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

Abstract

Objective: Evidence for using continuous glucose monitoring (CGM) as an alternative to oral glucose tolerance tests (OGTTs) in presymptomatic type 1 diabetes is primarily cross-sectional. We used longitudinal data to compare the diagnostic performance of repeated CGM, HbA1c, and OGTT metrics to predict progression to stage 3 type 1 diabetes.

Research design and methods: Thirty-four multiple autoantibody-positive first-degree relatives (FDRs) (BMI SD score [SDS] <2) were followed in a multicenter study with semiannual 5-day CGM recordings, HbA1c, and OGTT for a median of 3.5 (interquartile range [IQR] 2.0-7.5) years. Longitudinal patterns were compared based on progression status. Prediction of rapid (<3 years) and overall progression to stage 3 was assessed using receiver operating characteristic (ROC) areas under the curve (AUCs), Kaplan-Meier method, baseline Cox proportional hazards models (concordance), and extended Cox proportional hazards models with time-varying covariates in multiple record data (n = 197 OGTTs and concomitant CGM recordings), adjusted for intraindividual correlations (corrected Akaike information criterion [AICc]).

Results: After a median of 40 (IQR 20-91) months, 17 of 34 FDRs (baseline median age 16.6 years) developed stage 3 type 1 diabetes. CGM metrics increased close to onset, paralleling changes in OGTT, both with substantial intra- and interindividual variability. Cross-sectionally, the best OGTT and CGM metrics similarly predicted rapid (ROC AUC = 0.86-0.92) and overall progression (concordance = 0.73-0.78). In longitudinal models, OGTT-derived AUC glucose (AICc = 71) outperformed the best CGM metric (AICc = 75) and HbA1c (AICc = 80) (all P < 0.001). HbA1c complemented repeated CGM metrics (AICc = 68), though OGTT-based multivariable models remained superior (AICc = 59).

Conclusions: In longitudinal models, repeated CGM and HbA1c were nearly as effective as OGTT in predicting stage 3 type 1 diabetes and may be more convenient for long-term clinical monitoring.

Graphical abstract

Figure 1

Baseline ROC and survival analyses. A: ROC AUCs for prediction of rapid development of stage 3 type 1 diabetes (T1D) in ascending order. CGM metrics are indicated with a †. B: Survival for overall progression to stage 3 T1D in the entire cohort (n = 34). Tick marks indicate censoring. C–K: Kaplan-Meier curves (95% CIs), stratified according to values below (blue) or above (red) the Youden-based cutoff for the significant predictors. Log rank P values are shown in the lower left corners of each graph. *P < 0.50. ROC curves and cutoff values and their sensitivity, specificity, PPV, NPV, and diagnostic efficiency are shown in Supplementary Fig. 2 and Supplementary Table 4. CV, coefficient of variance.

Figure 2

Longitudinal spaghetti plots. Evolution of OGTT variables (rows 1–5), HbA_1c (row 6), and CGM metrics (rows 7–10) in progressors to stage 3 type 1 diabetes (T1D) (n = 17) vs. non- or slow progressors (n = 17) as a function of time to stage 3 T1D or last visit, respectively. OGTT metrics are fasting glucose (row 1), glucose at T120 (row 2), AUC glucose (row 3), AUC C-peptide (nmol/L · min) (row 4), and the ratio of AUC C-peptide and AUC glucose (row 5). CGM metrics are mean glucose levels (row 7), IQR (row 8), percentage of time ≥120 mg/dL (6.7 mmol/L) (row 9), and percentage of time ≥140 mg/dL (7.8 mmol/L) (row 10). Glucose levels (rows 1–3, 7–8) and HbA_1c (row 6) are shown in mg/dL (per minute for AUC) and percentage, respectively, on the left y-axes and in mmol/L (*) (per minute for AUC) and mmol/mol (‡), respectively, on the right axes. For the ratio of C-peptide and glucose, values are expressed in pmol/L and mmol/L (per minute for AUC), requiring a correction factor of 10⁻⁹ (†). Observations with concomitant dysglycemia on OGTT are indicated in red. Values at diagnosis or confirmation OGTT were excluded.