Analysis of the Relationship Between Electrodermal Activity and Blood Glucose Level in Diabetics

Diabetes is a chronic disease associated with impaired regulation of blood glucose level (BGL). A drop in BGL below the critical level of 70 mg/dl indicates hypoglycemia, and above 200 mg/dl is a sign of hyperglycemia. Many attempts have been made to noninvasively determine BGL based on signals measured by wearable devices. One of them is electrodermal activity (EDA). The aim of this study was to investigate which of EDA features recently used in the context of BGL fluctuations, change statistically significantly in the states of hypo- (H) and hyperglycemia (P) compared to normal BGL (N). To this end, data including BGL and EDA from 12 diabetics were collected for a few weeks and analyzed. These signals were synchronized and divided into episodes representing the N, H and P states. All episodes were separated into tonic (SCL) and phasic (SCR) components using discrete wavelet transform, and then their power and statistical features were determined. Finally, the Mann-Whitney U-tests were performed to check whether these features change significantly in N vs. H and N vs. P states. Generally, the connections between EDA and BGL in the examined subjects were not entirely consistent. Only some of the features proved to be sensitive to hyperglycemia. These are: power of the mean-free SCL component normalized by the power of EDA, average frequency of SCR arousals, and the standard deviation of the SCL. None of the analyzed features turned out to be significant enough in distinguishing hypoglycemia.