Characterization of the skin temperature signal for the definition of novel digital biomarkers of glycemia

Diabetes mellitus is becoming increasingly prevalent worldwide, making blood glucose monitoring essential both diabetes management and prevention. Currently, invasive and minimally invasive methods are the most widely used and reliable approaches for glycemic control. However, there is growing interest in non-invasive monitoring techniques due to their potential to enhance patient comfort. Wearable non-invasive devices, such as smartwatches, and smart bands, once able to provide a non- invasive assessment of glycemia, could encourage more individuals to monitor their glucose levels regularly. In this context, digital biomarkers for glycemic control are emerging as a promising research focus to contribute to the deployment of such non-invasive solutions. This study investigated skin temperature signal measured by a wearable device to identify novel digital biomarkers of glycemia. The analysis was conducted on 16 subjects spanning different glucose tolerance status (healthy and with prediabetes) wearing the Empatica E4 device, from which skin temperature signals were recorded. Standard metrics and a set of novel metrics describing both current and retrospective signal behavior were defined and extracted. For each subject group (All, Prediabetes, Healthy) and for each time interval (daytime, nighttime and whole day), a statistical comparison was performed using the Wilcoxon rank sum test between metric values within the tight glycemic range (70-140 mg/dL) and those out, above or below this range. During the day, the metric SDT hhmm reflected glycemic fluctuations across all groups. During nighttime, SDT hhmm remained a robust indicator, while additional metrics, including meanT, medianT, CVT SD, JT, M1T, and M2T demonstrated sensitivity to glycemic variations in all groups. In 24-hour period, SDT hhmm, M1T, and M2T consistently reflected glucose fluctuations in all subjects. The proposed metrics demonstrated their potential as digital biomarkers for glycemia and may represent a promising step ahead for future development of non-invasve glucose monitoring solutions.