Repeated OGTT Versus Continuous Glucose Monitoring for Predicting Development of Stage 3 Type 1 Diabetes: A Longitudinal Analysis Free

Presymptomatic type 1 diabetes is defined by the presence of multiple (two or more) islet autoantibodies (mAAbs), conferring a 90% 20-year risk of symptomatic (stage 3) disease both in children and adults (1–3). This presymptomatic phase starts with normoglycemia (stage 1) but progresses to dysglycemia (stage 2) as clinical onset approaches (2). However, the duration and course of these asymptomatic phases vary greatly. Consensus guidelines recommend counseling and metabolic monitoring in individuals who are AAb⁺ (4), facilitating early diagnosis and treat-to-target insulin therapy of stage 3 (5), reducing the incidence of inaugural diabetic ketoacidosis (6), and providing opportunities for early interventions (7,8). However, no disease-modifying therapies for early-diagnosed stage 3 type 1 diabetes have been approved beyond research settings. Screening for presymptomatic type 1 diabetes, through both familial and population-based programs, has gained momentum with the U.S. Food and Drug Administration approval of teplizumab, the first therapy that can delay insulin dependence during stage 2 (9). Therefore, disease staging and monitoring will be necessary for a growing number of individuals with asymptomatic disease (4).

Presymptomatic disease staging is currently based on findings from oral glucose tolerance tests (OGTTs) and increasing HbA_1c values (2,10). OGTT-derived variables of glucose dysregulation and insulin secretion, integrating C-peptide, can refine predictions of imminent stage 3 type 1 diabetes (11–13). In addition to time constraints of families and caregivers, OGTTs require overnight fasting and intravenous access, and may provoke nausea, limiting adherence to repeat testing (14), particularly in younger children. A 10% rise in HbA_1c is a specific indicator of impending clinical onset (15,16) but can be affected by nondiabetes-related factors (4).

Several CGM metrics, particularly time ≥140 mg/dL (7.8 mmol/L) during the daytime, have been shown to predict disease progression in cross-sectional studies in at-risk children (17–22) and cohorts including adults (23–25), especially when combined with age and AAb type or number (25). This highlights their potential for home-based monitoring of individuals who are AAb⁺ (4). In smaller studies, CGM metrics outperformed glycemic variability assessed by self-monitoring of blood glucose (18,23) and performed similarly to hyperglycemic clamp–derived variables, the gold standard for β-cell function (23). However, a larger TrialNet study found that OGTT-derived metrics outperformed CGM, which is deemed important if intended as entry or outcome criteria for disease-modifying therapies, especially in research settings (26). Given the limited availability of longitudinal CGM studies in presymptomatic type 1 diabetes (19,22,26), the aim of the current study was to compare the diagnostic performance of serial CGM, HbA_1c, and OGTT metrics in predicting clinical onset, based on longitudinal data collected by the Belgian Diabetes Registry (BDR), over a wide age and time range.

Data between March 2012 and April 2022 were obtained from a longitudinal multicenter metabolic monitoring study (NCT01402037) coordinated by the BDR. It was conducted in accordance with the Declaration of Helsinki (2013 revision) and approved by the ethics committees of Vrije Universiteit Brussel (VUB)-Universitair Ziekenhuis (UZ) Brussel (BUN143201422342) and participating centers. Informed consent or assent was obtained from all participants and from the legal representatives of minors.

First-degree relatives (FDRs) aged 5–39 years with confirmed mAAb (insulin autoantibody, GAD autoantibody [GADA], IA-2 autoantibody [IA-2A], and/or zinc transporter 8 autoantibody [ZnT8A]) positivity could participate, unless pregnant or lactating, treated with immunomodulating or diabetogenic medication, or having a relevant medical history according to the investigator. Participants were monitored semiannually for ≥2 years or until clinical diagnosis or discontinuation of participation. Collected data included age, weight, height, AAb profile, HbA_1c, and OGTT, followed by 5 days of iPro2 CGM (Medtronic). A repeat OGTT was proposed after 3 months for dysglycemia or within 2 weeks for hyperglycemia. Participants were asked to maintain their regular diet and activities during CGM (23). OGTT involved intravenous sampling for glucose and C-peptide before and 15, 30, 60, 90, and 120 min after an oral glucose load of 1.75 g/kg (maximum 75 g) after an overnight fast (13,23). The CGM device was placed at the OGTT visit, with participants blinded to CGM results. Participants logged sleep and wake times, dietary intake, activity, and medication and measured at least three preprandial values and one pre-bedtime glucose value per day with the provided Contour Next Link glucometer (Bayer) for calibration (23).

