Determination of biochemical parameters in human serum by near-infrared spectroscopy

Abstract

NIR offers multiple advantages for serum analysis, permitting a fast and direct determination of several parameters simultaneously, with low sample handling and without the need for reagents during the measurement step. The aim of this paper was to provide an evaluation of this technique in a real world scale, for the simultaneous determination of several parameters and based on a considerable number of samples. Direct near infrared (NIR) absorbance measurements were used to determine the concentration of clinical parameters in human serum that are required in routine biochemical tests. Total protein, albumin, total cholesterol, high-density lipoprotein (HDL cholesterol), low-density lipoprotein (LDL cholesterol), and very low-density lipoprotein (VLDL cholesterol), triglycerides, urea and glucose were determined in 447 serum samples obtained randomly from the clinical laboratory of the University Hospital Doctor Peset in Valencia (Spain). NIR spectra from 12 500 to 4000 cm⁻¹ obtained with a 1 mm optical path length were evaluated by using partial least squares regression models (PLS) built from the spectra of samples with known concentrations provided by the hospital. Root mean square error cross-validation (RMSECV) was used for selecting a number of factors, spectral regions and spectral preprocessing considered to build the models, that were evaluated from their prediction capability using the relative root mean square error of prediction (RRMSEP) of a series of around 30 independent samples, not used for calibration. For some analytes such as total protein, albumin, total cholesterol and triglycerides, errors obtained were 2.3, 4.4, 5.1, and 6.2% respectively, evidencing that the proposed methodology could compete with the enzymatic reference methodologies. However in the case of urea, glucose, HDL and LDL, average errors obtained were 16.0, 16.2, 18.0 and 11.0% respectively, and therefore the NIR methodology proposed is limited as a screening tool. With the use of a considerable number of samples for calibration, this study confirms that the proposed green and cost-effective methodology is ready for scaling up from the bench to the real world.