Modeling approaches to the estimate of plasma variables from exhaled volatile organic compounds

Exhaled volatile organic compounds (VOC) are potential biomarkers of biological processes, such as glucose (GL), insulin (INS), and lipid (FFA) metabolism. The accuracy of VOC‐derived estimates of plasma variables, though, depends on analytical methods and mathematical modeling. Here we report evolving modeling approaches in consecutive studies with simultaneous collection of plasma and exhaled gases at multiple time points, during 2‐4 h fluctuations of plasma GL, INS and FFA. a) Plasma GL estimate from multi‐linear regression (MLR) of exhaled ethanol + acetone, after GL ingestion; avg R of measured vs estimated GL = .70 b) Plasma GL estimate from MLR of ethanol, acetone, methyl nitrate + 3 aromatic gases after i.v. GL bolus (scaling back models from 6 to 4 gases, see figure); avg R = .91 c) Plasma GL, INS, FFA estimates from MLR of 4‐gas clusters during continuous i.v. GL infusion; avg Rs = .86 (GL), .92 (INS), .89 (FFA) d) Plasma INS estimate from MLR of a 5‐gas cluster + time, using residual values for gases (instead of breath‐room air differences used in a, b, c); avg R = .91 (same regression equation for all subjects, unlike a, b, c, where individual equations were used). MLR models of 4‐6 residual gas values and time can therefore accurately estimate blood variables related to glucose metabolism, providing the developmental basis of a non‐invasive methodology for diabetes screening, diagnosing and monitoring.