The cohort was selected based on mAAb positivity, no overweight (BMI SD score [SDS] <2), at least one baseline nonhyperglycemic OGTT with aligned CGM data, and a minimum of one subsequent study visit. Data were obtained from 319 OGTTs and 216 associated CGM recordings in 34 participants, excluding 11 diagnostic and 7 confirmatory OGTTs. Progression status and follow-up time were updated for 19 relatives included in our previous cross-sectional analysis (23). CGM data were unavailable after October 2020 due to discontinuation of the iPro2 system.

Dysglycemia during OGTT was defined by a plasma glucose level of 100–125 mg/dL (5.6–6.9 mmol/L) after fasting, 140–199 mg/dL (7.8–11.0 mmol/L) at 2-h postglucose load (T120), or ≥200 mg/dL (11.1 mmol/L) at intermediate time points (2,10). Having two or more consecutive abnormal OGTTs was considered as stage 2 type 1 diabetes. Participants were referred to clinical care for insulin therapy at diagnosis of stage 3 per American Diabetes Association criteria (10). The date of the first hyperglycemic OGTT was used as time of diagnosis if confirmed on repeat OGTT. Progression status during and after study discontinuation was verified through repeated contacts with the relatives and Belgian diabetologists, self-reporting, and a link with the BDR patient database for patients with new-onset disease (13). Participants diagnosed with stage 3 type 1 diabetes at a study visit or within 2 years of their last visit were considered progressors (n = 17). The others (n = 17) were considered nonprogressors and censored at their last study visit with available OGTT data (Supplementary Fig. 1). The median follow-up time of the entire cohort was 3.5 (interquartile range [IQR] 2.0–7.5) years.

Whole blood, serum, and plasma samples were stored frozen and analyzed centrally at the VUB-UZ Brussel clinical biology laboratory. Supplementary Table 1 lists previously described methods for AAbs, HLA-DQ polymorphisms, HbA_1c, plasma glucose and C-peptide, and relevant external quality-control schemes.

BMI-SDSs were calculated according to Flemish growth curves (27). C-peptide and glucose areas under the curve (AUCs) were calculated using the trapezoidal rule (13,23) and expressed per minute. C-peptide was expressed in nmol/L, and all glucose variables in mg/dL, except for ratios where C-peptide and glucose were converted to pmol/L and mmol/L (per minute for AUC), respectively, necessitating a correction factor of 10⁻⁹.

CGM data were included starting from the first evening meal after sensor placement, as CGM measurements were deemed unreliable in the hours immediately following the OGTT. Recordings were completely discarded if <96 h of data remained after processing (17,19,20,23,26) (n = 16 of 216). Entire 24-h periods were excluded for >20% missing CGM data or fewer than three valid calibrations (22 recordings with 96 h of usable data among the remaining 200 recordings). Only the missing records were excluded from days with <20% missing data (applied in eight 5-day recordings). CGM metrics were computed using GlyVarT version 1.0 (Medtronic Bakken Research Center) (23) and included mean sensor glucose; percentage of time above, below, and between various glycemic cutoff values; glucose IQR; overall SD; and coefficient of variation (28). Daytime was defined as the time between awakening and bedtime, as documented in the log books. Metrics were computed overall and for day- and nighttime. Overall metrics were highly correlated with daytime metrics (Spearman r range of 0.83–0.97, except for time <54 mg/dL [3.0 mmol/L], where r = 0.65). Weaker, but still significant correlations were found between overall and nighttime metrics (r range of 0.45–0.91). Overall and daytime metrics performed similarly. We report the most simple, overall metrics.

SPSS version 29.0 (IBM Corporation), Stata 18 (StataCorp LLC) and GraphPad Prism 10 (GraphPad Software) statistical software were used. Two-tailed tests were performed, with P < 0.05 considered significant. No adjustments were made for multiple testing. Baseline differences between progressors and nonprogressors for unpaired continuous and categorical data were assessed by Mann-Whitney U and χ² tests. Correlations between CGM variables were assessed on all recordings by Spearman rank order tests.

