Quantitative assessment of the effect of cholesterol on blood glucose measurement using near infrared spectroscopy and a method for error reduction.

Abstract

There is a growing body of evidence suggesting that the accurate measurement of blood glucose concentration can be perturbed by many factors. Current literature is limited in describing the influence of cholesterol on non-invasive blood glucose measurements by near-infrared spectroscopy (NIRS). This study aims to investigate the influence of cholesterol on blood glucose measurement through clinical oral glucose tolerance test (OGTT) and NIRS. Further, a method to reduce the prediction errors induced by cholesterol is proposed, facilitating the clinical application of non-invasive blood glucose sensing by NIRS. We obtained clinical data of glucose and cholesterol concentrations at specific time points (0, 0.5, 1, 2, and 3 h) during OGTTs from 115 subjects. The subjects were grouped into: Norm for normal control, IGT for Impaired Glucose Tolerance, and Diabetes. In addition, spectral data between 1200 and 1800 nm were collected from 130 phantom samples, which are separated into seven groups depending on glucose and cholesterol levels. Statistical methods including One Sample T-test (OSTT), Pearson Correlation Analysis(PCA), and Unary Linear Regression (ULR) were used to analyze clinical data and spectral data to determine the relationship between glucose and cholesterol concentrations with the time course of OGTT. Reference wavelength-based method (RWM) was introduced to diminish the influence of cholesterol on glucose measurement and further the prediction error induced by cholesterol was reduced when using partial least square (PLS) model. Clinical results statistically show that there is a strong negative correlation between the changes of glucose and cholesterol concentrations in the diabetes group. The spectra of cholesterol exhibit similar absorbance peaks to those of glucose within NIR range. PLS modelling results demonstrate that glucose prediction is influenced by cholesterol concentrations in a calibration model. Furthermore, a model expression (ΔCg=0.0356Cc+1.0129 R(2) = 0.993) is fitted to quantitatively describe the glucose prediction increment (ΔCg) due to cholesterol concentration (Cc). The results show that glucose prediction accuracy can be improved up to 38.36% by using RWM when using NIRS. The cholesterol has an effect on blood glucose sensing. RWM is useful to help realize non-invasive blood glucose sensing by NIRS.