Training and Evaluation of a Breath-Based Glucose Estimation Algorithm in People With Type 2 Diabetes: The BOYDSense-GM Study

Abstract

Background:

Type 2 diabetes (T2D) is a growing global health concern, with self-monitoring of blood glucose (SMBG) being essential for optimal management. Current SMBG methods are invasive and often underutilized due to discomfort and inconvenience. Non-invasive alternatives, such as breath-based glucose estimation using volatile organic compounds (VOC), could improve compliance and quality of life.

Methods:

A prospective, cross-sectional, single-center clinical study enrolled 130 adults with T2D not requiring prandial insulin. The study was conducted in two phases: algorithm training (n = 100) and validation (n = 30). Participants underwent a standardized meal test with venous, capillary, and breath VOC measurements at seven time points. A portable VOC analyzer prototype (MIB) was used, and algorithm performance was assessed using Parkes error grid analysis, calibration indicators, and reproducibility measures. User feedback was collected via a questionnaire.

Results:

The VOC-based device demonstrated high reliability and safety, with 99% of measurements valid on the first breath. In the validation phase, 98.7% of blood glucose estimates fell within clinically acceptable Parkes grid zones A or B (T2D grid: 99.7%). Best accuracy was observed in fasting and early post-meal states, with lower performance at higher glucose levels. User feedback indicated strong enthusiasm for a non-invasive device, with most participants expressing willingness to use it regularly.

Conclusion:

This study validates the potential of breath-based VOC analysis for non-invasive SMBG in T2D, demonstrating reliability, usability, and patient acceptance. Further optimization is needed to improve accuracy, particularly at high glucose levels, but the technology shows promise for enhancing diabetes management and screening.