To identify rapid progressors, we used a subset of 27 participants followed for at least 3 years from baseline without development of stage 3 type 1 diabetes (n = 19) or who developed stage 3 type 1 diabetes within 3 years (n = 8). Logistic regression and receiver operating characteristic (ROC) analyses (29) were used to evaluate the individual performance of baseline HbA_1c, fasting and OGTT-derived variables, and CGM metrics as continuous predictors of rapid onset of stage 3 type 1 diabetes, as well as combinations of two CGM metrics. ROC AUCs with 95% CIs were calculated (29), and the cutoff for optimal sensitivity and specificity was identified using the Youden index (30). Positive predictive value (PPV) and negative predictive value (NPV) were based on the stage 3 prevalence of 29.6% (8 of 27 participants) at 3 years. The diagnostic efficiency was calculated as the fraction of correctly classified individuals (those with true positive and true negative results) among the 27 participants in this subgroup (13). In the entire cohort (n = 34), stage 3 diabetes-free survival from baseline was assessed using the Kaplan-Meier method, with log-rank comparison of survival between groups stratified according to values below or above ROC-derived cutoff values.

Time to stage 3 type 1 diabetes was analyzed using baseline Cox proportional hazards (PH) models (31) with age, BMI, BMI-SDS, HbA_1c, fasting and OGTT-derived variables, and CGM metrics as continuous individual predictors and sex and AAb positivity as categorical covariates. The PH assumption was checked based on Schoenfeld residuals and log-log survival plots.

Significant (P < 0.05) OGTT-derived univariable predictors were entered for conditional forward selection in a multivariable model. C-peptide/glucose ratio, whether fasting or stimulated, was excluded from entry in multivariable models to avoid overfitting, as it was considered an interaction between covariates. Similarly, conditional forward selection was used to assess whether combined CGM metrics would improve model performance and fit. The predictive performance of models without time-varying covariates was assessed by Harrell's concordance index. To calculate the 95% CI, 1,000 bootstrap replications were used.

In progressors and nonprogressors, absolute values of HbA_1c, OGTT, and CGM metrics were plotted as a function of time to development of stage 3 type 1 diabetes or the last visit, respectively. The plots also show the available OGTT and HbA_1c data without associated CGM recordings but exclude data obtained at diagnosis of stage 3 type 1 diabetes or from confirmation OGTT.

Following transformation of longitudinal data into a counting process data format and ensuring lack of missing data in 197 OGTTs with subsequent CGM, Cox PH regression with time-varying covariates in multiple record data (32) was performed with robust variance estimation to adjust for correlations between intraindividual observations. Missing HbA_1c data (n = 3) were imputed by averaging the values of the preceding and following visit. Conditional forward selection of significant predictors (P < 0.05) was performed to determine a multivariable model combining only fasting parameters and HbA_1c, a model containing OGTT metrics and HbA_1c, a model with CGM metrics alone, and a model combining CGM and HbA_1c.

The goodness of fit of all regression models was tested using the corrected Akaike information criterion (AICc) (33), which represents an estimate of information not explained by the model and includes a correction for small sample sizes. A smaller AICc indicates a better fit.

The data set is available from the corresponding author upon reasonable request.

In this cohort of 34 FDRs who were not overweight, were mAAb⁺, and had a median age of 16.6 (IQR 13.4–23.4) years at baseline, 17 progressed to stage 3 type 1 diabetes after a median follow-up of 40 (IQR 20–91) months. Stage 2 type 1 diabetes was observed in 13 progressors (2 at baseline and 11 during follow-up) and 2 nonprogressors (Supplementary Fig. 1).

Progressors (n = 17) and nonprogressors (n = 17) did not significantly differ in follow-up time, sex distribution, or HLA-inferred risk (Supplementary Table 2). All but one participant tested positive for IA-2A and/or ZnT8A at baseline. All participants remained positive for at least one islet AAb type throughout the study. At baseline, progressors had lower BMI and BMI-SDS, stimulated C-peptide levels, and C-peptide/glucose ratios compared with nonprogressors. No differences were found for baseline age, CGM metrics, or HbA_1c according to overall progression status (Supplementary Table 2).

