (45.5 ± 9.7 in youth and 38.9 ± 11.1 days in adults, P < .001). The mean absolute difference in HbA1c was higher in youth than adults: 0.21% (2.27 mmol/mol) for youth and 0.09% (0.97 mmol/mol) for adults (P < .001 comparing groups). The mean absolute difference in fasting glucose between the 2 OGTTs was similar in both age groups (youth: 0.45 mmol/L, adults: 0.41 mmol/L; P = .38), while that for 2-hour glucose was lower for youth than adults (youth: 1.19 mmol/L, adults: 1.50 mmol/L; P = .051).

Between screening and baseline, in youth and adults, variation was smallest for HbA1c compared to the fasting and 2-hour glucose measurements. This variation was greater in youth than adults when we examined absolute difference, absolute percent difference, and CV (P < .001 for all measurements). For 2-hour glucose, the absolute difference, absolute percent difference, and CV were lower in youth than adults (P = .051, P = .033, P = .041 respectively), while measurements of variability for fasting glucose did not differ between youth and adults.

Figure 1 depicts the variation between the 2 visits for HbA1c, fasting, and 2-hour glucose for the youth and adults. The Pearson correlation coefficients for HbA1c were similar for youth and adults (r = 0.92 and r = 0.93, respectively, P = .6 comparing groups, Panel A), whereas for fasting glucose and 2-hour glucose the correlation coefficient was higher in youth than adults (fasting glucose r = 0.75 in youth and r = 0.59 in adults, P = .031 comparing groups, Panel B; 2-hour glucose r = 0.80 in youth and r = 0.62 in adults, P = .004 comparing groups, Panel C). The slopes of the correlation for all measures and among youth and adults were not significantly different from 1. Bland-Altman plots were used to evaluate the agreement between the measurements performed at screening and baseline for HbA1c, fasting glucose, and 2-hour glucose for youth and adults (Panels D, E, and F, respectively). The mean difference between the 2 visits was close to 0 for all measurements in both groups, indicating that there was no systematic improvement or worsening between the 2 visits.

Figure 1.

Correlation for HbA1c (A), fasting glucose (B), and 2-hour glucose (C) for youth (blue) and adults (green). On the correlation plots, the line of utility (slope of 1) represents identical values with both tests (dashed line). The Bland-Altman plots show the difference against the mean value for the 2 visits for HbA1c (D), fasting glucose (E), and 2-hour glucose (F) for youth (blue) and adults (green). The solid reference line at 0 represents no difference between the measurements at screening and baseline. The corresponding dashed lines represents the mean differences and the solid lines represent the 95% CI (±1.96*SD) of the mean of the differences. Correlation is stronger in youth than adults for fasting and 2-hour glucose (P = .031 and P = .004, respectively) but not significantly different for HbA1c. Abbreviations: HbA1c, hemoglobin A1C.

Variation of Glycemic Classification in Youth and Adults

Table 2 lists concordance for the glucose tolerance categories in youth and adults based on the OGTT glucose values. Of those classified as IGT or diabetes at screening, 10.6% of youth and 16.1% of adults were subsequently classified as normoglycemic or having only IFG at the baseline visit.

Table 3 summarizes classification concordance using different combinations of criteria for diabetes diagnosis. The criteria for defining each glucose tolerance category (normoglycemia, prediabetes, and diabetes) based on fasting glucose, 2-hour glucose, and HbA1c are listed in Supplementary Table S1 (13). Among adults, glucose tolerance status concordance was highest using HbA1c criterion alone and lowest with 2-hour glucose criterion alone. Among youth, the status concordance was highest using a combination of either glucose value and HbA1c and lowest with either the fasting or 2-hour glucose alone. Classification concordance was significantly lower in youth than adults using the HbA1c criterion alone (P = .004) but was not statistically different using fasting or 2-hour glucose criterion alone (P = .47 and P = .11, respectively). Using 4 categories defined based only on the fasting and 2-hour criteria (NGT, IFG, IGT, diabetes), 73% of youth and 61% of adults would have been classified in the same IGT and diabetes categories at both visits. Based on the combined fasting, 2-hour, or HbA1c criteria with 3 categories (normoglycemia, prediabetes, diabetes), 80% of youth and 72% of adults would have been classified in the same categories at both screening and baseline. Lastly, classifying participants based on combined OGTT (fasting or 2-hour) and HbA1c criteria resulted in 80% of youth and 89% of adults having concordant results. Results were similar when eliminating the 2-hour glucose and using fasting glucose and HbA1c combined as criteria for classification with 80% of youth and 85% of adults having concordant results. Supplementary Table S2 (13) describes the detailed frequency of glycemic status concordance between the screening and baseline visit by category for each set of classification criteria.

