Selectivity and Sensitivity of Near-Infrared Spectroscopic Sensing of β-Hydroxybutyrate, Glucose, and Urea in Ternary Aqueous Solutions

The next-generation artificial pancreas is under development with the goal to enhance tight glycemic control for people with type 1 diabetes. Such technology requires the integration of a chemical sensing unit combined with an insulin infusion device controlled by an algorithm capable of autonomous operation. The potential of near-infrared spectroscopic sensing to serve as the chemical sensing unit is explored by demonstrating the ability to quantify multiple metabolic biomarkers from a single near-infrared spectrum. Independent measurements of beta-hydroxybutyrate, glucose, and urea are presented based on analysis of nearinfrared spectra collected over the combination spectral range of 5000-4000 cm(-1) for a set of 50 ternary aqueous standard solutions. Spectra are characterized by a 1 mu AU root-mean-square (RMS) noise for 100% lines with a resolution of 4 cm(-1) and an optical path length of 1 mm. Calibration models created by the net analyte signal (NAS) and the partial least squares (PLS) methods provide selective measurements for each analyte with standard errors of prediction in the upper micromolar concentration range. The NAS method is used to determine both the selectivity and sensitivity for each analyte and their values are consistent with these standard errors of prediction. The NAS method is also used to characterize the background spectral variance associated with instrumental and environmental variations associated with buffer spectra collected over a multiday period.