Supplementary Table 3 provides baseline characteristics of 19 nonprogressors for at least 3 years and those who developed stage 3 type 1 diabetes within 3 years (8 rapid progressors with a median follow-up time of 20 [IQR 12–26] months). At baseline, rapid progressors had higher glucose levels during OGTT (P = 0.004) and increased CGM-derived SD (P = 0.017), IQR (P = 0.029), and mean glucose levels (P = 0.004) compared with nonprogressors but similar C-peptide or HbA_1c values. In rapid progressors, glycemia was 4.3% of the time ≥140 mg/dL (7.8 mmol/L) and 14.5% of the time ≥120 mg/dL (6.7 mmol/L) compared with 0.0% and 1.1% in nonprogressors, respectively (all P ≤ 0.001).

Using ROC analyses, we compared the diagnostic performance of fasting and glucose-stimulated OGTT-derived variables, HbA_1c, and CGM metrics as standalone markers for prediction of clinical onset within 3 years from baseline. Significant ROC AUCs (lower limit of 95% CI >0.5) ranged between 0.75 and 0.86 for OGTT metrics and 0.77 and 0.92 for CGM metrics (Fig. 1A and ROC curves in Supplementary Fig. 2). Similar ROC AUCs were observed, albeit with larger CIs, when excluding the two rapid progressors with baseline stage 2 type 1 diabetes (data not shown). The best predictors for rapid progression were OGTT-derived AUC glucose and CGM-derived mean glucose, time ≥140 mg/dL (7.8 mmol/L), and time ≥120 mg/dL (6.7 mmol/L), reaching ROC AUCs >0.85 (all P < 0.001) and the lowest AICc values (data not shown). HbA_1c alone was not significant (P = 0.233), but the combination with fasting glucose improved predictive performance (P = 0.008; ROC AUC = 0.79). Combining two CGM metrics did not improve ROC AUC beyond 0.92 (data not shown). The Youden cutoff of 7.8% for time ≥120 mg/dL glucose (6.7 mmol/L) achieved 100% specificity and PPV, with 75% sensitivity and 91% NPV (Supplementary Table 4). In comparison, 0.6% of time ≥140 mg/dL (7.8 mmol/L) offered higher sensitivity and NPV (88% and 94%, respectively) but lower specificity and PPV (84% and 70%, respectively) (Supplementary Table 4). Higher cutoff values of 3.0% or 5.0% for time ≥140 mg/dL (7.8 mmol/L) improved specificity to 100% but reduced sensitivity to 63% and 38%, respectively (data not shown).

The median survival time in the full cohort was 88 months (Fig. 1B). Applying the ROC-derived cutoff values for optimal prediction of impending stage 3 type 1 diabetes (Supplementary Table 4) to the full cohort’s baseline metabolic assessments (Fig. 1C–K) effectively distinguished faster progressors from slower progressors in Kaplan-Meier survival analyses for almost all variables considered significant in the ROC analyses (Supplementary Table 4). Exceptions were CGM-derived IQR and SD (Fig. 1D and F). For participants with glucose levels above (n = 23) or below (n = 11) 101 mg/dL (5.6 mmol/L) at T120 of the OGTT, the median stage 3–free survival time was 57 vs. 112 months, respectively (P = 0.010) (Fig. 1E). For FDRs with AUC glucose above (n = 16) or below (n = 18) 127 mg/dL · min (7.1 mmol/L · min), median stage 3–free survival time was 40 vs. 98 months (P = 0.010) (Fig. 1H). For participants with AUC C-peptide/glucose ratio below (n = 8) or above (n = 26) 226 × 10⁻⁹ median stage 3–free survival time was 20 months vs. 98 months (P < 0.001) (Fig. 1G). Median survival time was 27 months in participants with CGM-derived mean glucose levels ≥96 mg/dL (5.3 mmol/L) (n = 11 of 34), interstitial glucose levels ≥120 mg/dL (6.7 mmol/L) for ≥7.8% of the time (n = 10 of 34), or ≥140 mg/dL (7.8 mmol/L) for ≥0.6% of the time (n = 14 of 34) compared with 98 months for the other relatives (all P ≤ 0.001) (Fig. 1I–K).