Table 3.

Classification concordance summary

	Rates of Concordance
	Youth (n = 66)	Adults (n = 354)	P-value
HbA1c alone (%)	51 (77.3)	318 (89.8)	.004
Fasting glucose alone (%)	47 (71.2)	267 (75.4)	.470
2-hour glucose alone (%)	47 (71.2)	215 (60.7)	.107
Fasting glucose and HbA1c (%)	53 (80.3)	299 (84.5)	.400
OGTT criteria (fasting and/or 2-hour glucose) (%)	48 (72.7)	214 (60.5)	.059
OGTT or HbA1c (%)	53 (80.3)	255 (72.0)	.163
OGTT and HbA1c (%)	53 (80.3)	314 (88.7)	.059

Abbreviations: HbA1C, hemoglobin A1C; OGTT, oral glucose tolerance test.

Discussion

The RISE studies randomized youth and adults with IGT or recently diagnosed type 2 diabetes into 3 different protocols. Screening using a standardized OGTT and HbA1c was performed ∼6 weeks before the baseline visit, which included repeat OGTT and HbA1c measurements. These paired measurements provided a platform for evaluating differences in the reproducibility of these common diagnostics within and between the 2 age groups. For all measures evaluated, the slopes relating absolute values of the paired measures were not significantly different from 1 for either youth or middle-aged adults; however, the correlation for HbA1c measured at the 2 visits was higher and the variability was lower than fasting and 2-hour glucose for youth and adults. When comparing youth and adults, variability was slightly higher in HbA1c and lower in 2-hour glucose for youth compared to adults. The paired classification (test-retest) of participants when proximal to glycemia cutpoints highlights the potential difficulty of using a single test as an inclusion criterion or outcome for clinical trials. Classification concordance was statistically significantly lower in youth than adults using the HbA1c criterion alone, but the concordance was not statistically significantly different between the 2 age groups using the other criteria or combinations of criteria. Depending on the criteria used, classification concordance ranged from 71% to 80% for youth and 60% to 90% for adults, an important finding when designing future clinical trials recruiting cohorts with early diabetes.

In youth, there is limited previous research about the repeatability of OGTT (7). Libman et al examined a large group of at-risk, treatment-naïve, overweight youth similar to the RISE Pediatric Medication Study sample. They reported poor reproducibility of the OGTT glucoses and, in particular, noted that the 2-hour glucose value had poor test-retest correlations compared to the fasting value (9). In the current study, we also found the largest variation in the 2-hour glucose value and observed the most stability in HbA1c. Consistent with observations in adults, Libman et al found, among youth classified with IGT on the first OGTT, more than 50% would have been classified as NGT based on the second test (9). In contrast, among our youth classified with IGT at screening (n = 53), only 4 (7.5%) would have been classified as NGT based on the second test and 3 as IFG only, possibly explained by the RISE Youth Medication Study requiring a fasting glucose at screening of ≥5.0 mmol/L.

Poor reproducibility of 2-hour glucose values from an OGTT has been well documented in adults (15, 16). In 1 study of individuals from across the glycemic spectrum, among those with IGT on the first OGTT, less than half were confirmed as IGT on the second (17). Similarly, a study of middle-aged Swedish women found that nearly half who were classified as IGT on the first test had NGT based on the results of the second test 2 weeks later (18). In the National Health and Nutrition Examination Survey adult data, variability was higher for the 2-hour glucose compared to fasting glucose and lowest for the HbA1c (4). Thus, they suggested that applying clinical guidelines of a confirmatory test for a diagnosis may result in different prevalence estimates of disease than do epidemiologic studies that use only a single measurement (4). In a Chinese adult population, the reproducibility of the OGTT for diabetes classification was less than 60% even among participants with high HbA1c, BMI, or waist-to-hip ratio (3). Our findings confirm these previous studies in adults, documenting the higher variability in 2-hour glucose measurements relative to fasting plasma glucose and higher variability in both OGTT glucose variables relative to HbA1c measurements (4, 19).

