Analysis of Fasting Skin Spectra to Characterize Background Spectral Variance for Non-Invasive Glucose Sensing in...

Background spectral variance (BSV) is a critical parameter that strongly impacts calibration models for noninvasive glucose measurements. To date, the ability to characterize such variances is limited. A protocol is described for collecting skin spectra under fasting conditions and using these spectra to characterize how the BSV impacts measurement accuracy.

Transmission noninvasive skin spectra were collected over the combination region of the near infrared spectrum. These spectra were collected continuously over a period of 270 minutes while maintaining fasting conditions. A glucose transient was created by linearly adding a set of concentration-scaled pure component glucose spectra to each fasting spectrum.

Quality of the noninvasive fasting spectra was evaluated by root mean square noise analysis of 100% lines, resulting in a noise level of 26 micro absorbance units (µAU). Evaluation of the BSV was done by (1) creating a set of net analyte signal calibration models for glucose with different subsets of the fasting spectra and (2) using each calibration model to predict the concentration of glucose from a set of synthesized skin spectra that represent the transient glucose concentration profile. Results reveal that the glucose transient can be predicted when the BSV is captured within the calibration dataset. Specific time points were identifiable, however, when the calibration dataset no longer fully represented the BSV in the prediction dataset.

This preliminary report proposes a fasting protocol suitable for characterizing the BSV for noninvasive near infrared spectra. Net analyte signal calibration models are used to illustrate how unaccounted for BSV negatively impacts analytical accuracy. This protocol is general and can be applied to other approaches for noninvasive glucose sensing.