In univariable Cox PH models, fasting glucose (concordance = 0.69), glucose at T120 (concordance = 0.73), and glucose AUC (concordance = 0.78) were confirmed as the best OGTT-derived significant predictors for stage 3 type 1 diabetes development from baseline (Table 1). GADA positivity (concordance = 0.57) and higher C-peptide/glucose ratios, whether at T120 (concordance = 0.68) or as AUC (concordance = 0.78), were significantly associated with a reduced risk of developing stage 3 type 1 diabetes.

CGM-derived time ≥120 mg/dL (6.7 mmol/L) (concordance = 0.73) and time ≥140 mg/dL (7.8 mmol/L) (concordance = 0.74) performed similarly to stimulated OGTT-derived variables, whereas glucose SD (concordance = 0.68), IQR (concordance = 0.66), and time ≥160 mg/dL (8.9 mmol/L) (concordance = 0.64) matched with fasting glucose (concordance = 0.69) (Table 1). Using a conditional forward approach, combinations of OGTT-derived variables (Table 1, model A) or CGM metrics (Table 1, model B) or adding GADA positivity (data not shown), could not further improve model performance and fit. AUC glucose and time ≥140 mg/dL (7.8 mmol/L) remained the best selected baseline predictors for OGTT and CGM, respectively.

Longitudinal spaghetti plots (Fig. 2) revealed that despite notable intra- and interindividual variation, stimulated glucose during OGTT (rows 2 and 3), HbA_1c (row 6), CGM-derived time ≥120 mg/dL (6.7 mmol/L) (row 9), and time ≥140 mg/dL (7.8 mmol/L) (row 10) generally increased in progressors within the last 3 years before their diagnosis compared with most nonprogressors. This rise was accompanied by a decrease in AUC C-peptide and AUC C-peptide/glucose ratio (rows 4 and 5). More subtle differences between both groups were observed for fasting glycemia (row 1), CGM-derived mean glucose (row 7), and IQR (row 8). Excursions of increased CGM-derived glycemic variability did not always correspond with either persistent or transient dysglycemia on OGTT, although their association became more apparent on study visits closer to the end of follow-up (rows 7–10).

We extended our Cox PH model to include longitudinal, complete data of 197 OGTTs with subsequent CGM from the 34 relatives who were mAAb⁺ and adjusted for correlations between intraindividual observations. HbA_1c now also emerged as a significant individual predictor for progression to stage 3 type 1 diabetes (AICc = 80.4) (Table 2). Fasting glucose (AICc = 87.6) was outperformed by stimulated glucose at T120 (AICc = 77.4) and glucose AUC (AICc = 71.1), as indicated by the lower AICc. Fasting glucose was not withheld in a multivariable model assessing its combination with HbA_1c (model A) or in the model combining OGTT-derived variables and HbA_1c (model B). The interaction between glucose dysregulation and insulin secretion is illustrated by the low AICc values for models using the stimulated C-peptide/glucose ratio (T120: AICc = 64.8; AUC: AICc = 60.4) as predictors and the inclusion of a glucose and C-peptide parameter in multivariable OGTT + HbA_1c models (model B: AICc = 62.7; model B [replacing glucose at T120 by AUC glucose]: AICc = 58.9).

In univariable prediction models, CGM-derived mean glucose, SD, IQR, time ≥120 mg/dL (6.7 mmol/L), and time ≥140 mg/dL (7.8 mmol/L) achieved similar AICc values to or lower AICc values (range 75.1–79.9) than HbA_1c (AICc = 80.4) or OGTT-derived glucose at T120 (AICc = 77.4). The goodness of fit of the best univariable CGM model, time ≥120 mg/dL (6.7 mmol/L) (AICc = 75.1), did not match that of OGTT-derived AUC glucose (AICc = 71.1). The latter was almost equal to multivariable CGM model C (AICc = 72.5), which selected mean glucose and IQR out of all significant CGM metrics using conditional forward selection. An even better model (AICc = 68.4) was obtained by combining CGM metrics with HbA_1c (Table 2, model D).

In this first in-depth longitudinal analysis to our knowledge, repeated intermittent CGM metrics, especially when combined with HbA_1c, were nearly as effective as repeated OGTTs for predicting progression to stage 3 type 1 diabetes in FDRs who are mAAb⁺. Rising trends in CGM metrics close to onset parallelled changes in OGTT-derived variables, though both showed considerable intra- and interindividual variability over time and some between-method inconsistencies. For cross-sectional predictions of clinical onset, CGM metrics equaled OGTT-derived variables, outperforming HbA_1c.