While some clinicians have suggested the OGTT is no longer clinically useful (20), others have maintained that, despite its flaws, including poor reproducibility and patient burden, it remains a valuable tool for the early identification of individuals with diabetes or those with IGT who are at an increased risk of developing diabetes in the future (21). HbA1c has been suggested as a convenient alternative to OGTT as a diagnostic tool within the clinical setting and as a simple screening tool for clinical trials, with some even positing that measures of HbA1c may more accurately identify those at increased risk for diabetes and cardiovascular disease than fasting glucose (22). This work has led others to recommend that using both HbA1c and fasting glucose may be best for identifying adults at risk for diabetes (23-25). Our findings that using HbA1c in combination with fasting and/or 2-hour glucose generally results in more concordant classification than using any single measure alone compliment this previous work. Further, inclusion of the OGTT allows for classification based on glucose tolerance categories (NGT, IGT, diabetes). On the other hand, if the goal is a simple classification of normal, prediabetes, and diabetes, HbA1c alone may be sufficient.

In the current study, the test-retest correlation of HbA1c was stronger than correlations of fasting and 2-hour glucose. While the correlation was similar between youth and adults for HbA1c, it was significantly stronger in youth than adults for fasting and 2-hour glucose. These results stand in contrast to the CV, which showed more variation in youth than adults for HbA1c and less for 2-hour glucose, highlighting the weakness of using the Pearson correlation coefficient alone to assess reproducibility. The higher variation in youth than adults that we observed for HbA1c is consistent with a previous study suggesting HbA1c alone is a poor diagnostic tool in obese youth (26).

Given that we recruited youth and adults with IGT and recently diagnosed type 2 diabetes, RISE was well positioned to examine classification concordance using consecutive OGTT-based and HbA1c measurements. The RISE Consortium identified individuals for participation who were near the cutpoints for classification and, thus, by design may have been more susceptible to discordance in classification. In youth and adults combined, among those with prediabetes results at screening, 74.1% would have confirmed clinically as prediabetes, and of those classified with diabetes at screening, 71.4% would have confirmed as diabetes when using either OGTT or HbA1c for classification [Supplementary Table S2C (13)]. These findings of lack of classification concordance in 19.7% of the youth and 28.0% of the adults using HbA1c or OGTT (Table 3) underscore the limitations of one-time testing for epidemiologic studies and clinical trials, where this approach is commonly used. We attempted to simulate a confirmatory test using the HbA1c as confirmation to the OGTT (requiring OGTT and HbA1c for classification). This resulted in the highest overall classification concordance [367/420 (87.4%)] concordance among all participants) but at the expense of classifying more participants as normal and fewer as diabetes. This strict approach utilizing multiple measures or duplicate tests may be optimal for clinical trials recruiting a very specific patient population. Among adults, the HbA1c criterion alone results in the highest rate of concordance; however, among our sample of dysglycemic youth, including the OGTT improved the concordance rates, possibly due to the greater HbA1c and lower 2-hour glucose variability in youth compared to adults.

The current analysis has a number of strengths and limitations. We performed repeat measurements in a large number of youth and adults in 7 different clinical centers using the same laboratory, thus for the first time in a clinical trial setting an opportunity to compare the reproducibility of the OGTT and HbA1c in these 2 age groups and determine diagnostic concordance of these tests. Further, given the criteria for participating in RISE, individuals with 2-hour glucose values on the first test that were consistent with IGT or diabetes are included in the current analysis, a fitting population to evaluate test-retest variability and concordance when glucose tolerance is disturbed. However, due to study eligibility criteria, the current analyses lack participants who were normoglycemic at screening, which forfeits the opportunity to analyze false-negative outcomes and reduced the sample size considerably. We examined the differences between participants included in the current study and those with only a screening visit and found no significant differences in demographic characteristics, weight, or BMI among youth [Supplementary Table S3 (13)]. Among adults, participants who were eligible to continue to the baseline assessment were on average a year younger (P = .021) and more likely to be non-Hispanic White (37% of screened, 47% of baseline; P < .001) [Supplementary Table S3 (13)]. While we studied youth and adults, our study population of youth was considerably smaller than that of adults and, based on the study designs, did not include many participants in the 20- to 40-year age range, inclusion of whom could have provided a deeper understanding of differences across the age continuum. Taken together, these limitations created by the inclusion criteria somewhat limit the generalizability of our findings. While no intervention took place between the 2 visits, the length of time between the tests and the run-in period may have introduced some variability if any changes in participant behavior or lifestyle change occurred; however, the direction of variability appears to be randomly distributed. Finally, we did not utilize continuous glucose monitoring for screening as this is not a standardized approach to diagnose prediabetes. Future research efforts may consider determining whether continuous glucose monitoring may be more useful than OGTTs in identifying people with prediabetes for clinical studies.

