A multiparameter model for non-invasive detection of hypoglycemia

Abstract

Objective: Severe hypoglycemia is the most serious acute complication for people with type 1 diabetes (T1D). Approximately 25% of people with T1D have impaired ability to recognize impending hypoglycemia, and nocturnal episodes are feared. Approach: We have investigated the use of non-invasive sensors for detection of hypoglycemia based on a mathematical model which combines several sensor measurements to identify physiological responses to hypoglycemia. Data from randomized single-blinded euglycemic and hypoglycemic glucose clamps in 20 participants with T1D and impaired awareness of hypoglycemia was used in the analyses. Main results: Using a sensor combination of sudomotor activity at three skin sites, ECG-derived heart rate and heart rate corrected QT interval, near-infrared and bioimpedance spectroscopy; physiological responses associated with hypoglycemia could be identified with an F1 score accuracy up to 88%. Significance: We present a novel model for identification of non-invasively measurable physiological responses related to hypoglycemia, showing potential for detection of moderate hypoglycemia using a wearable sensor system.