Abstract
Methods
Cross-sectional CGM data from 100 individuals with T2D were collected over 4 h following a standardized meal consumed twice. Dynamic PPGR features—glucose peak, incremental area under the curve (iAUC), rise and fall rates, final vs. fasting glucose—were used for K-Means clustering, with stability assessed using a Random Forest classifier trained on the first meal. In 50 participants, postprandial plasma glucose and insulin were measured, and clinical/metabolic parameters compared across clusters using one-way ANOVA.
Results
Three CGM-defined PPGR clusters were identified. Cluster 1 (n = 19) showed the highest peak and iAUC, with post-meal glucose remaining persistently above baseline. Cluster 2 (n = 56) and 3 (n = 25) had lower peaks and iAUCs, but Cluster 3 exhibited higher rise and fall rates than Cluster 2. Clusters did not differ in age, sex, BMI, or diabetes duration, but metformin use was lower in Cluster 3. Cluster 1 showed significantly lower insulin secretion (HOMA2-B%: 77.42 ± 25.64 vs. 104.96 ± 43.94) and higher insulin resistance (HOMA-IR: 7.94 ± 3.27 vs. 4.84 ± 2.78) than Cluster 3, with intermediate values for Cluster 2, confirmed by postprandial indices. Cluster 3 had a higher early insulin response than Cluster 1 and 2 (60-min insulinogenic index: 1.67 ± 1.07, 0.84 ± 0.31, 0.84 ± 0.58, respectively; p < 0.05).
CGM-derived PPGR features could identify T2D subtypes with similar clinical profiles but distinct insulin secretion and sensitivity impairments, supporting targeted interventions.