Conclusions

In summary, current criteria for the diagnosis of diabetes are based on values from the fasting glucose, 2-hour OGTT glucose, or HbA1c. Our findings further highlight the potential adverse impact of relying on a single test and indicate that there are some differences between youth and adults. The findings in youth are particularly novel and thus important as studies are designed to develop approaches that could be useful in slowing progression of beta-cell function in this younger population. Our findings also have implications for epidemiological studies, where prevalence estimates could vary, and in clinical trials, where incorrect classification of a glucose-based outcome may result. Thus, we would recommend that duplicate tests be performed or that more than a single parameter, ie, glucose and HbA1c, be utilized for screening and diagnosis in youth and middle-aged adults, especially when classification category is important (eg, for study inclusion or stratification).

Abbreviations

BMI

body mass index

HbA1C

hemoglobin A1C

IFG

impaired fasting glucose

IGT

impaired glucose tolerance

NGT

normal glucose tolerance

OGTT

oral glucose tolerance test

RISE

Restoring Insulin SEcretion

Funding

RISE is supported by grants from the National Institutes of Health (U01-DK-094406, U01-DK-094430, U01- DK-094431, U01-DK-094438, U01-DK-094467, P30-DK-017047, P30-DK-020595, P30-DK-045735, P30-DK-097512, UL1-TR-000430, UL1-TR-001082, UL1-TR-001108, UL1-TR-001855, UL1-TR-001857, UL1-TR-001858, UL1-TR-001863), the Department of Veterans Affairs, and Kaiser Permanente Southern California. Additional financial and material support from the American Diabetes Association, Allergan, Apollo Endosurgery, Abbott Laboratories, and Novo Nordisk A/S was received.

Author Contributions

Members of the RISE Consortium recruited participants and collected study data. A.H.T., S.L.E., and S.E.K. proposed the analysis. A.H.T. performed the analysis, interpreted the data, and wrote the first draft. All members of the writing group interpreted the data and reviewed and edited the manuscript. The RISE Steering Committee reviewed and edited the manuscript and approved its submission. A.H.T. and S.L.E. are the guarantors of this work and, as such, had full access to all of the data and take responsibility for its integrity and the accuracy of the data analysis.

Disclosures

S.A.A. and S.E.K. serve as paid consultants on advisory boards for Novo Nordisk. S.E.K. is a member of a steering committee for a Novo Nordisk–sponsored clinical trial. S.A.A. is a participant in a Novo Nordisk–sponsored clinical trial. K.J.M. holds an investigator-initiated research grant from Novo Nordisk and is currently employed at Eli Lilly at the time of acceptance. No other potential conflicts of interest relevant to this article were reported.

Data Availability

In accordance with the National Institutes of Health Public Access Policy, we continue to provide all manuscripts to PubMed Central, including this manuscript. RISE has provided the protocols to the public through its public website (https://rise.bsc.gwu.edu/web/rise/collaborators). The RISE Consortium abides by the National Institute of Diabetes and Digestive and Kidney Diseases data sharing policy and implementation guidance as required by the National Institutes of Health/National Institute of Diabetes and Digestive and Kidney Diseases (https://repository.niddk.nih.gov/studies/rise/).

Prior Presentation

Parts of this study were presented in June 2019 at the 79th Scientific Sessions of the American Diabetes Association, San Francisco, CA.

Clinical Trial Registry

• RISE Adult Medication Study (RISE Adult), NCT01779362

• RISE Pediatric Medication Study (RISE Peds), NCT01779375

• Beta Cell Restoration Through Fat Mitigation (BetaFat), NCT01763346.