FT-NIR combined with chemometrics versus classic chemical methods as accredited analytical support for decision-making: Application to chemical compositional compliance of feedingstuffs

This research was aimed at assessing whether analytical results achieved by Fourier-transform near-infrared spectroscopy (FT-NIR) can be used for decisions on checking the chemical compositional compliance of animal feed samples, instead of decisions based on univariate data from classic reference methods. Results yielded by both analytical approaches (FT-NIR and chemical methods) on several measurands and matrices were compared. Samples (n = 330) were collected to inspect the guarantee levels declared on labels, yielding 3050 assessments in total. Each approach was also assessed in terms of the estimate of the acceptable risk of making a wrong decision. Parametric, non-parametric, and robust statistics were used for data pre-treatment from independent or joint measures. The distributions nature was assessed in order to choose a statistical test to compare them. The correlation and linearity of the model yielded R-2 values that ranged among 0.73 and 0.99. The worst value of R-2 (0.73) was related to the ash measurement model in ruminant feed. The best values of R-2 (0.99) refer to the measurement of protein in wheat bran and the measurement of ash in soybean meal. The values estimated by the multivariate models had both internal (using reference materials) and external (using proficiency tests) accuracy. The FT-NIR measurement uncertainty was lower compared to the reference methods. Both FT-NIR and reference methods yielded satisfactory performances. FT-NIR can be used in decision making regarding sample compliance, replacing several univariate data-based analytical approaches.