To our knowledge, longitudinal patterns of CGM metrics and similar temporal fluctuations have only been previously illustrated in a small cohort of children who were AAb⁺ (n = 16 of 23) (19) or over a limited 6-month period (22,26). By using data of adults and children with presymptomatic type 1 diabetes over a longer time span, we observed that the increasing trend in CGM-derived glycemic variability paralleled changes in OGTT-derived variables, consistent with findings from longitudinal OGTT studies (34–37). The baseline lower stimulated C-peptide release in progressors may reflect prolonged preexisting β-cell dysfunction, as previously described (37). Our longitudinal prediction models indicated that repeated CGM recordings alone provide less predictive information compared with repeated OGTTs, aligning with TrialNet’s primarily cross-sectional report (26). Despite the limited utility of baseline HbA_1c, serial HbA_1c measurements improved the predictive ability of repeated intermittent CGM, confirming evidence for the good predictive value of rising HbA_1c (15,16).

Our baseline analyses support earlier findings (17–26) that single CGM metrics, such as time ≥140 mg/dL (7.8 mmol/L) and ≥120 mg/dL (6.7 mmol/L), effectively predict clinical onset. Cross-sectionally, our best individual CGM metrics performed similar to OGTT-derived variables, contrary to a larger TrialNet report that favored OGTT (26). Combining CGM metrics did not improve predictive value, likely due to their collinearity as previously demonstrated in the Autoimmunity Screening for Kids (ASK) cohort (20) but contradicting findings from the Fr1da study (22). Most studies have suggested cutoff values ranging between 5 and 18% of time ≥140 mg/dL (7.8 mmol/L) (17–21,23–26) for predicting impending stage 3 type 1 diabetes. Our cutoff values and observed ranges for various CGM metrics were generally lower, particularly compared with Fr1da data (22), but nonetheless, effectively stratified participants into progressors and nonprogressors, similar to previous studies (20,22,24). This might relate to population characteristics such as age, weight, dietary habits, and use of different CGM systems. Moreover, we defined rapid progression as development of clinical onset within 3 years after metabolic testing, an interval deemed relevant for prevention studies, and derived Youden-based cutoff values with optimal sensitivity and specificity. However, metrics and/or cutoff values with high(er) diagnostic specificity may be preferred for participation in intervention studies, while those with high sensitivity may be used for prevention of diabetic ketoacidosis.

We included FDRs who were not overweight and had persistent mAAb⁺ (without considering cytoplasmic islet cell antibodies), leading to a relatively homogeneous cohort comprising both adults and children. This is relevant because most patients develop stage 3 type 1 diabetes in adulthood (3), while adherence to repeat OGTTs is especially challenging in children (14). Limitations are the predominant European-Caucasian descent of the cohort and potential selection bias by excluding individuals who were overweight, warranting further studies. The start of metabolic follow-up did not align with the moment of seroconversion, reflecting real-world scenarios. Although our cohort consisted solely of FDRs and was smaller in size (n = 34) than TrialNet’s cohort (n = 93; average of 2.3 CGMs over a maximum of 12 months) (26), our follow-up time, and number of prediagnostic CGM recordings with concomitant OGTT (on average 5.8 per participant) exceeded that of other CGM studies. Consequently, we identified a high ratio of progressors to stage 3 type 1 diabetes, whereas the progression rate to stage 2 may have been underestimated by requiring confirmed dysglycemia (38). Nonetheless, disease heterogeneity and variable follow-up times prevented observation of progression in all participants, despite all being expected to progress at some point. At least two nonprogressors were classified as having stage 2 type 1 diabetes at their last visit. Their spaghetti plots resembled those of progressors, although no progression to stage 3 was recorded within the next 2 years. A key strength is the consistency in using the same central laboratory and the same blinded CGM system throughout this multicenter study. However, the iPro2 system is now obsolete and was discontinued by Medtronic toward the end of the study, leading to missing CGM data before the last follow-up visit in some nonprogressors. Hence, confirmation studies are needed using blinded, present-day, factory-calibrated CGM devices, capturing data over longer, more representative periods (22,39). Variability between 10-day Dexcom G6 recordings was recently reported (22). Finally, the extended Cox models could not be adjusted for clinical or immunologic parameters because of partially missing data. Compliance with the event-to-variable ratio of Vittinghoff and McCulloch (40) also prevented the building of more complex models, including a model combining OGTT and CGM metrics.

