990-P: Developing a Breath-Based Glucose Estimation Device for People with Type 2 Diabetes

• Published in: Diabetes (New York, N.Y.) Vol. 73; p. 1
• Main Authors: Gourdy, Pierre, Cazals, Laurent, François-Marsal, Clarisse, Calvo, Sophie Blanc, Isz, Sandrine, Lepage, BenoÎt, Flacke, Frank
• Format: Journal Article
• Language: English
• Published: New York American Diabetes Association 01.06.2024
• Subjects: Algorithms; Diabetes; Diabetes mellitus (non-insulin dependent); Glucose; Glucose monitoring; Sampling; VOCs; Volatile organic compounds
• ISSN: 0012-1797; 1939-327X
• DOI: 10.2337/db24-990-P

Introduction and Objectives: BOYDSense is developing a breath-based monitoring system for chronic conditions starting with glucose monitoring for people with type 2 diabetes (T2D). These are results of a proof-of-concept study to evaluate the correlation of Volatile Organic Compounds (VOCs) in exhaled human breath with SMBG glucose values. The aim of this study, performed at CHU-Toulouse, was to develop a first glucose estimation algorithm for the BOYDSense breath-based technology and to assess its performance. Methods: The study combined 2 successive phases and enrolled in total 130 subjects with T2D on OAD and/or GLP-1 and/or basal insulin. In phase 1 100 subjects underwent a standardized meal challenge and, at each sampling point, a reference glucose value (venous blood) and 3 breath data-sets were collected to train the algorithm supported by AI. In phase 2 the developed algorithm was used to display glucose estimates. 30 subjects underwent the same procedure as described above and also SMBG readings were taken at the sampling points. The study center collected the data and conducted the data analysis. Results: In phase 2, 630 data-pairs (breath/SMBG) were collected. For the complete dataset, a Consensus Error-Grid-Analysis (C-EGA) for T2D was calculated. zone A contained 79.7% and zone A+B 99.7% of all readings. Only 2 readings were in zone C. A comparison of the C-EGA of the 3 breath strokes at each sampling point showed no obvious differences between the strokes (zone-A/zone-A+B): breath-1 80.0%/99.5%, breath-2 78.1%/100%, breath-3 81.0%/ 99.5% The overall MARD was estimated to 19.6%. The auto-sampling prototypes demonstrated a high reliability (99.9%). Conclusion: The prospective performance of the first algorithm iteration of the BOYDSense breath analyzer showed in comparison to real-life SMBG readings very promising results towards a new glucose estimation system supporting people with T2D. The high reliability speaks to an already advanced stage of the hardware development.