Abstract
The purpose of this study was to perform a comparative biochemical analysis between conventional spectrophotometry and Raman spectroscopy, techniques used for diagnoses, on the urine of healthy (CT) and diabetic and hypertensive patients (DM&HBP). Urine from 40 subjects (20 in the CT group and 20 in the DM&HBP group) was examined in a dispersive Raman spectrometer (an 830 nm excitation and a 350 mW power). The mean Raman spectra between both groups showed a significant difference in peaks of glucose; exploratory analysis by principal component analysis (PCA) identified spectral differences between the groups, with higher peaks of glucose and proteins in the DM&HBP group. A partial least squares (PLS) regression model estimated by the Raman data indicated the concentrations of urea, creatinine, glucose, phosphate, and total protein; creatinine and glucose were the biomarkers that presented the best correlation coefficient (r) between the two techniques analyzed (r = 0.68 and r = 0.98, respectively), both with eight latent variables (LVs) and a root mean square error of cross-validation (RMSecv) of 3.6 and 5.1 mmol/L (41 and 92 mg/dL), respectively. Discriminant analysis (PLS-DA) using the entire Raman spectra was able to differentiate the samples of the groups in the study, with a higher accuracy (81.5%) compared to the linear discriminant analysis (LDA) models using the concentration values of the spectrometric analysis (60.0%) and the concentrations predicted by the PLS regression (69.8%). Results indicated that spectral models based on PLS applied to Raman spectra may be used to distinguish subjects with diabetes and blood hypertension from healthy ones in urinalysis aimed at population screening.