Assessment of the feasibility of determination of cholesterol and other blood constituents by near-infrared reflectance analysis.

Near-infrared reflectance spectrometry of blood serum can yield values for serum cholesterol that correlate reasonably well (r = 0.96) with those from common reference analytical methods. However, the variability of serum can cause ostensibly validated calibrations to fail on new samples. The determination of blood components such as cholesterol and triglycerides by near-infrared reflectance is complicated by their low concentrations, the variety of forms in which they appear, and by the natural variability of the blood matrix. These difficulties, when combined with the problems encountered in obtaining a representative sample from a given individual, can make it almost impossible to select, by a regression procedure, a wavelength combination that is characteristic of the complete blood matrix. The failure of the regression process to find characteristic wavelengths generates a false-sample problem in which even small changes at the analytical wavelengths produce a grossly unreliable cholesterol or triglyceride determination.