Multivariate calibration techniques applied to derivative spectroscopy data for the analysis of pharmaceutical mixtures

The prediction power of different chemometric methods was compared when applied on ordinary UV spectra and first-to-fourth order derivative spectra. Principal component regression (PCR) and partial least squares with one dependent variable (PLS-1) and three dependent variables (PLS2) were applied on spectral data of pharmaceutical formulations containing paracetamol, propiphenazone and caffeine. Derivatization ability in resolving spectral overlapping was evaluated when the multivariate methods are adopted for analysis of multicomponent mixtures. The chemometric models were tested on an external validation dataset and finally applied to the analysis of commercial formulations containing two or three drugs. The models were optimized by selecting the wavelength regions to be used in calibration through a new method which ensured either an acquisition of useful information or a removal of redundant or noisy data. This procedure provided to evaluate the analytical information of the wavelengths by using the component regression coefficients calculated in multivariate regressions. Significant advantages were found in analysis of all the analytes when the calibration models from third-order derivative spectra were used, showing relative standard errors less than 1.4%. In contrast, the other derivative orders displayed higher variance and their use gave inaccurate results.