Background: Poor data availability and accessibility characterizing some research areas in biomedicine are still limiting potentialities for increasing knowledge and boosting technological advancement. This phenomenon also characterizes the field of diabetes research, in which glycemic data may serve as a basis for different applications. To overcome this limitation, this review aims to provide a comprehensive analysis of the publicly available data sets related to dynamic glycemic data. Methods: Search was performed in four different sources, namely scientific journals, Google, a comprehensive registry of clinical trials and two electronic databases. Retrieved data sets were analyzed in terms of their main characteristics and on the typology of data provided. Results: Twenty-five data sets were identified including data from challenge tests (5 of 25) or data from Continuous Glucose Monitoring (CGM, 20 of 25). As for the data sets including challenge tests, all of them were freely downloadable; most of them (80%) related only to oral glucose tolerance test (OGTT) with standard duration (2 h), but varying for timing and number of collected blood samples, and variables collected in addition to glucose levels (with insulin levels being the most common); the remaining 20% of them also included intravenous glucose tolerance test (IVGTT) data. As for the data sets related to CGM, 7 of 20 were freely downloadable, whereas the remaining 13 were downloadable upon completion of a request form. Conclusions: This review provided an overview of the readily usable data sets, thus representing a step forward in fostering data access in diabetes field.
This study exploited the skin temperature signal derived from a wrist-worn wearable device to define potential digital biomarkers for glycemia levels. Characterization of the skin temperature signal measured through the Empatica E4 device was obtained in 16 subjects (data taken from a dataset freely available on PhysioNet) by deriving standard metrics and a set of novel metrics describing both the current and the retrospective behavior of the signal. For each subject and for each metric, values that correspond to when glycemia was inside the tight range (70–140 mg/dL) were compared through the Wilcoxon rank-sum test against those above or below the range. For hypoglycemia characterization (below range), retrospective behavior of skin temperature described by the metric CVT SD (standard deviation of the series of coefficient of variation) proved to be the most effective both in daytime and nighttime (100% and 50% of the analyzed subjects, respectively). On the other side, for hyperglycemia characterization (above range), differences were observed between daytime and nighttime, with current behavior of skin temperature, described by M2T (deviation from the reference value of 32 °C), being the most informative during daytime, whereas retrospective behavior, described by SDT hhmm (standard deviation of the series of means), showed the highest effectiveness during nighttime. Proposed variability features outperformed standard metrics, and in future studies, their integration with other digital biomarkers of glycemia could improve the performance of applications devoted to non-invasive detection of glycemic events.
Nocturnal hypoglycemia (NH) following exercise represents a critical challenge in the management of type 1 diabetes (T1D), particularly in pediatric populations, where its occurrence is associated with severe adverse outcomes and increased caregiver burden. This study aimed to identify an interpretable early signature based on CGM-derived digital biomarkers of post-exercise NH risk in children and adolescents with T1D. CGM data from 49 pediatric subjects (DirecNet cohort) were used to extract several CGM metrics across two temporal configurations: (i) Exercise + Cumulative, where features were computed over the exercise window and over an extended window spanning from exercise onset through recovery (16:00–17:00 and 16:00–22:00); and (ii) Exercise + Post-exercise, where features were computed separately over two non-overlapping intervals, capturing the exercise phase and the subsequent recovery phase (16:00–17:00 and 17:00–22:00). A Random Forest classifier was trained within a Leave-One-Out Cross Validation framework, incorporating variance inflation factor (VIF)-based multicollinearity filtering, minimum redundancy–maximum relevance (mRMR) feature selection, and SMOTE-based class balancing. The Exercise + Post-exercise configuration achieved superior performance: balanced accuracy (BA) = 76.9%, F1-score = 0.71, Area Under Receiver Operating Characteristic Curve (ROC-AUC) = 0.75, outperforming the Exercise + Cumulative configuration; this result was achieved using only five features: CONGA-15_EX (short-term glucose variability during exercise) emerged as the most robust predictor, alongside below_54 and above_250 (time spent in hypoglycemic and hyperglycemic ranges), MAG (mean absolute glucose change), and GRADE_hypo (hypoglycemia risk score). The generalizability of the temporal framework was further supported by independent validation on the OhioT1DM free-living cohort, where the Exercise + Post-exercise configuration (BA = 76.3%, ROC-AUC = 0.804) again outperformed the cumulative approach. These results suggest that a small set of interpretable CGM-derived features, extracted from the exercise and recovery windows, can effectively discriminate pediatric T1D subjects at risk of NH, supporting the development of lightweight CGM-only decision support tools for safer exercise management.