In conclusion, intermittent CGM, especially when combined with HbA_1c, may serve as a minimally invasive alternative to OGTTs for long-term monitoring of individuals with presymptomatic diabetes. Further validation through more diverse, longitudinal studies incorporating newer CGM technologies and/or increased CGM frequency is needed, alongside direct comparisons of psychological impact, user acceptance, practical implementation, and cost-effectiveness. Further research should focus on predicting stage 2 type 1 diabetes and assessing the need and timing of confirmatory tests when considering teplizumab for stage 2, as most participants in the TrialNet study named Teplizumab for Prevention of Type 1 Diabetes in Relatives “At Risk” (TN10) (n = 68 of 76) had confirmed dysglycemia before randomization (9,38). Based on the variability in test results of both OGTT and CGM (22,24), and in disease progression, we recommend confirmatory testing before initiating interventions. While OGTTs, especially with integrated C-peptide measurements, remain valuable in clinical trials and for assessing disease stage for disease-modifying therapies, CGM and HbA_1c may be particularly useful for young children and in ambulatory settings.

Acknowledgments. The authors sincerely thank all participants and their family members and the study teams for their contributions to the recruitment and follow-up in the multicenter study (Supplementary Appendix A). The authors thank the researchers and coworkers of the central unit and the clinical biology reference laboratory of the BDR, the Department of Clinical Chemistry and Radio-immunology at UZ Brussel, and the Diabetes Research Center at VUB (Supplementary Appendix B) for involvement in central study coordination and technical assistance. The authors also sincerely thank all members of the BDR who have contributed to the screening of relatives in Belgium and their teams, particularly those who contributed to this cohort. The authors thank Å. Lernmark (formerly of the University of Washington), M. Christie (King’s College School of Medicine and Dentistry), and the late J.C. Hutton (Barbara Davis Center for Childhood Diabetes) for gifts of cDNA for the preparation of the radioligands used to measure GADA, IA-2A, and ZnT8, respectively. The authors thank A. Arrieta (Medtronic Bakken Research Center BV) for developing the GlyVarT program. Last, but not least, the authors are grateful to W. Cools (Biostatistics and Medical Informatics Research Group, VUB) for statistical advice.

Funding. Medtronic provided the iPro2 devices and Enlite sensors free of charge, and the Contour Link glucometers were donated by Bayer. This study was supported by Fonds Wetenschappelijk Onderzoek - Vlaanderen (FWO) project grant G.0868.11 and junior research fellowships 1SD8122N (to A.K.D.) and 11D1214N (to A.V.D.) and by Agentschap voor Innovatie door Wetenschap en Technologie (IWT) grant 130 138 and Strategic Research Programs SRP42-Growth and SRP55-Spearhead. The BDR received funding from the Center for Medical Innovation Flanders and by the Wetenschappelijk Fonds Willy Gepts, UZ Brussel. W.S. holds FWO senior clinical investigator grant 1806421N and BreakthroughT1D career development award CDA-2024-1491-S-B.

Medtronic and Bayer were not involved in the study design or in the collection, analysis, and interpretation of data; the writing of the report; or the decision to submit the manuscript for publication.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.K.D., B.K., U.V.d.V., A.V.D., B.L., C.D.B., P.G., N.S., E.V.B., W.S., S.V.A., M.d.B, S.D., J.M., H.K., and F.K.G. contributed to recruitment and/or critical discussions. A.K.D., B.K., and F.K.G. designed the study. A.K.D., U.V.d.V., E.V.B., and E.R.V.V. retrieved, statistically analyzed, and interpreted data. A.K.D. and F.K.G. wrote the first draft of the manuscript. All authors critically reviewed and edited the manuscript and approved the final version. A.K.D. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in poster form at the 20th Immunology of Diabetes Congress, Bruges, Belgium, 4–8 November 2024.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Cheryl A.M. Anderson and Rodica Pop-Busui.