Improved predictive models for plasma glucose estimation from multi-linear regression analysis of exhaled volatile organic compounds

Abstract

Exhaled volatile organic compounds (VOCs) represent ideal biomarkers of endogenous metabolism and could be used to noninvasively measure circulating variables, including plasma glucose. We previously demonstrated that hyperglycemia in different metabolic settings (glucose ingestion in pediatric Type 1 diabetes) is paralleled by changes in exhaled ethanol, acetone, and methyl nitrate. In this study we integrated these gas changes along with three additional VOCs (2 forms of xylene and ethylbenzene) into multi-linear regression models to predict plasma glucose profiles in 10 healthy young adults, during the 2 h following an intravenous glucose bolus (matched samples of blood, exhaled and room air were collected at 12 separate time points). The four-gas model with highest predictive accuracy estimated plasma glucose in each subject with a mean R value of 0.91 (range 0.70–0.98); increasing the number of VOCs in the model only marginally improved predictions (average R with best 5-gas model = 0.93; with 6-gas model = 0.95). While practical development of this methodology into clinically usable devices will require optimization of predictive algorithms on large-scale populations, our data prove the feasibility and potential accuracy of breath-based glucose testing.